• •

• •

t •








2.1. HUMAN VISION ··· 7



2.4. COLOR RENDERING ··· .. · · · .. · 15







3.5.1. Mercury Lamps · .. · 30

3.5.2. High-Pressure Sodium Discharge Lamps 32




3.5.3. Metal Halide Lamps 32





4.1.1. L ED Performance 38

4.1.2. Recombination of Electrons and Holes 39

4. 1.3. Injection in a p-n Junction LED .46

4.1.4. Heterostructures and Quantum Wells .48



4.2.1. Requirements 55

4.2.2. A1GaAs Materials System <o •••••••••••••••••• 57

4.2.3. AIGalnP Materials System 59

4.2.4. AllnGaN Materials System 61

4.2.5. Heterostructure Growth Techniques 66


4.3.1. Electroluminescent Structures 70

4.3.2. Contacts and Current Spreading 75

4.3.3. Emissive and Electrical Characteristics 79



5.1 .. 1. Escape Cones 84

5.1.2. Distributed Bragg Reflectors 87

5.1.3. Absorption Losses and Photon Recycling 89



5.2.1. AIGaAs Red LEDs 90

5.2.2. AIGalnP LEDs 92

5.2.3. AllnGaN LEOs 95




NONPLANAR GEOMETRIES ···· ······· .. ············· .. ··· 99

5.3.1. Shaped Chips ·· ·· ··· ···· .. ··· .. · 100

5.3.2. External Outcouplers ·· .. ······ .. · .. ···· .. ·· .. · 102

5.3.3. Nonresonant Cavity LEOs 104


5.4.1. Resonant Cavity LEOs..... . 106

5 A.2. Surface Plasmon-Enhanced LEOs 110

5 A.3. Photonic Crystals 112



6.1.1. Trade-offbetween Luminous Efficacy and Color Rendering 118

6.1.2. Dichromatic Systems ·· ·· · .. ·· 119

6.1.3. Polychromatic Systems 121

6.2. PHOSPHOR CONVERSION LEOs ······· ·····I22

6.2. I. Dichromatic Phosphor Conversion LEOs 122

6.2.2. Polychromatic Phosphor Conversion LEDs 126

6.3. MULTlCHIP LEOs 128

6.3.1. Dichromatic Multichip LEDs · · .. ·· ···129

6.3.2. Polychromatic Multichip LEDs · · .. ·· ···· 130



7.1.1. LED Strings · ··· .. · · .. · · ·· 134

7.1.2. Battery-Powered LEOs 137

7.1.3. High-Power Pulsed Drive ·· ···· .. · · .. · .. 137


7.2.1. Traffic Lights 139

7.2.2. Automoti ve Signage ·· · ······· .. ··· 141

7.2.3. Miscellaneous Signage · .. ······· .. ·· ···143

7.3. DISPLAYS 144



7.3.1. Alphanumeric Displays 144

7.3.2. Full-Color Video Oisplays................... 145


. -- 148

7.4. I. Phototherapy of Neonatal Jaundice 8

.. 14

7.4.2. Photodynamic Therapy.... .. ..

__ ... 149

7.4.3. Photopolymerization of Dental Composites 150

7.4.4. Phototherapy of Seasonal Affective Disorder.. lSI


7.5.1. Plan! Growing -- 153

7.5.2. Photobioreactors .... __ ..

............................................... .. -., 155


", ••••••••• " " ••••••••• " ••• - •• "'0<, 156

7.6.1. Fluorescence Sensors ...

---.- ", -- " , 156

7.6.2. Time- and Frequency-Domain Speclroscopy

7.6.3. Other Optical Applications __ 161

7.7. ILLUMfNATION -- 162

7.7.1. Local Illumination __ .

............ , ,', -.-._ , 163

............. 159

7.7.2. General Lighting......... .. 165

7.7.3. The Future of Solid-State Lighting 167

REFERENCES ... _............................................... _ 169



Supplem~ntary files ae~ompanyjng this book can be accessed at: ftp://ftp,wlley.com/publlc/sci_teeh_med/lighting

.. when touched, the light will split into an infinite amount of small lights that will COme down the hill and spread over your home,

Jonas Biliunas, Lithuan ian writer "Light of Happiness" (1905)


Breakthroughs in artificial light sources-a piece of burning wood "invented" more than 500,000 years ago, gas lighting (1772), electric lighting (1876), and fluorescent lamps (I938)-have led to the development of modern lighting sources. These sources are tungsten incandescent and compact fluorescence lamps for residential use, fluorescence lamps for work environments, and sodium lamps for street lighting. Today, 21 % of electric energy use is in lighting, and perhaps half of this energy could be saved by switching to efficient and cold solid-state lighting sources. Projected cumulative financial savings from solid-state lighting might reach $115 billion by year 2020. Solid-state lighting will use visible and ultraviolet LEDs that are expected to reach lifetimes exceeding 100,000 hours. At present, LEOs are the most efficient sources of colored light in almost the entire visible spectral range. White phosphor conversion LEDs already surpassed incandescent lamps in performance, and their efficiency is expected to triple by the year 2010. From traffic lights to road signs, from automobile taillights to outdoor displays, from landscape to accent lights, solid-state light sources that are harbingers of the next lighting revolution have already arrived.

Semiconductor physicists, chemists, materials scientists, opticians, lighting engineers and business administrators work jointly on solid-state lighting so that humankind can reap the benefits of this exciting technology. Various issues of bright semiconductor lamps have been described in the scientific literature. Modern LEDs based on AIGaAs, AIGalnP, and AllnGaN semiconductors are described in the collections of articles edited by Stringfellow and Craford (1997) and by Mueller (1999a). Nakamura and Fasol (1997) and Nakamura and Chichibu (2000) reviewed blue AlInGaN-based light emitters that are the most recent achievement




of LED technology. These books focus on semiconductor physics. technology. device fabrication, and first mass applications related to a new generation of lightemitting diodes: high-brightness LEOs.

Our book treats solid-state lighting as an inherent part of lighting technology.

Here, high-brightness LEOs are considered within a wi der context. We present the high-brightness LED topics in the contents of I ight sources and solid-state lighting technology. We also introduce historical aspects of lighting. describe the characterization of visible light, discuss conventional lighting devices, and consider the problems of generation of white light by LEOs and of light extraction from a solid state. Another unique feature of our book is a comprehensive review of existing and emerging applications of sol id-state I ighti ng.

We hope that the book wi II be useful for technologists. scientists, and engineers. as well as for businesspeople and students who are interested in the development and applications of sol id-state I ighting. The book can also be used as a textbook for graduate and senior undergraduate courses on solid-state I ighting and as an additional reference text for courses in semiconductor physics. materials science, electronic device design. lighting engineering, and optics.

Supplementary files accompanying this book can be accessed at: ftp:!/fip,wiley.com/public/sci_tech_med/lighting.

We are grateful to our wives for their infinite patience and understanding and for encouragement and support.

ArtCiras Zukauskas. Michael S. Shur, and Remis Gaska

Troy. NY, 2001




To introduce the subject, a brief historical survey on the technology of lighting devices is presented in this chapter. The survey is based on some encyclopedic data, a recent historical book of Bowers (1998), and the references therein. Some details come from the Encyclopedia Britannica (1911), which offers an authentic review of lighting technology achievements as they were in the first decade of the twentieth century. More details on present-day lighting devices, most of which were engineered in the later decades of the twentieth century, are given in Chapter J.

All light is produced by electronic transitions from higher to lower energy states. Excitation to higher states may be achieved by a variety of techniques. Generally, the history of lighting technology describes discoveries of more and more efficient and convenient methods of electronic excitation and radiatiue recombination. However, with the exception of the last two hundred years, lack of basic knowledge resulted in extremely slow progress, and many generations of people saw no noticeable improvement in lighting in their lifetimes.

Up to the beginning of the nineteenth century, flame produced by combustion Was the only source of artificial light Combustion results in high levels of excitation of atoms and molecules. The emission is caused by pyroluminescence, which is due to radiative transitions in excited atoms and ions, recombination of ions to form molecules, and incandescence of solid particles in the flame.

The first artificial lighting source was fire, which was widely used as long as 500,000 years ago. Probably at the same time, humans invented portable light by picking up a piece of burning wood, which became the first torch. For thousands of years, the most primitive utilization of combustion was employed in braziers




(dishes contai nil:g fire), cressers (fire-baskets on poles) and torches (made of vegetables l.reated with flammable substances-pitch, wax, resin, tallow, oil), which now exist mostly as antiques and ceremonial accessories,

The next important prehistoric discovery was a fiber burning in a pool of molten fat. That discovery gave .bInh to the Wick, a capillary cord that draws fuel up to a flame .. Probably the first lighting device, this greatly improved the efficiency of pyrolurninescencc-based lighting. The wick became a key component of oil lamps and .candles. The lamps were shallow vessels with a dipped wick. Archeoloaical findings and cave paintings prove that stone lamps began to be used JO.o()O to 7Q,OOO y~ars ag~. ~,ater, lamps n~ade of shell, pottery, and metal appeared, but WIthout substantial Improvement III the emission process, Candles, which are beIl~ved LO have arriv:d in the days of Roman times, employ the same principle of a Wick but with the fuel (beeswax, tallow, and later, paraffin wax) melting in the heat of the flame,

An the end of the eighteenth century, the first notable improvement in the oil lamp marked the beginning of modem lighting-device engineering. That milestone was due t? ~ml Agrand of Geneva. who devised an oil lamp with a tubular wick placed within two concentric rubes and a glass ch i mncy around the burner (Fig, 1. I). A tenfold gain in ! ight was achieved, OWl ng to a better supply of air to the name and the resu lting higher burn i. ng efficiency. Bes i des, the devi ce inc I uded



FIG. 1.1. Agrand lamp, the first lamp designed based 011 research, The glass chirnnev has b~ell removed. (Courtesy of- tile Science Museum! Science and Societ Picture

LIbrary, London.) Y



a control lor winding the wick up and down, The design was ba~ed on the rescar~h conducted by A. L. Lavoisier, who discovered that combustion IS due to oxygen lTI the air. The lamp was demonstrated to King George Ill, and Agrand was grantc~ an English patent (No. 1425 of 178~) '. In the nineteenth eent~ry, numerous improvements in the oil lamp (means oj oil supply, design of the wick an? bu_r~er, - t duction of mineral oils) were proposed. Kerosene lamps, introduced m 1 ~)OS,

In.fO. . hanai th

L. arne widelv accessible lighting devices and had a huge Impact on c angmg e

oec J f lizhti .

after-dark activities of civilized man. They are sti II being produced or rg ung 111

areas when: electricity is unavailable. . , .

Gas lighting was introduced by Scottish inventor Wilham Mu~doch ." 1772.

Soon, at the beginning of the nineteenth century. luminous combustion ot gaseous by-prod ucts of coke production was rap idl y adopted. in t?wns _ [or Indoor domestic, i~dustrial. and public lighting as well as for street lLghlrng., For more than a .hundred years, gas lighting was advanced successfully and mtrast.ructure compnsmg

distribution mains. remote sources. and maintenance was established. .

The first solid-stale lighting device was the limelight (Fig. 1.2). The device, introduced by Thomas Drummond in 1826, consisted of a cylinder of lime (calCium oxide), which was brought to a state of dazzling brilliancy ~y the flame of an oxyhydrogen blowpipe. Emission was due to a novel e[Je_ctCO/u{oiliminescence-discovcred by Goldsworthy Gurncy in 1820. Candolummescence is caused bv thermal excitation of ions, which em it in excess of blackbody incandescence. Limelight was used in theaters in the 18605 and I 810s until superseded by the electric arc. Nevertheless, in 1886, the candoluminesccnce-bascd

FIG. 1.2. Lime! ight, the first sot id-state I ighting device, (Courtesy 0 r the Science Museum / Science and Society Picture Library. London.)



light Source was revived b A

cotton fabric soaked in Yl' uer von Welsbach, who developed the gas mantle

a so utlOn of a m t II" I ( ,

using a mixture of ceri id e a IC sa t the best results were obtained

JUm OXI e and tho ' ide i ,

Was heated by h 'h . rrurn OXI e In the ratro I :99), The mantle

a Ig -temperature. I' fl

(named after R W B . non ummous arne from a Bunsen burner

. . unsen m 1855) Th B b

gas with a certain amount of air be' e unsen. urner operated by mixing coal

used widely in the first third fth fore combustIOn. The Welsbach mantle was

, ,0 e twentieth centur I d II "

to electric lighting. It can still b f d . y, on y gra ua y glvmg place

Th ' " e oun 111 kerosene and gas lamps

e marn prmcipleg of electric Ii htil1' '

of the nineteenth centu h ,g g were discovered at the very beginning

ry, w en SIr Humphry D d ' .

tween two rods of carb ( avy emonstrated discharge be-

on an arc) and the I ' f ,

electric current (incandes . ) H g owing 0 a prece of wire heated by

cence. e used a batte d f 20 '

and zinc elements invented b Al d ry rna e 0 00 pairs of copper

nous discharge of static electrYI'c't essan ro Volta In 1800. The effect of the lurni-

, I Y III mercury vapor wa di d '

III the seventeenth century Alth hI.. S iscovere even earlier,

undertaken since Davy's d~mons~u~, a ,ot of research and engineering has been nomically feasible The cha c: ra 10fln, 19h! dependent on batteries was not eco-

. nge rrorn arne to elect ' ",

ogy happened only in the 1870s, when Z. T Ora nco power m hghtlllg. techno]-

lmuous-current generator (dynamo). . mme Introduced an efficienj con-

. Paul Jablochkoff (Pavel Yablochkov) f bri t .

hghting device in 1876 (Fig I 3) It . a fica ed the first practical electric

rated by a thin layer of insul~ti~' conSlsjted of two parallel carbon rods sepag gypsum piaster, which crumbled as the carbons

FIG. 1.3. lablochkoff candle, the first electric Ii htin devi

Museum I Science and Society Pictu Lgb g Levlee, (Courtesy of the Science re I rary, ondon.)



bumed away (hence, no mechanism for adjusting the gap between the electrodes was required), Despite a small lifetime (just a few hours), Jablochkoff's candles were immediately adopted for street illumination and provided good publicity for electric lighting. In a few years, they were replaced by higher-performance, longer-lifetime (up to 1000 hours) carbon-arc devices with glass-enclosure, electromagnet-based self-regulation works and chemically treated electrodes. Carbonarc lamps were used widely for street I ighting up to the second decade of the twentieth century and for aircraft floodlighting during both world wars. They are the ancestors of modern high-intensity discharge lamps.

Thomas Alva Edison and Joseph Wilson Swan were the most successful and prominent inventors of the incandescent filament lamp. Edison demonstrated his device at the end of 1879 and obtained a patient for a filament lamp (U.S. patent No, 223,898 of 1879). Swan demonstrated an incandescent lamp early in 1879; however, the first topics of his patients were the methods of evacuation and prevention of fracture of the gl ass at seals (British patents No. 18 and No. 250 of 1880), Both started with carbonized-paper filament in an evacuated glass bulb. Later, Edison used a fiber of a particular type of Japanese bamboo. He also launched the first electrical distribution system. The invention of the incandescent filament lamp was accompanied by famous patent trials. Subsequent rapid commercialization of the carbon-filament lamp promoted companies that are presently giants of the electric and lighting industry: General Electric Company (OE, descendant of the Edison Electric Light Company), [British] General Electric Company (GEC), AEO, Siemens, and Philips, At the end of the nineteenth century, the carbon-filament lamp was the preferred lighting device for indoor use. Nevertheless, great efforts were made to replace carbon with a material that could operate at higher temperature and with reduced deposition on glass (carbon darkened the bulb). In 1897, Nernst developed a filament made of cerium oxide-based solid electrolyte. The origin of the emission was the same as in the limelight and gas mantle, but the thermal excitation was due to electric current. The Nernst lamp had the highest efficiency and lifetime. However, it was soon superseded by metal-filament lamps. Although metals with the highest melting points are brittle and difficult to draw into wires, this obstacle was overcome by gradual introduction of osmi um, tantalum, and final! y, tungsten (A. Just and Fe Hanaman, Germ an patent No, 154,262 of 1903). The present design of the incandescent tungsten lamp was basically completed in the first third of the twentieth century.

Discharge in low-pressure gases was investigated intensively and gas discharge lamps were experimented with throughout the nineteenth century, But not until 1900 did Peter Cooper Hewitt patent the mercury vapor lamp. The lamp was started by mechanical tipping and emitted bluish-green light that distorted colors. However, it exhibited efficiency much higher than that of the carbon-filament lamp, To improve color rendition in the red part of the spectrum, C. O. Bastian and A. E. Salisbury combined the lamp with a low-temperature incandescent lamp (1904). Discharge lamps using air were introduced commercially by D. M. Moor in 1904 (carbon dioxide and nitrogen were used somewhat later). Moor's tubes were long (-10m), contained gas at about 0.00 I atm pressure, and were fed by high voltage.



P. Claude developed technology for liquefying air and separating its constituents. He was the first to filJ discharge lamps with inert gases (1910). Further development of gas discharge lamps involved hot cathodes (similar to those developed for radio tubes) to reduce operation voltage, startup circuitry, introduction of sodium vapor, and increased pressure for widening the emission spectrum.

In 1938, GE and Westinghouse Electric Corporation put on the market new colored and white lamps that were nothing more than low-pressure mercury discharge lamps with the inside of the tube coated with a fluorescent powder. The fluorescent lamp employs photoluminescence excited by ultraviolet emission of mercury. The first phosphors were of mineral origin. They were superseded by synthetic inorganic materials, and starting in 1948, the most commonly used phos-

phor in white fluorescent lamps has been calcium halophosphate Cas(P04b(F,CI),

. db Sb3+ 2+.

activate y and Mn Ions.

By the end of the I 990s, at the dawn of semiconductor-device lighting, the major part of residential lighting is provided by tungsten incandescent lamps. A compact fluorescent lamp is being promoted actively because of its higher efficiency. Most work environments employ fluorescent lamps, and street lighting is dominated by sodium lamps. However, all this is about to change because of the explosive development of solid-state lighting based on high-brightness visible light emitting diodes (LEDs), which have already found numerous niche applications.



In this chapter we offer a brief survey of vision, ph~tometry,. and colorimetry in terms of the basic topics that are most relevant for solid-state lighting, More atte~tion is paid to color rendering, which is an important property of "cold" illuminants, More details can be found in specialized books (see, e.g., Coato~ and Marsden 1997, Wyszecki and Stiles 2000, Rea 2000, and references. therein) as well as in literature on light-emitting diodes (Bergh and Dean 1976, Gillessen and Schairer 1987, Stringfellow and Craford 1997).


Lighting technology relies on the properties of human visi~n. Vision is a com~lex multistage process which yields meaning as to the changm~ pattern of a.mblent luminance and chromaticity. The properties of vision determine the quantity and

quality requirements for lighting. . . .

The primary processes of vision occur in the eye, where the Image IS projected onto the retina. The retina comprises detector cells (receptors), which convert the energy of light into nerve impulses, The receptor.s are of two t~pe~, rod~ .and cones. Rods exhibit higher sensitivity and play an Important role In night vIsl~n, when the eye adapts to dark (scotopic vision). However, rods are no~ able to dl~tinguish between colors since they contain only one type of photopigrnent. lt IS worth noting that these receptors are concentrated outside the central part of the retina, thus being responsible for peripheral vision.




Under conditions of high luminance, the response of rods is saturated, and vision is mediated entirely by cone receptors (photopic vision). Cones may contain one of three pigments: erythrolabe (L-type or long-wavelength cones), chlorolabe (M-type, middle-wavelength cones), and cyanolabe (Svtype, short-wavelength cones). Owing to different spectral sensitivity, these photopigments allow us to distinguish colors. The density of cones is highest at the central part of the retina (fovea) and drops in the periphery. Photopic vision has lower sensitivity but higher spatial resolution.

Due to the different photo receptors involved, the spectral sensitivities of scotopic vision and overall photopic vision differ (Fig. 2.l.l). The spectrum of scotopic sensitivity, which is determined by the photoresponse of rods and the transmittance of pre-retinal media, peaks in the blue-green region at a wavelength of 507 nm in air. The photopic spectrum is red-shifted with respect to the scotopic spectrum. The peak of the photopic sensitivity resides in the yellow-green region at a wavelength of555 nm in air.

Most human activities involve photopic vision, which is the most important from the point of view of lighting technology. Therefore, much effort has been expended to calibrate and digitize the spectral response and color resolution of photopic vision. In 1924, the rnt~rnational Commission on Illumination (CIE; Commission Internationale de l'Eclairage) introduced the relative luminous efficiency function, V(J), for photopic vision. In 1951, a similar function, V'(J), was introduced for scotopic vision. Both functions are depicted in Fig. 2.1.1. The function V(J) is defined in the range 380 to 780 nm. This wavelength interval of electromagnetic radiation is ultimately defined as the visible spectrum (Rea 2000).


~ 10·'


Q) 1 D·' (J)



ia 10-3 (i)



V'("A.) , V ... , .... -, V(lc)
<, " I , ~
/( • \

~ .:/ , ,
\ \

• , \
I •

I \ '\
• 400





Wavelength (nm)

FJC.2.1.1. Normalized spectral sensitivity curves for photopic vision [solid line, V (-1.) , from 1924 CIE] and scotopic vision [dashed line, V'(,.t) , from 1951 Clli].




Light is electromagnetic radiation. Radiometry easures th~ quantities associat~d with radiant energy. These quantities are desi ated as radiant .and e~ploy umts that refer to energy (joules). For example, e radiant flux ¢ e IS the time rate of

flow of radiant energy measured in wa ; the radiant intensity Ie = d¢>e/dw (W/sr) is the radiant flux per u~it so ·0 ang.le i~ a given direction. Photometry deals with the visual sense of bright ess, which IS actuated by light, Hence, photometry differs from radiometry i hat it measures visual respon~e. The relevant quantities are designated as luminous. The luminous flux, ¢J- (.!, IS related to the

spectral density of the radiant flux, <D e,.l, ;;; d<D e / dit [also called the spect ral power distribution (SPD), S(A)] through the 1924 CIE luminous efficiency function v (it ) (solid curve in Fig. 2.1.1) and is measured in lumens (lm):


Here the integral extends over the entire visible spectrum. Consequently, the luminous intensity Iv is the luminous flux from a point source per unit solid angle:


where leX;;; dIe / d)' is the spectral density of the radiant intensity. The luminous intensity is measured in candelas (cd) or lrn/sr [the ca~dela i~ a ba.sic unit of the Systerne International (d'Unites)-SI, equal to the luml~~us intensity of a so~rce that emits monochromatic radiation of frequency 5.4 x 10Hz and that has radiant intensity of 1/683 WIST)' Usually, intensity values are measured with a small solid angle (~1 0-3 sr) at different angular positions. To convert from luminous intensity to luminous flux, the intensity is integrated numerically over the entire sphere using special procedures (Hodapp 1997).

The concept of luminous intensity is not directly applicable to an exte~ded

source of light that cannot be treated as a point source. Such sources a.re described by luminance, which is the quotient of the luminous flux propagatmg from an element of the surface dA and observed at an angle f} per unit solid angle:


where dA' is the area projected in the direction of the observation. The luminance is measured in candelas per square meter (cd/m"). Sources with a higher luminance appear brighter than those with a lower luminance. The visual sense of brightness of an object that does not emit but, rather, reflects light can be characterized by the reflected radiation that is emitted from a certain source. Scotopic



vision dominates at a luminance below 10-2 cdlm2, while above IO cdlm2 the vision is completely photopic. The sun viewed from sea level exhibits an average luminance of 1.6 x 109 cdlm2, and the luminance of the moon is approximately 2500 cdlm2

Following Eq. (2.2.1), the radiant flux of I W at a wavelength of 555 nm produces a luminous flux of 683 1m. A broader or shifted spectrum results in a lower luminous flux, since visual sensitivity drops upon moving away from the central 555-nm wavelength. A measure of the ability of the radiation to produce a visual sensation is luminous efficacy, which is measured in Im/W

{SO V(1)s(1)dA.

K '" <l>lJ = 683 hn/Wx_38_O _

¢e fa S(A.}d..l


The highest possible efficacy is 683 Im/W, as in the example described above. Note that luminous efficacy characterizes the radiation spectra rather than the source. Also, Eq. (2.2.4) suggests that the luminous efficacy can be calculated by using the spectral power distribution presented in any relative power units.

To describe how efficient the source is in converting the energy to light, radiant and luminous efficiencies are introduced. Dimensionless radiant efficiency designates the ability of the light source to convert the consumed power Pinto radiant flux:


The radiant efficiency may range from zero to unity. The luminous efficiency is the ability of the source to convert the consumed power into actuation of the vision:


Luminous efficiency is measured in Im/W and is not to be confused with efficacy, which is described by the same units.

Practical applications of lighting usually deal with illuminance, which is the density of the luminous flux incident on a surface:


Here dA is the element of the surface, .9 is the angle of incidence, and r is the distance from a point source to the illuminated plane. The measurement unit for illuminance is lumen per square meter, also called lux (Ix). Again, the sun generates the illuminance on the earth's surface from 104 to 105 lx, depending on cloudiness; illuminance by the moon does not exceed 0.1 Ix.



Table 2.2.1. Recommended illuminance ranges for different types of activity

Type of Activity Illuminance (lx)
Oricntation and simple visual tasks (public spaces) 30-100
Common visual tasks (commercial, industrial, and 300-1000
residential applications)
Special visual tasks, including those with very small or 3000-10,000
very low-contrast critical elements The higher the illuminance, the higher the ability of the ~y~ to disti~guis~ details, small contrasts, and color hues. Therefore, ~iffer~nt activities require different levels of illuminance. Table 2.2.1 presents t1IUlmnance. ranges for different types of activity recommended by the Illuminating Engineering Society of North America (Rea 2000).


Colorimetry deals with measurements of color. The seJ~sation of. c~lor is much more complex than sensation of the brightness. A numerical descnpt.lOn o~ colors relies on a very simplified model of human vision and therefore might disagree with certain subjective observations. Nevertheless, the basic concepts of colorimetry, such as tristimulus values, chromaticity coordinates, color temperatu~e, and color rendering, are well formulated at present. These concepts are of crucial importance in describing light sources for lighting applications.

Describing colors by certain numbers, tristimulus values emerged. fro~ the experimental fact that most colors can be accurately imitated by a combination of not more than three appropriate primary colors (stimuli), such as red [R1, green [G1, and blue [B]. This makes it possible to specify colors in amounts ~f three stimuli. However, some colors, which are close to monochromatic, fall to be matched by using only positive amounts of these stimuli (i.e., by [R], [G], and [6]) and require negative amounts (color subtraction). This inconvenience was removed by introducing imaginary stimuli [X], [V], and [Z]. The tristimulus valu~s X, Y, and Z [i.e., the amounts of each stimuli in a color represented by a certain SPD S(,1.)] are obtained by integrating the spectrum with the standard color-matching junctions x(,t) , Y(..l), and ':(1), which are characteristic of an ideal observer (introduced by Cl E in 1931 and shown in Fig. 2.3.1):

X= Jx(,l,)S(A)dA,


y = fY(..l)S(1)d..l,


Z;: f Z(,t)S{,t} dk


> 1.0


0.0 L_~~~':::::"""L..3II~~ ........ ......I. .................. .;;:iD:m=. _ _' 350









Wavelength (nm)

FIG.2.3.1. 1931 erE color-matching functions: purple 1(,1.); grecn y{l}; and blue


The 1931 CrE Standard Observer is the basis of the trichromatic system of modern colori metry (defined by erE in 193 I). 193 1 CIE color-rnatchi ng functions are tabulated at different wavelength intervals (Wyszecki and Stiles 2000). The 1931 CIE green matching function Y(A) was matched completely with the

1924 eIE luminous efficiency function V(A) for photopic vision (compare with Fig. 2.1. I). The functions X(A), Y(A), and Z(A) were defined for narrow matching fields (2 degrees) to avoid any participation of rod vision. An alternative set of color-matching functions xlO (it), )710 (1), and zlO (1) is recommended for angles above 4 degrees (see the 1964 erE Supplementary Standard Observer, in Wyszecki and Stiles 2000).

For convenience, the chromaticity coordinates (x, y) of a light source with a spectrum S(it) (measured in power units, watts) were introduced:







y= .



The third coordinate,


z= =I-x-y,




. dditional information Therefore an in-plane description of colors by

contams no a .,

eans of two chromaticity coordinates (x,y) may be made.

m F. 2 3 ? depicts the 1931 CIE chromaticilY diagram with the (x, y) coor-

igure .. - . I

dinates of imaginary tristimulus [XYZ] and a.n arbitrary set of real pnrnary co ors

[ROB]. CIE standards include monochromatIc-color coordmat~s (x, y) !ocate~ on a horseshoe-shaped curve closed by a straight bottom line (purple line). Table 2.3.1 presents wavelengths, frequencies, and photon energies for mono-

chromatic colors.

, [Xl
0.2 0.4 0.6 0.8 1.0 [Yl

1.0 ~",:"",_",----"-_,,,,-_"""-,.----r-,.----r---,






'6 0.6


o o U


0.0 ~~~:'L.:~_L_.---'--_L._.......L..-__...I-_'__----:'I 0.0

x Chromaticity Coordinate

nG.2.3.2. \931 CJE chromaticity diagram.



Table 2.3.1. Monochromatic colors

Color Wavelength, A. frequency, v Photon energy,
(run) (THz) hv(eV)
Violet 380 < A < 455 659 < v < 789 2.72 < h v < 3.26
Blue 455 < it < 492 609 < v < 659 2.52 < h v < 2.72
Green 492 < 1 < 577 520 < v < 609 2.15 < h v < 2,52
Yellow 577 < A < 597 502 < v < 520 2.08 < h v < 2.15
Orange 597 < it < 622 482 < II < 502 1.99 < h v < 2.08
Red 622 < A < 780 384 < v < 482 1.59 < h v < 1.99 The area embraced by the contour comprises the coordinates of all real colors.

Inside the contour, a locus of points for blackbody radiators of different temperatures tPianckian locuss is shown. The region in the vicinity of the blackbody radiator locus (starting at approximately 2500 K) defines the white color. Red, green, and blue hues reside within regions that span from the white region toward the corresponding corners of the diagram. Orange, yellow, blue-green, purple, and pink regions are situated between the corners.

Sources with chromaticity coordinates very close to the Planckian locus may be described by color temperature (CT). If the chromaticity of a source is not exactly equal to any of the chromaticities of a blackbody radiator, a correlated color temperature (CCT) may be assigned to the source using chromaticity match with isoternperature lines (Wyszecki and Stiles 2000),

CIE data also include four standard sources: A (tungsten at 2856 K), B (direct sunlight, approximately 4870 K), C (overcast sunlight 6770 K), and D65 (daylight, 6504 K). Point E marks equal energy (x == 0.3333 , Y = 0.3333 ).

The 193 J CIE chromaticity diagram provides a simple means of color mixing.

The principle of color mixing follows from the makeup of the diagram. A set of n primary sources with the chromaticity coordinates (xi' Yi ) and radiant fluxes cDei will produce a color with the chromaticity coordinates

n / n

X,,=LXicDei LcDei,

1=1 i=1



To apply Eqs: (2.3.3) properly, the radiant tluxes of the primary sources should be normalized so that Xi + Yj + Z, equals the same constant (unity) for all i.





G .

$ •...

. S 0.6

"0 '-

o o o >-

~ 0.4







>- 0.2

x Chromaticity Coordinate

FIG. 2.3.3. Mixing of two (blue and yellow) and four (orange, yellow-green, gre~n, a~d pink) colors for white (standard source C) using 1931 Cl E chromaticity dia-


For two primary sources, any color with the coordinate~ I~cated on ~ straight line that connects the coordinates of the sources can thus be imitated, For instance, white color (standard source C) may be composed of two colors (blue and yellow; see Fig. 2,3.3). For three and more sources, the resulting coo.rdinates can b~ produced within the top-area polygon with the apices at the coordinates of the primary Sources, Again, the chromaticity of the standard source C may be obtained from three colors (red, green, and blue; see Fig. 2.3.2), from four colors (orange, yellow-green, green, and pink; see Fig. 2.3.3), and so on.


Chromaticity coordinates describe quantitatively the color of the radiating light SOurce. Because of vision properties, the same coordinates may be obtained for a Planck radiator for a combination of a few monochromatic sources, or for a Source that irradiates a set of certain spectral lines. Sources of a different spectrum but the same chromaticity are called metameric. After the radiance is reflected from an illuminated object, the spectrum is altered in accordance with the reflectivity spectrum, resulting in a shift of the chromaticity coordinates (colorimetric



shift)· .How~ver, an object illuminated by metameric sources may appear visually as having different colors, since the reflectivity spectra will produce different colorimetric shifts for different spectral composition of the sources. Metamerism has presented a serious problem for lighting technology since "cold" sources such as discharge and fluorescent lamps were introduced. It should be noted that a quanti. tative description of lighting quality is even more complicated because of the chromatic adaptation of the human eye (i.e., the ability of the human eye to correct for colors). The mechanism of this phenomenon is not fully understood.

To estimate the quality of lighting, chromatic coordinates of lest samples are measured under illumination by the source to be tested and by a reference source. Then the colorimetric shifts are evaluated and graded with respect to chromatic adaptation. However, the 1931 CIE diagram has been shown to be unsuitable for color discrimination via colorimetric shifts, since the density of hues that can be resolved by human vision is highly nonuniform in this plot. For instance, when m.oving from the white center of the diagram toward the green angle (see Fig. 2.3.2), one resolves much fewer hues than when moving toward the red or blue. angle. ']_bis shortcoming of the 1931 CI E diagram was improved considerably b~ introducing a uniform chromaticity scale (UCS) diagram (1960 CIE, see Fig. 2.4.1). The new UCS coordinates can be obtained from the 1931 CIE chromaticity coordinates by using the transformations



2 .~ 0.3.


o o U

z·13 0.2

~ E s

..c U > 0.1




u Chromaticity Coordinate

F1G. 2.4.1. 1960 CIE uniform chromaticity scale diagram with Planckian locus.



4X 4x


X+15Y+3Z -2x+12y+3'

(2.4.1 a)

6Y 6y

(2.41 b)

v= ==

X+15Y+3Z -2x+12y+3

Using UCS chromaticity coordinates, the lighting quality can be rated in terms of color rendering indices (1974 C1E, updated in 1995; see CIE 1995), which grade colorimetric shifts obtained in test samples. This method of grading (ClE test-color method) accounts for the chromatic adaptation of the human eye. In most cases, a general color rendering index (CR I) Ra, which integrates the data for eight specified samples, is used. Additional information on lighting quality may be obtained from special color rendering indices, Ri, which refer to each test


Figure 2.4.2 depicts retlectivity spectra of the eight test-color samples

(1964 Cl'E), which are selected from the color palette introduced by Munsel (Wyszecki and Stiles 2000). In these samples, spectral reflectance has only moderate variation, which is applicable for a general-purpose light source.

Calculation of the general CRt consists of number of steps. First, a reference illuminanl with the spectrum S r (.-1.) and chromaticity coordinates (X,. , Yr) is selected (for color temperatures below 6000 K, a Planck blackbody radiator is recommended, unless specified otherwise). Then the reflected spectra S r (A.) Pi (J) (i == I, ... ,8 ) from each of the eight test samples are determined and the chromaticity coordinates (xri ,Yn) are calculated using Eqs (2.3.1) and (2.3.2).

Second, the spectrum of the test source S k (,.l) is measured, the chromaticity coordinates (xk, Yk) are calculated, the reflected spectra S k (,.l) Pi (,.l) for each test sample are determined, and the chromaticity coordinates (xki ,Y ki ) are calculated. Third, all the 1931 ClE coordinates obtained are transformed into the 1960 CIE UCS coordinates: (u, ,VI ), (uri, v-. ), (Ilk, Uk ), and (uki' vk;), respectively. Fourth, to account for the adaptive color shift due to different chromatic adaptation for the reference illuminant and the source 10 be tested, the coordinates for reflected spectra of the test source are modified:





400 500 600 700 Wavelength (nm)

FI(;.2.4.2. Reflectivity spectra of eight test samples used for calculation of the color rendering indices (1964 eIE).

c = (4-u -IOv)/v,

d = (I .708v +OA04-1.48Iu )/v.

0.8 Light grayish red
0.0 I
0.8 Dark grayish yellow
z- 0.0
Q) 0.8 Strong yellow-green
400 500 600 700 where for all relevant sets ofUCS coordinates,

Light bluish green

Light blue



Now the special color rendering indices Ri, which refer to each test-color sample, are calculated:

s, == 100-4.60 {[Wki -WriY +132[wduki -ur)-Wn(Uri -ur))2 + 1 32 [Wki(uh - vJ- Wn(Uri - vr))2 }1/2.




Here the lightness indices, Wri and Wki, are to be determined for all reflected spectra under consideration:

w == 25 Y 1/3 - I 7 ,


where values of Yare obtained from Eq. (2.3.1 b). To implement the Y values properly, the initial spectra Sr(..1,) and Sk(,l.) are to be normalized so that

Yr == Yk = 100.

Finally, the value of the general color rendering index, Ra, is obtained through averaging the values of the special color rendering indices

] 8

s, =- LRi· 8 i=1


The color rendering properties of a lamp with Ra =] 00 are identical to those of a standard reference illuminant, at least for the eight test samples speci fled. A source with Ra = 50 has color rendering properties that shift the chromaticity coordinates in average as much as they are shifted by a "warm white" halophosphate fluorescent lamp.

The rating of the color rendering properties of a lamp can be supplemented by special indices that refer to six additional test samples (i == 9, ... ,14) defined as strong red, strong yellow, strong green, strong blue, light yellowish pink, and moderate olive green.

It is to be noted that the method described above is applicable for light sources which have chromaticity close to the reference illuminant, Accurate values of color rendering indices are achieved under conditions of a small chromaticity difference between the source to be tested and the reference illuminant, so that

r( )2 ( )? Jl/2 -3

rUk -ur + Uk =v- - <5.4x]0.


When one reads about this procedure, one cannot help but think that there must be a better way. At the present time, there is none, even though quite a few of these standards and procedures have historical roots, and, in principle, might be improved in the future. However, proposing changes in existing standards is a difficult road to follow.



In this chapter we consider existing light sources, which compete and will continue to compete with semiconductor lighting. The nonsemiconductor lighting technology is very versatile and, in many ways, defines the outlook for advances in semiconductor lighting devices. We present the basics of the physical operation principles and device design and review the efficiency and quality of light. We also discuss the main advantages, disadvantages, and applications of nonsemiconductor lighting (for more details, see Coaton and Marsden 1997, Rea 2000, and references therein). Finally, we address the economy of lighting briefly in Section 3.7.


The emission of incandescent lamps is due to blackbody radiation. In modern incandescent lamps, filaments are made almost exclusively of tungsten, since of all metals it has the highest melting point (3683 K) and the lowest evaporation rate (lowest vapor pressure). However, the high melting point is insufficient to bring the peak wavelength to the visible, and even at high operation temperatures, most of the emission is in the infrared. Fortunately, tungsten is a selective emitter (i.e., it differs from an ideal blackbody in that the peak emission wavelength is blueshifted and the luminous efficacy is thus higher; see Fig. 3.1.1). The blue shift does not affect the color rendering, which remains almost perfect compared to that of an appropriate Planckian radiator.






' .... '. " .. , ,

,I "

It , •

i. \ I

:. \ ~

.t " 'to

. \ .

, . \

. . ,


, ,

, , ,


.. ..


I •

1000 1500 2000 2500 3000 Wavelength (nm)

FlG.3.1.1. Normalized spectral power distribution of the blackbody (dotted line) and tungsten radiator (dashed line) at 3000 K. Solid line is the spectral sensitivity of the human eye.



In a typical general lighting service (GLS) incandescent lamp, the filament is wound into a helix to reduce the heat conducted from the wire surface into the surrounding gas (further reduction of gas loss is achieved by secondary coiling of the primary coil-a coiled coil). The filament is supported by molybdenum wires and connected electrically to leads made from nickel or nickel-plated wires. The design usually includes a copper-nickel fuse, the bulb is made of soda-lime silicate glass, and the cap is made of aluminum or brass. Matching the thermal expansion and stabiJ i ty of the gl ass-metal seal is prov ided by Dumet wires (a composite material consisting of a nickel-iron core covered with copper). To reduce the rate at which tungsten evaporates from the filament, the bulb is filled with highatomic-weight inert gas (argon and, rarely, krypton). A small amount of nitrogen is included to prevent the formation of an arc during filament failure. To reduce blackening of the bulb, some getter is added for absorption of any remaining oxygen and moisture. Small bulbs of lamps below 40-W power are usually evacuated. To diffuse and direct the light, inside frosting and integral reflectors are used, respectively. When one realizes that such a complex product is sold (after retail markup) for as little as 50 cents, one better understands the uphill battle that semiconductor lighting will face in trying to replace this nineteenth/twentieth century technology completely.

Additional to the fact that the main portion of the radiation is in the infrared (IR), some power is lost in the supports and lead wires (1.5 to 2%), some is conducted from the wire surface into the surrounding gas (6 to 20%), and some is removed from the filament by convective gas flow «1%). The overall luminous efficiency for 120- V lamps thus ranges from 8 Im/W at an operation temperature of 2400 K (lOW) to 23 ImlW at 3100 K (1500 W). In a 120 VI100 W lamp, only



about 7% of the energy consumed is converted to visible radiation. The efficiency of lamps depends on the voltages they are designed for. For instance, 220- to 240-V lamps are about 20% less efficient; lower-voltage (12- to 75-V) lamps used in vehicles are more efficient than 120- V units of the same wattage. To improve the efliciency, sophisticated design might comprise IR reflecting films, which redirect the long-wavelength radiation to the filament, thus maintaining its higher temperature .

The principal cause of incandescent bulb failures is loss of tungsten from the filament by evaporation. The filament melts at a "hot spot" that develops at geometrical and structural inhomogeneities. Higher operational temperatures result in a shorter lifetime but higher efficiency. This trade-off yields an economic life for an incandescent lamp for a minimum cost per lumen-hour of





where r Lis the life (hours), P L is the wattage, and C Land C [ kWh are the cost of the lamp and of I kWh of electricity, respectively. Typically, the lifetime is 750 to 1000 hours (with 220- to 240-V lamps having a longer lifetime). Lamps with lifetimes up to 2500 hours are available for applications where replacement costs are high. Higher lifetimes, as well as some other special requirements (e.g., vibration resistance), are obtained at the expense of efficiency.

At present, tungsten incandescent lamps serve most of the needs of residence lighting. They are also used widely for signal and panel lighting in vehicles, as special-service lamps, and in old-fashioned traffic lights (where they are being replaced rapidly by more efficient and economical LED traffic lights; see Section 7.2.1).


If the tungsten evaporation rate were reduced, the filament of an incandescent lamp might operate at higher temperature and the lifetime of the lamp might increase. In tungsten halogen lamps, the addition of halogen to the gas fill ing is known to establish a chemical transport cycle in which tungsten forms halides When diffusing from the hot filament toward the cooler wall. Tungsten halides diffuse in the opposite direction and dissociate at the filament. The transport cycle results in nearly zero concentration of the tungsten at the bulb wall and in increased concentration at the filament. As a result, filament temperatures as high as 3450 K Can be achieved, with consequent improvement in efficiency. The details of the underlying chemical reactions are not yet fully understood. They probably inVolve some residual gases (water vapor, oxygen, hydrogen) and impurities.

For the halogen cycle to work, considerably higher bulb temperatures are required (i.e., >500 K). Therefore, the bulbs employed are of smaller dimensions and are made either of alkali-free hard glass or, for higher power, of fused si lica. Since no metal matches the low coefficient of thermal expansion of fused silica, her-



metic seals are made by pinching the molten bulb onto molybdenum foils. Smaller bulbs may be made mechanically stronger and thus may operate at a higher pressure of filling gas. In addition, utilization of more expensive higher-atomicnumber filling gases (krypton or xenon) is more economical because of the smaller amount of gases required. Halogens (iodine, bromide, chlorine, and fluorine) are

usually introduced into the gas tilling as halogenated hydrocarbons (e.g., CH3Br).

The failure mechanism of the tungsten halogen lamp is similar to that for a conventional incandescent lamp. However, typically, tungsten halogen lamps have at least twice the lifetime of an incandescent bulb at the same efficiency, or higher efficiency at the same lifetime. Also, tungsten halogen lamps may require much lower wattage for efficient emission of light with high color temperatures. This makes them very attractive for many applications. One of the disadvantages is a long heating and cooling time (i.e., -I s) that often prevents halogen lamps from use in all kinds of signal lights. Also, these lamps cannot be dimmed, since reduced temperatures break the halogen cycle.

Tungsten halogen lamps are convenient (but expensive) compact nonflickering sources of bright I ight and are widely used in floodlights, motor vehicle headlights, photographic and television studio lighting, film projectors, and even in residence lighting (mostly as spotlights).


When a large enough electric field is applied to a gas, the gas breaks down and partially ionizes. The resulting conductive plasma comprises electrons as well as a mixture of ionized and neutral particles, some of which are excited. By limiting the electric current (by introducing a ballast in the circuit), the discharge is prevented from avalanche ionization and stabilized. The fluorescent lamps utilize low-pressure discharge, in which electrons are accelerated to effective temperatures typically of I 1,000 to 13,000 K, while ions remain almost in thermal equilibrium with the environment (;::; 310 K). Fast electrons inelastically relax by exciting atoms, molecules and ions, which might emit light. At present, two efficient lowpressure discharge emitters are utilized-vapors of mercury and sodium. Sodium emits yellow light, which is used directly, mostly for street lighting (see Section 3.4). At low pressures, the major part of the emission from mercury atoms is in the ultraviolet (UV), owing to the radiative transition from the excited 3p1 state to the ground state (4.886 eV /253.7 nm). In a fluorescent lamp, visible radiation is produced by photoluminescence in phosphors, which are deposited on the wall of a tubular bulb. UV photons reach the wall via radiative transport (i.e., by multiple reabsorption and reernission by other mercury atoms),

Only ionic phosphors are suitable for operation in fluorescent lamps. They comprise a robust crystal lattice with activator ions at typically -I % concentration. The process of the photon down-conversion in an activator ion is depicted in Fig. 3.3.1a. The potential energy of the activator ion is plotted on the vertical axis. The effective displacement of the host ions is represented by a configuration coor-





Configuration Coordinate

FIG. 3.3.1.

Configuration-coordinate diagrams of an activator ion in a phosphor for radiative conversion process (a) and nonradiative conversion process (b).

dinate Q, which is plotted on the horizontal axis. A parabolic potential curve is the locus of vibrational states of the ion. Owing to strong electron-phonon coupling, different occupations of electronic states in the activator ion result in different displacements (i.e., the parabola of the nonexcited ion differs from that of the excited one in the ground-state configuration coordinate), The electronic transitions (vertical arrows) occur so fast that the host atoms have no time to change their positions (Frank-Condon principle). Therefore, optical transitions from the ground states (bottoms of the parabolas) involve vibrationalexcited final states, which relax via excitation of local modes and, further, lattice phonons. Eventually, a part of the energy of the absorbed photon is converted to heat and the photons emitted are of lower energy than those absorbed (Stokes shift)·

At nonzero temperatures, both emission and absorption spectra are usually broadened. However, sharp emission lines occur in rare-earth ions for 4f-4ftransitions (since these levels are weakly coupled to the lattice), In cases when the incident-photon energy is too large to be absorbed efficiently by the activator, the process requires mediation by a second impurity ion (sensitizer) that converts the i nci dent radiation to photons 0 f appropriate energy.

The excited activator might also lead to a nonradiative decay (Fig. 3.3.1 b). In this case, the electron slides to the initial state while the ion is brought to a highly excited vibrational state. At low temperatures, the transition occurs because of tunneling in the configuration space (horizontal arrow in Fig. 3.3.1b) and the rate of the nonradiative process is almost temperature independent. At high tempera-



tures, the harrier at the crossover of the parabolas can be surmounted and the probability of nonradiative decay increases exponentially with temperature.

Phosphors used in fluorescent lamps must. satisfy requirements for structural and chemical stability under UV radiation and in the environment of mercury and plasma discharge. They should also exhibit high quantum efficiency (typically, >85%). Calcium halophosphate phosphor, Ca5(P04h(CI,F):(Sb3+, Mnl '), became widely used soon after the appearance of the fluorescent lamp. The antimony ion act both as sensitizer and activator. II absorbs the UV radiation and emits a broad band peaked near 480 nm. Part of this radiation is transferred to the Mn2+ ions, wh ich cmi I lit 580 n m. A typ iea I spectrum of hal op hosphate phosphor em ission is shown in Fig. 3.3.2a. By changing the Sb3+/Mn2+ and CI/F ratios, different hues of white color are produced. However, because of deficiency of the red color, the light is of poor color rendering (R(I == 50 to 76). Higher rendering indices

( Ra - 90) arc obtained in deluxe phosphors, which are mixtures of strontium ha!ophosphate, Sr5(P04h(CI,F):(Sbh, Mn2+), and strontium orthophosphate, SfJ(P04h:Snh. However, deluxe phosphors yield somewhat lower luminous efficiencies than those of standard halophosphate phosphors.

In the 19705, a new category of phosphors (triphosphors) was developed. The idea employed is that good color rendering can be obtained by using emitters, which produce narrow lines at the peaks of the CI [ color-matching functions (which occur at 600,550, and 450 nrn; see Fig. 2.3.1). Triphosphors may comprise





Wavelength (nm)

FJG. 3.3.2, Spectral power distributions of typical fluorescent lamps. (a) halophosphate, (b) triphosphor blend. (c) multiband phosphor. (After Coaton and Marsden 1997.)



a variety of matrices doped with rare-earth ions. Y 203 :Eu3 + is widely used for red;

3+ 3+ 3+ 3+ 3+

CeMgAI11019:Tb , LaP04:Ce , Tb ,and GdMgBSOIU:Ce ,Tb for green;

2+ 2+

and BaMg2AI16027:Eu and Sr5_x_yBaxCay(P04)3CI:Eu for blue color. A

lypical emission spectrum of a triphosphor blend is displayed in Fig.3,J.2b. Triphosphors are much more expensive, but they surpass halophosphates in quantum efficiency, lifetime, and color rendering (Ra = 80 to 85). However, three narrow emission lines still distort colors. Higher color rendering indices ( R(I :> 90) are achieved by using combinations of up to five phosphors (multi band phosphors). The multiband phosphors comprise triphosphor blue and green phosphors, halophosphates, and Mn2+-activated pentaborate for the broad red emission peaking at ~620 nm (Fig. 3.3.2c).

The design of fluorescent lamps is based on a tube of soda-Ii me silicate or low-sodium-content glass. In a hot-cathode lamp, two electrode mounts are sealed a: each end and caps with pin connectors are attached. The electrode mount holds the cathode, which is made of tungsten wire coiled in two or three stages (coiled coil or coiled-coiled coil). The wire is coated with low-work-function materials such as barium oxide to make electrons escape easily at the typical operation temperature of 1400 K. The lead wires are composed of the inner iron-nickel alloy wire. Dumet wire is used for seals, and the outer part is made from copper or copper-plated iron. (Cold cathodes are also available for certain appl ications.) The inside wall of the tube is coated with a layer that has to protect the phosphor from the sodium in the glass and to reflect UV (different precoatings are being used). The pressure of the mercury vapor is typically around 0.5 to 0.8 Pa (4 to 6 xlO-3 torr) at the operating temperature. To reduce the diffusion of electrons and ions toward the wall, the tube is filled with an inert gas at some optimum pressure (high pressure reduces the electron temperature). The inert gas also assists starting and reduces the energy of ions hitting the cathodes. The gas commonly used is argon at pressures around 260 to 670 Pa (2 to 5 torr). Mixtures containing krypton and neon are also utilized. Since the most efficient generation of UV is achieved at an electric field typically of about I Vzcm, the length ofthe tube has to be about 1 m at an operating voltage of 100 V. The diameter of the tube is also critical (in wide tubes, radiative transport of UV photons experiences losses; in narrow lubes, losses occur because of electron collisions with the walls). Linear fluorescent lamps are most commonly made of three diameters (38, 25, and 15 rnm) with a variety of lengths (150 to 2400 mm). Single-ended lamps are bent and thus twice as short. The European design (220 to 240 V) allows a simple inductor to be used for the ballast. The starting circuitry preheats the cathodes and employs the inductor to produce a high-voltage kick. The North American design (120 V) requi res an autotransformer to increase the voltage. Recentl y, highfrequency ballasts, operating at 20 kHz or more, were introduced. They provide benefits in energy saving, reduce flickering, and increase the lifetime of the cathodes and phosphor.

At present, compact fluorescent lamps (CFLs) are heavily promoted as energy-saving light bulbs. CFLs are designed to substitute for incandescent lamps in



common applications such as domestic lighting. They are made using a multiply folded small-diameter (10- to 16-mm) tube, contain miniature control gear within the unit, and feature improved color rendering.

At optimal conditions, the energy consumed in a fluorescent lamp is converted to UV radiation with about 63% efficiency (the remainder is dissipated as heat in the discharge, electrodes, and at the wails). Only about 40% of the UV radiation is further converted to visible because of the large Stokes shift and some nonradiative losses in the phosphor. In addition, approximately 3% of the energy is converted directly to visible radiation in the discharge. Therefore, the overall efficiency of the conversion is about 28% (i.e., four to five times higher than that of a 100- W incandescent lamp). The luminous efficiency ranges from 35 to 50 ImlW in low-power units (4 to 5 W) to 75 to100 Im/W in high-power linear (70 10 125 W) and electronically ballasted (10 to 60 W) lamps. However, the high luminous efficiency of some types of lamps is achieved at the expense of good color rendering.

The failure of fluorescent lamps is commonly determined by deterioration of the cathodes. Depending on the construction, the lifetime ranges from 5000 to 24,000 hours. At the end of the life, the output of the lamps drops to 70 to 80% of its initial value because of deterioration of the phosphor material.

One disadvantage is strong dependence of the light output on the ambient air temperature, which influences the pressure of the mercury vapor (see Fig. 3.3.3). This is also the reason that fluorescent lamps require warming up for a few minutes to reach peak output. Flickering at 100- or 120-Hz frequency, insufficient color rendering, radio-frequency interference, and sometimes, audible noise from the ballast are still shortcomings of the old-design lamps and some of the cheaper eFLs.

LL 60
I:: 40
20 OL...----'-_--'-_L._--'-_...J...._L._--L_...J

-20 -10 0 10 20 30 40 50 60

Temperature te)

FfG.3_3.3. Light output of a typical fluorescent lamp as a function of ambient temperature. (After Coaton and Marsden 1997.)



Like almost all discharge lamps, fluorescent lamps contain mercury. Collection and recycling of end-of-life lamps is a serious problem which requires national and international infrastructures. However, the overall mercury load of fluorescent lamps on the environment is smaller than that for incandescent lamps, since the relevant mercury emission from the power stations is much larger than the amount consumed in production of the lamps.

In ind ustri al co untri es, fluorescent lamps prov i de the maj or portion of artificial lighting. They dominate in general lighting of industrial and commercial buildings and offices. Multiband phosphor lamps are suitable for many appl ications where high color rendering is necessary. This is one of the reasons that CFLs are beginning to supplant incandescent lamps in domestic lighting.


In lew-pressure sodium (LPS) lumps, discharge is of the same origin as that in mercury lamps. Sodium emits almost monochromatic yelJow light, a doublet consisting of 589.0- and 589.6-nm lines (D-line). However, the melting point of sodium is higher than that for mercury, and the optimal operating temperature is about 530 K. Also, hot sodium is highly reactive chemically. This imposes additional requirements on the materials the lamp is made of and requires more means of preventing heat losses.

The design of lamps is based on a U-shaped arc tube made of sodium-resistant ply-glass (soda-lime glass with a l Ofl-urn-thick inside layer of aluminoborate glass). The outer surface of the arc tube is dimpled to make sodium condense uniformly. At operating temperatures, the pressure of sodium vapor is kept at 5 to 8 x 10-3 torr (0.7 to 1 Pa). The cathodes are made from triple-coiled oxide-coated tungsten. In contrast to the mercury fluorescent lamp, the cathodes are heated to an electron-emissive temperature by ion bombardment. The tube is filled with neon at a pressure of 3 to 15 torr (400 to 2000 Pa). Small additions of argon, xenon, or helium are introduced to facilitate starting. To reduce heat losses, the arc tube is enveloped in the outer bulb. The space between the arc tube and the outer bulb is evacuated and contains getter to maintain vacuum throughout the life of the lamp. Additionally, the inside of the outer bulb is coated with indium oxide film that reflects IR. The lamp requires a control gear for ballasting and starting. When the lamp is switched on, discharge starts in the neon. The sodium vapor pressure reaches its optimal level after 10 to 15 minutes of warming up,

Since the luminous efficacy of the D-line is about 530 Im/W, the theoretical efficiency of the LPS lamp is extremely high. Despite the fact that 60 to 80% of the power is wasted through infrared radiation and heat losses, the luminous efficiencies (100 to 200 Im/W) are highest among present practical lamps. The wattage of the LPS lamps marketed is from 18 to ]80 W, with the light outputs ranging from 1800 to 33,000 1m. The failure is due to deterioration of cathodes, and typical lifetimes are 14,000 to 18,000 hours.

The main disadvantage of LPS lamps is very poor color rendering. The standard calculation discussed in Section 2.4 yields a value of Ra = -44. Another



problem is the long warm-up time. These drawbacks restrict the use of LPS lamps in street and road lighting as well as in security lighting.


Physically. high-pressure discharge differs from low-pressure discharge in that the heavy particles (atoms and ions) are heated to almost the same temperatures as the electrons, owing to the high rate of elastic collisions. At pressures of around I atm, the temperature of the plasma is typically in the range 4000 to 6000 K. Because of interaction with the surrounding walls, a radial temperature gradient is established in the arc. Most of the I ight is generated in the hot center of the arc. However, because of the temperature gradient, heat flows out of the center, decreasing the radiative efficiency to approximately 60%. High pressure leads to collision broadening of the line spectra. The resulting wide emission bands considerably improve the color rendering ofthe light. However, the outer part of the arc, which contains cooler plasma, may reabsorb the radiation so that dips occur at the centers of the wide bands (called self-reversed spectral lines).

The design of high-pressure discharge lamps uses an arc tube with robust electrodes. Commonly, to reduce heat losses, the tube is enveloped into an outer bulb. The electrical gear of the lamps consists of a ballast and high-voltage ignition circuitry to start discharge under high pressure. High-pressure discharge lamps, also called high-intensity discharge (HID) lamps, are mainly of three types: mercury lamps, high-pressure sodium discharge lamps, and metal halide lamps.

3.5.1. Mercury Lamps

Mercury vapor discharge may efficiently emit in the visible only under high pressure. At operating pressures of 2 to I 0 atm (0.2 to l MPa), the emission spectrum is shifted toward broad longer-wavelength lines (405, 436, 546, and 578 nm) and contains some continuous background (see Fig. 3.5.1 a), To permit hightemperature, high-pressure operation, high-pressure mercury vapor (HPMV) discharge lamps are manufactured with an arc tube made from fused silica. The tube contains a mercury dose, which is necessary for operating pressure. The electrodes are made from tungsten and impregnated with electron-emissive materials. To facilitate starting, argon pressure of 18 to 36 torr (2400 to 4800 Pa) is maintained and an auxiliary electrode is introduced. Pinching the softened silica onto the molybdenum foil produces the seals, The outer bulb, commonly made from borosi I icate glass, is 51 led with nitrogen or a nitrogen-argon mixture to prevent oxidation of the internal structure and to suppress internal arcing.

Because of deficiency of red color, a clear mercury lamp has a very poor color rendering ( Ra = 16). The improved-color lamps are produced by coating the inner wall of the outer bulb with phosphor, which converts the remainder of UV into red light. The phosphor, (typically, europium-activated yttrium vanadate) raises the color rendering index up to 50 and improves the luminous efficiency (see the spec-

..... (c) (d)





500 600 700 400 500 600 700

Wavelength (nm)

FIG.3.5.1. Typical power-distribution spectra of high-pressure discharge lamps: (a) clear mercury lamp; (b) improved-color (phosphor-coated) mercury lamp; (c) sodium lamp; (d) metal halide lamp with rare earth (Dy/Ho/Tmj-Na-Tl dose. (After Coaton and Marsden 1997.)

trum in Fig. 3.5.1 b). Another approach is to introduce a tungsten filament in series with the discharge (the filament also acts as a ballast resistance). However, improvement in color rendition in such a "blended" lamp is achieved at a considerable decrease in efficiency.

In HPMV lamps, more than half of the power consumed is wasted to produce heat. Other losses comprise IR and UV radiation and, for phosphor-coated lamps, photon down-conversion losses. The luminous efficiency averaged throughout the lifetime depends on wattage and is 20 to 45 Im!W in clear mercury lamps (75 to 1000 W) and 20 to 50 Im/W in phosphor-coated lamps (45 to 1000 W). Failure of HPMV lamps is normally due to loss of emission from the electrodes, and the Ii fetime may reach 24,000 hours. However, because of gradual loss of light output, these lamps are commonly replaced after 8000 to 10,000 hours.

The main use of HPMV lamps is in street and road lighting supplementary to sodium lamps, Improved-color lamps are also suitable for some commercial interiors.



3,5,2, High-P.ressure Sodium Discharge Lamps

High-pressure sodium (HPS) discharge lamps emit a broad self-reversed D·line (Fig, 3,5,1 c). The arc tube contai ns sodium at pressure of about 50 torr (7000 Pal. The operating temperature of the plasma is about 4000 K at the center and 1500 K at the wall, respectively, To reduce heat loss and to increase the power per unit volume, the tube is filled with a buffer gas, a mixture of mercury vapor (450 torr/60,000 Pa) and xenon (150 torr/20,000 Pa), To withstand such harsh conditions, the arc tube is made from translucent polycrystalline alumina (PCA), The electrodes are made from tungsten, impregnated with alkali-earth oxides, and supported by niobium leads. The hermetic end seals are fused from separate components with a highly chemically resistant substance (calcium-aluminum oxide glass or alloys containing titanium, vanadium, or zirconium). The arc tube is mounted on a frame made from nickel, manganese-nickel, or nickel-plated iron. The arc and the frame are enveloped in the evacuated outer bulb (borosilicate glass or fused silica)

Standard HPS lamps are produced for wattages in the range 50 to 1000 W.

The luminous efficiency increases with wattage from 60 to I 30 !m/W. The typical lifetime is 24,000 hours (the failure occurs because of deterioration of the arc-tube components). Color rendering is much higher than in the LPS lamp, but still too poor (Ra = 20 to 25) to extend the range of applications beyond road lightening. Color improvement is possible with increasing sodium pressure. The penalty is reduced luminous efficiency and lifetime.

3.5.3. Metal Halide Lamps

The luminous efficiency and color rendering of high-pressure mercury discharge lamps may be improved considerably by introducing other metals into the arc in addition to mercury. However, the metals required have insufficient vapor pressure and react with the arc-tube material. Therefore, metal halides are used to implement this idea. As soon as the tube wall reaches sufficient temperature, a metal halide evaporates and starts a transport cycle similar to that in the tungsten halogen lamp but with a somewhat different function. At the hot core of the discharge, the halide dissociates and produces metal atoms that contribute to emission. When the metal atoms diffuse toward the cooler region at the wall, they recombine with the halogen to form the halide, which does not react with the wall material. The operati on pressure a f the additive metals is in the range 10 to 100 torr (I 300 to D,OOO Pa). Although this pressure is small in comparison with the pressure of mercury (typically, I to 20 atm), additive metals produce a considerable part of light because their excitation energy (around 4 eV) is lower than that of mercury (7.8 eV). Mostly, line spectra are obtained by using sodium, scandium, thallium, indium, cesium, and rare-earth iodides. For more continuous spectra, tin or tin-sodium halides (iodides with addition of bromides and chlorides) are used. By varyi ng the hal ide cornposi t ion, the output spectrum might be ta i lored to obtain desirable characteristics (see Fig. 3.5.1 d).



The basic design of metal halide (MH) lamps is similar to that of HPMV lamps. The arc tubes are made from fused silica and PCA. The length of the arc may be much shorter than in HPMV lamps (MH lamps with arcs as short as I mm are available). For higher chemical resistance, thorium metal or thoriated tungsten is often used for the electrodes. To facil itate starting at high pressure, high-voltage igniters are used, and sometimes, radioisotopes are introduced into the gas filling. In MH lamps with reduced warm-up time, xenon at pressures of several atmospheres is utilized as the starting gas. Some compact lamps are produced without the outer envelope; others employ envelopes made from doped or thin-film-coated fused silica for UV blocking.

MH lamps are still being actively developed. They are produced in a wide range of wattage (from 20 to 18,000 W). The initial luminous efficiency is 70 to 110 ImlW, depending on the wattage and additive metals used. The general rendering color indices commonly are above 60, and may reach 95 for some rare-earth doses. Depending on the design and materials used, the lifetime is 2000 to 30,000 hours.

Since the metal halides are maintained in a partially dissociated regime, the density of additi ve-metals vapor is strongly dependent on temperature. This is one of the reasons for poor lamp-to-lamp color consistency of MH lamps. The color al so changes th rough I ife. A nother problem is that segregati on of the dose speci es in the arc results in color nonuniformity that is undesirable for image projection. Besides, all MH lamps emit UV radiation, which passes through fused silica and PCA. Therefore, some degree of protection is usually required. Finally, because of the high pressures employed, MH lamps may undergo a violent (explosive) failure and thus be dangerous.

Owing to diverse wattage, dimensions, and other specification, MH lamps find a wide range of use. High brightness and good color renderi ng characteristics make them appl icable for general lighting serv ices in offi ces, supermarkets, large stores, and in a lot of industrial and social environments. High-l umi nosity units are indispensable for floodlights. Low-power short-arc MH lamps gave birth to a new kind of economical, precise, small-dimension, long-lifetime vehicle headl ights.


The main reason for failure of all kinds of discharge lamps is deterioration of electrodes. Electrodeless discharges were invented more than 100 years ago. However, only recently, electrodeless lamps became available, owing to the progress in manufacturing reliable and inexpensive high-frequency sources. In practical electrodeless discharge lamps, the discharge is excited either by induction or by rnicrowave radiation.

The induction lamp utilizes magnetic coupling between the plasma in the lamp and the induction coil. The initial breakdown of the gas is due to the voltage applied to the coil and is maintained using parasitic capacitance. After the discharge is initiated, it forms a single tum around the coil. At frequencies above I MHz, high coupling efficiencies are achieved. Typically, the lamp contains a -1 O-~lH ferrite-cored coil placed inside a bulb. The bulb is filled with an inert gas



and contains a source of mercury. To convert the UV radiation of mercury to the visible, the inner wall of the bulb is coated with phosphor. A built-in or external radio-frequency (RF) oscillator drives the coil at the frequencies allowed for industrial applications (13.56 MHz or 2.2 to 3 MHz). The lamp is equipped by means that prevent large currents in the fixtures and suppress RF interference.

The induction lamps are designed to substitute for incandescent lamps, especially in the areas of high-cost maintenance. The wattage varies from 23 to 85 W at a luminous efficiency of 47 to 71 Im/W. The general color rendering index (R" '" 80 ) is sufficient for many general-lighting-service applications. Lamps are available with lifetimes as high as 100,000 hours.

Another kind of electrodeless lamp, the sulfur lamp, utilizes microwavedriven discharge in sulfur, a novel dose, which is impossible to usc in a design \ .... ith electrodes. The microwave radiation is produced using a magnetron operating at 2450 MHz, the frequency allotted for microwave ovens. The discharge is excited in a bulb filled with sulfur and argon and placed into the cavity, At present, a 1425- W lamp with the luminous efficiency of 95 Im/W ( Ra = 79) is marketed. It exhibits very consistent unit-to-unit performance. The lamp is the only mercuryfree discharge-based source of white light The lifetime of the system is determined by deterioration of the magnetron, which is to be changed each 20,000 hours. The sulfur lamp is suitable for many applications, including large-area lighting (warehouses, airplane hangars, meeting halls) and light-pipe technology.


The most important data on the typical practical lamps discussed are summarized in Table 3.7.1 (some numbers may vary [or different manufacturers). From an economical point of view, the price of light produced by different sources is of high interest. The cos! oflight can be estimated roughly from the cost of the lamp and the electric power consumed divided by the number of lumens produced over the lifetime. For I Mlm-h, this yields a cost of


where CL is the cost of the bulb corrected for the cost of the external circuitry if required for the lamp in question, C1kWh is the price of I kWh of power, "L is the luminous efficiency corrected for losses in the ballasts, and PL and t L arc the wattage and I ifetime of the lamp, respectively. The first term on the right-hand side of Eq. (3.7. I) accounts for the price of the lamp, and the second term represents the price of the power consumed per 1 Mlm·h. Note that the expenses for maintenance and disposal or mercury-containing bulbs are not accounted for in Eq. (3.7.1).



Table 3.7.1. Parameters of practical lamps and tubes

Type Wattage Luminous Em- Ra CT Life- 1 Mlm·h
(W) flux, init. ciency (CCT) time price ($)
(avg) (Irn) (lm/W) (K) (hours)
Incandescent 60 865 14.4 100 2790 1000 7.4
(120 V)
Tungsten 50 590 11.8 100 2750 2000 12
(120 V)
Flu 0 rcscenr 32 2,850 84 78 (4100) 24,000 1.6
triphosphor (2,710)
Compact 15 900 51 82 (27UO) 10,OOD 3.9
fluorescent (765)
Low- 90 12,750 123 -44 (1800) 16.000 1.6
pressure ( 11,095)
High- 250 11,200 34 50 (3900) 24,000 3.8
pressure (8.400)
Iligh- 250 28,000 108 22 (2100) 24,000 J_J
pressure (27,000)
Metal halide 400 36,000 60 65 (4000) 20,000 2
Induction 55 3,500 64 80 (3000) 100,000 2
Microwave 1,425 135,000 95 79 (5700) 20,000 ?
sulfur The rightmost column in Table 3.7. j displays the results of the estimation.

The correction for the cost of external ballast amounts to 25% of the price (the price of the igniters is neglected). The losses in the external ballast are typically around 20% of the lamp's wattage. The price for the electricity was rated at $0.10, and OEM (original equipment manufacturer) prices are used for the bulbs and ballasts.

From Table 3.7.1 one can see that the lowest cost is that of the lowest-quality light, produced by sodium lamps. For general-lighting sources with high color rendering indices, the fluorescent lamp is noncompetitive. The CFL's light is still two to three times cheaper than that of incandescent lamps. However, a twofold reduction in light price can be expected with further improvement and cheapening of the eFL (at present, the price of the unit contributes 30 to 50% of the light cost). Incandescent lamps are sources of the most costly light Their use is justified



by the high quality of the emission, which is preferred for residence lighting and by the deceptive cheapness of the bulbs.

According to Lidow (1999), the estimated total global consumption of elec-

.. . b I 13 k

tricity IS a out x 10 Wh per year with 21 % of that consumed for lighting.

Thus, an improvement of luminous efficiency by 1 % may save $2 billions per year. However, the efficiency of conventional bulbs and tubes is not likely to improve considerably in the future, since the physical principles of light generation employed are essentially exhausted. The introduction of efficient new light sources based on other physical principles is therefore very important for energy saving. What about semiconductors?



Conventional lamps rely on either incandescence or discharge in gases. Both phenomena are associated with large energy losses that are essentially inherent because of the high temperatures and large Stokes shifts involved. Semiconductors offer an alternative way of light generation. Spontaneous light emission in semiconductors is due to radiative recombination of excess electrons and holes. Excess electrons and holes are produced by current injection with small energy losses. Subsequent radiative recombination of the injected carriers may attain quantum yields close to unity. This phenomenon, called injection luminescence, is the basis of operation of all light-emitting diodes (LEDs) used in all-solid-state lamps. The development of visible and near-UV LEOs that exhibit efficiencies comparable or higher than those of conventional lamps (high-brightness LEOs) gave birth to new I ighting technology (solid-stale fighting).

In this chapter we deal with the basics of injection luminescence as well as with materials systems, electric design, and the electroluminescence characteristics of high-brightness LEDs. Optical design, which is the most advanced feature of solid-state lamps, is considered in Chapter 5.


Electro{uminescence, which is emission in excess of blackbody radiation excited by an electric field, occurs in a variety of systems under different conditions. Ex-




arnples are gas discharges (see Chapter 3) and emission due to impact ionization in thin films (see Mueller 2000b for recent reviews). The most efficient kind of electroluminescence is that caused by carrier injection in semiconductors. This phenomenon was discovered by Round (1907), who injected carriers into silicon carbide from a metal contact and observed a yellowish light.

Since the invention of a red LED by Holonyak and Bevacqua (1962), the process of injection luminescence received considerable attention. It is considered both in the early books on conventional LEOs (Bergh and Dean 1976, Williams and Hall 1978, and Gillessen and Schairer 1987) and in more recent review literature dealing with high-brightness LEOs (Stringfellow and Craford 1997, Den Baars 2000, Mueller 2000a). In this section we consider LED performance determined by the processes of light generation in semiconductors, injection in a p-n LED, and the properties of heterojunctions and quantum wells that are used in the active layers of modem high-brightness LEOs.

4.1.1. LED Performance

An injection-electroluminescence device is characterized by its radiant efficiency (also called wall-plug efficiency)


where 17ext is the external quantum efficiency and 'l/ is the feeding efficiency.

The luminous efficiency of LEOs is related to the radiant efficiency via the luminous efficacy [see Eq. (2.2.6)]. The luminous efficacy can be calculated from Eq. (2.2.4) by using the emission spectra measured in relative power units.

External quantum efficiency is the ratio of the number of photons emitted and the number of electrons passed through the LED. Explicitly, it is the product of the internal quantum efficiency (radiative efficiency), 'lTad; injection efficiency, 'linj;

and optical efficiency (light-extraction efficiency), 'lopt:

'lext ee 'linj x'lrad X 'lop! .


Injection efficiency is the fraction of the electrons passed through the LED that are injected into the active region, where radiative recombination takes place (see Sections 4. J .3 and 4. I .4). internal quantum efficiency is the ratio of the number of electron-hole pairs that recombined radiatively to the total number of pairs that recombined in the active region (see Section 4.1.2). Finally, light-extraction efficiency is the fraction of the photons generated that escape from the device (see Chapter 5).

Feeding efficiency is the ratio of the mean energy of the photons emitted, h v , and the total energy that an electron-hole pair acquires from the power source when passing through the LED:



( 4.1.3)

where V is the forward voltage drop across the LED and q is the elementary charge (1.6022 x 10-19 C). The feeding efficiency can exceed unity provided that the appl i ed voltage is smaller than h Ii / q (this is a common situation observed at small driving currents). The paradox is due to the availability of high-energy electrons in the thermal distribution (i.e., photons emitted with the energy h v > q V "cool" the crystal) Actually, an ideal LED with external quantum efficiency

'l '" I and with negligible series resistance is a refrigerator that converts a porext

tion of the thermal energy into optical radiation. However, in real LEDs, the cool-

ing is counteracted by the internal heating, since 'lex! < 1 and some heat is produced because of the voltage drop across the series resistance of the contacts and semiconductor structure. Typically, at nominal driving currents, the value of the lost voltage V - hli!q ranges from 0.05 to 1.0 V. As the result, the feeding efficiency ranges from 0.75 to values in excess of 0.97.

4.1.2. Recombination of Electrons and Holes

Excess carriers can recombine both radiatively and nonradiatively. Competition between radiative and nonradiative recombination processes determines the internal quantum efficiency of an LED. An intrinsic mechanism of radiative recombination is band-to-band transitions, in which an electron-hole pair recombines, emitting a photon (Fig. 4.1.1 a). Furthermore, provided that the temperature is not too high, the electron and hole can form an exciton, a hydrogenlike structure with a binding energy in the range of millielectronvolts. Radiative annihilation of excitons is the second intrinsic mechanism of light emission (Fig. 4.1. I b). In some alloys used for the fabrication of LEOs (e.g., InGaN), the nonuniformity of the spatial distribution of constituents may cause considerable fluctuations of the band potential. Carriers that are localized at such fluctuations recombine radiatively with large probability, since they are no longer able to reach sites of nonradiative recombination. Actually, a localized carrier of one type endures until a carrier of different type is localized at a distance small enough to overlap the wavefunctions forming a localized exciton (Chichibu et al. 1998). Localized-exciton recombination is the third intrinsic mechanism of radiative recombination (Fig. 4.1.1 c).

Other mechanisms of radiative recombination are linked to impurities caused by defects and/or by intentional or unintentional doping (Fig. 4.1.2). The impurity levels in the bandgap trap free carriers that may contribute to photon emission. For instance, a radiative transition between the conduction band and an acceptor state (Fig. 4.1.2a) or between a donor state and the valence band (Fig. 4.I.2b) might occur. Also, electrons trapped at donor states can recombine radiatively with holes trapped at acceptor states (donor-acceptor recombination, Fig. 4.1.2c). Finally, a



= Conduction band =

= Conduction band =


I, \

I , , , , ,



I I I , \ I



~ r

== Valence band ==

= Valence band ==

FIG.4.J.t. Intrinsic radiative transitions in semiconductors: (a) band-to-band transitions; (b) free-exciton annihilation; (c) recombination of excitons localized at bandpotential fluctuations.

trapped carrier can form an excitcniccornplex with a carrier of different type (impurity-bound exciton). In many semiconductors, radiative annihilation of bound excitons (Fig. 4.1.2d) is the main emission mechanism at low temperatures and at low densities of excess carriers.

During a radiative transition, the energy and momentum have to be conserved.

The energy conservation results in the photon energy being equal to the distance between the levels occupied by the electron and the hole, respectively. For bandto-band transitions, photons with the mean energy close to the bandgap energy, E s» are produced. For transitions involving excitons and impurity levels, the

mean photon energies are somewhat lower. Also, the energy conservation determines the width of the emission lines. For instance, in the recombination of delocal i zed electrons and holes (Fig. 4.1.1 a), the line width is related to the thermal distribution of carriers and, typically, equals 1.8 k B Tc' where k B is the Boltzmann constant and Tc is the temperature of the carriers. For localized exci-

Ec----4 .... --


ED ED ED , ,
I ,
(b) (d) I ,
(e) I I
I 1
I ,
, ,
EA I ,
, ,
J Ev

FIG. 4.1.2. Radiative recombination involving impurity levels: (a) donor-stare -e-valcnceband transition; (b) conduction-band-e accepror-state transition; (c) donoracceptor recombination: (d) bound-exciton recombination,



tons (Fig. 4.1 .1 c), the emission line is shaped by a magnitude of the fluctuations of the band potential.

Momentum conservation imposes strict requirements on the energy-band structure of semiconductors used in the active regions of LEDs. A photon wavenumber, kph == Zstv] c , where v is the frequency and c is the velocity of

light (2.9979 x 108 m!s), is very small compared to typical wavenumbers of electrons and holes (k '" n -I ~ 2m * k 8 Tc , where m * is the effective mass). This means that the electron transition from the conduction band into the valence band looks practically vertical on a band diagram E(k) (see Fig. 4. 1.3 a). That is why direct-gap semiconductors (e.g., GaAs, GaN, ZnSe), which have the top of the valence band and the bottom of the conduction band aligned in the same k-point, are better suited to emit or absorb light and are generally better suited for applications in optoelectronic and light-emitting devices. In indirect-gap semiconductors (e.g., GaP, SiC), where the band extrema are separated in the k-space. band-toband radiative transitions require a third particle (e.g., a phonon, a plasmon, or a carrier with an appropriate momentum) to facilitate the photon emission. Usually, the three-particle processes are of lower probability and indirect-gap materials are incapable of emitting light efficiently via intrinsic recombination routes. Fortunately, momentum conservation can be violated, owing to Heisenberg's uncertainty 11k - n/ t.x . This is important for emission that occurs at localized states. For instance, in GaP doped with isoelectcnic impurities (N, 0), an electron can be trapped from the conduction band by an impurity level, Subsequently, a bound exciton with high probability of radiative recombination can be formed (Fig. 4.1.3b).

The rate of the radiative recombination, Rr, is one of the crucial parameters for electroluminescent materials. Most high-brightness LEDs employ intrinsic






(a) Vertical band-to-band radiative transition in a direct-gap semiconductor; (b) impurity-assisted radiative transition in an indirect-gap semiconductor.



mechanisms of radiative recombination (Fig. 4.1.1). For direct band-to-band transitions, the rate of radiative recombination can be estimated by using the principle of detailed equilibrium (Van Roosbroeck and Shockley 1954). Under conditions of thermodynamic equilibrium, the rate of the spontaneous emission equals the rate of the absorption. For nondegenerate carriers, the latter can be estimated numerically as


where nr (v) is the refractive index, a(v) is the absorption coefficient, and T is the temperature of the crystal. When excess electrons and holes with the densities 6.n and 6.p, respectively, are injected, the net emission rate is

(no + 6.n )(po + 6p) - nopo

s, == RO,


( 4.1.5)

where 110 and Po are the equilibrium densities of electrons and holes, respectively. Usually, the emission rate is expressed via the coefficient of radiative recombination, B,

( 4.1.6)



Here nj "" ~ nopo is the intrinsic density of carriers.

Another way to calculate the radiative recombination rate is to sum over the quantum-mechanical probabilities of photon emission (Varshni 1967). This yields an expl icit temperature dependence of the recom bination coeff cient of nondegenerate carriers. In a three-dimensional crystal,


whereas in a two-dimensional structure (quantum well; see Section 4.1.4),


The emission rate and carrier lifetime determined by the band-to-band radiative process depend on the type of injection and the injection level [see Eq. (4.1.6)]. For excess minority carriers (such as electrons injected into the



p-region of a p-n junction LED), the recombination is called monomolecular. The radiative lifetime is inversely proportional to the equilibrium density of the majority carriers (in this case, holes)

I 're == 6.n/ R, ==--.




When both types of carriers are injected into the same region of the lightemitting structure (e.g., in a double-heterostructure LED; see Section 4.1.4), Sn= 6.p and the recombination is called bimolecular. For low injection levels

( !1n < 110 + Po ),



For high injection levels (6.n > flO + Po), the lifetime depends on the excess carrier densi ty:


Tre '" 'rh ==--.




In the case of radiative transitions via impurity levels (see Fig. 4.1.2), the rate of the radiative recombination depends on properties of the luminescent centers that are introduced by doping. Usually, transitions between the conduction band and acceptor impurity levels [e.g., zinc (Zn) in group IIJ-V materials] are employed (FiK 4.1.2a). Provided that the rate of the hole capture is high enough, the electron radiative lifetime can be roughly estimated as


, ""~---

re BN A '


where N A is the concentrati on of the I uminescent centers.

Some of the carriers injected into the active region can recombine nonradiatively, thus reducing the internal quantum efficiency. An intrinsic nonradiative recombination route is the Auger process (Landsberg 1970). Auger recombination is a process reciprocal to the impact ionization. In the act of Auger recombination, the energy released by the electron-hole pair is absorbed by a majority carrier (i.e., an electron in n-type material and a hole in p-type material). In wide-bandgap semiconductors emitting visible light, the Auger process is relatively inefficient and non radiative recombination is due primarily to the deep impurity levels.

Typically, nonradiative capture of free carriers by deep levels involves the excitation of a deep center to a high vibrational state with subsequent relaxation via multiphonon emission (Henry and Lang 1977). The process illustrated in the configuration diagram (Fig. 4.1.4) proceeds in a way similar to that described in Sec-



tion 3.3. The potential curve C depicts the initial state of a free electron-hole pair and the vibrational state of the center. The potential curve T is the same diagram but with one of the carriers, say electron, captured. The transition from state C to state T can occur in the vicinity of the intersection of the potential curves. This means that before the capture, the empty center must have some vibrational energy, We' acquired from phonons, and after the capture, the center experiences the transition into a highly excited vibrational state that relaxes by multiple phonon emission. The second stage of the recombination is the capture of the hole (i.e., transition from the state T to the final state V). The latter process requires surmounting the barrier Wh. Eventually, the entire energy of the electron-hole pair

( "" E g ) is converted to lattice vibrations.

The nonradiatiue lifetime of the electron-hole pair determined by capture by a deep center via multi phonon emission is


where TOe and TOh are the electron and hole lifetimes in the infinite-temperature limit, respectively. Typical barrier heights are of the order of 0.1 eV and depend on the origin of the nonradiative center. It should be noted that exponential decrease

FIG. 4.1.4.

Configuration coordinate diagram of a deep center of nonradiative recombination. Potential curves C, T, and V correspond to the electron in the conduction hand, at the trap, and in the valence band, respectively.



of the nonradiative lifetime with temperature occurs under condition kBT > ftwO' where hwo is the optical phonon energy. At low temperatures (kBT < "wo), surmounting the barriers becomes improbable, and transitions between the potential curves are more likely to occur because of the tunneling in the configuration space (horizontal arrow in Fig. 4.1.4). This yields a temperature-independent nonradialive lifetime

Tnr == const.


For localized carriers, the nonradiative capture should be preceded by a supplementary process that brings the carrier into the proximity of the nonradiative center. Usually, tunneling in the real space (Street 1976) is involved. In the hightemperature limit, the relevant nonradiative lifetime is

Tnr oc exp(-T/ToL


where TO is a characteristic temperature of the process.

The overall lifetime of the excess carriers, r , is determined by both radiative and nonradiative recombination processes:



T Tr T nr


The internal quantum efficiency can be presented as the fraction of the radiative recombination rate in the total recombination rate:


In some important materials systems, such as AIGaAs and AIGaInP (see Section 4.2), a substantial fraction of the conduction electrons may reside in indirect valleys X and L that are close in energy to the direct valley F. Since the predominant recombination of electrons in the indirect valleys is nonradiative, this causes an additional decrease in the internal quantum efficiency (Archer 1972). The internal radiative efficiency that accounts for this effect is (Steranka et at, 1995)



where mr, mx ,and mL are the density-of-states effective masses for the r, X, and L valleys, respectively, and t..Exr and Mu are the energy differences between the X and rand Land I" valleys, respectively.

Typically, the internal quantum efficiency is a decreasing function of temperature, since the nonradiative and radiative lifetimes decrease and increase with temperature, respectively.

The overall lifetime determines the characteristic response time of an LED.

The LED cutoff frequency is given by

I !r=--.



Typical lifetimes in direct-gap materials used in high-brightness LEOs are on the nanosecond seal e. Such LEOs can operate at frequencies of hundreds of megahertz.

4.1.3. Injection in a ~n Junction LED

The basic element of an LED is a semiconductor electroluminescent structure that comprises, at least, a region of radiative recombination and regions of different conductivity type (p and n) that supply the recombining carriers. In the simplest design, the structure relies on a junction between a p-type semiconductor and an n-type semiconductor of the same kind (p-n homo junction) with one or both conductivity regions employed as the radiative-recombination region or regions. Although this LED design is somewhat obsolete and not used in modem highbrightness LEOs, it is considered here to illustrate the basic principles of the carrier injection in an electroluminescent device.

Figure 4. I .5a depicts a band diagram of a Jrn homojunction. Under zero bias (Fig. 4.I.5a), the majority electrons from the n-region diffuse into the p-region and the majority holes diffuse in the opposite direction. This process creates depleted regions on both sides of the interface. The space charge of the depleted regions creates an internal electric field that counteracts the diffusion. In equilibrium, when the potential barrier is somewhat smaller than the bandgap energy, the diffusion current is counterbal anced by the reverse current of rninori ty carri ers that dri ft in this internal electric field.

The reverse currents for minority electrons and holes, respectively, are


(4. !.20b)





FIG.4.1.5. Band diagram of a p-n homojunction LED: (a) zero-biased junction; (b)junction with voltage VappJied in the forward direction. Solid vertical arrows, radiative transitions; dotted arrows, nonradiative transitions.

where Dn and Dp are the diffusion coefficients and Ln = (DnTn)1I2 and L p '" (D p r p)'/2 are the diffusion lengths of electrons and holes, respectively, N A and N D are the acceptor and donor concentrations in the p- and n-regions, respectively, and Sis the effective device cross section. Equations (4.1.20) are valid only for a low injection regime and for homojunction diodes with long neutral regions (much longer than the diffusion lengths ofthe minority carriers).

When a voltage V is applied in the forward direction (Fig. 4.i.5b), the reverse currents of the minority carriers change negligibly. Meanwhile, the barrier for majority carriers decreases by qV Consequently, the majority-carrier diffusion current increases by a factor of exp(qV /kBT). Enhancement of the diffusion due to the electric field, injection, results in an excess density of minority carriers on both sides of the j unction. The injected carriers recombine both radi atively and nonradiatively (solid and dotted arrows in Fig. 4. I .5b, respectively).

The net injection currents are

(4. 1.21 a)


(4.1.21 b)

for electrons and hoi es, respectively. Here, V T '" k B T / q .



In real LEOs, the decrease in the potential barrier by the applied voltage is somewhat smaller, because some voltage drops across the LED series resistance, Rs' Also, an additional current is caused by nonradiative surface recombination at the junction perimeter (Henry et at. 1978; see Fig. 4.1.5b). Therefore, the forward LED current-voltage (I-V) characteristic is described by a nonideal diode equation,

( 4.1.22)

where I nrO is the reverse nonradiative recombination current and r is the ideality factor for the recombination current (1 s r :S 2 ).

For an LED with the radiative recombination taking place in both the p- and n-regions, the injection efficiency is


Typically, the luminescent centers are introduced in only one region (say, in the p-region) and the injection to the other region is minimized using an appropriate doping profile ( N D » N A ). In this case,

In lJillj == l'


In practical homojunction LEDs, the injection efficiency, 1}inj, ranges from 0.3 to 0.8 (Bergh and Dean 1976).

4.1.4. Heterostructures and Quantum Wells

Conventional p-n diodes utilize doping profiles to co ntro I carrier injection. The potential barriers for electrons and holes are created by the charges of ionized donors and acceptors in the depletion region near the boundary between the n- and ptype semiconductors. LEOs based on p-n homojunctions have important shortcomings that limit their application in solid-state lighting. First, the light generated in the active region is reabsorbed, to a considerable extent, in the conductive regions. This reduces the light-extraction efficiency (see Section 5.1.3). Second, since high internal quantum efficiency is attainable in only one conductive region (typically, the p-region), a low injection level of holes into the n-region is required (I n » I p ). Equations (4.1.20) imply that this problem could be solved by using a



highly asymmetric doping profile, N D » N A' However, high doping levels result in increased reabsorption. Also, extensive doping can lead to the formation of undesirable impurity complexes that serve as additional nonradiative recombination centers. Finally, the radiative recombination in such LEOs is monomolecular, so that only increasing the doping level can increase the emission rate.

Changing the material composition as a function of distance provides new opportunities for governing injection, radiative recombination, and reabsorption. A change in the material composition results in a change in the energy gap that enables one to modify the potentia! profile. Structures composed of semiconductors (hat have different bandgaps due to different chemical composition are called heterostructures. By using heterostructures, improved injection and internal quantum efficiencies are attained in modem high-brightness LEOs. A special case of heterostructures is quantum wells, in which additional benefits are reaped due to the quantum confinement of carriers.

An example of such bandgap engineering is an LED based on a single heterostructure (SH; also called p-n heterojunction) (Casey and Panish 1978a). The potential profile of an SH LED is shown in Fig. 4.1.6. The p-type conductive region is made of a semiconductor with bandgap E g2 which is narrower than that

of the n-type region, E gl . The band discontinuity increases the potential barrier for holes that diffuse to the n-type region by the valence-band offset, Mv. Depending on the abruptness of the interface, the potential barrier for electrons can decrease by a value ranging from 0 to Me (Kressel and Butler 1977,

Grinberg et at. 1984). As the result, the ratio of the injection currents / nil p increases by a factor proportional to exp(I1E/kBT), where MV :5 11£ :S Mv + Me . Another important advantage is that the n-type region is transparent to photons generated in the p-type region. This minimizes reabsorption of light propagating toward the n-!ype pol e of the structure.


P Eg2


FIG. 4.1.6. Potential profile in e p-n heterojunction LED.



Practical high-brightness LEOs use a double heterostructure (OH), which implements the benefits offered by bandgap engineering to a larger extent (Kroemer 1963). The potential profile of a DH LED is shown in Fig. 4.1.7. The structure comprises a narrow-gap active p-type layer sandwiched between wide-gap conductive regions of n- and p-types, respectively. This allows for bidirectional inj ection of excess carriers into the active layer, where electrons and holes recombine. Moreover, the minority carriers that diffuse through one of the heterointerfaces are trapped in the active layer by the second heterointerface and cannot diffuse away. This increases the excess carrier density and, consequently, the rate of radiative recombination. Also, in such a structure, both conductive layers are transparent to the light emitted, and the reabsorption effect is minimized for both directions. However, reabsorption is stilt present in the active layer.

It should be noted that both SH and DH LED structures require materials with good lattice matching. If lattice constants differ too much, a high density of defects (typically, threading dislocations) is produced at the heterointerfaces, giving rise to nonradiative recombination. The requirement of lattice matching imposes severe limitations for the materials systems used in heterojunction LEDs (see Section 4.2).

Thinning the active layer is the way to further increase the rate of radiative recombination and reduce the reabsorption. In addition, using very thin active layers enables one to overcome some lattice-matching problems, since such layers are able to conform to the thick confining layers without defect generation (pseudomorphtc layers). However, when the thickness of the active layer becomes comparable or smaller than the de Broglie wavelength of electrons in the crystal, the energy spectrum of the carriers is modified (quantum confinement). Such double heterostructures are called quantum well (QW) structures. Single quantum wells (SQWs) and multiple quantum wells (MQWs) offer the most versatile structures for high-brightness LEDs.

P Eg1
n FIG. 4.1.7. Potential profile in a double-hetcrojunction LED.



The optical properties of QWs di ffer from those of the bulk material in that instead of free motion along the direction perpendicular to the heterointerface (x), discrete energy levels, E", occur. For an infinitely deep rectangular quantum well,

these levels split from the bottom of the conduction band, Ec , by


where 11 = I, 2. 3, ... (any positive integer number) is the quantum number, a is the width of the quantum well, and me is the electron effective mass. 1J1 the y-.: plane, which is parallel to the heterointerface, the electronic motion is not quantized, so that the electron energy within a subband n is given by

( 4.1.26)

where k is the two-dimensional wavenumber. The electron wavefunction can be presented as

(7)l/2 ( J

<D(x,y,z)= ~ sin: x exp(iky)exp(ilc).


In real device structures, the potential well is never infinitely deep. For a finite symmetrical potential well, the electron energy is given by (see Landau and Lifshits 1965)


where the value of gil is found from solution of the following equation


Here Uo is the depth of the well. The same considerations are valid for holes with the effective mass mh .

Figure 4.l.8 depicts a band diagram of QW structure composed of a thin layer of a semiconductor with the bandgap energy E g2 sandwiched between thick clad-

ding layers of a semiconductor with bandgap energy Egl. Typically, the electron mass is much smaller than the hole mass, and the energy distance between the energy levels for electrons is more pronounced.



u1 n=3


Egl Eg2
/ -
<, n=3 FIG. 4.1.8. Band alignment diagram of a quantum well structure. Bold arrows indicate optical transitions between quantum-confined electron and hole states.

It should be noted that the solution given by Eqs. (4.1.28) and (4.1.29) is valid when the effective mass, me' is the same for wide- and narrow-gap semiconductor layers comprising the QW structure. In reality, the effective mass in the cladding layers is usually larger. The space dependence of the effective mass in heterostructures results in even more complex descriptions of the energy spectrum (see, e.g., Dyakonov and Shur 1998 and references therein). In this case, the effective depth of the quantum well (and, hence, the position of the subbands and the degree of the electron localization within the quantum well) depends on the wavenumber, k (i.e., on the kinetic energy of the electron motion along the heterointerface).

Intersubband optical transitions occur between the electron and hole levels with the same quantum numbers (ne == n~; see Fig. 4.1.8). A strong overlap of electron and hole wavefunctions in a QW leads to a highly efficient bimolecular recom binati on. Hence, dopi ng of the acti ve region is not req ui red. Also, the intrinsic recombination yields emission lines with smaller line width than recombination via impurity levels. Furthermore, the exciton binding energy is much larger in quantum wells than in bulk crystals (Dingle et al. 1974, Bastard et al. 1982), and in many materials, two-dimensional-exciton optical transitions enhanced by confinement-reduced electron-hole distance (Weisbuch et al. 1981, Deveaud et al. 1991) can be observed even at room temperature. Also, a characteristic feature of QWs is an energy-independent density of states in the subbands (in bulk crystals, the density of states equals zero at the band extrema and increases with energy as

£1/2). A constant density of states results in a weaker temperature dependence of the radiative recombination coefficient [see Eqs. (4.1.8)], and the peak wavelength of the emission equals the intersubband separation and is nearly temperature independent.



FIC. 4.1.9. Potential profile in a quantum well structure with applied electric Held. Electron and hole states reside within triangular wells and the wavefunctions arc spatially separated.

When an electric field exists in a QW, the bands are tilted (Fig. 4.1.9). The tilt can be due to an external bias or can be caused by the piezoelectric effect in a strained structure (Smith 1986) and/or by spontaneous polarization in a noncubic lattice (Bykhovski et a1. 1993, Fiorentini et al. 1999). All these effects are relevant to QWs employed in LEDs. The energy levels in an infinitely deep triangular potential well are given by (Kroemer 1994)

_ [ n2jl/3(3l«JFS )2/3( 3)2/3

En - -- --- n+-

2me 2 4


where n is the quantum number, qFs is the potential slope, and Fs is the surface field.

One of the most important consequences of the applied or built-in electric field is the spatial separation of the electron and hole, shown in Fig. 4.1.9. This may result in a reduced rate of radiative recombination because of a reduced wavefunction overlap, and in a red shift of the emission line (quantum-confined Stark effect; Miller et al. I 984a).

Light-emitting structures based on QWs require optimization of the injection efficiency. The drawback is that in materials used for LED fabrication, electron mobil ity is very high and electrons can leak into the p-type cladding layer wi thout being captured by the QW states. The leakage is less probable for holes, which have much less mobility. Usually, to prevent leakage of electrons into the p-type conductive layer and thus to improve injection efficiency, an electron blocking layer (EBL) made of a wider-gap p-material (Eg3 > £g2 > Eg1) is introduced between the QW and the p-conductive layer (Fig. 4. 1.10).







FIG. 4.1.10. Band alignment diagram of an electroluminescent structure based on a quantum well w ith an electron blocking layer (asymmetric QW).

The active layer shown in Fig. 4.1.10 is an asymmetric QW with the rightand left-hand barriers V, and U2, respectively. The electron energy levels in an

asymmetric QW are given by Eq. (4.1.28), where, however, the values of q" are now given by (Landau and Lifshits 1965)


The forward current-voltage characteristics of DR and QW LEOs are more complex than that of a p-n LED [Eq. (4.1.22)]. First, the diffusion current can no longer be described in terms of the diffusion length [Eqs. (4.1.20)] since most such LEOs operate in a high injection regime and since the typical thickness of the active layer is much smaller than the diffusion length. Second, in OH and QW structures, the recombination current is determined by both radiative and nonradiative transitions in the active layer and by nonradiative losses at the heterointerfaces. Third, in OH and QW LEOs with an active layer containing deep levels, the tunnel current from the cladding layers into the deep levels can be significant (Casey et at. 1996, Perlin et al. 1996). Generally, the forward current-voltage characteristics can be described by three terms

where the diffusion current is





the recombination (both radiative and nonradiative) current is


and the temperature-independent tunnel current is

[ (V - JR., J ]

I, = '10 e.xp~. - I .

( 4.1.33c)

Here E, is a characteristic energy constant with the magnitude - 0.1 eV (Perlin et al. 1996). The reverse currents I DO, I RO, and '10 have no general explicit form and are usually introduced as empirical fitting parameters.


4.2.1. Requirements

In this section, semiconductor materials systems that are used for fabrication of high-brightness LEOs are discussed. The requirements imposed on these materials by solid-state lighting applications are more stringent than those for conventional LEDs.

First, a universal requirement is matching of the semiconductor bandgap energy with visible and near-UV photon energies. The energy and wavelength (in air) of a photon are related as

r (

1239.5 A(llm)=-(~).

hv eV


However, a high rate of radiative recombination can be achieved only in semiconductors with a direct gap. The bandgap energies of some semiconductors are compared with the spectrum of relative eye sensitivity in Fig. 4.2.1. Elementary semiconductors such as Ge (germanium), Si (silicon), and C (diamond) are indirect and do not match the spectral region required. SiC polytypes (group !V-!V) and some group Ill-V materials (AISb, AlAs, and AlP) can emit in the visible region; however, they are not suitable for solid-state Jamps because of the indirect bandgap.

Another requirement is lattice robustness and high resistance in respect to the formation of nonradiative recombination centers. This requirement is the main issue that makes the solid-state lighting use of the entire family of IJ-Vl binary





SiC(4Hl SiC(6H

ZnSe CdS AlP CdO SiC(3C) GaP ZnTe AlAs InN edSe AISb CdTe GaAs InP


GaSb Ge

InAs InSb



Wavelength (nm) 500 400 300



~ \

~ -; 10-5

L-~~~~ __ ~~-L~~~L-~~~~

o 1 2 3 4 5 6 7 Bandgap Energy (eV)

FIG.4.2.1. Energy gaps of elementary and binary semiconductor materials compared with the spectral sensitivity of the human eye. Gray bars represent indirectgap semiconductors. (After ShUT 1996.)

compounds doubtful. For many years, they were considered as the most natural candidates for light-emitting devices in the visible and near-UV region: CdSe (red), ZnTe and CdO (green), CdS (blue-green), ZnSe (blue), and ZnO and ZnS (near UV). Despite some encouraging results, mostly for ZnSe-based materials (see, e.g., Nakayama et al. 1996), low thresholds of defect formation seem to be an inherent property of II-VI compounds. Nevertheless, searching for possible solutions of the problem (e.g., the introduction of beryllium (Be) in order to improve lattice rigidity; Waag et al. 1997) continues.

The other requirements are the possibility of tailoring the bandgap by alloying, the availability of both p- and n-type, and the possibility of fabricating heterostructures with a predetermined potential profile.

Direct-gap binary group J][-V compounds (lnP, GaAs, InN, GaN, and AIN) have bandgaps that overlap the required spectral region, and these materials can form robust direct-gap alloys even with the indirect-gap binary group III-V com" pounds. Ternary and quaternary alloys that contain a mixture of aluminum (AI), gallium (Ga), and indium (In) cations and one of arsenic (As), phosphorus (P), or nitrogen (N) anions are the bases of the present high-brightness LED industry. The three relevant systems are Al GaAs (In is not requi red here, since it shi fts the



emission to infrared), AIGalnP, and AlInGaN. Mature growth and fabrication technologies enable one to produce all these alloys within wide ranges of composition and with both P: and n-type doping. Also, heterostructures that are required for high levels of carrier injection and efficient radiative recombination can be fabricated for each system. Below, the most relevant properties of AIGaAs, AIGaJnP, and AlInGaN systems are presented.

Here we do not discuss a rapidly evolving field of organic small-molecule and polymer-based LEDs (OLEOs). Despite a lot of advantages (cost-efficiency, mechanical flexibility, large area), to date, inherent reliability issues restrict OLEO applications to specific niches (e.g., indoor displays and LCD backlighting; for a recent review of OLEDs, see, e.g., Sato 2000).

4.2.2. AIGaAs Materials System

GaAs. AlAs, and their alloys crystallize in a cubic (zinc blende) lattice. Each atom of group III is tetrahedrally bonded to four closest neighbors of group V (and vice versa), resulting in a close-packed crystal structure. The essential feature of the crystal structure of this ternary system is almost ideal lattice match of alloys in the entire range of Al molar fraction x. The lattice constant of AlxGa l-xAs changes with composition in accordance with Vegard's law (i.e., linearly; see Levinshtein et al. 1999):

a(nm) '" 0.56533 + 0.00078x .

( 4.2.2)

Owing to such excellent lattice matching, AlxGal-xAs layers with different composition can be grown epitaxially on GaAs substrates in any sequence with very few defects at heterointerfaces, This was the reason that AIGaAs was the first widely used semiconductor heterostructure system (Kressel and Butler 1977, Casey and Panish 1978a,b). Consequently, AIGaAs heterostructure red LEDs were the first LEDs with the luminous efficiency exceeding that of filtered incandescent light bulbs (see Sections 4.3.1 and 5. 2.1).

The physical properties of AIGaAs system are well known (see, e.g., Casey and Panish 1978a, Adachi 1994, Steranka 1997, Levinshtein et al. 1999). The energy gap varies between 1.424 eV (for GaAs) and 2.168 eV (for AlAs). For x < 0.45, AlxGal_xAs has a direct gap:

Er(eV) == 1.424 + 1.247x.


For 0.45 < x < I , the energy gap is indirect:

EX (eV) =1.9+0.125x+0.143x2.


The crossover from direct-gap material to indirect-gap material occurs at a bandgap energy of 1.985 eV, which corresponds to the emission wavelength of 624 nm.



In pure GaAs, the internal quantum efficiency may be in excess of 99% (Schnitzer et al. 1993a). However, approaching the point of direct-to-indirect gap crossover, the internal quantum efficiency decreases rapidly because of increased occupation of indirect valleys [see Eq. (4.1.18)]. Because of this factor and with respect to the eye-sensitivity spectrum, the optimal luminous efficiency is achieved in the red region around 640 to 660 nm for x "" 0.35 to 0.4. In this wavelength region, the internal quantum efficiency is around 50% (Nishizawa et at. 1983).

For x < 0.41, the electron effective mass varies with composition as

mr = (0.063 + 0.083x )mo '

( 4.2.5)

where mo is the free electron mass. The heavy and light hole effective masses are given by Levinshtein et al. (1999):

mil" =(0.51+0.25x)mo, mil, = (0.082 + 0.068x )mo



The direct-direct bandgap discontinuity at the heterointerface is distributed between the conduction and valence bands with a ratio close to 60:40 (Miller et at. I 984b, Watanabe et al. 1985). For direct-indirect band discontinuity, this ratio shifts in favor of the valence band. This property is important for the prevention of the hole injection across the heterointerface (Steranka 1997; see Fig. 4.1.6).

Typical donors used in the n-conductive layers of A1GaAs LEOs are tin (Sn) and tellurium (Te). Zn or magnesium (Mg) are used as acceptors in the p-conducting layers; also, Zn is used to dope the active layer to invoke emission due to the radiative transitions from the conduction band to the acceptor levels.

The radiative recombination coefficient B is 1.3 x 10-10 cm3/s at room temperature (see Steranka et al. 1995 and references therein). The nonradiative losses are due mostly to penetration of electrons into the indirect X valley, with subsequent recombination via deep centers. The nonradiative lifetime has typical values of around 100 ns. However, it does not exhibit the exponential dependence on temperature, typical for multi phonon processes [see Eq. (4.1.13)], and the nature of the nonradiative traps is not completely understood (Steranka et al. 1995).

Aspnes et al. (1986) performed a detailed investigation of optical properties of AlxGal_xAs alloys with compositions ranging from x = 0.00 to x == 0.80 in steps of 0.1. Typical values of the refractive index for x = 0.3 to 0.4 at the photon energies corresponding to bandgap energies are around 3.6. In the relevant transparency region, the refractive index of the cladding materials with x = 0.6 to 0.7 is in the interval between 3.3 and 3.4. The absorption coefficient at the edges of the direct and indirect bandgaps is around 1 x 104 ern" 1 and ~ 1 0 ern - I. respective Iy.

An advantage of AIGaAs materials system is the low-cost production technology using liquid-phase epitaxy (LPE; see Section 4.2.5). However, this technology is unable to assure sufficient oxygen (0) decontamination. Also, a serious issue of the AlGaAs produced by LPE is an environmental degradation of alloys that in-



creases with increasing Al content because of the tendency of AI to form oxygenrich compounds through hydrolysis (Dallesasse et at. 1990). This imposes restrictions on high-temperature, high-humidity operation of AIGaAs LEOs (Steranka 1997). One of the solutions for this problem is the formation of a stable passivating oxide film (Richard et al. 1995).

4.2.3. AIGalnP Materials System

Group III-P materials (also having zinc blende cubic lattice) exhibit larger bandgap energies than group HI-As materials do and can be used in electroluminescent devices for visible spectral regions with shorter wavelengths than red. However, both AlP and GaP and their ternary alloys have an indirect gap. Fortunately, by alloying of AlP and GaP with the direct-gap InP, direct-gap quaternary AIGalnP crystals can be produced. The unique feature of AIGalnP materials system is the availability of (AlxGal_x)ylnl_yP alloy, which is nearly perfectly lattice matched to GaAs at y - 0.5 and has a very close coefficient of thermal expansion (Casey and Panis h 1 978b, Stri ngfell ow 1978). By changi ng the molar fraction of A I, x, the direct-gap energy can be varied within the red-to-green portion of the visible spectrum (see Fig. 4.2.2). To date, high-quality (AlxGal-x)osInosP films grown on GaAs substrates are one of the most important heterostructure materials systems for solid-state lighting.

Under certain conditions of epitaxial growth, AIGaJnP shows a tendency for atomic ordering (formation of a sequence of column III atomic planes; Gomyo et al. 1987) that is undesirable for LED fabrication (e.g., the bandgap en-

2.4 :§_ 2.2


1? 2.0 (ll


w 1.B a.


.g 1.6

c ro

III 1.4

·600 s:
700 (jj
5.9 5.4





FIG. 4.2.2.

Lattice Constant (A)

Bandgap energy vs. lattice constant in the AIGalnP system. Solid dots and solid lines represent direct band gaps; open dots and das-hed lines represent indirect bandgaps. (After Chen et al. 1997.)



ergy can be reduced). Therefore, most investigations of the A1GainP system refer to crystals with disordered sublattices, However, data dealing with accurate determination of the bandgap energy and direct-to-indi rect crossover point are still controversial. First estimates of the bandgap energy of disordered

(Al-Ga l-x)oslno.sP were based on the low-temperature photoluminescence spectra (Cao et al. 1990, Prins et al. 1995). More accurate data obtained at room temperature by spectroscopic ellipsometry (Schubert et al. 1999) indicate that the direct gap of the disordered AIGalnP lattice matched to OaAs varies with composition as

Edev) =1.899+0.543x+0.12x2,


whereas the indirect gap is (Ozaki et al. 1996)

EX (eV) = 2.20 + 0.16x .


Equations (4.2.8) and (4.2.9) imply that crossover from the direct to the indirect gap occurs at x - 0.65, corresponding to a bandgap energy of 2.30 eV. Hence, emission in the range 656 (red) to ".540 nm (green) can be obtained. Other estimates yield a crossover x value of 0.53 and a critical wavelength of 555 nm (Kish and Fl etcher 1997).

Ozaki et al. (1996) derived a precise condition for lattice matching between

(AlxGal-x)ylnl-yP and GaAs:

y = 0.516/{1- O.027x).


For direct-gap alloys, Eq. (4.2.10) yields y values from 0.516 and 0.525, depending on AI/Oa ratio.

The effective mass in the lowest (I") valley of the conduction band is

mr = (0.11 + 0.00915x - 0.0024x2 )mo.


The heavy and light hole effective masses are given by


mlh = (0.1 I + 0.03x )mo


(see Kish and Fletcher 1997).

Kish and Fletcher (1997) averaged data on the conduction- and valence-band

offsets in (Al-Ga, _x)oslnosP obtained by several groups. The results for the conduction-band discontinuity are given by

( ) {0.369X 6E eV -

C - 0.285-0.157x

for x:O; 0.53, for x> 0.53.




The valence-band discontinuity is given by

Mv{eV)= 0.24Ix.


AIGalnP doping issues were reviewed by Chen et al. (1997). Typically, n-type doping is easily performed by using Te or Si as donors, Typical acceptors for p-type doping are Zn and Mg. However, p-type doping is associated with some difficulties that become more pronounced with increasing Al content. First, the difficulty inherent in wider-bandgap materials is the increasing impurity ionization energy. Acceptors, which have ionization energies higher than those of donors, are influenced by this effect to a greater extent. As a result, high hole densities become harder to achieve because of a low level of acceptor ionization. The second difficulty follows from the tendency of acceptor impurities to compensation. Oxygen, which is a typical contaminant for Al-containing compounds, generates deep levels that compensate shallow acceptors. Acceptor compensation is also characteristic of hydrogen that is unintentionally incorporated during the growth process (see Section 4.2.5). Fortunately, the reduction in background oxygen in the growth process and postgrowth annealing offer a fairly good technological remedy for the compensation.

The luminescence intensity drops down rapidly for wavelengths shorter than 590 nm (Chen et al. 1997), which is the inherent consequence of the electron population in the indirect band minima (Patel et al. 1999). At room temperature, the nonradiative recombination in AIGaInP heterostructures is probably due to capture of carriers at the heterointerfaces with typical nonradiative lifetimes on the order of IOns (Domen et al. 1994). The estimated value of the radiative recombi-

. ffici I ~ - 10 3/ f (K I

nanon coe icient IS . .J X 10 cm s or x == 0 llll1 et or 1997).

The optical properties of AIGalnP materials system are well known. The refractive index of (AlxOaj -x)05InOSP is around 3.6 at the photon energies correspondi ng to transiti ons between the extrema of the conducti on and valence bands. In the transparent region of cladding layers, the refractive index varies from 3.2 to 3.5, depending on the wavelength and on the Al molar fraction x (Adachi et al. 1994, Schubert et al. 1999). The absorption coefficient is around 5 x 104 cm -I in the relevant opaque region (Adachi 1999).

In contrast to AIGaAs, AIGalnP material cannot be grown by LPE. To date, the mature technology for growth of this quaternary compound is metalorganic chemical vapor deposition [MOCVD, also called organometallic vapor-phase epitaxy (OMVPE); see Section 4.2.5]. This method allows one to maintain a high level of compositional control and minimizes contamination.

4.2.4. AlinGaN Materials System

For many decades, application of group III nitride semiconductors was hindered by the absence of appropriate substrates and by the unavailability of p-type material. In the last decade of the twentieth century, the All nGaN materials system un-



derwent a phenomenal evolution that resulted in the creation of high-brightness LEDs in the blue, green, yellow, and near-UV region (Nakamura and Fasol 1997). A breakthrough in group III nitride materials fabrication is of crucial importance for solid-state lighting technology, since it resulted in producing efficient electroluminescent devices over the entire visible and even near-Uv spectrum. The properties of AlTnGaN materials system are described in numerous books and review articles (see, e.g., Nakamura and FasoI 1997, Pankove and Moustakas 1998, Gil 1998, Pearton et at. 1999, lain et at. 2000, Nakamura and Chichi bu 2000, Levinshtein et al. 2001). However, contrary to AIGaAs and AIGaInP materials systems, AlinGaN is a more intricate material, which is less understood. The properties of this materials system that are the most relevant to high-brightness LED technology are discussed briefly below.

The thermodynamically stable phase of binary compounds InN, GaN, AIN and their ternary and quaternary alloys is the wurtzite structure, although zinc-blende and even rock-salt structures might be produced under certain growth conditions. Similar to the zinc-blende structure, wurtzite crystals have each atom tetrahedrally bonded to its four nearest neighbors. However, the relative orientation of penetrating tetrahedrons is different, and the wurtzite unit cell has hexagonal symmetry with two lattice parameters, a (perpendicular to the optical axis) and c (along the optical axis). The physical properties of wurtzite-type cryst.als parallel to the c axis differ from those perpendicular to the c axis (see Table 4.2.1). Typically, epitaxial growth is parallel (or at a small angle) to the c axis.

The most striking achievement of group III nitride technology is that materials grown on lattice-mismatched substrates and, therefore, containing a high density of threading dislocations can still exhibit high internal quantum efficiency. Epitaxial growth of high-quality GaN over a lattice-mismatched substrate was facilitated by introduction of low-temperature-grown buffer layers of AIN and AIGaN by Yoshidaetal. (1983) and Amanoetal. (1986) and GaN by Nakamura (1991). The most extensively used substrate is sapphire (AI20]). The period of the hex-

agonal O-sublattice of sapphire (as/ J3 ) that has an epitaxial relationship with the III plane of a nitride (Dovidenko el al. 1996, Nakamura and Fasol 1997) differs from the lattice constant a of GaN by - 16%. Another widely used substrate, 6H-SiC, and a potential substrate candidate, ZnO, are mismatched from GaN by 3.5% and about 2%, respectively (Lin et al. 1993, Morkoc et al. 1994). Other substrates used for these LEOs include spi nel (Khan et al. 1997) and silicon (Yang et al. 2000). Figure 4.2.3 depicts the bandgap energy vs, lattice constant diagram for AlinGaN system along with data for the lattice constants of sapphire, 6H-SiC, and ZnO.

The lattice constants 1 (f = a or I = c) of an AlxlnyGal_x_yN alloy are given by Vegard's law:

I = xl AIN + yi'nN + (1- x - Y)/GaN '


where lAIN, lInN, and 'GaN are the relevant lattice constants of AIN, InN, and GaN, respectively (see Table 4.2.1). However, because of the growth over lattice-



6 lAIN]' 200
:> E
~5 m- E-
>- 0 ..c
'- 300 rn
(].I 4 (II
c: ::0 c:
V> Qj
a. Q. 400 >
ro 3 ro
OJ !!? IlnNI 5:
"0 , 500
c: :E
ro ,
2 a.. 0:
In 0.. 800
(II c:,
V> N,
2.6 2.8 3.0 3.2 3.4 3.6 3.8 4.0
Lattice Constant a (A) FIG. 4.2.3. Bandgap energy vs. lattice constant in the AllnGaN system Dashed lines show lattice constants of the most appropriate substrates.

mismatched substrates and post growth cooling, the layers are atTected by in-plain stress. Also, hydrostatic stress can be introduced by the presence of point defects (Kisielowski et al. 1996, Gorgens et al. 2000). In heterostructures, additional inplain stress occurs in thin layers that are pseudomorphically accommodated to cladding layers. The stress results in strain (i.e., in the altered lattice constants).

The binary, ternary, and quaternary compounds of the AllnGaN materials system have direct bandgaps in the entire range of molar fractions. The indirect mini rna reside well above the direct mi ni ma (>0.7 e V; Levi nshtei n et al. 200 1) and, in contrast to AIGaAs and AIGalnP crystals, are occupied negligibly in the reasonable range of temperatures.

Bandgap energies and carrier effective masses of binary group III nitrides are presented in Table 4.2.1. For most parameters, linear interpolations between these values similar to that given by Eq. (4.2.16) are used for crude estimates in AlTnGaN system-based ternary and quaternary compounds. However, usually, linear interpolation is insufficiently accurate for the bandgap energy. For instance, the bandgap ofa ternary alloy composed of A and B binary compounds is given by


where bAB is the bowing parameter (-1 eV). It should be noted that the values of the endpoint bandgap energies depend on strain (Kisielowski el al. 1996). Strain, as well as composition inhomogeneity, is the probable reason for large discrepan-

cies in determining the values of the bowing parameters (for In,.Gal-xN alloy, see, e.g., Teles et al. 200 I and references therein).



Table 4.2.1. Basic parameters of InN, GaN, and AIN at 300 K

Parameter Units GaN AIN InN
Lattice constant. c A 5.186 4.982 5.693
Lattice constant, a A 3.189 3.112 3.533
Bandgap energy, E g cV 3.439" 6.2 \.97
Effective electron mass. m" rno o .19b(lll o .33c(ll) 0.11 b(lD
o d(J_) 0.25"(1) O.IOb (J_ )
Effective heavy hole mass, nlhh frio 1.76" (Ill 3.53" (II) 1 .56b (II)
1 .61 C ( J_ ) 10.42" l J_) 1.68b (J. )
Effective light hole mass, nllh frio 1.76 (Ii) 3.53 (II) 1.56 (II)
0.14 (J.) 0.24 ( J_ ) 0.11 (J.)
Piezoelectric constant, e31 elm 2 -0.33 -0.48 -0.57
P iczoelectric constant, e33 elm 2 0.65 1.55 097
Spontaneous polarization. PI d elm 2 -0.029 -0.081 -0.032
Radiative recombination cm3/s 4.7xlO-11 1.8x 10-11 5.2xIO-11
Refractive index at 555 nm 2.4 2.1 2.8
Absorption coefficient at the 105 cm-I 1 3 0.4
photon energy h v '" Eg Unless otherwise ind icated, data is from l.evinshtein et al. 2001

a Derived from the exciton position in unstrained GaN (Korona et at. 1996). byco et al. (1998)

C Suzuki et al. ( 1995).

d Bernard in i et al. (1997)

e Drrutricv and Oruzheinikov (1999)

Martin et al. (1996) measured the valence-band disconti nuities for various wurtzite GaN, AIN, and InN heterojunctions, The values obtained are

(1.0sev nEv == 0.70 eV 1.81 eV

for InN/GaN. for GaNiAIN. for lnNlAIN.


Using Eqs. (4.2.18), the conduction-band discontinuities for each heterojunction can be estimated as nEe == nEg - l1Ev, respectively.



The characteristic feature of AllnGaN-based heterostructures is that the potential profile is influenced greatly by the built-in polarization electric field. The field is caused by the inherent spontaneous polarization, Ps' that is present in wurtzite-

type crystals (Bernardini et al. 1997) and by the piezoelectric polarization, Ppz, that is caused by strain (Bykhovski et al. 1993). The latter is given by

where e31 and e33 are the piezoelectric constants and "'a and &c are the strains in the heterointerface plain and along the c axis, respectively. The actual value of electric field depends on the structure and doping (Sanchez-Rojas et al. 2000).

Typical n-type dopant in AlinGaN materials is Si (Nakamura et al. 1993a).

The most appropriate p-type dopant is Mg. However, p-type doping in an AllnGaN materials system encounters even more severe problems than in AIGalnP. Mg-doped films exhibited high electronic conductivity because of the formation of Mg-H complexes. Amano et at. (1989) succeeded in overcoming this problem by activation of the dopant by low-energy electron-beam irradiation. Later, Nakamura el al. (1992) developed an activation process by using thermal annealing. Nevertheless, because of high ionization energy of the Mg acceptor (~0.2 eV), only a small portion of acceptors are ionized at room temperature. This is the reason that group III nitrides have relatively low p-type conductivity even at the highest-possible doping levels.

The radiative recombination coefficient for direct band-to-band transitions is

- II 3 . h G II h .

on the order of 10 ern /s (see Table 4.2.1). However, III t e In aN a oy t at IS

widely used for LED fabrication, the emission mechanism is not fully understood and is probably affected by composition inhomogeneity that is characteristic for this material (sec Chichibu et al. 2000 and references therein).

The origin of nonradiative recombination in group [JJ nitrides is also not unveiled completely. Although the dislocations that are due to lattice-mismatched substrate are recognized as centers of the nonradiative recombination, it is argued that the carrier diffusion length can be shorter than the dislocation spacing (Sugahara et at. 1998). Nevertheless, the nonradiative lifetime in bulk materials has a tendency to depend on the substrate used for growth. For instance, in UIldoped OaN grown over sapphire, the carrier lifetime can be close to 250 ps at room temperature (K won et at. 2000), whereas in GaN films grown on latticematched substrate (bulk GaN), carrier lifetimes can be as high as 890 ps (Jursenas et at. 2001). In thick InGaN epilayers, the carrier lifetime is also on the order of 100 ps (Smith et al. 1996). Much smaller nonradiative lifetimes are obtained in Al-containing epilayers, For instance, room-temperature lifetimes in AlxlllvGal_x_yN (x = 0.09) are 30 to 60 ps (Tamulaitis et al. 2000).

In single and multiple quantum wells, typical carrier I ifetimes are higher. This can be caused by spatial separation of electron and holes by the built-in electric field (Takeuchi et al. 1997, Irn et al. 1998) and better materials structural quality, owing to pseudomorphic growth of thin layers. In InGaN QWs, the carrier lifetime



depends strongly on alloy composition. For instance, the electroluminescence decay time measured in InxGa l-xN single-quantum-well LEOs by Narukawa et al. (1998) is 10 to 20 ns for x = 0.45 (green LED), 2 to 2.5 ns for x = 0.35 (blue LED), and 1.5 ns for x = 0.15 (near-UV LED).

Optical properties of binary group III nitride compounds are reviewed by Levinshtein et al. (2001). Some values of the refractive index and absorption coefficient are given in Table 4.2.1.

Most commercially available nitride-based LEOs contain active layers made of ternary InGaN alloys and confining layers made of either binary GaN or AIGaN ternary alloys (sec Section 4.3.1). A more versatile approach is to use AlxlnyGal_x_yN quaternary alloys. Changing the composition profiles of In and Al allows one independently to control strain (and, hence, the built-in fields) and energy gap offsets. This approach, called strain energy band engineering (see Khan et al. 1999, 2000), offers an important tool for the optimization of AllnGaN materials used in high-brightness LEOs. Optical studies clearly showed improved materials properties of AlInGaN quaternary layers compared to conventional AlGa ternary films (Tamulaitis et al. 2000). High-quality p-n junctions (Chitnis et at 2000) and enhanced luminescence (Zhang et al. 2000) in InGaN quantum wells with quaternary barriers were demonstrated.

To date, the main fabrication technology for AlInGaN materials system is MOCVD.

4_2_5. Heterostructure Growth Techniques

Fabrication of high-brightness LEOs involves the growth of multilayer semiconductor films over appropriate substrates. Then the wafers are diced, the chips are equipped with contacts, and the LEOs are encapsulated into transparent domes or other types of packages.

Development of heterostructure growth techniques has required tremendous efforts by semiconductor physicists and materials engineers. These techniques must facilitate precise and reproducible control of the composition, doping level, and thickness of the layers. Also, the gro w 1h methods used in mass production must be cost-efficient. To date, only two heterostructure growth techniques meet these requirements. AIGaAs LEDs are produced by using a relatively inexpensive LPE technique. AIGalnP and AlInGaN LEOs used in the most advanced solidstate lamps require a more sophisticated method (i.e., MOCYO). LPE and MOCVD methods are described in brief below. [We do not discuss other growth methods, such as molecular-beam epitaxy (MBE) and vapor-phase epitaxy (VPE), as well as in situ monitoring and post growth characterization issues, which are described in specialized literature.]

LPE is the oldest epitaxy technique (Nelson 1963, Rupprecht et al. 1967), in which the material to be deposited is contained in a liquid. The basics of LED production by LPE are reviewed in numerous books (e.g., Bergh and Dean 1976,



Gillessen and Schairer 1987). Mature versions of LPE used for high-volume AIGaAs LED production are reviewed by Steranka (1997).

The initial solutions are prepared by introducing AI, GaAs, and dopants into a purified Ga melt with the composition of a solution being determined by A l-Ga-As phase diagrams. Typically, the epitaxial process is performed in a horizontal slider (Fig. 4.2.4). Solutions with a different content of AI and dopants are contained in separate cavities of a slider, and substrate is placed on the base plate. Both the slider and the base plate are made of high-purity graphite and are contained within a quartz tube filled with purified hydrogen. A heterostructure is grown by consecutive positioning of the cavities over the substrate. The initial temperature of the solutions is slightly above the growth temperature (800 to 900°C). When the solution is brought into contact with the substrate, the temperature is lowered and the solution becomes supersaturated. This results in growth of an AIGaAs layer. After a layer of the required thickness is grown, the slider delivers the next solution, and growth is continued by establ ishing an appropriate temperature regime. The temperature regime needed for the growth is set either by a slow decrease in the furnace temperature (slow-cooling LPE) or by establishing a vertical temperature gradient within the cavities containing the solution (temperature-difference LPE).

The advantages of the LPE technique are simplicity, purity of the layers produced (undesired impurities such as oxygen are pushed out to the top of the solution), and the possibi lity of growth ofthick layers that are required for light extraction (see Section 5.2.1).

As stated above, the growth of heterostructures based on AIGalnP and AlInGaN materials systems is incompatible with LPE. For AIGalnP, the main reason is the difference in the thermodynamic stability of AlP and InP, which makes compositional control difficult and leads to the segregation of AI. Similar problems occur in the case of AlinGaN, where, in addition to problems similar to those for AIGalnP, the melt is difficult to handle because of high melting temperatures


/ Quartz Tube

GaAs Substrate

FIG. 4.2.4. Basic configuration of the LPE horizontal slider method. (After Gillcssen and Schairer 1987.)



of the nitrides and because of the high equilibrium pressure of nitrogen. To date, the most appropriate technology for group III phosphide- and nitride-based highbrightness LEOs is MOCVD. Other techniques, such as MBE and hydride or chloride vapor-phase epitaxy (HVPE or ClVPE, respectively), are also being actively developed; however, they have not yet been used for large-scale production.

MOCVO is a nonequilibrium growth technique introduced for GaAs by Manasevit (1968). The technique utilizes vapor transport of source materials (precursors; see Fig. 4.2.5), subsequent reaction of these materials in the heated zone, and deposition of the final crystalline product on a substrate. The group III precursors are metalorganic compounds, alkyls that are either trimethyl or triethyl based. The most extensively utilized group III sources are trimethylaluminum AI(CH3)3 (TMAI), trimethylgallium Ga(CH3h (TMGa), and trimethylindium In(CH3)3 (TMln). The group V precursors are the hydrides PH3 (phosphine) and NH3 (am-

ffi Mass flow controller hl Run/vent valve

o Pressure controller ~

rl=h qJ 3-way valve

U Bubbler bypass valve

MOCVD Reactor

Process Exhaust



FIG. 4.2.5.

Schematic of the gas delivery system for an AIGalnP MOCVD reactor. (After Chen et at. 1997,)



monia) for phosphide and nitride growth, respectively. Typical dopant precursors are the metalorganic compounds diethylzinc (OEZn), dimethylzinc (DMZn),

bis(cyclopentadienyl)magnesium (Cpj Mg), and diethyltellurium (OETe), as well as hydrides [silane (SiH4) and disilane (SiZH6)]. Examples of the basic MOCVD reactions are


for AIGalnP and AlinGaN materials systems, respectively. The basic reactions contain intermediate stages that are not completely understood. Nevertheless, highquality epitaxial layers are obtained.

MOCVD growth issues related to AIGalnP LEOs are reviewed by Chen et at. (1997) and Chui et al. (2000). An example of a schematic of the MOCVD gas delivery system for an AlGainP MOCVO reactor is shown in Fig. 4.2.5. A carrier gas (N2 or H2) flows through bubblers that contain metalorganic precursors. The run/vent arrangement ensures rapid, efficient, and controllable switching of the precursor gases into the reactor. The reactor contains the GaAs substrate, which is placed on a graphite plate (susceptor). The susceptor is heated to 700 to 800°C by means of radio-frequency induction, resistance, or exposure to an infrared lamp. A typical growth rate is 2 urn/h at a group VIIII precursor partial pressure ratio of 250.

Details on AlinGaN materials system MOCVO growth are described by Nakamura and Fasol (1997), Briot (1998), and DenBaars and Keller (1998). The gas delivery system is similar to that shown in Fig. 4.2.5 except that some precursors are different. An example of an MOCVD reactor developed by Nakamura (1991) is shown in Fig. 4.2.6. A stainless steel chamber contains a rotating susceptor with a sapphire substrate. The precursors are delivered by the horizontal flow through a quartz nozzle. Another flow of inactive gases is directed vertically and serves to bring the precursors into contact with the substrate. Prior to the highquality nitride layer, a nucleation layer is grown at reduced temperature (450 to 600°C). Then growth proceeds at higher temperatures (> I OOO°C), a typical growth rate being 4 urn/h. Growth takes place at atmospheric or lower pressure.

The MOCVD growth technique is highly versatile. Heterostructures containing multiple thin layers with abrupt interfaces can be grown reproducibly. Also, precise control of the composition and doping profiles can be real ized. The disadvantage is nonavailability of the thick window layers required for high efficiency of light extraction; hence, combining MOCVO with other growth techniques and/or wafer bonding is required (see Section 5.2.2). As many precursors used in the MOCVD growth are hazardous gases, environmental and safety issues must also be addressed.



IR Radiation Thermometer ~



¥ Quartz Tube

Stainless Steel .- Chamber


Quartz Nozzle


Vacuum Exhaust

FIG. 4.2.6. Schematic of a two-flow MOCYD reactor for growth of group III nitride semiconductors. (Aller Nakamura 1991.)


In this section, basic features of electroluminescence in high-brightness LEDs are described. In Section 4.3.1, A1GaAs, AIGaInP, and AlInGaN heterostructures that provide high carrier injection and internal quantum efficiencies are discussed. Section 4.3.2 is devoted to contacts and current-spreading issues, (The complete design of LEDs, which includes means for efficient light extraction, is described in Chapter 5.) Emissive and electrical characteristics of injection electroluminescence in practical high-brightness LEOs are presented in Section 4.3.3.

4.3.1. Electroluminescent Structures

The electroluminescent structures of high-brightness LEOs employ either a double heterostructure with a thick active layer or a quantum well. Single-DH or SQW design can be used; also, the active region can contain multiple layers that confine carriers. These are thick-layer-based multiple-wet! heterostructures (MWHs) and MQWs.

The first visible DH LED was fabricated by lshiguro et al. (1983). The LED was based on an AIGaAs materials system that was well developed at that time and relied on mature LPE technology. The bandgap alignment in the electroluminescent structure of the first AIGaAs-based DH LED is shown schematically in Fig. 4.3. I a. The active layer is made of undoped direct-gap AlxGal-xAs with an


(a) (b)

> >
OJ <lJ
Q) (0
0 CC!
N > <lJ

> Q)




o N


------------~-----~ ~----------------------

p-(AIQ7Gao 3)oslno_sP I n-(AlolGao3)o,lno.sP

----- r-0.sllm

{Alo2 Gaos)osl no.sP

FIG. 4.3.1. Schematic of bandgap alignment in typical DH A1GaAs (a) and AIGuinP (b) LEDs.

AI molar fraction x == 0.35 that yields the peak wavelength at 660 nm. The p- and n-Iype cladding layers are made of indirect-gap AlosGaO,2As that is transparent to the light emitted, The entire structure is grown on lattice-matched AlxGal_xAs (x> 0040) substrate and is well suited for high light-extraction efficiency (see Section 5,2 _ I).

The basic structure of the active and cladding layers of the A!GaAs DH LED shown in Fig. 4.3.1 a underwent minor improvements in mass-produced devices (Cook et al. 1988, Ishimatsu and Okuno 1989). In the active layers, the AI molar fraction ranges between 0.35 and 0040 depending on the emission wavelength (see Section 4.2.2). To provide a higher potential barrier for the electrons injected, the AI molar fraction in cladding layers is at least 0.75 to 0.8 and the active layer is p-doped to match the valence-band potential with that in the p-type cladding layer

(see Fig. 4.1.7). .

The DH concept developed for AIGaAs LEDs was adapted successfully for AIGaTnP LEOs grown by MOCVD (Fig. 4.3.1b). In typical AIGalnP DH LEDs, introduced by Kuo et at. (1990) and Sugawara et al. (1991), the light-emitting structure is lattice-matched to GaAs substrate and contains an active layer made of

direct-gap (AlxGal-x)osIno_sP with the relative content of AI, x, between 0 and -0.5. Typical values of the active layer thickness are between 0.3 and 1.0 urn, Further reduction in the active layer thickness to :0;0.2 urn and the introduction of an MWH improved the LED performance (Gardner et at 1999). The active layer is unintentionally doped and, in practice, can be slightly n- or p-doped or even contain a p-n junction (Chui et al. 2000). The cladding layers have a higher Al



content (x> 0.7), which corresponds to indirect-gap material. The DH is combined with light-extracting windows as discussed in Section 5.2.2.

At peak wavelengths above 590 nm, (AI.rGal-.r)o.slnOjP DH LEOs exhibit carrier injection efficiencies close to unity and negligible leakage currents (Kish and Fletcher 1997). However, at as Iowa value as 570 nm, a reduced confining potential results in considerably smaller hole injection efficiency and in noticeable electron leakage over the confining barrier. These drawbacks become even more pronounced with increasing operational temperature. Also, with increasing AI content in the active layer, the internal quantum efficiency drops because of enhanced nonradiative recombination caused by the proximity of direct-indirect band crossover and by the oxygen-related traps.

To achieve penetration of AIGalnP LEDs into the yellow-green spectral region and to facilitate their high-temperature operation, improved electroluminescent structures were proposed. One possible solution is to use QWs in the active layer. Quantum confinement leads to higher energy of the photons emitted in a QW compared to that in a thick DH with the same Al molar fraction. Sugawara et al. (1994a) fabricated MQW structures composed of 50-A

(Alo2Gao.s)o.5Ino.5P quantum wells separated by 40-A (Alo5Gao5)osIno.sP barriers, with the number of periods from 3 to 40. The structures exhibited improved emission efficiency compared with OH control samples. Chang and Chang (1998a,b) succeeded in increasing the rate of radiative recombination by using compressively strained MQW structures composed of lattice-mismatched 80-A

Gao47InO.S3P wells and lattice-matched 200-A (Alo6GaoA)o.slnosP barriers. The thickness of the wells was small enough to accommodate to the barriers compressively with no misfit dislocations. The compressive strain resulted in reduction of the hole effective mass and increased the probability of interband radiative transitions.

To prevent the leakage of electrons over insufficiently high confining barriers, Chang et al. (1997b) inserted a thin lattice-mismatched EBL between the active layer and the p-cladding layer (Fig. 4.3.2). The layer was made of p-AIO.65InO.35P, which had a bandgap much wider than in lattice-matched AloSlnoSP and was tensile strained (tensile strain results in even greater separation of the conduction and valence bands). Again, due to the small thickness of such a layer (250 A), no misfit dislocations were generated, but the confining potential for electrons was increased considerably. The structure that had a tensile strain barrier cladding (TSBC) layer exhibited a twofold increase in the emission intensity at 573 nm relative to the control structure, which lacked the blocking layer.

Another way to prevent electron leakage is to lise the multiquantum barrier (MQB) proposed by Iga et al. (1986) for laser diodes, where leakage is an even more serious issue because of higher injection currents. Figure 4.3.2b is a schematic of an electroluminescent structure with an MQB placed between the active layer and the p-type cladding layer. The period of the MQB is adjusted so that the electrons experience reflection. Chang et al. (1998) demonstrated a 573-nm AIGalnP LED with a chirped MQB (CMQB) that did not require precise adjust-




(b) •• ~ __ :;:::=_

01 c

"0 "0 m "0 6..

active layer MOB

Ff(;.4.3.2, Schematic of bandgap alignment in advanced AIGalnP l.EDs: (a) structure with tensile strain barrier cladding (electron blocking) layer (alter Chang et al. 1997b); (b) structure with a multiquantum barrier (MQB) (alter Chang et al. 1998).

men! of the period. The CMQB is made of p-Alo.5Ino.sP barriers and p-(Alo6Gao4)osIno.sP wells of linearly increasing thickness. The electroluminescence intensity in the structure with the CMQB was twice as high as a similar structure without CMQB.

AlInGaN-based heterostructures exhibit reduced sensitivity of the internal quantum eff ciency to the presence of misfit di slocati ons (Lester el al. 1995). A I so, there is no direct-indirect gap crossover. This allows for more versatility in the LED design than with AIGaAs and AIGalnP LEDs. Actually, in most commercially available nitride LEOs, the MOCVD-grown electroluminescent structures are composed of layers that are somewhat lattice-mismatched with each other, and confinement potentials are high. An important feature of these LEDs is a large increase in the internal quantum efficiency that comes with incorporation of In into the active layer. Although this phenomenon is not completely understood, there are strong indications that this enhancement is due to In-related composition inhomogeneity. This inhomogeneity leads to carrier localization, which counteracts the spatial separation of the electrons and holes caused by the built-in electric field and reduces the nonradiative recombination (Chichibu et al. 1996, 1997). Detailed discussions of the electroluminescent structures of commercially available nitridebased LEOs are presented by Nakamura and Fasol (1997) and Kern et at. (1000).

Group 111 nitride-based OH LED was introduced by Nakamura et al. (1993b).

High-brightness blue 450-nm LEOs (Nakamura et al. 1994a) and blue-green 500-nm LEDs (Nakamura et al. 1994b) employed a OH structure with widebandgap AIGaN cladding layers (Fig. 4.3.3a). The blue LED contained a Zn-doped Inoo6GaO.94N active layer of about 500 A thickness. The cladding layers were



_(a) __ ~I.......-_



n-GaN -




i p-GaN




FIG. 4.3.3. Schematic of bandgap alignment in AllnGaN LEOs. (a) Dll-based structure with two wide-bandgap cladding layers; the radiative transitions occur between donor-acceptor pairs (after Nakamura et al. 1994b). (b) SQW structure with asymmetric confining layers; the radiative transitions occur between quantum-confined levels of electrons and holes (after Nakamura et al. 1995c).

made of Alo.1 SGao85 N doped with Si and Mg, respectively. The thickness of the cladding layers was 0.15 11m. To improve electrical conductivity at the contacts, the electroluminescent structure was equipped with p-type (O.S-l1m) and n-type (4-jlm) contact layers. The green OH LED had the same structure, with the exception that the In molar fraction in the active layer was 0.23 and the layer was codoped with Si and Zn (co-doping invoked intense radiative recombination via donor-acceptor pairs). Since holes require no high confining potential, the DH structure shown in Fig. 4.3.3a can be simplified by removing the n-AIGaN cladding layer. Such asymmetric DH (AOH) GaNlInGaN/AIGaN/GaN LEDs were described by Koike el al. (1996).

However, highly efficient green and longer-wavelength InGaN OH LEDs could not be fabricated because of degradation of the thick active layer with increasing In molar fraction (Nakamura and Fasol 1997). Also, impurity-based emission suffered from a pronounced shift in the peak wavelength and saturation of the donor-acceptor electroluminescence with increasing current, and the emission line was quite wide (typically, 70 to 80 nm). These drawbacks stimulated the development of new-generation nitride-based LEOs which employed thin layers and relied on intrinsic radiative transitions. The first high-brightness SQW LEDs were fabricated by Nakamura et at. (l99Sa,b) for the violet, blue, green, and yellow spectral regions. The most recent design of SQW LEOs relies on an asymmetric QW with an n-GaN contact layer used as the n-confining layer and a 1000-A-thick p-Alo.2Gao.SN layer used for p-side confinement (Nakamura et al. I 99Sc; see



Fig. 4.3.3b). SQW LEDs exhibit improved characteristics over those of OH LEOs (Mukai et al. 1999).

To improve injection efficiency at high operational currents in AllnGaNbased LEDs, use of MQW structures (Nakamura et al. 1993c) was proposed. First nitride LEOs containing MQW as an active layer were fabricated by Koike et al. (1996) and Kozodoy et at. (1997). A mature MQW device described by Lester et al. (1998) was composed of 12 InGaN!GaN QWs clad by p-AIGaN and n-GaN layers. The device exhibited high-power output characteristic superior to those of conventional SQW LEOs.

Quatemary Alxln,Ga l-x-yN alloys offer even more versatility in the design of nitride-based high-brightness LEOs. Lattice and energy band engineering in quaternary group III nitrides (see Section 4.2.4) allows one to fabricate heterostructures with controllable strain and spontaneous polarization. LEDs with the active regions made of InGaN MWQs clad by quaternary AlinGaN barriers have been demonstrated (Adivarahan et al. 2000, Shatalov et al. 200 I). However, obtaining AllnGaN alloys within a wide range of Allin ratio is a challenging task because the incorporation of In into AIGaN requires reduced growth temperatures that cause degradation of the material. New growth techniques, such as pulsed atomic layer epitaxy (Zhang et at. 2001), allow one to reduce the growth temperature, thus solving this problem.

4.3.2. Contacts and Current Spreading

Important elements of an all-solid-state lamp are contacts (electrodes) through which the electroluminescent structure is fed. The contact design has to achieve several objectives in view of optimization of both electric and optical properties of the LED. First, the LED series resistance, Rs, which reduces the LEDs' feeding efficiency (see Section 4.1.1) has to be minimized. This requires high conductivity of the contact layers. Also, the current has to be spread over as large a portion of the chip cross section as possible. Second. the contacts should facilitate escape of the light generated and introduce small optical losses (see also Chapter 5). Again, for optical optimization, the distribution of current over the LEOs' cross section plays an important role. Tn this subsection, contact design in high-brightness LEOs is discussed briefly with a glance at current spreading.

The basics of the electrical properties of LED contacts are described by Bergh and Dean (1976) and Gillessen and Schairer (1987). Deposition of a metal on a semiconductor results in the formation of a Schottky junction, which builds a potential barrier at the metal-semiconductor interface. For ideal contacts, the height of the barrier should be determined by the metal work function and the electron affinity of the semiconductor. However, in most group III-V compounds, the bar-

rier height often depends only weakly on the metal and is on the order of '" t E g

and '" t E g for n-type and p-type semiconductors, respectively. A thick barrier can only be surmounted by carriers that have a high thermal energy; hence the



current is due to thennoionic field emission, with the I-V characteristics similar to those for a Jrn junction. Highly conductive (ohmic) contacts require thin barriers, so that charge carriers could tunnel through, assuring low resistance and linearity. Thin barriers can be obtained by doping the semiconductor to high levels (_1019 em -3).

In AIGaAs-based LEOs (Steranka 1997), the contacts can be formed using gold (Au) metallization. To dope the semiconductor region adjacent to the contact, the metal is alloyed with an appropriate dopant. Most frequently, Au+Ge alloys (0.5 to 12% Ge) are used for l1-type contacts, whereas Au-Zn alloys (I to 6% Zn) are used for p-type contacts. The dopant diffuses into the semiconductor during heat treatment. Also, a p-contact can be formed by introducing a thin layer of heavily doped GaAs (p +-GaAs) and capping it with pure Al (see Fig. 5.2.1a).

Using Au-Ge /Hype and Au-Zn (also Au-Be) p-type electrodes is a standard approach in LEOs based on an AIGalnP materials system (Chui et at. 2000). However, because of the low hole mobility and the saturation of p-doping, highly conductive p-AIGalnP is difficult to produce, especially with the high Al molar fraction that is required for transparent cladding layers. This is one of the reasons for the introduction of p-GaP (Kuo et at. 1990) and p-AIGaAs (Sugawara et al. 199\)

conductive layers over the (AIyGal-x)o.slnOSP electroluminescent structures. These layers are also important for current spreading (see below) and light extraction (see Section 5.2.2).

In AllnGaN materials, heavy doping from the alloyed electrodes is difficult to implement. Fortunately, in this materials system, the Schottky barrier height exhibits a pronounced dependence on a metal work function (Foresi and Moustakas \993). Also, formation of the conductive metal nitride under the contact may be important. Therefore, ohmic contacts can be obtained by empirical selection of an appropriate metal (for a contact review; see, e.g., Pearton et al. 1999 and references therein). The simplest approach relies on AI and Au metallization for n- and p-type contacts, respectively, and was used in first heterostructure LEOs (Nakamura et at. 1993b). An improved specific contact resistance was obtained for l1-titanium (Ti)1 AI bilayer contact (Lin el at. 1994a) and p-Au/nickel (Ni) contact (Nakamura et al. 1993d). To date, these electrodes are standard for A I InGaN LEOs.

Generally, both transparent and nontransparent electrodes are employed. The area covered by nontransparent contacts must be as small as possible, to minimize shielding of the light emitted. On the other hand, the shape of the contacts must provide for current spreading over the LED cross section. Figure 4.3.4 illustrates possible current flow paths in LED chips with nontransparent lop contacts. At a distance x from the contact, the current density is given by (Thompson 1980)


where )0 is the current density under the contact and Ls is the spreading length. The latter is given by




Bottom Contact


FIC.4.3.4. Current flow paths in LED chips. (a) Thin/low-conductivity current-spreading layer (CSL). The current crowds under the top contact. (b) Thick/highconductivity CSL. The current uniformly spreads over the entire cross section.


where a and t is the conductivity and thickness of the current-spreading layer (CSL), respectively, and r d is the diode ideality factor. Equations (4.3.1) and (4.3.2) imply that current uniformity improves with increased CSL thickness and conductivity, as shown in Fig. 4.3.4.

In AIGaAs-based LEDs, highly conductive CSLs of both n- and p-type are available. Therefore, the top contact can have a simple circular shape of 80 to 100 urn minimal diameter, which is required for wire bonding (Fig. 4.3.5a). In AIGainP LEOs, the top CSL is usually of p-type with lower conductivity because of common p-doping constraints in wider-bandgap materials. A fair solution is top

100 urn



(a) (b)

350 urn



FIG. 4.3.5. Possible top-contact geometries for high-brightness LEDs: (a) simple circular contact for a standard chip; (b) contact with finger projections: (c) and (d) complex patterns used for larger chips. (Aller Kish and Fletcher 1997.)



contacts with finger projections that prevent current crowding in the peripheral area (Kish and Fletcher 1997). Even more complex top-contact patterns are used for large-area high-power chips (Fig. 4.3.5c and d).

Another current-spreading approach relies on use of a transparent electrode that covers the entire top of the chip. Conductive and transparent indium tin oxide (ITO) was demonstrated to be a good candidate for use in AIGalnP LEOs instead of thick CSLs (Lin et al. I 994b, Aliyu et al. 1995). ITO layers as thin as - 100 A exhibit transmittance in excess of 90% and produce an improved emission pattern (Morgan et al. 2000). However, this approach is still under development and has not yet been used in large-scale production of AIGalnP LEDs.

In contrast to AIGalnP LEDs, the transparent-contact approach is the basic solution for group ITI nitride-based LEOs, where growth of thick CSLs is difficult and high levels of p-type doping are even more difficult to achieve. The common design ~ of high-brightness AlInGaN LEDs contains the aforementioned thin (nanometer-scale) Ni/Au electrode evaporated onto the p-type conductive layer (Nakamura et al. 1993d, 1994a), The Nil Au bilayer metal film exhibits transmittance of about 85% at 470 nm (Sheu et al. (999). ITO electrodes (Margalith et. al. 1999) and thin-film platinum (Pt) contacts (Huh et al. 2000) were also used.

Although the transparent p-electrode facilitates the uniform current distribution in AlinGaN chips, the current-spreading issue is still important for LEOs grown on an insulating substrate (typically, sapphire). The reason is that the lateral path of the current to the n-contact is usually formed in the etched mesa as shown in Fig. 4.3.6a. The drawback of the conventional asymmetric structure is that current crowds toward the n-pad (Kim et al. 2000). Krames et al. (2000) described a symmetrical nitride-based LED with a ring n-electrode that diminishes the lateral field and improves current spreading (Fig. 4.3.6b). This geometry is especially suitable for high-power LEOs with a large chip area.




FIG. 4.3.6. Current paths in AlInGaN chips grown on sapphire: (a) asymmetric design with current crowding toward (he n-pad: (b) symmetric design with a ring n-pad. (After Krames et al. 2000.)



4.3.3. Emissive and Electrical Characteristics

In. this section, emission spectra, I-V ~nd output characteristics of practical highbnghtne:s ~E~s are presented. Typical electroluminescence spectra (spectral power dlst~lbutJons) of SOme commercially available high-brightness LEOs are shown 111 FIg. 4.3.7. The spectra were recorded using an optical multichannel analyzer with a calibrated spectral response. The 648-nrn AIGaAs red and 594-nm AIGaTnP amber LEDs exhibit almost symmetrical emission lines with a full width at half magnitude (FWHM) of 16 and 13 nm, respectively. These values are in excellent agreement with theoretical calculations of the line width,


that is characteristic of emission lines due to tree-carrier band-Io-band recombination, which have an energy broadening value of L8kBT .

The emiss.j~n li?es in AllnGaN-based LEOs are distinctly asymmetrical, due ~o the .composltlo~ inhomogeneity of the InGaN alloy used in the active layers. The Width ~f the lines IS much higher than that predicted by Eq. (4.33). The blue D~ LED With the co-doped active layer features a broad line (FWHM of 57 nm) With a multiple structure on the long-wavelength wing (optical-phonon replicas),

AlinGaNSaw green (517nm)

400 450 500 550 600 650 700 Wavelength (nm)


Examples of cicctroluminescence spectra in high-brightness LEDs at a forward current of20 mA. The peak intensity is normalized for clarity.



which is typical for donor-acceptor recombination. The line width in the blue SQW LED is much smaller (22 nm). However, an increased In molar fraction in the green SQW LED results in broadening of the emission line to 34 nm.

Figure 4.3.8 shows the peak position of the emission lines as a function of the forward current. In AIGaAs and AIGaInP LEOs, the increased current results in a slight shift of the peak position toward longer wavelengths. Such behavior can be attributed to the bandgap renormaJization that occurs in dense carrier systems because of many-body interaction (Banyai and Koch 1986). The second factor is the bandgap shrinkage caused by heating the chips at a high forward current.

In AllnGaN-based LEOs, the behavior of the peak position is different from that in A1GaAs and AIGalnP LEDs_ As one can see in Fig. 4.3.8, the peak position shi fts toward shorter wavelengths with increased forward current. The shift is more prominent in the green LED than in the blue LED (i.e., it is enhanced with increased In content in the active layer). The cause of the blue shifting is band filling of the localized energy states that are due to composition inhomogeneity of InGaN alloy (Mukai et at. I 998c).

Figure 4.3.9 depicts the output intensity of typical high-brightness LEOs as a function of forward current. In AIGaAs and AIGaInP LEOs, the output intensity depends on the forward current superlinearly for small currents. This behavior can be understood in terms of the increased radiative recombination rate and saturation of the nonradiative centers with increasing carrier density. For high currents, the slope of the output characteristi cs decreases and the output intensity tends to saturate, probably because of increased carrier leakage through the confining layers.

In AlinGaN-based SQW LEOs, the output characteristics (Mukai et at. 1999) are similar to those of arsenide- and phosphide-based LEOs, with the difference that superlinear behavior is observed at lower currents (not shown in Fig. 4.3.9). Above 1 mA, the output characteristics are slightly sublinear, probably because of

650 r ~-o-o-O

E -S600

___c ...

O"l C

"* 550 f-

~ (). AllnGaN green

~ v--o--o-~-o_~~~~~~

---><: 500 -v <»




AIGalnP amber


AllnGaN blue

... --...__._.__._..._ ... __.__.

10 100

Forward Current (mA)

FIG. 4.3.8. Emission line peak position vs. forward current in high-brightness LEDs.


10' ~AIGaAsred
-AIGalnP amber
~ -:::-AlinGaN green
......Ao-AlinGaN blue
::J 10°
IJ) 10"
-5 1O.~
10 100
Forward Current (mA) FIG. 4.3.9. Output characteristics in AIGaAs-, AlGalnP-, and AlinGaN-based LEOs. The dependences are arbitrarily shined along the vertical axis.

carrier overflow from localized energy states. The overflowed carriers are delocalized and can reach the nonradiative recombination centers, thus reducing the internal quantum efficiency. In addition, in SQW LEOs, high current can result in electron I~akage into the confin ing I ayers despite the presence of wi de-bandgap carrier blocking layers. This drawback can be removed by using MQW structures (Lester et al. 1998, Adivarahan et at. 2000, Shatalov et al. 2001).

Figure 4.3. 10 depicts I-V characteristics measured in typical high-brightness LEDs. AIGaAs and AIGaJnP LEOs exhibit I-V characteristics that are in good agreement with single-exponent behavior and saturation due to LED series resistance. The ideality factor r '" 2 implies that the current is dominated by the recombination [see Eqs, (4.1.32) and (4.1.33b)]. The series resistance extracted from the I-V characteristics is 5 to 6 Q, which is a typical value for this type of LED (3 to 15 Q) .

The I-V characteristics for AllnGaN-based LEDs differ from those for A1GaAs an~ AIGalnP LEDs. In a OH LED, the low-current characteristic is exponential and JS probably dominated by tunnel current with a characteristic energy of -0.19 eV (see Eqs. (4.1.32) and (4. I. 33c)]. Tunneling is caused by the presence of numerous deep levels in the co-doped InGaN active layer (Perlin et al. 1996). ln ~QW LEOs, the low-current I-V characteristic can be described by a recornbinanon teml. with an ideality factor close to 2. However, at higher currents, the nonexponential I-V characteristics in AIInGaN-based LEOs are quite complicated and ~a~not be understood within the approach used in Eq. (4.1.32). The I-V charactertstics Imply that ni~ridc LEDs have higher series resistances (typically, 20 to 40 n) than those of arsenide- and phosphide-based LEOs_




0.1 I
0.01 I~
Y =2 I 0 ~
~ 1E-3 \ 1°
E 1E-4 I 0.
!!! I o·
'- iE-S I 0:
U I q,
"E 1E-6 I q,
I 'lI
ro q, 0 AIGaAs DH
~ 1E-7 I q, ..
a I {JI • AlinGaP DH
l..L.. 1 E-8 I i:!. AlinGaN DH
... AiinGaN SQW
1.0 1.S 2.0 2.5 3.0 3.S 4.0 4.5
Forward Voltage (V) FIG. 4.3.10. Currenl-voltage characteristics of AIGaAs-, A1GalnP-, and AllnGaN-based high-brightness LEDs.

Temperature characteristics of high-brightness LEOs are discussed bv St~ranka (1997) and Mukai et al. (1998c, 1999). Generally, all LEOs exhibit a red shi ft. of the peak wave I en.gth with increasi ng temperature because of the temperatu.re-mduced bandg~p shrinkage: ~s a rule, the output power decreases with opera~ln~ temperature, since the radiative recombination rate decreases and the nonradiative recomb!n~tion rate increases. However, the decrease in output power is mO.re. characteristic of AIGaAs and A1Ga[nP LEOs than of nitride-based LEOs. ThIS IS caused by the. absence of the close indirect valleys in the band structure of AJInGaN-based serniconductnrg in contrast to arsenide- and phosphide-based materials.



For many years, most of the LEOs manufactured employed a primitive design that featured a planar structure on an absorbing substrate encapsulated in an epoxy dome. Such design resulted in extremely poor light-extraction efficiency ('loP! "" 4%), much less than typical internal quantum efficiencies. Since the ex-

ternal quantum efficiency of such LEOs was just a few percent, they were able to compete with other light sources only for applications in color indicator lamps and miniature numeric displays, where small dimensions and extended lifetimes were the main advantage. Lighting applications require LEOs with greatly improved efficiency that is typical for modern high-brightness LEOs.

The main physical reason that light extraction is difficult is the large ratio of the refractive indices of the semiconductor and the surrounding media. Consequently, a major part of the light generated inside a chip is reflected back into the semiconductor. High-brightness LEO designs implement additional means for easy photon escape. In this chapter we review the LED photonics, which is the key issue for solid-state lighting technology. \Ve start from basic considerations determining the design and performance of high-brightness LEOs (Section 5. I). Practical designs of conventional (planar rectangular) high-brightness LEOs are discussed in Section 5.2. Section 5.3 deals with supplementary classical-optics means of extracting light from semiconductors (i.e., outcouplers), which requires unconventional technology. Finally, recent achievements in light extraction, based on modification of the photon density of modes, are discussed in Section 5.4.





5.1.1. Escape Cones

Photons emitted in the acti ve layer of an LED escape into the surroundings after complex travel within the chip. These photons get lost at each stage of the journey via absorption in the substrate, semiconductor layers, contacts, and in the epoxy dome. Estimation of the extraction efficiency is highly involved and requires the use of statistical methods for photon gas. (Joyce et al. 1974, Schnitzer et al. I 993a,b, Boroditskyand Yablonovitch 1997, Lee and Song 1999). Below, we present a simpi i fi ed approach based on convent ion al geometrical considerations (Lee 1998 a, b).

Consider a planar rectangular light-emitti ng structure grown on an absorbing substrate and encapsulated in an epoxy dome-an absorbing-substrate (AS) LED. The refractive indices of the layers in the semiconductor structure are usually very close and one can use a single average value "s for these layers. The refractive

index of the dome epoxy is n e. The struct ure is depicted i 11 Fig. 5.1 .1. whi ch abo shows some relevant rays. The photons generated in a certain point of the active layer (where isotropic generation is assumed) may escape into the epoxy dome only for directions contained in a cone with an apex of 2ee· The critical angle Be is given by Snell's law

(5. 1.1)


Epoxy ne

Absorbing Substrate

tJG.5.1.1. Schematic or an escape cone in a conventional LED encapsulated in all epoxy dome.



Some of the I ight propagating within the escape cone is reflected from the semiconductor-epoxy interface. Additionally, some of the photons that escaped into the dome are reflected from the epoxy-air interface. The photons that are incident to the semiconductor surface with the angles e > Be propagate outside the escape cone and undergo the total internal reflection. ln the structure shown in Fig. 5.1.1, most of the light which did not escape into the air on the first pass is lost due to absorption in the substrate and electrode area. This design, based on one escape cone, is typical of low-brightness LEDs with thin epitaxial structures grown on a light-absorbing substrate (Fig. 5.1.2a).


FIG. 5.1.2. LED designs with different numbers ofescape cones.



One of the basic approaches employed in high-brightness LEOs is to increase the number of escape cones (Stringfellow and Craford 1997). In a rectangular configuration, up to six escape cones may be opened. A twofold improvement in extraction efficiency is achieved by removing the absorbing substrate or by introducing a transparent substrate (Ishiguro et al. 1983). Figure 5.1.2b shows the hypothetical design of a thin-structure two-cone transparent-substrate (TS) LED with an additional downward cone opened. Since the chip requires some thick support, the two-cone configuration might be imitated by introducing a mirror between the AS and the emitting structure (see Section 5.1.2 and Fig. 5.1.3).

Opening of the lateral cones is based on introduction of thick transparent layers on both sides of the electroluminescent structure. For instance, if the thickness of the upper transparent layer is increased, four lateral semicones emerge (Huang et at. 1992). For an appropriate thickness of the upper layer, often call~d the window layer (WL), the sernicones are opened completely and three full cones are available for light escape. A three-cone WL LED schematic is shown in Fig. 5. 1.2c. Similar thickening of the transparent substrate (Cook et al. 1988) adds four more semicones to the initial two-cone configuration, and the total number of escape cones becomes four (Fig. 5.1.2d). A five-cone design is available by growing both lower and upper WLs of sufficient thickness on an absorbing substrate (Fig. 5.1 .2e). Finally, the most efficient six-cone design (Fig.5.1.2f) is implemented by sandwiching the thin electroluminescent structure between two thick transparent layers, substrate and window (Kish et al. 1994a).

The extraction efficiency of an LED, in which an integer number of cones, N, is employed, is given by


The first multiplier on the right-hand side of Eq. (5.1.2) accounts for the Fresnel reflection losses at the epoxy-air interface under the assumption that the photons reflected within the epoxy dome are not utilized. For a typical value of ne = 1.50, this yields 0.96. The second multiplier is the fractional solid angle for a single escape cone [(1- cosBJ/2]. The third multiplier is the sum of the transmittances for

each of the N cones (0::; '0 ::; I ) at the semiconductor-epoxy interface. Generally, some escape cones are partially shielded by the contacts that absorb the light. Also, the lateral cones may not be opened completely (truncated) because of insufficient layer th~ckness. However, when the bases of the opposite cones are separated by a low-optical-density structure, these two cones are coupled so that a photon reflected at the base of one of the cone might escape through the opposite cone (Lee 1998a).

Multiple reflections might lcad to negligible Fresnel losses in a coupled cone pair since only an optically thin layer separates the escape surfaces. With the shielding neglected, transmittance of the completely opened cones is given by


for coupled opposite cones, [or uncoupled cones.


Here the average reflectivity in the cone-base area is assumed equal to that for normal incidence. The cones next to the absorbing substrate and the opposite cones operating in the one-pass regime (i.c., with bases separated by an optically dense layer) must be treated as uncoupled cones.

In AIGaAs materials (ns "" 3.5), Eqs. (5.1.2) and (5.1.3) yield '70pl "" 0.04 for

a one-cone AS LED and 'l"pt "" 0.093 for two coupl ed cones .. F or three pa irs of uncoupled cones (six cones). 'lopl "" 0.23 for A IGaAs and 'lopl "" 0.25 for AIGalnP (ns '" 3.4), respectively.

The average refractive index of A IlnGaN materials ( "s "" 2.5 ) is smaller than that of AIGaAs and AIGalnP. Thus, even the primitive one-cone AS LED design yields 'loPl '" 0.092. In practical AllnliaN-based LEOs, the concept

of escape cones is modified, since the semiconductor structure is placed on a transparent substrate with a refractive index ni that satisfies the con-

dition lie < IIi < lis (Lee 1998b). In this case, the downward cone is enhanced. Additionally, a fraction of the photons that undergo total internal reflection from the top surface might escape through the transparent substrate (see Section 5.2.3).

5.1.2. Distributed Bragg Reflectors

The tran sparent substrate, wh ich is one 0 f the key elements in incrcas i ng the number of escape cones, is difficult to implement in some material/substrate systems. Fortunately, the downward escape cone can he replaced by an appropriate mirror positioned between the active layer and the absorbing substrate. An approach suitable for planar structures is the lISC of distributed Bragg reflectors (OBR;;). A DBR structure consists of a number of alternating layers of high- and low-refractive-index materials (n Hand n I. , respectively) with an optical thickness of a quarter of the em ission wavelength. The thickness of the layers is given by (Sugawara et al. 1993)

hIJ = (Jo/4)/(IIH cosBH), h: ;; (..lo/4);( nL COSBL)'



where Jo is the emission wavelength in air, and ef! and Of. are the incident angles at each individual layer. The higher the number of periods and the difference in refractive indices, the higher is the reflectivity of the DBR structure. A large




difference in the refractive indexes is also desirable for wide-band operation. 1l1e schematic design of a DBR LED is shown in Fig. 5.1.3. Katoetal. (1991) and Saka et al. (1993) were first to develop this approach for increasing the efficiency of infrared GaAs!AIGaAs LEOs grown on an absorbing GaAs substrate. Soon after, high-brightness AIGaInP LEOs in the orange-to-green color region were devised using OBR structures consisting of AlInPl AlGalnP layers (Sugawara el al, I 992a) or AlInP/GaAs (Itaya et 01. 1994) layers (see Section 5.2.2). An AlGaAs red LED with an AlAs! AIGaAs OBR (Thomas et al. 1995) and an AllnGaN blue LED with a ?aN/AlGaN DBR (Nakada et al. :WOO) were also reported, Typically, 10- to 20-peflod DBRs are employed.

As can be seen from Eq, (5.1.4), a simple OBR is a resonant structure operating in a limited range of incident angles for a particular wavelength. This is ideal for laser diodes, but it does not always work well for LEDs, which isotropically emit a broad photon spectrum. To provide high reflectivity in a wide-angle, broad-spectrum regime, more sophisticated OBRs have been developed, A cascade of AlAs! AIGaAs quarter-wave reflector stacks designed for different wavelengths was demonstrated by Lee (1995). This design expanded the high-reflectance band in the red region, Combined (hybrid-type) DBR structures, consisting of different multi layers, were introduced for green AlGalnP LEOs. A typical hybrid OBR structure consists of two stacks of multilayers, either AllnP/AIGalnP or AIInP/GaAs (Sugawara et al. 1994b) or AlInP/AlGalnP and GaAs/AIAs (Chiou et of. 2000). Another way to attain a wide reflectance band is to utilize a chirped OBR structure, where the optical thickness of the alternating layers varies monotonically (Saka et 01_ 1993), A yellow-green AIGalnP LEO with a chirped OBR structure was fabricated by Chang et al. (1997a).

n-DBR Structure

p-Current-Spreading Layer p-Cladding Layer

Active Layer

n-Cladding Layer

n-Absorbing Substrate


FIG. 5.1.3. Schematic design of a DBR LED.


5.1.3. Absorption Losses and Photon Recycling

The photons that do not esc~~e through the cones are eventually ~bsorbed in the chip. The irreversible (parasitic) losses are due to the absorption 111 bthe substrate

d the contact area. Substrate losses are avoided by rernovmg the su strate, using ~ransparent substrate, or redirecting the emission as describedin Secti~n ~,1.2.

Another important issue in the utilization of escape cones IS to rmrurmze para-

ltic absorption in the contact area. One possible solution is use of the aforernen:;~ned transparent electrodes (see Section 4,3.2). This idea is implemented in practical AIInGaN LEOs, where a thin Ni/Au ohmic metal is de~osited on top of the chip (Nakamura et al. 199Jd, 1994a, Lee I 998b;, ~ee also Secl~on 5.~J), Also, low optical absorption h~ been achieved bl depositing :ndlUl1l tin OXide (IT?) contacts on AIGalnP (Lin et al. 1994b, Aliyu et al. 199) and GaN (Margalith et al.

J 999) as well as indium oxide (In20J) on AIGaAs (Thomas et al. 1995). An advantage of a transparent contact is that it also serves as a current-spreading layer (CSL). However, highly conductive transparent contacts might not be available or be difficult to implement in a particular materials system. In these cases, this problem has to be solved by other means, One possible solution is to avoid the light generation underneath the contact. A current-blocking layer (eBL) under the top electrode (Sugawara et al. 1992a; see Fig. 5 ,2.2a) or, in LEOs with a top ring contact a current aperture (Windisch et al, 1998a) is introduced to implement this idea, Another way to reduce parasitic losses is to use electrode patterning

(Ishiguro et al. 1983; see Figs. 5.2.lb and 5.2.2b). . .

However, in a well-designed LED, most of the losses are due to absorption III the active layer and, to some extent, in the surrounding transparent cladding and window layers. The losses in the active layer depend strongly on the probability of light reemission, since the photons absorbed can experience reincarnation and get a new chance to find the escape cone. If the internal quantum efficiency were high, the photon would be recycled many times until it escapes [the theory of photon recycling in LEOs is discussed by Baba et al. (1996) and in references therein]. Hence, light-extraction efficiency is a function of internal quantum efficiency, as shown by Boroditsky and Yablonovitch (1997). For instance, extremely high external quantum efficiency (72%) was demonstrated in an optically pumped AlGaAs/GaAs double heterostructure mounted on a high-ref1ectivity surface (Schnitzer et al. 1993a). Multiple photon recycling was due to internal quantum efficiency as high as 99,7%, A GaAs LED utilizing photon recycling with an external quantum efficiency of 12.5% was reported by Patkar et al. (1995). It should be noted that internal quantum efficiencies close to unity are needed in order to exploit the advantages of photon recycling. Therefore, the recycling technique is extremely sensitive to material quality and parasitic losses. A straightforward way to use photon recycling is the utilization of a thick active layer with a high internal quantum efficiency in order to catch and recycle as many nonescaped photons as possible. An example of this approach is a highly efficient AIGaAsfGaAs LED integrated with a quantum-well infrared detector (Dupont et at. 2000, Dupont and Chiu 2000). Nevertheless, in practical LEDs, the internal quantum efficiency is lower than J 00% (and it decreases with aging) and the absorption in the active




layer is often considered as parasitic. Therefore, thin active layers (homogeneous or comprised of multiple wells) are often preferred. An optimum activelayer thickness involves a trade-off between active-layer reabsorption and electron confinement and depends on the emission wavelength (Gardner et al, 1999).


The design of high-brightness LEOs is aimed at the high efficiency of light extraction and uses the escape-cone concept and low-parasitic-loss approaches (Stringfellow and Craford 1997). Cost pressures (Craford 1996) result in diverse schemes for commercial high-brightness LEOs utilizing different material systems. In this section we discuss the designs of practical AIGaAs, AIGalnP, and AllnGaN high-brightness LEDs that are manufactured using conventional planar rectangular technologies.

5.2.1. AIGaAs Red LEOs

Steranka (I 997) described the photonics of practical AIGaAs LEDs. As we mentioned in Section 4.2.2, the main advantage of an AIGaAs/GaAs materials system is a very small lattice mismatch (lattice constants ofGaAs and AlAs differ by less than 0.2% at 25°C). This makes feasible the growth of high-quality AIGaAs films on GaAs substrates. Thick and transparent layers (with high AI content) that are required for opening the lateral escape cones are easily produced using liquidphase epitaxy (LPE) (Rupprecht el al. 1967).

Starting with the introduction of a thick window layer to improve light extraction in a single-heterostructure (SH) AIGaAs LED by Nishizawa et al. (1977), the quantum efficiency of the device was gradually increased (Varon eta!' 1981, Nishizawa et al. 1983). Although an SH-AS version of the LED is still available, it exhibits moderate performance comparable to that of a red-filtered incandescent lamp. Implementation of a double heterostructure (DH) and transparent substrate by Ishiguro et at. (1983) and Cook et al. (1988) led to 21 % external quantum efficiency (Ishimatsu and Okuno 1989). Since the internal quantum efficiency in AIGaAs is - 80% (Steranka et al. 1995), this value for the external quantum emciency is close to the highest value obtainable in a six-cone design. Therefore. the design of AIGaAs red LEOs has undergone no significant improvements ~ver since.

Figure 5.2.1 shows two basic designs of commercial high-brightness AIGaAs LED chips with a DH active layer. The active layer is sandwiched between the wide-bandgap cladding layers. A version of a OJ-I-AS LED is depicted in Fig. 5.2.1 a. Figure 5.2.1 b presents the design of a DH- TS LED. In the latter case, the GaAs substrate is removed by selective etching, and a patterned back contact is deposited.





30 11m p-Alo.75Ga025As Cladding Layer

2 11m -.1===========9 +- p-Alo.37GaO.63As Active layer

10 urn n-Alo.7sGaO.2SAs Cladding Layer

200 11m

n-GaAs Substrate

(b) AuGe Top Electrode


30 um n-Alo.soGao20As Cladding Layer

2).l.m .-..1============9,__ p-Alo3SGa06SAs Active Layer

120 11m

p-Alo8oGao2oAs Cladding Layer (Transparent Substrate)

I~ ~ ~I

~ AuZn Patterned Back Electrode


l ; --- GaAs Substrate (Removed)

FIG.5.2.1. Typical chip structures of high-brightness AIGaAs double-heterostructure LEOs: (a) absorbing substrate (DH-AS) LED; (b) transparent substrate (DH-TS) LED (After Steranka 1997.)

The attainable thickness of the high-quality upper window layer is about 30 urn. This may be insufficient for complete opening of the upper lateral semicones. Other losses occur because of shielding of the vertical escape cones by the contacts. Photon absorption in the ohmic region, produced by metallization, is due mainly to the amorphous nature of the crystal in this region (Lee 1998a). The shielding extends over a significant area, since the top contact must be at least of 80 tol 00 urn in diameter for high-speed wire bonding and the patterned back contact covers about 30% of the surface. With these losses taken into account, our estimate of the extraction efficiency in a TS LED for a typical chip width of 210 j.tm is about 19%, which is somehow lower than that observed. TI,e actual light extraction efficiency is probably increased due to the randomization of photon trajectories at the chip walls formed by sawing (Schnitzer et al. 1993a, Lee 1998a).



Since the radiant efficiency in AIGaAs decreases for wavelengths shorter than 700 nm (because of the proximity of the direct-gap to indirect-gap crossover), the highest luminous efficiency is achieved around 650 nm (Nishizawa et al. 1983). At this wavelength, the 21% external efficiency yields a J5-lm/W performance. Commercial AIGaAs LEDs exhibit somewhat lower performances, around 10 Im/W (Craford 1997). which are still three times higher than that of red-filtered incandescent lamps.

5.2.2. AIGalnP LEOs

The (Al-Ga l-x)O.5InOSP alloy, which is lattice matched to GaAs and exhibits a direct bandgap in the range 1.9 to ~2.3 eV, is the most favorable material for red to yellow-green high-brightness LEDs. However, AIGalnP LEDs employ photonics different from that of AIGaAs LEDs (Kish and Fletcher 1997, Yanderwater et al. I 997b). The reason is that LPE or vapor-phase epitaxy (YPE) suitable for the manufacture of thick window layers are incompatible with the growth of AIGalnP alloys. Metalorganic chemical vapor deposition (MOCYD), which is the mature epitaxial technique for AIGalnP growth, is unable to provide thick enough layers. After the first high-brightness A1GalnP LEDs were reported by Kuo et al. (1990) and Sugawara et al. (1991), these technological constraints have led to branching of the subsequent chip design in two principle directions. One of these directions (initially promoted by Toshiba) aims at exploiting thin-layer MOCVD technology to the maximum extent. Another direction (initiated by Hewlett-Packard) relies on MOCYD in combination with other technologies (YPE and wafer bonding) to implement thick window layers.

The first Toshiba AIGalnP LED comprised a DH grown on an absorbing GaAs substrate. The DH was capped with a lattice-matched Alo7Gao_3As lowresistivity CSL, which was transparent to emission light and also served as a WL. Furthermore, an n-type AIGalnP CBL (Sugawara et al. 1992a,b) was introduced. The CBL does not allow current to enter the DH underneath the top contact. Hence, this layer addresses two important issues of light extraction. First, shielding of the upward escape cone is minimized. Second, the lateral semicones are less truncated by the thin WL, since light generation is shifted to the peripheral area of the chip. The photonic design of thin-layer AIGalnP LEDs was completed by adopting a DBR (Sugawara et al. 1992a, 1993; Itaya et al. 1994) (Fig. 5.2.2a). We estimate that this design utilizes up to three effective escape cones and the lightextraction efficiency may reach 12% for one-pass escape (again, the actual efficiency is somewhat higher, owing to sawed walls and multiple reflections between the top surface and the DBR).

Further development of this kind of LED involved optimization of the DBR structure (Sugawara et al. 1 994b, Chang et al. I 997a, Chiou et af. 2000), an attempt to introduce a transparent contact (Lin e/ al. I 994b, Aliyu et al. 1995), and thinning the active layer using a multiple-quantum-well (MQW) structure (Chang ef al. 1997b, Chang and Chang 1 998a, Li e/ al. 1999). Quantum efficiencies


A1GalnP LEOs

n-GaAs Substrate

_,-- AuGe/Au Back Electrode

(b) Top Electrode


p-GaP Window Layer

~~§ii~~~~~~~)- ....... AIGalnP OH

211m --

--- Wafer Bond

200 urn

n-GaP Wafer-Bonded Transparent Substrate

/ Back Electrode



. . h (·AI G ) In P LEDs: (a) DHR

Typical chip structures of high-brig tness x _ 31->: 0.5 0.5 ..

LED with a current-blocking layer; (b) TS LED WIth a wafer-bonded trans-

parent substrate. (After Kish 1997.)

as high as 7% at 610 nm (orange, 24 Im/W) (Itaya et al. 1994) and 5% at 590 run (amber 26 Im/W) (Chiou et al. 2000) were reported for AIGalnP.DBR.LEDs.

The approach developed by Hewlett-Packard is based on using thick, conductive (heavily doped) GaP windows (Fletcher et al. 1991) that ar~ transpa~ent throughout the red to yellow-green spectral regions, where AIGalnP IS an efficient

light emitter. Although GaP has a 3.6% lattice mismatch to (AlxGal-x)OslnosP

I . th twork of dislocations that occurs at the interface does no~ penetr~te ayers, . e neD II D h oh-quallty into the DH and the LED reliability is preserved. MOCV a ows . or 1.,

GaP layers of only up to 15-fllll thickness to be grown over AIGalnP: The resulting WL is too thin to open the lateral escape cones comp~etely. The IInplcmentation of combined technologies led to the elimination of this .shortcom1l1g. Following the MOCYD growth of an AIGalnP DH on the absorbing GaAs substrate, a



GaP WL!C~L as thic~ as 63 11m was grown by hydride vapor-phase epitaxy (HVPE). This resulted III light extraction through three escape cones (Fig. 5.1.2c) and LEDs With an external quantum efficiency of 6% (-20 Im/W) in the range 59() to 620 nrn were fabricated (Huang et al. 1992).

Kish et al. (I ?94.a) im~lemented a thick GaP TS in order to exploit the rest of the escape-cone f~mdy. Figure ~.2.2b shows a typical design of a practical sixcone A1GainP chip. The absorbmg GaAs substrate is removed from the urown hybrid AIGalnP/GaP structure by means of conventional selective chemical etchrng, and a transparent GaP wafer is fused instead to the revealed AIGalnP DH. The wafers are ?Onded by annealing at elevated temperatures under applied uniaxial pre~sure (Liau and ~full 1990) '. In this way, optical coupling of the wafers is a.chleved. Ho:vever, In:plementatlOn of a reliable interface with low electrical resrstance required considerable effort. Matching of the orientation of the b d d

f (whil . on e

sur aces w I e simultaneously maintaining rotational crystallographic alignment

of t~e waf~rs) was shown, to be of crucial importance for the electrical properties req.U1red. Irrespective of lattice mismatch between the AIGalnP and GaP (Kish et at. 1995, Vanderwatcr et al. I 997a). Bonding of large-diameter A1GalnP/GaP ~tructures With mechanical strength exceeding that required for .. manufactUring LED chips was adopted for large-scale production (Hofler et at. 199~: ~aranowski et al. 1997). The bond exhibits high reliability With the pr~Jected lifetime oft~e LED in excess of 120,000 hours for degradation do:vn to, 70 Yo of the average initial iigh; ou~put (Ki~h et al, 1996). The quality and ~llIfOnnlty o,f b_ondmg were characterized via scanning acoustic microscopy, white light transmlss~~n measurements, full-wafer mapping of parametric performance, and operation life tests (Tan et af. 2000).

. Due to t~e .six ~scape cones opened, the AIGalnP/GaP DH- TS LEDs exhibited luminous efficiencies (external quantum efficiencies) of 50.3 Im/W (13%) at 607 nrn ~nd 35.51rn/W (24%) at 637nm (Kish et al. 1996, Maranowski et al 1997). High-power LED lamps based On AIGalnP/GaP DH-TS structures wer~

demonstrated. A monolithic LED bar (375 x 4500 11m2) exhibited a luminous flux of 841m at 610 to. 615 nrn (Kish et al. I 994b). However, the device suffered a threefold decrease 111 extraction efficiency because of the truncation of the side escape cones and a 40% efficiency decrease due to heating, A somewhat smaller lamp with a 500 x 500 11m2 junction area with improved thermal resistance emitted 10 to 201m of flux while maintaining luminous efficiencies exceeding 20 Im!W (HOfler et at. 1998).

The versatility o.f the combined te~hnology made different kinds of lightes.cape structures feasible, A four-cone chip was fabricated by bonding a thick GaP wll1do,:,", on the top of the AIGalnP OH with the subsequent removal of the GaAs absorbing substrate (Chang et af. 1996). The thick window substitutes for the substrate and allows. one to avoid handling thin epitaxial layers. Zeng et al. (1998) reported on the elimination of the wafer bonding process in fabrication of the OHTS LED, After the GaAs substrate was removed from the AIGalnP/GaP st t

th I G P . d rue ure,

e ower a WIl1 ow was grown by HYPE. Wafer bonding of mirror substrates

(AuBe/glass: Horng et al. r 999a and Au/AuBe/Si02/Si: Horng et al. 199%) to the

AllnGaN LEOs


top of the MOCVO-grown AIGaTnP OH with subsequent removal of the GaAs substrate was also demonstrated.

Of recent improvements in planar rectangular AIGalnP LEDs, the most significant is replacement of the DH by a multiple-well heterostructure (MWH) that resulted in reduced thickness of the active layer (Chang and Chang I 998b, Gardner et al. 1999). The MWH- TS LED exhibited luminous efficiencies (external quantum efficiencies) of74 lm/W (24%) at 615 nrn and 54 Im/W (32%) at 632 nm that to date are the highest values reported for escape-cone planar rectangular LEDs.

Record-performance A1GalnP LEDs are fabricated using a nonrectangular device shape, These LEDs are descri bed in Section 5.3.1.

5,2.3. AlinGaN LEOs

The InxGal-xN alloy has a direct bandgap which varies from 1.9 to 3.4 e V, depending on the In molar fraction, whereas Al-containing nitride alloys can be used for wide-bandgap cladding layers. This allows one to fabricate high-brightness LEDs that cover the spectral range from red to near U V. At present, an AlInGaN materials system offers the most efficient LEDs in the blue-to-green region. AIInGaN-based blue, violet, and near-UV LEDs are indispensable for the fabrication of white LEOs, which operate by radiation down-conversion in phosphors (see Chapter 6).

As discussed in Section 4.2.4, for many years, development of group III nitride materials was hindered by the lack of a suitable substrate. In spite of this, a mature technology for MOCVO growth of nitrides over substrates that are mismatched in lattice constants and coefficients of thermal expansion was developed by using low-temperature buffer layers. The substrate used most exten-

sively is sapphire (Ah03). Light-emitting AlTnGaN structures have also been grown on 6H-SiC (Kuga et al. 1995) and spinel (MgAI204; Kuramata et al. 1995, Khan et al. 1997). Fortunately, all these substrates are transparent to the light generated in blue and green LEOs.

Since the transparent substrate is the inherent component of most AlInGaN light-emitting structures, the very first high-brightness blue AllnGaN OH LED introduced by Nichia (Nakamura et al. I 994a) featured an efficient light-extraction design, The photonic design underwent minor improvements in subsequent development of blue-green DH LED (Nakamura et al. 1994b), blue-violet, blue, green, amber, and red single-quantum-well (SQW) LEDs (Nakamura et al. I 995a,b,c, Mukai et at. 1999), and blue and green MQW LEOs (Koike et al. 1996, Kozodoy et al. 1997, Lester et al. 1998) on sapphire substrate. Figure 5.2.3a shows a typical chip design for a practical AilnGaN/A120J LED. The thin emitting structure (OH or SQW) is sandwiched between n- and p-GaN contact layers. A thin transparent-metal layer is used for current spreading. This design is similar to that of an A1GaAs TS LEO (Fig. 5.2.1 b), the difference being that the




Ni/Au p-Electrode

"'" / Transparent Metal (Au/Nil

r"'""' ..... _....:/:....-, / p-GaN Contact Layer ~~~~~~~} .-- InGaN/AIGaN OH, SQW or MQW Structure

I y TilAI n-Electrode

,_----------I n-GaN Contact Layer

'" Buffer Layer

Sapphire Transparent Substrate

0-51lm 0151lm 41lm

-100 urn


...------ Au Top Electrode

Contact Layer


Contact Layer

_._ Shorting Ring

Insulating Buffer Layer

SiC Transparent Substrate

)- InGa
...._____ I----
........._.,_. .....-....0= N,j Back Electrode

FIG. 5.2.3. Typical chip Structures ofTS AlinGaN-based LEOs: (a) AlinGaN/A1203 LED (after Nakamura and Fasol 1997); (b) AlinGaN/SiC LED (after Edmond et at 1997b) -

~-elec~rode is exposed upward (by means of etchi ng a mesa) because of the insulating substrate. However, the TS concept, which was described in Sec-

tion 5.1. I, is to be reconsidered for AlinGa NI Ab03 LEOs (Lee 1998b) as d i s-

cussed below. '

As menti oned in Section 5 _1.1, the extracti on effi cien cy for AI! nGaN LEDs grown on TS is much higher than the efficiency for AIGaAs and AIGalnP LEDs First, the refractive index of AlInGaN materials (ns "" 2.5 ) is smaller than that of GaAs and GaP Therefore, about 9.2% of the photons em i tted can escape through the upward cone with the apex of 2BcI (ns' ne) "" 740 (see Fig. 5.2.4). Second, the transparent sapphire substrate has even a smaller refractive index i n, = 1.78 ), which satisfies the condition ne < n; < n,. This results in a larger apex of the

AlinGaN LEOs




J ~/


FIG. 5.2.4. Escape cones in an AlInGaN~based LED chip 011 sapphire substrate. The apex of the upward cone is 74° and the apex of the inner downward cone is 90°. The downward subconical structure with the ring base (inner apex 74° and outer apex 90°) supplements the upward cone to a full escape cone with an apex of 90°.

downward cone 20c2 (11,_,11/) "" 90° . A new feature of this asymmetric system is that photons that undergo total internal reflection from the upper surface within the range of angles eel < e < eel can escape to the su bstrate through a subconical

structure with an outer apex of 20c2 and an inner apex of 20c1, as shown in Fig. 5.2.4. The upward cone and the subconical structure supplement each other; thus a second full escape cone with an apex of 20c2 occurs. It should be noted that the photons escape from the sapphire to the epoxy through a downward cone and four lateral serniconcs with apexes of 20<"3 (n;, nJ "" I 15° (not shown in Fig. 5.2.4). These five cones overlap completely, so that all photons may escape from the sapphire to the epoxy.

Owing to thin semiconductor layers used in this structure, Fresnel losses in the vertical-cone chip arc negligible, since the majority of the photons reflected may escape on the second and third passes: Consequently, the light extraction effieiency is given by two coupled (i.e., lossless) cones with an apex of 20c2 and Fresnel losses for the epoxy-air interface:

170p! = 4n" 2 [1-~I-(n/ln,.)2J""O.29. (n" + I)


This efficiency is larger than that for a six-cone design in GaAs/GaP-based chips. Even higher efficiency is expected if some light can escape from the semi-



conductor structure in the lateral direction (Lee 1998b). However, this path requires multiple (-100 in average) reflections within the structure Even for active layers of -30 A thickness, the absorption losses are too high for this path to contribute.

Another type of practical AlinGaN LED structure is grown over a 6H-SiC substrate (Den Baars 1997; Edmond et al. 1997a,b; Adivarahan et al. 2001). SiC has the advantage of smaller lattice mismatch, and the use of SiC substrates leads to improved quality of the nitride materials. However, the AlInGaN/SiC LEOs (Fig.5,2.3b) might have somewhat lower light extraction efficiency than the AlinGaN/AI103 LEOs. The reason is that the refractive index of SiC (n, = 2.7; Ninomiya and Adachi 1994) is higher than that of AlInGaN, Almost all the downward-directed emission from the semiconductor structure (50% of the total emission) may penetrate into SiC substrate after some multiple reflections (no total internal reflection occurs). Actually, the device operates in a four-cone regime as shown in Fig, 5 .1.2d. However, the three cones in the SiC TS (one downward cone and four lateral semicones) allow only 11,5% of the photons emitted to escape because of the high refractive index of SiC. With the upward escape cone (coupled with the downward one) taken into account, the total light-extraction efficiency is about 0.21, which is somewhat smaller than that of AlinGaNI A120) LEOs.

The reported luminous efficiency (external quantum efficiency) of topemitting devices on a sapphire substrate is 7lmIW (14%) for blue (465 nm), 60lm/W (11%) for green (525nm), and 21lm/W (4,5%) for amber (595nm) (Mukai et at. 1999). Even higher efficiencies of AiinGaN/AJ203 LEOs were achieved by exploiting the photonic configuration depicted in Fig. 5.2.4 to the maximum extent. As can be seen in Fig. 5.2.4, a major part of emission in the AilnGaN/AI103 LED is directed downward, Hence, it is reasonable to flip the chip to reduce losses caused by the built-in reflector contained in the plastic dome. Mass production of flip-chip (FC) AllnGaN LEOs was reported by Koike et at. (2000). Blue and green AllnGaN FC LEOs are claimed to exhibit 20% external quantum efficiency. Wierer et al. (2001a,b) demonstrated a high-power AlInGaNbased Fe LED with a large emitting area (0.70 mml) The device (see Fig. 5.2.5) emitted at 435 nm and exhibited 21 % external quantum efficiency and 20% radiant efficiency with a record light output power of 400 m W.

Despite lower light-extraction efficiency, AIGaN/SiC LEOs exhibit perform-

ances that are similar to that of AIInGaN/AI]03 LEOs or even higher. Typical external quantum efficiencies of commercially available blue AIGaN/SiC LEOs are 20% (~15% radiant efficiency).

The present performance of AlinGaN LEOs is not to be considered as ultimate, since the possibilities of an increase in the internal quantum efficiency in AllnGaN have not been exhausted. Furthermore, the light-extraction efficiency can still be improved, since the upper transparent window with two additional escape cones has not yet been exploited. A lot of techniques are shown to possess potential for improvement of the photonics in AllnGaN LEOs. For instance, thick upper WLs may be grown by HYPE (Usui et al. 1997, Smith et at. 1998) in combination



n-GaN Contact Layer

1- __ +-- --t- __ -1!lnGaN/GaN

Multiple Quantum Well

p-GaN Contact Layer ~iiiij:;::;;- Solder


FIG. 5.2.5. Schematic cross section of high-power flip-chip AllnGaN-ba~ed LED. (After Wierer et al. 200 lb.)

with MOCVO. Epitaxial lift-off and bonding of AlinGaN light-emitting structures to host substrates have been explored (Floyd et at. 1998; Wong et at. 1998, 1999; Song el al. 1999a,b). Incorporation of DBR structures in the devi~es was demo~strated by Song et al. (1999b) and Nakada el al. (2000). Further Improvement III the transparent contacts (Margalith et al. 1999, Sheu el al. 1999, Huh el al. 2000) and introduction of epitaxially laterally overgrown GaN substrates (Kato et at. 1994, Mukai et al. 1998a, Nakamura 1999) was performed. A Si02 CBL was inserted underneath the p-electrode of InGaN/GaN MQW LED to improve the lightextraction efficiency by reducing the total area of the top contact (Huh el al. 2001). To completely eliminate the top semitransparent electrode and highly ~esist!ve p-AIGaN cladding layer, Takeuchi et at. (2001) introduced a tunnel junction

grown on InGaN/GaN MQW.

New possibilities are offered by molecular-beam ep~taxy (MBE) t~chnology (Riechert et af. 1997; Grandjean et al. 1997, 1998; Grandjean and Massies 1999). New substrates for growth of IlI-N materials are still being sought. Bour el at. (2000) have demonstrated polycrystalline LEOs deposited on quartz substrates. An improved quantum efficiency in LEOs is offered by epitaxial growth of GaN on

tetragonal LiAIOl substrate in a nonpolar direction (Waltereit et al. 2000).


Most practical LED chips are made by planar technologies and employ rectangular geometry. Solution of the light-extraction problem in such chi~s relies on ~he concept of escape cones, discussed above. Because of geometncal constraints, the quantum efficiencies of planar and rectangular LEOs are unlikely to exceed values far above 30%. In this section we consider nonrectangular and non planar ap-




preaches to light extraction, Both conventional and advanced t h . .

cussed E t I .ec ruques are dIS, . ven ec ml~ues regarded as obsolete or no! suitable for lightin a lica-

nons ~re reviewed, since further boosting of LED lurnin ffici g pp ,

recycling of some of these ideas. ous e iciency may require

5.3.1. Shaped Chips

A straight~or.ward wa~ .to make all the emitted photons arrive at the semiconduct surface within the critical angle is to use a spherical chip The id ,or ment d ' ' .. . e I ea was IlTIpleTh e ,)~ars ago 10 th~ first hIgh-power infrared LEDs (Carr and Pittman 196"')

sha~ee~l~t~n~ stI_"Uc~ure ~s f07ed on a thick GaAs wafer which is sUbsequen~~): as hi h 40~nlsP ere .y po ishing (see Fig. 5.3.la). External quantum efficiencv

Ig as Q was achieved in archaic hemispherical LEDs cooled to r id .-

trogen temperature (Carr 1965). The same technique Was applied to h. ul:UI 1:1- A IGaAs LEDs by Kurata et al (1977) A I 10 power

L ED ' h . . . n A GaAs double-heterostructure DI-J

. WIt. an efficiency of 45o/c t

(Kurata et al. 1981 . 0 a .room. tempe~a(.urc was demonstrated . . ). AI~o, a coruc LED With a hght-emIttmg structure at the a ex

was devised by Franklin and Newman (1964) Carr (1966) id d h P

t . ffici . .. consi ere t e light

ex raction e JClenCles of the sources shaped as h . h .

(W . ermsp ere, truncated sphere

ererstrass sphere), truncated ellipsoid, truncated cone, and paraboloid in term

of figures of ment and compared with those of flat structures. S

whiC~~~~er old approach is ~o .shape a. thick wafer into a parabolic reflector outcouples the ennssion and Improves its directionality (Fig. 5.3.1 b)'


Hemispherical Chip

/ / ~ Electroluminescent Structure



~ ./ Electrol"m;ne"enl Structure

ms ¥ ..-- Antireflection Coating

~ :::::<

......__ Au Coating

HG.5.3.1. (a) Hemispherical LED chip: (b) I ED chip shaped as a b I' f1

• • J , para 0 Ie rc ector.


GaAs parabolic LEDs with a chip coated with a reflecting film were reported by Galginaitis (1965). To reduce absorption losses, the parabolic wafer is made from

a graded-band GaAs l-xPx alloy. Recently, an InGaAsllnP heterostructure with an JnP substrate polished into a parabolic reflector (7 mm in diameter and 1.8 rnm in beight) was demonstrated to emit in the IR with an external quantum efficiency exceeding 60% (Gfroerer et al. 1998).

However, the polishing of semiconductor chips is a costly and inconvenient technology. A new method for the fabrication ofGaAs/AIGaAs hemispheric LEDs by formation of monolithic microlenses on the AIGaAs surface was proposed by Xu et al. (J 998). The method combines crystal growth, ion etching and steam oxidation with wet chemical removal of the oxide. Also, a nonpolishing technique for manufacturing LED structure within an integral concave mirror was demonstrated by Schmid et al. (1999). The InGaAs structure is grown by MBE and shaped to a laterally tapered disk by means of photolithography and chemically-assisted ion beam etching. Subsequently, the structure is coated with a gold mirror. Schmid et al. (2000, 200 I) reported on tapered LEDs with external quantum efficiencies around 50%.

A radially symmetric shape is not the only one that may facilitate the outcoupiing of photons from an LED. It is possible to increase the light-extraction efficiency by utilization of geometrically deformed (nonrectangular) chips with plane walls. Some polyhedron chips (rhomboidal and triangular) with parallel bases were considered by Lee and Song (1999). Figure 5.3.2a depicts a horizontal cross sec-




~Wafer Bond ""'E!E5!E!E5!~i!S!5!l5!/ ......... AIGalnP Active Layer


FIG. 5.3.2. Geometrically deformed chips: (a) photon trajectories in a rhomboidal horizontal plane (after Lee and Song 1999); (b) vertical cross section of a truncated-inverted-pyramid LED (TIP LED) (after Krames et al. 1999).




tion of a proposed rhomboidal-geometry chip with the plane deformation angle ali '# 900. The photons that travel parallel to the horizontal plane inevitably escape, since each internal reflection reduces the incidence angle by aJ,. The optimal deformation angle at which only a couple of flights are required for most photons to escape is close to ah "" 2B,.(n"nJ. If, in addition, the sidewalls were slanted similarly, most of the photons generated would find escape cones regardless of their traveling directions, Increased light-extraction efficiency (up to 120% compared with rectangular geometry) was proved by a statistical tracing of the photon trajectories. A technique for the fabrication of geometrically deformed chips by slanted sawing of wafers was proposed (Lee and Song 1999).

A practical geometrically deformed LEO was reported by Krames et at. (1999) and Holcomb et 01. (2000). The device was made from an epitaxial AIGalnP structure wafer bonded to a thick GaP substrate (see Section 5.2.2 and Fig.5.2.2b). By using a beveled dicing blade, chips with sidewall angles of 35° with respect to the vertical were fabricated. The schematic cross section of the resulting truncated-inuerted-pyramid (TIP) LED is shown in Fig. 5.3.2b. The TIP geometry is seen to improve light extraction by redirecting totally internally reflected photons from the sidewalls to the top surface or from the top surface to the sidewalls at small incidence angles.

To date, an AIGalnP TIP LED is the record semiconductor visible-light source. In the orange region (61 I nm) it exhibits the highest reported luminous efficiency exceeding 102 Im/W; high-power chip provides a peak luminous flux of 60 1m. In the red region (652 nrn), external quantum efficiency as high as 55% is achieved under dc operation. Measured luminous efficiencies for AIGalnPiGaP rectangular and TIP LEDs and conventional lamps are compared in Fig. 5.3.3.

5.3.2. External Outcouplers

Extraction of the light from an LEO chip might be facilitated by external. design, using, for example, an encapsulating dome, antireflection coatings, reflectors, or lenses. An example of an external outcouplcr is the plastic dome (Nuese et at. 1969) (see Fig. 5.1.1). Owing to the smaller ratio of refractive indices of the plastic and the semiconductor, escape cones are enhanced compared with those at the semiconductor-air interface. The closer the refractive index of the dome is to that of the semiconductor, the higher the extraction efficiency. Years ago, Fischer and Nuese (1969) proposed high ly refractive glasses (n = 2.4 to 2.9) as encapsulants to improve the extraction efficiencies of red and infrared LEDs. Despite severe thermal expansion mismatch, stable glass domes were demonstrated using glass compositions that permit strain relief by thermoplastic tlow. Quantum efficiencies as high as 32% were achieved in GaAs LEDs by using high-index-glass encapsulation (Ladany 1971). However, because of a variety of optical, thermal, and mechanical limitations of glasses, most practical LEDs arc encapsulated into transparent plastic with a somewhat smaller refractive index (-1.5). For instance, bisphcnol-Avbased epoxy casting resin transferred thermally to highly transparent



1000 .---..--..,-..,....--.--.---r---r-r-"'-""-"'-l


... ..



... ,



High-pressure Sodium 250 W ..

Fluorescent 32 W ..

CFL15W" ... High-pressure Mercury 250 W ..

\ •

\ \

Tungsten 60 W .. Amber-filtered .. Tungsten 60 W





• ,

\ •

Red-filtered ..

Tungsten 60 W

, •

, ,

1 550



FIG. 5.3.3.

650 700 750 Peak Wavelength (nm)

P/GaP TIP LEOs standard AIGalnP/GaP

t~~~~o::d ~~i~~~;~i~~ll~~S.~I~a~~hC~ envelope cu~e shows luminous effi-

ciency for 100% radiant efficiency. (After Holcom b et al. 200 I .)

polyester networks is developed for LED encapsulation (Bogner el al. 1?~~'0). T:~,

( _ 1 57) exhibits stable properties and an operation I e ime

epoxy ne - . .' d f high ower LEOs is based

100 000 hours Another approach, wh. ich IS propose or I -p ofl I '1998)

,. . I "h d" lens (Ho er et a . .

on using a "soft" gel encapsulant With an externa ar . e The dome

Improved extraction efficiency is not the only functhlOn o!.t~~ dO~t~ a narrow

. I sha ed into a spherical lens to focus t e ra ra 1011 I

~ com~o~'g\_bri~htness LEOs, which utilize more than an upward escape cone,

earn. n 1 1976) is built into the encapsulant.

an external reflector (Bergh and Dean . . 1 made from photoresist.

her t e of external outcoupler IS a micro ens . .

Anot yp . d . I h hotoresist lenses might be fabricated

Large arrays o~ LED chips. eappe Wit ~~~ec f~rication uses the photoresist reflew using conventIOnal photolithography. . . I d i heated to tempera-

I 1991) A ircular photoresist IS an IS

technique (Hutley et a . .' CI . f tension to draw the resist into

tures above the reflow temperat~re, asl!~i;jl~;sS~~eac~icrolens. Since the refractiv.e a truncated-spher~ s~ape. Coo 111~ lower than that of a semiconductor), the rminde. x of photoreSist IS about 1.6 (i.e., I' the light extraction effi-

di h . 'on but a so mcreases -

crolens not only re irects t e errussi t d shroom-shaped photoresist mi-

eiency. Heremans e/ al. (1997) demonstra e mu

crolenses integrated with ~EO arrays. t ti efficiency relies on antireflection

An old approach to increasmg ex rae Ion . . rformance of

coatings fabricated using transpar~nt d~electrics. A 35% II1c~:~se (I~~~nreich 1963, GaAs LEOs coated with a thin SIO layer was repo

Yamamoto and Kawamura 1966).



5.3.3. Nonresonant Cavity LEOs

The principal problem of light extraction from LEDs is the inability of photons, propagating in particular directions, to escape from the semiconductor chip. Although multiple reflections may facilitate the escape, the length of the path, which is required for acquiring an appropriate incidence to a surface, might be too large to avoid absorption. One possible solution to this problem relies on randomization of the propagation directions, while maintaining a short path for the photons. A simple approach was introduced by Bergh and Saul (1973), who proposed increasing emission from an LED by making its surface rough. It should be noted that conventional technology of LED fabrication usually involves dicing by sawing (Williams and Hall 1978). The sawed walls of the chip are rough and cause some degree of photon trajectory randomization. In rectangular planar LEOs, this effect is known to contribute to the light-extraction efficiency to an extent that compensates approximately for internal losses due to absorption (Lee 1998a). By using optimized surface texturing, Linder et al. (200 I) achieved wall-plug efficiencies exceeding 10% in AIGaInP absorbing-substrate LEDs.

A light-emitting device that implements this idea to a large extent employs a randomly nanotextured surface for chaotization of the photon trajectories and a thin semiconductor layer mounted on a mirror (Schnitzer et al. 1993b). This kind of LED is called a nonresonant cavity (NRC) LED Windisch et al. (l998a, 1999b) used NRC LEOs for high-speed parallel short-distance optical interconnects. Figure 5.3,4 shows a schematic structure of the MBE-grown NRC LED. The device consists of a 120-nm GaAs active layer sandwiched between 130-nm AIGaAs layers. A current aperture made of oxidized AlGaAs is introduced to reduce the generation of light below the top contact. The semiconductor structure is separated from the GaAs substrate and van der Waals-bonded (Yablonovitch et al. 1990) onto dielectric-coated Au mirror evaporated onto a Si wafer. The top surface is texturized using a technology known as natural lithography (Deckman and Dunsmuir 1982). The surface is coated with polystyrene spheres with a diameter of 300 nm, which act as a mask for chemically assisted reactive ion beam etching to a

Oxidized AJGaAs


Si Substrate

Polyimide/Au Mirror

FIG. 5.3.4. Outline of an NRC LED. (After Windisch et al. 1999b.)



. . et al 1998b). The etching results in the for:nat~on of

depth of 160 nm (Wmdls.ch. ·.11 which lead to angular randomizatIOn of randomly dist~ibuted .cY~JIldncal PI2~~OC) resented details on optimizat~on of the the scattered light. W lIld.lsch et al. ( ( h . P of the spheres, the distribution of the

amelers of the texturmg process t e size

~;~eres on the s~rfac~, and the et~~~:~:e~t:;. are internally reflecte.d at the top

As shown In Fig. 5.3.4~ P d h ge the angle of propagation. After re-

db the pIllars an C an . c.

surface are scattere .y ulti le chances to escape. The escape 1.5 last

flection at the back mirror, t~ey get m (Jeremans et al. 1999). GaAs IR (865 11m) enough and no photon recycill1g oc~urs f .... O t 54% (Schnitzer et at. 1993b, LEDs exhibit external efficlen~~~sa ~ . ~ind~sch et al. (2000a,b) dem?nstrated Windisch et al. I 999a, 2000b,c, 2. ,) f NRC LED bv using spherical glass

improvements in the quantum efficiency 0 an ,

microlenses. . . d· ible InGaP/AIGalnP LEOs emitting at

Rooman et al. (2001) mtroduce VISl 0

655 nm, with external quantum efficiency of 31 Yo.



. xtract Ii ht that is generated isotropically.

Up to now we hav~ discussed how to e 5 ro: ating chaotically from the active The main problem IS to push the photon, p / ~ape-cone engineering, thin-layer layer, into narrow escape .cones. B_y mea:s ~o es ically 30% might be fabricated. devices with light-extractIOn efficle~ghCY bP hr:P d using more sophisticated deHigher efficiencies of 50 to 60% ml t e ac ieve

signs. .. .. sotro icaH generated photons into escape

However, instead of guiding I. . P Y s itself so that the generation of

.. the emissron proces

cones, it IS pOSSible to ,govern. . d des and suppressed in other mod~s

Photons is enhanced In certam deSire. d DmoL R 1999) The feasibility of this

. 1996 Krauss an e a ue .

(Weisbuch and Rarity , h t m mechanical rate of spontaneous

approach follows inherently from \ ed {u~eu el~ctromagnetic field (Dowling and emission in a two-level system coup e 0

Bowden 1992):

_ 21! l( f 1 I H \ i 0 k )12 p(tW)k' r) .

Wfi - - . 'k mt l '> I I Ii


. h h t ns of wave vector

Here Ii 0 ) is the initial state of the system without t e P 0 0 . .

, k h H is the mteractlOn

k (I 1 \ is the final state after the release of a paton, mt. .

, , k to the electromagnetic field, (Uk 15

part of the Hamiltonian that couples the.system (to. ). the density of opti-

, f the photon emitted, and p "alk, r IS

the angular frequenc) 0 If f g·ven desizn the mode density

cal modes (photon states) at that frequency. or a I ( )_ O2'/ 2.3 the emis-

.. rn p ca - (;)k 1! C ,

. . ic mode density In vacUU 0 k

differs from the isotrop E (_ 4 1) improved light extraction from an

sion rate is altered. As seen from q.:l.. ,



LED might be achieved by increasing the density of the optical modes coupled to the outside modes and decreasing the density of the outside-uncoupled modes.

At present, photon density-of-states engineering is being applied to LEOs utilizing a variety of techniques. One technique is based on spontaneous emission enhancement due to cavity modes as suggested by Purcell (1946). Purcell enhancement gave birth to the resonant cavity (RC) LED (see Section 5.4.1). Another technique employs the COupling of spontaneous emission to surf(lce plasmon-polariton (SPP) modes that occur at the semiconductor-metal interface (Section 5.4.2). The third lechnique relies on photonic crystals (Yablonovitch 1993, Joannopoulos et al. 1995) that are optical analogs of conventional sol id-state crystals. A photonic crystal approach which offers light-extraction efficiencies up to 100% is diSCussed in Section 5.4.3.

5.4.1. Resonant Cavity LEOs

Optical mode density may be altered by interference due to the optical environment (Benisty et al. 1998a). A simple example is a mirror placed close to an emitter. If the wave emitted outward from the mirror and the reflected wave interfere constructively within the entire escape cone, a twofold increase in light-extraction efficiency might be achieved. By contrast, destructive interference results in inhibition of emission perpendicular to the mirror plane. LEDs with active layers coupled to metal mirrors were demonstrated by Deppe et al. (1990), Hunt et al. (l992b), and Dill eta/. (1999).

The RC LED employs a much stronger modified photon state density, which is caused by multiple-beam spatial interference in a small Fabry-Perot cavity (microcavity) (Schubert et a/. 1994). The cavity is formed of two coplanar reflectors separated by a wavelength-order distance. Along the optical axis of the cavity, propagation of optical waves is allowed only at the fundamental resonance and its hannonics. At the resonance wavelength, the emission along the optical axis is enhanced by a factor


Where Rj and R2 are the reflectivities of the front and back mirrors, respectively, and c; is the anti node enhancement factor (Hunt et al. 1993). The latter depends on the position of the active layer with respect to the standing optical wave inside the cavity (c; = 2 for the exact antinode location and q ~ 0 for the exact node

position). Typically, the radiative lifetime inside the LED cavity 'cav differs from the free-space lifetime T by less than 10% (weak-coupling regime).

The first RC LED demonstrated by Schubert et al. (1992) consisted of an MBE-grown GaAS/AIGaAs structure sandwiched between a semitransparent


d AIAs/AIGaAs OBR structure (R2 =0.99).

. (R - 0 9) an an .

Ag/CdSnOx nurror 1 - . ., ith the thin GaAs active

. h d t ture IS about 11<, WI .

The thickness of the sandwic e .s r~c The top-emitting device produced a line

layer positioned in one of the antino les. idtl of 17 meV that is much smaller than

2 . th a spectra WI 1

eaked at A. == 86 nm WI .' RC LEDs utilizing InGaAsP

P L t bottom-emitting 1

that for conventional LEDs. a er, I G A IAIGaAs (A. == 940 nm; Hunt et a.

/ 199?a) and n as. . .

(1==1.3J1ll1; Hunt er cr/. - depi tsanoutlineofabottom-emlttlllgunen-

d F' e 5 4 1 a eptc d. an

1993) were develope. igur .. 994 1996) The emission generate 111

capsulated RC LED (SChu.bert et al. 1 DB'R (R ., 0.9) and the GaAs substrate

apes through the 1

MQW structure esc (R _ 0 96) serves

. ti film The silver (Ag) reflector 2 - .

coated by a Zr02 antireflec ion I ·.h . The entire structure is covered

A A ap protects t e mirror. I

as an ohmic contact. n II c . I r and an Au heat-sink/contact ayer are

by borosilicate glass. A TI bonding aye

deposited over the glass.


p-Electrode ~


InGaAs/AIGaAs A-Cavity

~~~~~~~~~~il--- n-Electrode


GaAs Substrate


FIG. 5.4.1.

. IDHR bottom-emitting resonant-

. tu of (a) metal-mirror .. C LI-J)

Typical chip struc re . (b) DIlR-DAR top-ermrnng R :

cavity LED (after Schubert et at 1994),

(after Carlin et al. 2000).



Schubert and Hunt (1998) have shown that th . . .

able, with an expected lifetime of 78 000 h e planar structure IS hIghly rei i-

. t ity) Th ' ours (for a 50% decrease' '.

In ensuy . e enhancement factor calculated f th In emISSIon

60 at ,; =: i. 7. However the ex t d . or . e structure described is around

, pee e increase In total extract" ffici .

about one order smaller, since it depends not onl ion e rciency IS

factor but also on the ratio of widths f th . Y on the resonance enhancement sion spectrum (in semiconductors th 0 t~ cavrty resonance and the natural emisnot increased at ali). Actually the'e tIS rat' 10 caffin be so small that the efficiency is

, x rae Ion e iciency might b

cause of field penetration into the fl (H e even smaller be-

with the RC LED is a strong depen~en:~t:;th unt " al. ~ 992a~. Another problem due primarily to the variation of . d e ermssron mtensity on temperature, 1996) semlCon uctor bandgap energy (Schubert et at

. Benisty et al. (1998a) considered the Ii h _ . . .

rmcrocevines. The constructive interf g ~ ~xtractlOn efficiency III metaJ-DBR

highest possible efficiencies are pre~~c~~~S f~:ng 0~1t 15 to 50% of the light. The moderate reflectivities. Competition between e ~ptlmlz~d DBR structures with leaky modes (escaping through a OBR tsid hmlsslOn into Fabry-Perot modes. (propagating laterally between the rni ou Sl e t e escape cone), and guided modes Benisty et at (1998b) considered the I~~;~) was analyze~. '" a subseq.uen! paper, accounting for photon recycling L' ~t nnan.ce of rea.lIstl~ planar mlcrocavi!ies

350/,' III P .' Ig -extractIOn effiCienCIes of 45% in ITI-N

o In - and ~50% 111 III-As! AI 0 . '

main channels of loss were found to xb: sy~tems were shown to be feasible. The

and absorpti?n in the active layer (~IO%).emlsslon II1to the guided modes (~30%)

Theoretical and experimental eff .

DBR RC LED structures involved the ~n~o~:c~~::~e;~ per:~nn:ce of mel.alof the guided modes (Blondelle et '. cavi res or suppression 1995), and tuning the cavity reson:~c~ ~4 ~ th(;r"m: ~Ihe substrate (De Neve et al. 1 998a, D. ill eta' 1998) Th high 0 e on e e et al. 1995, Benis. ty et a},

. . e I est external quant "fi .

achieved in an MOCVO gr I G' um eli rciency of 22% was

- own n aAs! AIGaAs 5..1./4 tr .

of the guided d '. s ucture using absorption

. mo es propagatmg 111 the lateral di .

recycling (De Neve et ai 1997 B k t I I' rrection and subsequent photon

bv serri ., oc s ae eel a 1999b) This ffi .

y settmg the cavity width to an 0 t' I I f' . e ect was achIeved

. . pIma va ue a 1.5 mm

Bottom-emJttmg LEDs have some limitati (. .

stricted to the spectral region wh h on~ e.g., their operation is re-

ere t e su bstrate IS trans ) A

structure might be formed by insert' . . . parent. top-emitting

(y k mg an active layer between t DBR

o oyarna et al, 1990), as shown in F' 5 . . wo . structures

LEDs were realized in MBE-grown rnG~gA.sI1/b. Top-~mlttJng OBR-DBR RC M?CVD-grown AIGaAsl AlAs (Bockstaele As (Carlin et "': 1999, 2000) and chips consisted of a high-reflecrivn b et al. I999a) materials systems. The

DBR . . VI Y ottom n-OBR and low-ref tivi

p- . The emlttmg structure (1,1. or A./2) . ec rvrty top

Th . was sandWiched between the reflec

tors. e device, exhibited about 14% external . -

ternal quantum effi ciency of 270/, h. . q~antum effici ency, A record ex-

Go was ac ieved 1I1 a DBR-DBR RC LED . h

aAs/InGaAs cavity and native-oxide-based A '. wit

1999). AI 0 IGaA . IxO/GaAs mirrors (Wierer et al.

x y S rrurrors feature reduced penetration of tile optical field due to a



higher contrast. Because of the insulating mirrors, the device was furnished with tunnel contact junctions, making lateral electron current excitation possible.

Lett et al. (1993a) reported on the first visible RC LEDs. The device demonstrated emitted in the red (660 nm) and consisted of an AIGalnP strained quantumwell 3A.-thick active region surrounded by AIAs/AIGaAs OBRs. The structure was grown by MOCVD on a GaAs substrate. Lott et at (1993b) reported on a similar structure with AIGaTnP OBRs. Jalonen et al. (1997, 1998) fabricated structures with a lA-thick AIGaInP optical cavity active region and oxide current aperture (using an MBE growth technique). The improved device emitted at 660 nm with a wall-plug efficiency of 3% (Orsila et al. 1999). Aging tests revealed no degradation in output power after 36,000 hours of operation (Sipila et al. 2000). Highperformance 660-nm RC LED was reported for an AIGaInP/GalnP 1 A.-thick quantum-well structure with AIGaAs/AIAs DBRs (Streubel et al. 1998). TIle device exhibited the external quantum efficiency of 4.8% with output power exceeding 8 mW. A similar structure devised by Gray et al. (2000) exhibited an even higher peak efficiency of 6% at 650 nm. Dumitrescu et. al. (200 I) reported on RC LEOs emitting in the 650 to 655 nm range, with external quantum efficiency of9.5% and operation lifetime of 93,000 hours.

Introduction of OBRs and microcavities into AlinGaN structures (Song et al. 1999a,b) demonstrated that blue and IN RC LEOs are plausible, Recently, Song et at. (2000) have fabricated an RC LED based on InGaN/GaN quantum-well heterostructures. The structure comprised one metallic chromium (Cr)/ AI mirror

and one Hf02/Si02 OBR or two DBR stacks. The device emitted in the range from about 430 to 480 nrn,

At present, most potential applications of RC LEOs (such as optical fiber communications, high-resolution printing, etc.) are targeted to employ the benefits of narrow line width, improved directionality, and high brightness of these devices. RC LEOs are more reliable than vertical cavity surface-emitting lasers (VCSELs) and might be able to replace VCSELs in many applications. In particular, they are very good candidates for plastic optical fiber (POF) applications, where optical losses are lowest at 660 11111. However, with further penetration into the visible region and increased performance, they might compete for lighting applications as well. The light emission from a single facet reduces the packaging constraints and allows the monolithic integration of LEOs into arrays (Carlin et al. 2000).

Further improvements in RC LEOs are in progress. A temperature-insensitive regime was demonstrated using cavity detuning (Takarnori et al. 1999, Pratt et al. 2000). Kitson et al. (1998) proposed means for reducing losses in metal mirrors. To suppress leaky modes, Bienstman and Baets (2000) developed a novel design with a symmetric resonant cavity added to the outcoupling reflector (RC2LED).

Efforts to overcome the poor overlapping of broad natural emission spectra with narrow Fabry-Perot resonances in RC LEDs are under way. Quantum dots feature narrow luminescence lines, and are therefore suitable for matching the emission spectrum with the cavity resonance (Marzin et al. 1994, Notorni et al. 1996). GaAs planar cavities with InAs (Gayral et al. 1998, Gerard et af. 1998) and



InGaAIAs (Graham et al. 1998) quantum dots are being investigated. Visible emission from InP quantum dots in InGaP cavities is expected (Zwiller et al. 2000).

Another shortcoming of planar Fabry-Perot microcavities is that the optical field is confined along single spatia! direction. This results in guided modes and makes spontaneous-radiation lifetime difficult to control, More complete control of the emission properties requires three-dimensional photon confinement. This may be implemented in a three-dimensional microcavity-a photonic dot, Huffaker and Deppe (1995) and Graham et al. (1999) reported on suppression of guided modes by lateral confinement in apertured Fabry-Perot microcavities. The confinement of optical modes in laterally patterned DBR rnicrocavities (Gerard et al. 1996, Reithmaier et al. ) 997, Gutbrod et at. 1999), rnicrospheres (Nagai et al. 1997), microdisks (McCall et al. 1992, Mair et al. 1998), and micropyramids (Jiang et al. (999) was observed. Jin et 01. (2000a,b) demonstrated InGaN/GaN microdisk LEDs.

Ultimate implementation of a light emitter that employs cavity modes is a quantum dot in a photonic dot with coupled electronic and photonic states (Artemyev and Woggon 2000). Huang and Deppe (2000) have calculated efficiencies for quantum-dot apertured microcavity LEOs. Efficiency in excess of 80% was predicted for a submicron device.

5.4.2. Surface Plasmon-Enhanced LEOs

Another approach to outcoupling the light from a thin layer is based on coupling of the spontaneous emission with SPP modes with subsequent scattering of the SPP modes out of the structure (Barnes 1999). The SPP modes are guided optical modes that exist at the interface between a dielectric and a metal. Utilization of the SPP modes in a light-emitting device involves two basic issues. First, semiconductor emission has to be strongly coupled to SPP modes within the metal-capped layer, with a thickness on the order of a wavelength or less. In this case, emission to SPP modes is enhanced. A light-emitting microcavity structure should be used to enable maximum utilization of this enhancement. Second, means of outcoupling SPP modes from the structure should be provided.

The dispersion relation for an SPP is given by


where (J) is the angular frequency and t:;'vl and E: D are the dielectric constants for the metal and dielectric, respectively. As can be estimated from Eq. (5.4.3), the SPPs have wavenumbers larger than those of the relevant photons. Therefore, these modes are, in general, nonradiative. A periodic patterning of the metal layer is introduced to facilitate both coupling of the photons emitted to the SPP and the scattering of SPP modes out of the structure (Raehter 1988). The metal grating




of the Spp mode to the phot?n ,:",avenumber both III the matches the wavednulllb~~e former matching condition IS given by

. onductor an air.


UJ . 21r _ k: (5.4.4)

n -sll19s ±m- - spp'


. ductor 9 is the photon angle of

fi . . ndex of the semi con ,s

,.,here ns is the re racnve I .. d A is the grating period. The

,.., d t In IS an mteger, an

. id nee in the semicon uc or,

mCI e .. .

latter matching condition IS


21r UJ. 9

k +1-=na-s1n (1'

spp - A c

are the refractive index and incidence angle subjected to the

where na and 9a . A mple of such an SPP-enhanced

. I d I is an mteger. n exa I 1996)

ambience, respective y, an . . a silver grating (Kitson et a. .

emitter is that of dye molecules depo~lted ~~trated enhancement of emission frolll

Kock et al. (1988, 1990) have emo di odes and AIGaAs/GaAs LEDs via metal_surface-patterned AgfGa~s sc~::t~ed Isemiconductor structures were relacoupling to surface plasmons. mce d . of states was almost unaffecte~. tively thick, the overall pho~on ensl~y of s ontaneous emission from a thin Gontijo et al. (1999) observed direct cou~I~~se_p~sjtioned Ag film with plasmon InGaN/GaN quantum-well structure ~o . . to SPP modes was found to be larger energy around 3 eV_. spontaneo~l~:~~:~~:~~ factor or around 55. Gianordol~ et al. than uncoupled emission by an. f SPP modes can be optimized by placmg an

(2000) have shown that outcoupltng 0 .

additional dielectric layer over the metal g~atJ.ng~urface plasmon-enhanced lightVuckovic et at. (2000~ dem~nstra~e a dicaily patterned metal capping (see emitting structure with a ml~rocavlty a~a::;~~aAsiInGaAS sub-AJ2 semiconducFig 542) The cavity consisted of a . . d ] tl e middle of the structure. The

. ... . I tum well posltlOne III 1 f I s

tor wafer with a smg e quan . f I The top semitransparent I m wa

. h d b t en two Silver I ms. f

wafer was sanowic e e we d A + . on milling. Enhancement 0

patterned using electron-beam lithography ;n t ~f ~ 5 was estimated. The photo-

spontaneous emission into SPP modes by a tac or .


FIG. 5.4.2.

d LED (After Vuckovic et al. 2000.)

Surface plasmon-enhance .



luminescence intensity measured was u to 46' .

essed wafers. Such a structure can be de~el d 1.1mes higher than ~hat of unproc. with metal mirrors used as contacts. ope Into an electrolumlOescent device

5.4.3. Photonic Crystals

In structures with a periodically patterned ref ti .

. . . Hac Ive index photon xhihi

propertres Similar to that of electro . ,s e I It wave

structures can be implemented in 0 ns I~ Crystal~ Call~d photonic crystals, Such

Similar 10 what is done for electronll:av:~'i or t ree ~lmenslOns (see Fig. 5.4.3).

tion Oflelectromagnetic Waves ill photonic cr~s~~~~:n~~~~:i!b~J~tyalCS'onthce PtroPt~ga-

ciproca space Brl'II' disnerei ep s 0 re-

. . ' oum zones, IsperslOn relations Bloch funcn

singularities, and effective masses (Y bl . h ]'9 wave uncttons, Van Hove

I h . a onovitc 93, Joannopoulos et al 1995)

nap otonic crystal, photon modes disa - D' . ' .

quencies and directions while th d . fPpear Or particular (forbIdden) fre-

, e ensrty a modes can in D h

lowed) frequencies and directions (John 1987 Y bl . hcrease or ot er (al-

. . , a onovrtc 1987) As a It

spontaneous emrssion can be eithe . hibi d _. resu ,

1988). A defect introduced into a r In. I ite or enhanced (Yablonovitch et al.

the bandgap (Yablonovitch et a/I:~~l~~ c7~.tal produces a localized state within emission can be supported only . ,,' rbl means that natural Spontaneous

In one ravora e mode and d '

modes. Actually, a DBR-DBR RC LED (F" 54' suppresse In other

di . I g, . .1 b) Implements the concept of

one- ~men.slOnal photonic crystal with a defect (,t or ,t/2 cavity) included. How~

ever, In this LED, the advantages of hotonic c I .

only a single dimension is used. p rysta are poorly exploited because

Two-dimensional (20) h t . .

M d di 9 P oronic crystals (Ph hal et al. 1991 Plihal d

ara u In I 91) offer more possibilities f h . ' an

Commonly the structu . f . or _p oton denslty-of-states engineering ..

, res consist 0 periodic arrays fill li .

(e.g., semiconductor pillars in air or cylindrical h I ~ pafla e cy: indrical rods Piche 1992) as shown in F' ,0 es III a ayer; Villeneuve and 19. 5.4.4. Large photonic bandgaps for all lateral direc-




L _/ .r'

,zi ~ ;J
:.:.' FIG. 5.4.3.

Simple. ex~ples of photonic crystals: (a) periodic in one . " , (bi periodic III two directions; (c) periodic in three directions. Gray ~~e~~~~, co nrs represent matenals with differe fi.... . c Joannopoulos et al. 1995.) nt re ractrve indices, (After




FIG. 5.4.4. Typical structure of2D phoionic crystals: (a) periodic array of semiconductor pillars; (b) air rods in a semiconductor layer.

tions and for any polarization are obtained in arrays of triangular (hexagonal; Meade et al. 1992) and honeycomb (graphite; Cassagne et al. 1996) symmetry provided that the refractive index contrast is high. A variety of techniques for fabrication of 2D photonic crystals already exist (lithography and etching, electrochemistry, vertical selective oxidation, epitaxial lift-off, etc.; see the review paper by Krauss and De La Rue 1999). Semiconductor-based 2D photonic crystals with the bandgap in the near-IR range were demonstrated in GaAs/AIGaAs (Gourley et al. 1994, Krauss et al. 1994) and InGaAs/lnP (Baba and Koma 1995, Baba and Matsuzaki 1996) materials systems. In the visible, a 20 photonic bandgap was observed in nanochannel-glass structures (Lin et al. 1996).

Light extraction from a thin slab of 2D photonic crystal was considered by Fan et al. (1997a). The extraction efficiency was predicted to exceed 90% due to the elimination of guided modes. A schematic design of the proposed lightemitting device that exploits a 2D photonic crystal (Baba 1997, Fan et at. 1997b) is shown in Fig, 5.4.5. The chip features a semiconductor slab, which contains an active layer sandwiched between cladding layers. To inhibit lateral emission, a triangular lattice of air holes is patterned into the slab. The slab may be placed Over a OBR structure that directs emission only through the top surface. Enhancement and suppression of spontaneous emission in a thin-film lnGaAs/lnP photonic crystal was demonstrated by varying the period of the hexagonally arrayed holes (Boroditsky et al. 1999b). Similarly. the device may be composed of an array of dielectric rods in air. Baba et al. (1999) demonstrated the latter version of a 2Dphotonic-crystal emitter by fabricating a honeycomb array of InGaAsP/lnP micropiHars. The emitter exhibited a more than tenfold increase in photoluminescence extraction compared with a planar wafer.



20 Photonic Crystal

Active Region

FIG. 5.4.5. Outline of a semiconductor light-emitting device comprising a 20 photonic crystal and a DBR structure. (AJ1er Saba 1997 and Fan et al. J 997b~)

One of the limiting factors for these nanometer-scaled devices is surface recombination, since patterning of the semiconductor layers (such as shown in Fig. 5.4.4) results in considerable enhancement of the surface-to-volume ratio. Boroditskyet al. (2000) investigated the surface recombination velocity in ITI-V candidate materials for nanostructure LEDs. A gallium nitride system and chemically passivated InGaAs were shown to possess relatively low surface recombination velocity (_104 cm/s), while an AIGalnP materials system exhibited a value that is an order of magnitude higher.

Boroditsky et al. (J 999a,b) reported on a promising thin-slab LED design which uses highly efficient external scattering of light by a 20 photonic crystal The light was generated in an unpattemed InGaAsllnP heterostructure of 20-l-lm width. The generation region was surrounded by a photonic crystal, which was a thin film patterned by a hexagonal array of air holes. The photonic structure did not influence spontaneous emission but improved light extraction via coherent scattering of internally trapped light. The device exhibited photoluminescence yield with 70% quantum efficiency. Ochoa et af. (2000) proposed a similar structure surrounded by circular concentric deep-etched trenches that acted as a 2D Bragg mirror. A sixfold enhancement in light extraction of an InGaP/TnGaAs LED was achieved by using a 2D photonic crystal inside the upper cladding layer (Erchak et at. 2001).

Fabrication of 3D photonic crystals with a lattice constant in the optical. range is being developed. For example, a face-centered-cubic photonic structure can be produced by etching three sets of holes 35.260 off the vertical direction into the semiconductor surface (Yablonovitch et al. J 991 a). GaAs structures with a photonic bandgap in the near JR were fabricated in such a way by applying chemically assisted ion-beam etching of the masked surface in three directions (Cheng et al. 1996, 1997). Another fabrication method relies on stacking selec-



. dir U' this method 3D stacks of

tively etched 20Ist;~~~ur~a:I;~t~retb;; 11~~8) :~I~g)nP (Noda et 'at I ?99) rods GaAs (Noda et a ., . .' d t r rods mto a 3D d A differcnt technique of stacking serrucon uc 0

were produce . . d (·L· I al 199X Fleming and Lin 1999)~ The stack

h t nic crvstal was propose . 111 e . ~, ~ . h

P OS? d - as produced bv repetitive dcposition, patterning, and etching trenc es of I ro s W ~ ~. ~, Thi t h' e allowed

. ~ la ers and filling of the trenches with polysihcon. IS ec llI~U ..

m S102. Y f. IR ~ 0 photonic crystal with single-mode detect cavities

fabncatlOn 0 an ~

(Lin et at. 19919b)"I't of'" 0 photonic crvstals [or operating in the visible spectral

The avat a I 1 Y oJ - • . he onlv ~ m

. _ f rs from a lack of workable technologies. At present, t e ~I:) I. -

range still su! ~. ~ 0 photonic bandgap in the visible range is based on utilization

plfem~~_~~gl~l~i~e~ ~ystems (John and Busch 1999). For example, Astratov et al. o se . ." S'O) 1 .. closelv packed III

(1996) used synthetic opals composed of Silica (0- I 2 sp teres . . - id

entered-cubic structure with a period of about 200 nrn. By fi!hng the VOl s ~ f:,:~~ the spheres with a semiconductor with a refractive index .hlgher that~. th~t

~ ilica photonic bandgaps [or some directions (pscudogaps) might be ~eall;~6' o S hotonic cr stals formed of opal filled with CdS (Astratov et a.. ' 3D p I 19~7 ) and InP (Romanov et ar 1997) were demonstrated. Both enVlasov ~f a. VI a t ( 1997b) and inhibition (Blanco et al. 1998) of op:i~al haocernent (CdsasqOUVaeI1t~l~ dots embedded inside the inlerstitials between the SIlica emission m

spheres were achieved. . . . . f f Ii. ht emitters

All in all, phoionic crystals remain an mmgumg Idea or uture ; rt will be

with external quantum efficiencies close .to unity .. Ho~.vever, a lot. of effo

required to make this idea feasible for solId-state lighting applicatIOns.




Creating sources of white I ight is the ultimate goal of solid-state lighting technology. The most challenging application for LEDs is the replacement of conventional incandescent and probably, even fluorescent lamps. Attempts to up-convert long-wavelength emission of IR LEOs to broader visible spectra were undertaken years ago by Galginaitis and Fenner (1968). Berggren et al. (1994), Dodabalapur et al. (1994), Kido et al. (1994, 1995), and Hamada et al. (1996) demonstrated white light-emitting devices made from electroluminescent organic semiconductors. However, practical white LEDs became feasible only after the development of high-brightness blue AllnGaN emitters (Nakamura and Fasol 1997). Based on short-wavelength LEOs, white LEDs that exploit the mixture of two or three colors (dichromatic and trichromatic LEDs, respectively) are being developed. In view of potential applications, the designs of these solid-state light emitters aim at a combination of high efficiency and high color rendering.

There are basically two ways to produce white LEOs (Nakamura and Fasol 1997, Eisert et al. 2000, Mueller-Mach and Mueller 2000). The first approach is to mix light of different colors emitted by different chips. A white multichip (MC) LED should comprise at least two electroluminescent emitters. Another way is to down-convert the emission from a blue or UV LEO to a longer-wavelength light using phosphors (in case of a blue LEO, a part of the initial emission is used as a component of the white light as well) The number of phosphors involved in a phosphor conversion (PC) LED may vary depending on the device characteristics required. In this chapter we discuss the principles of generation of optimal spectra of white light (Section 6.1) and describe the existing designs of white all-solidstate lamps (Sections 6.2 and 6.3).


6.1.1. Trade-off Between Luminous Efficacy and Color Rendering

As we discussed in Chapter 2, a white emitter used f; .. ".

characterized by two figures of merit. th I . . °fr iighting applications IS

I . e ununous e ficiency and the gene I

co or rendering index (CRl). The luminous efficiency, '7p' can be presented as the

product of the radiant efficiency '1 and th I . ffi .

E ').. . ' e' e urmnous e icacy, K, as given by

q. (_.2.6). Basically, the radiant efficiencv depends on the d .

sion perfo ' I . I . - eVlce power conver-

rrnance, w lIC 1 Improves as technology matures The eff

only on spectral power distribution (SPD) of th li ht .' d' icacy depends means of e .. F . . . e Ig emitted, Irrespective of the

mission. 'or a given SPO S(J) th I .

, ,e urrunous efficacy is given bv

Eq. (2.2.4). The general CRI Rd' I I -

. . .. ' a' an specia co or rendering indices, R, (which

pro.vlde. additional information), are obtained from the SPO b d

scribed III Section 2.4. y a proce LIre de-

Practical optimization of a white LED' I ' . . -.

the ~ame as for colored LEOs-atlainin~ tf~;~~gVh~S:~:dil~~~e:ffi::~en~m;~ Issue ~s

:;:I~~~~e 51')nt~;aI quandtum e~ciency and light extraction efficiency ~se~ 2::~~

. e secon I ssue IS the SPO opti mizati 0 f hi

values for efficacv and CRI Below d' . ! c n or ac ieving the highest all-solid-state lam-p. . we ISCUSS! re SPO optimization of the white

cacie:o a~,~n~~srat~ the p~ssibiljties of such optimization, let LIS first consider em-

Planckian spect:a ~n~~~it:~~~~P~~~~~i:n ~~~~o~~~·~ :;e:l~t~:f ~~t:e::~c~ll:~~~e~e

spectra that are metameric with three standard CIE . (A - )

sources ,B, and D6S). The



Tahle 6.1.1.

Characteristics of white light emitted by full-range Planckian and trimmed Planckian radiators

Spectral range Color Luminous
temperature (K) efficacy UmIW) General CRI
2856 17 100
Full (Planckian) 4870 79 100
6504 95 100
38010780 nm 2856 154 100
(trimmed-Planckian) 4870 196 100
6504 193 100
43010660 nm 2856 334 95
(trimmed-Planckian) 4870 320 95
6504 305 95 119


full-range Planck ian spectra have Rn ;e 100 but low efficacies (because of spreading out of the visible range, especially to [R at low color temperatures; see Fig. 3.1.1). The imaginary trimmed-Planckian spectra, which are cut at the edges of the visible range (380 to 780 nm), have much higher efficacies and still the same 100-point CRI. Additional trimming of "wasted" ranges below 430 nm and above 660 nrn (Einhorn and Einhorn 1967) results in efficacies in excess of 300 Im/W. The CR[ reduces to 95 points, which is still an excellent rating for most

lighting applications.

Further reduction of the emission in the spectral regions with poor eye sensi-

tivity can boost efficacies above 500 lrn/W (Macadam 1950, Ivey 1963). However, this improvement in efficacy is achieved at a considerable expense in color rendering. A general rule is that in a particular system. efficacy and CRI are somewhat incompatible, and a trade-off between these two figures of merit should be sought.

The present LED technology offers semiconductor and phosphor emitters with relatively narrow spectral lines. This implies that white light can be obtained only by mixing colors of a few solid-state sources. White color chromaticity corresponds to the Planckian locus or to its very close vicinity (see Fig. 2.3.2). The principles of color mixing (see Section 2.3) suggest that an infinite number of combinations of primary sources can be used to produce white light with a required color temperature. Furthermore, the number of primary sources might he arbitrary (~2). Optimization of efficacy and color rendering of a white lamp composed of an arbitrary number of colored emitters is a problem with 110 general solution. This problem can be formulated as finding global maxima of the

objective function


where n is the number of the primary sources and S i and I; (i == I, ... , n) arc the normalized SPDs and relative intensities of the sources, Here (J is the weight that controls the trade-off between efficacy and color rendering (0 $ (5 $1 ).

The trade-off between efficacy and CRI can be analyzed easily for an S PD composed or two lines (dichromatic system), since the spectra contain welldefined complementary pairs of lines (see Section 6.12). However, a dichromatic

system is incapable of offering high-quality light.

An SPO containing three. four, five. or even more primary sources

(trichromatic, quadrtchromauc. qllinlichrornalic. and higher systems, respectively) can be sufficient for many practical lighting applications However, optimization ofthese polychromatic systems is a more intricate problem (see Section 6.1.3).

6.1.2. Dichromatic Systems

Leverenz; (1940) has analyzed the efficiency of a dichromatic system in view of white-light production trom kinescopes. The efficacy of the white light was shown to be greatest (or the pair of spectral lines at 459 nrn (violet-blue) and 572 nrn



(yellow-green). Ivey (1963) corrected these data for phosphors by taking into account losses due to the Stokes shift in the down-conversion process. The preferred peak wavelengths were found to be 445 nm and 570 to 590 nm. It was also noted that peak :vavelengths of 500 to 505 nm are least desirable for high output in the dichromatic system,

. Koedam. and Opstelten (1971) simulated spectra composed of two narrow lines for optimum color rendering. They have noticed that at low color tempera~u~es (23~0 to 2800 ~), when .the blue-wavelength region becomes less important, It IS possible to achieve relatively high color rendering. However, their data referred ~o a 550 nm/612 nm system that produces chromaticities far away from the Planckian locus and the CRT cannot be defined for a white reference source.

Walter (1971) analyzed a trade-off between the efficacy and CRI of a twocomponent phosphor system with Gaussian line shapes. In standard fluorescent lamps, a typical decrease in efficacy of about 20% was shown to be sufficient to compensate for an increase in the general CRl from 50 to 90 points. White light was characterized by an empirical quality index, the arithmetic mean of the general CRl and the brightness index (the ratio of the actual efficacy with that for monochromatic primary sources). Sources with the highest quality index were shown to be capable of a brightness index of about 70% and a CRl of about 90 point.s. However, color rendering was poor in the red region (0 to 50 points).

Zukau_skas. et al. P002) considered a dichromatic LED system composed of two Gaussian lines with 30-nm full width at half magnitude (FWHM). Figure 6.1.1 depicts the phase distribution (K, Ra) for the color temperature 4870 K (CIE direct sunlight). The phase distribution was obtained by computations for the values of the first peak wavelength, AI' with a l-nm step. The important feature of the


/4961635 nm

. ,


4811580 nm-.__

489/591 nm :.~

-, ······.·r----:;>l

380/569 nm



c::: -20

450/571 nm





200 300 400 K (Im/W)

FIG.6.1.1. Phase distribution (K,R,,)for a dichromatic white lamp with a 30"nm line width for primary sources and a 4870-K color temperature. (After Zllkau~kas et al. 200n



di tribution is a boundary (shown by open points in Fig. 6. I .1) that links the high~~ values of K and R(J This optimal boundary comprises all maxima of the objective function [Eq. (6.1.1 )]_ . By moving along the boundary, a trade-off between eificacy and color rendering can be realized.

6.1.3. polychromatic Systems

Improved characteristics of white I ight can be obtain~d in t~e trichromatic sy.~tem (Koedam and Opstelten .1971, ~~o:nton 1971). The SPO ot ~hlte light :~at IS re-

uired to maximize luminous ettlclency and the color rendering index was shown ~ contain peaks near 450, 540, and 610 nm. A three-co~lponent white I ight should avoid wavelengths ncar 500 and 580 nm. Compared wl:11 the dIchromatIc system, the trichromatic system was shown to yield a 20% gam III luminous efficacy at the

~eal. - .' .

Thornton (1971) characterized the performance of white light obtained from "a

trichromatic system by an empirical figure of merit ( K x 0.4 7 W lim + Ra ). Typically, optimization by line broadening yields an efficacy of about 300 Im/W at R = 70 for a color temperature of 3000 K. In this region, a 10% decrease in effi-

cacy can be traded off against a IS-point increase in CRT.. . - .

Mahr (1972) utilized a trial-and-error method to optimize color rcndltlo~ and

luminous efficacy of a combination of four and five narrow spectral lines. HIgher CRl values were obtained at a reduced efficacy. Using nonlinear programmmg techniques, Walter (1978) showed that four narrow lines ?ositioned at approximately 460, 530, 580, and 620 nm yielded a CRl value as high as 90 to ?5 pomts. Doughty et al. (1998) proposed a general illumination system compnsmg four LEDs with a CR] of at least 80 points The central wavelengths were assumed to lie in the range 440 to 450 nrn for the first LED. and 495 to 505 nm, 555 to 565 nm, and 610 to 620 nm for the other three. Although these values do not necessarily represent the global maximum for Ra ' they are close to the optimal values

(460,530,580, and 620 11m) obtained by Walter (1978). . _. _ -

Zukauskas et al. (200 I, 2002) developed a method of opt1l11lZatIOn ot a \~'hltelight source containing an arbitrary number of primary sources WIth arbl~ary SPDs. The method, which employs stochastic routine, finds only the optimal boundary of the (K, R,J phase distribution. Figure 6.1.2 depicts the results ob-

tained for trichromatic, quadrichromatic, and quintichromatic systems of 30-nm lines at a color temperature of 4870 K (direct sunlight). The optimal boundary for

the dichromatic system (Fig. 6.1.1) is shown for comparison.. _ ,

As can be seen from Fig. 6.1.2, the optimal boundaries obtained for a different

number of primary sources partially overlap and a joint boundary is formed. Typically, in the overlap regions, the lines merge so that the number ?f primary sources reduces to a minimal value. Therefore, the joint boundary prOVIdes gUldeiJnes for

both K-Ra trade-off and selection of the number of primary sources (the crosses in Fig. 6.1.2 mark the suggested points or highest reasonable CRI values for each


ro a::

FIG. 6.1.2.


99,5 9 9

95 90

80 70 60 50 40 30 20

9r411 ~L -, I I
8 Systems:
• quintichromatic
~ 6 quad rich romatic
• trichromatic
~ I- a dichromatic
..._ ...
~u :::Jooo ~
'vc, ~ ;,n!
~2 f----I 10 5

320 340 360 380 400 420 440 K (ImlW)

Optimal boundaries of the (K, R,,) phase distribution for 4870·K white-light

sources containing two th f d f .

, ree, our. an ivc primary sources with ~O I'

widtl C ' ' . ~ .. .J ·nRl me

15, Tosses mark the pornts suggested for the h' h

val ' r. _ - ' Ig est reasonable CRI

20~~,) or each number of pr_lmary sources, (After Zukauskas et 01. 2001,

number of pri mary sources) AI the i - ,

duction of ddi _ ", ' so, e jornt optimal boundary suggests that intro-

I'd' ~n a ItlOn,al primary source does not change the general rule that the co or ren itron can be Increased only at the expense of the luminous efficacy,


6.2.1. Dichromatic Phosphor Conversion LEOs

Ah~tra;ghtfor~ard waY,to devise a dichromatic white LED is to utilize an AllnGaN c IP6 \l~t ~mlts blue .lIght an,d a phosphor that emits in the yellow region Figure '_', sows a typical ,deVice design. The blue chip is mounted in a built-in reflector cup and cO,ated With a converter layer, which is a mixture of e ox resin and phosphor particles, The entire structure is embedded in a trans Pt y, A

part f th bl I' h " paren resin

low ~ ht e Thue rg t IS absorbed I~ t~e phosphor layer and down-converted to ;el-

g . ae rest of the blue errnssion escapes into the transparent resin A vari ety of phosphors for the down-conversion of AlinGaN LED " ' . - ered White LED th 1 .. ermssron were consid-

, s at exp ott organic dyes (Schlotter et al 1997 1999) ,

gated polymers (Hide et al. 1997, Zhang and Heeger 1998), ~nd an 'AIGaJ~/:e%~~



Yellow Blue

Phosphor Layer

AlinGaN Chip

FIG. 6.2.1. Schematic structure of an AllnGaN-based luminescence-conversion white LED, (After Nakamura and Fasol 1997, and Schlotter e/ al. 1997, 1999,)

conductor (Guo e/ al. 2000) were also reported. However, the best choice seems to be yttrium aluminum garnets (Y AGs) doped with Ce3 + ions (Nakamura and Fasol

1997, Bogner et al. 1999).

Cerium-doped garnets were developed years ago for applications in cathode

tubes (Blasse and Bril 1967). The Y 3AIsO 12:Ce3+ phosphor has high thermal and chemical resistance, a low corrosive potential, a defect-free structure, and exhibits a quantum efficiency of nearl y 100%, The phosphor is produced by a class! cal oxide melting process at a typical temperature of 160QoC (Bogner el al. 1999),

Figure 6.2,2 shows the energy levels in Y3AlsOI2:CeH,

The relevant optical properties result from allowed dipole transitions between

the ground-state 4fl and excited-state 5i bands, The shielded ground-state 4fl level is spin-orbit split. The excited state 5d] features strong crystal-field splitting and vibronic coupling because it is no longer shielded by the environment (Loh 1967). The lowest absorption band at 460 nm is due to transitions from the lower 2FS/2 sublevel to the excited 2D band. The emission spectrum results from Stokesshifted transitions from the 2D band to the 2F5/2 (520 nm) and 2F712 (580 nm) sublevels. At room temperature, two emission lines overlap, resulting in a structure-

less band (Holloway and Kestigian 1969),

An important feature of the garnet is that substitution for AI3+ and y3+ ions allows one to tailor the emission and excitation spectra (Holloway and Kestigian 1969, Tien el at. 1973, Nakamura and Fasol 1997). For example, the (Y l_aGda)3(AII-bGab)SO 12:CeH system yields the peak of the emission band in the range 510 to 580 nm and the peak of the excitation spectra in the range of 450 to 480 nrn. Also, the spectral characteristics depend on Ce3 + concentration (Tien et al. [973, Batenschuk et al. 1999), Technological optimization of the AlinGaN/YAG:Ce3+ system is described by Nakamura and Fasol (1997),



B 6eV B.OeV

5d1(5s'5p6) 7,0 eV


6.2 ev----~~~~~~::': CB

'" ,60 eV

"'G==F=F> D 5.1 eV

o eV---------- VB

FIG. 6.2.2. Energy levels of Ce31 (4/) in yttrium aluminum garnet Y]AISOI2. (After Batenschuk et of. 1999.)

Batenschuk et al. (1999), Bogner et al. (1999), Schlotter et al. (J 999), and Wu et at. (1999).

The spectral properties of Y AG:Ce3 + almost ideally meet the requirements for a dichromatic white LED. First, the peak of the excitation spectrum around 460 nm coincides with the peak wavelength of the most efficient blue AlinGaN LED available (465 nm; Mukai et al. 1999). This wavelength is also close to that required for the upper component of the optimal-efficiency dichromatic system (445 nrn), Second, the emission spectra of the phosphor fit the complementary component (570 to 590 nm). Moreover, tailoring the emission spectra of the phosphor results in white light with various color temperatures.

The eTE 1931 chromaticity diagram displayed in Fig. 6.2.3 illustrates mixing of the emission from a 465-nm blue AlInGaN-based LED (30-nm line width) and that from the phosphor for peak wavelengths of 560, 570, and 580 nm (120-nm line width). Intersections of the Planckian locus with the lines that link the relevant chromaticities show that white light with color temperatures above 4000 K is feasible_

Figure 6.2.4 depicts typical emission spectra of AlInOaNI

(Y l_aOda)3(AII_bOab)sOI2:CeH white LEOs for the phosphor band peaked at 570 nm (solid line) and 580 nm (dashed line). The luminous efficacies of the spectra exceed 300 Im/W, while the general CRI is a bit less than 80 points. Analysis using individual standard test samples unveils a deficiency of color rendering for the yellowish-green (R4 '" 64) and reddish-purple (RS ~ 55 to 60) colors. Additional test samples show relatively low indices for strong green and strong blue (R11,12 = 50 to 60) and extremely low color rendering for strong red (R9 '" 0).







'E 0.6


a 500




~ 0.4 E


s: o

>- 0.2

640 700

FIG. 6.2.3.

~~~~4~0~_L __ ~_L __ ~~~~~

0.0 04 06 08 1.0

0,0 0.2 . . .

X Chromaticity Coordinate

Obtaining white light from blue emission of AlInGaN LED (465 n~) and ellow emission of cerium-doped garnet with different peak wavelcngt post-

;ion5 (CIE 1931 chromaticity diagram),

Red-shifting the phosphor emission by 10 nm r~sults _in a .significant change in color temperature and in the blue-to-yellow intensity ratio, WIth less of an effect on

efficacy and color rendering.

AlInGaNIY AG:Ce3+ white PC LEDs were commercialized soon after the de-

velo ment of high-brightness AlInGaN LEDs (Nakamura and Fasol l~997). By

'1' p. All GaN LED with 5 6% external quantum efficiency, the first white

uti IZlI1g an n '" ff . f 5 Im/W

LED h'bt'ted a luminous flux of 10 mlm With a luminous e iciency 0 . ' s ex I K S' th uantum efficiency CRJ of 85 points and color temperature of 8000 . ince e q k' I 1999' , .' dil (14Q/ in 1999' Mu me/a. r,

of the blue LED has been mcreasmg stea t y /0 '. h a lumi

. I 2000) hite PC LEOs WIt a urrunous

20% in the beginning of 2000: KOIke et a . , WI, .

efficiency exceeding that of incandescent lamps became feasible. F~~ mst~nce,

a blue 465-nm LED with a quantum efficiency of 20% should b~aw e t~~s ;:~i~ a white PC dichromatic emitter with a luminous efficiency of 50 I . . (

mate includes losses of about 20% due to the Stokes shift and nonradJatlve conver-

. . tl e hosphor: Mueller-Mach and Mueller 2000). The upper theoretical

~;J~~t l~f t~e fuminou~ 'efficiency of the LED und~ cO~Sird)er~~~~e i~~7~El;:~J;~~

100% quantum effiCiency of the blue LED and p osp 0 .



All nGaNIY AG:Ce3+ CT=3970 K
1.0 I' / Ra=77
~ K=3201rnJIN
-~ 0.8
t: I \
:::J CT=5750 K
I \
-e 0.6 Ra=79
~ K=310 ImfW
Q) 0.4
0 \
0.2 0.0 ~~~=-____'__l.__'__i~"':;'"

400 500 600 700

Wavelength (nm)


Model emiss ion spectra of All nCiaN/(Y Gd) (A I G) (J . C 3· .

IE' . . . I a a 3. I-h ah 5 12· e white

.. Os tor two, c~m~oslllOns_ of the garnet. Solid I inc, the peak wavelength of the phosphor emissron IS at )70 nm: dashed line, same lor 58011111.

iting performance around 14 I lUI I

2000). m .. are a ready being manufactured (Harle et al,

L Narendran.et al. (2000, 200 I) investigated the reliability of practical PC

EDs. The ~glllg tests of standard plastic-encapsulated white lamps revealed a reduced lifetime (typically, 50?0 to 6000 hours for 50% of the initial ~utput) compared t~ blue LEOs. Degradation of the phosphor and/or a change in the transmittance at the encapsulant might account for the reduced II' fetirne Novel hi h

PC LED . . . . IU -power

. s eqUIp~ed ':Ith heat sink and containing encapsulants that do n07 degrade

over tune exhibited Improved lifetime characteristics (Narendran et al. 2001) P - tential ?f such LEOs for general lighting applications (e.g., architectural ligh~inOg) was pointed out.

6.2.2. Polychromatic Phosphor Conversion LEOs

~richro~1atic PC L.ED~ offer improved efficiency and color rendering. Once again

he optimal com binati on of primary sources (4501" 401610 Th '

. h b . .' J I I1In; ornton I 97 I )

n~.g t e Implemented by using a partially absorbed blue light from an AllnOaN c I~ ~nd two appropriate phosphors for the green and orange-red regions BI eml~SIO~ from A1InOaN LEOs is capable of exciting a variety of organi~ and inour~ game p osphors (Sato et at. .1996, 1998). White trichromatic PC LEOs based on

luminescence down-conversion In organic dyes (Schlotter et at 1997) d .

conducto (0 I . . an semi-

c rs uo et a. 2000) were proposed. However, in view of technological



advances in practical PC LEOs, ionic phosphors with broadband emission are preferred.

Mueller-Mach and Mueller (2000) described an appropriate combination of

two ionic phosphors. They selected the SrGa2S4:Eu2+ phosphor for conversion of blue Light to green emission (-535 nm) and SrS:EuH for the conversion to orangered emission (-615 nrn). Figure 6.2.5 shows a model spectrum of white emission from a trichromatic AlinGaN/(SrGa2S4:E/+ + SrS:Eu2+) system with a 450-nm peak position for the AlInOaN chip. Compared to a dichromatic ALInGaNIY AG:CeH LED (Fig. 62.4), the trichromatic spectrum features higher values of the luminous efficacy and of the general CRY. Lower color temperatures are also available. It should be noted that all CRls for standard individual test samples (R;, i =:01 to 8) exceeded 80 points, whi le all "strong" colors ( i == 9 to 14 ) were rendered in excess of 50 points. This implies that "deluxe" lighting can be achieved based on this system.

Another approach to a polychromatic LED is based on exploiting a UV LED

for excitation of a set of phosphors. A P-/1 junction GaN U V LED was invented by Akasaki el al. (1992). Later, heterostructure UV LEDs based on GaN/AIGaN (Han et al. 1998, Ohba and Yoshida 1998), InGaN/A1GaN (Mukai et al, 1998b,c; Mukai and Nakamura 1999), AIGaN/A1GaN (Nishida and Kobayashi 1999, Kinoshita et al. 2000, Otsuka et at. 2000, Nishida el al. 200 I), and quaternary AlInGaN (Adivarahan et al. 2000, 2001; Zhang et al. 2001) materials systems were developed. At present, group IlJ-N system offers LEOs with the shortest wavelengths (up to 333 nm, Kinoshita et al. 2000 and 327 run, Zhang el all001).

~ CT=3700 K

§ 1.0 .ri




<ll 3: 0.5 o c,

400 500 600 700

Wavelength (nm)

FIG. 6.2.5.

Model white emission spectrum from an AllnGaN/(SrGazS4:Eu2+ + SrS:Eu2-) system. (After Mueller-Mach and Mueller 2000.)



CT=4s00 K Ra=90 K=310 ImIVV

ti i ~

: :

f i J i

: :

: !

= :

: :

: :

! 1

: :

J :

500 600 700 Wavelength (nm)


Model white emission spectrum from a trichromatic PC LED embracing a UV pump LED and three phosphors. (After Eisert et al. 2000)



For UV LE~s, the visible pa,: of the spectrum is completely generated by phosphors. Eisert et al. (2000) Simulated a single-chip device with several phosphor :onverters excited by a UV LED or UV laser diode (Fig. 6.2.6). A quas.l-contmuou.s spectrum With a CRI value in excess of 90 points can be obtained by usrng three phosphors with a spectral width between 70 and 120 nm. Due to the absence of narrow lines in the PC LED with th UV

r hi . . e pump,

supr~me-qua ity w ite It?ht With high values of all individual test samples can be

obtained. Ho~ever, m~vmg the pump SOurce into the UV spectral range results in

a reduced radiant efficiency because of enhanced Stokes losses in the .

process. conversion


Compared with a PC LED, a white MC LED should have better efficiency due to

the absence of energy losses caused by the Stokes shift and b di .

bi ti . h h Y noma iattve re-

com m~ Ion In t e p iosphor, It also avoids phosphor-related aging problems. One

o~ the dlsadv~ntages IS a more complex design that might increase cost. Another disadvantage IS narrow .emission lines. The latter fact requires using an increased number of LEOs to achieve high values of color rendering indices. However this m~ght not be too Important in lighting applications involving large LED as~embites.

Provided that the radiant efficiency of individual LEOs '7' I'S kn th

. , 'el' own, e

luminous efficiency of an MC LED composed of n colored LEOs is found from





where <!lei is the relative power contribution of the ith LED (I;~I<!lei = 1) and K is the overall luminous efficacy that is the function of the LEOs' spectra Si:

I;~I (:oOS/(2)V(2)d2

K '" ---------

I:J fa S,(2)d2


The upper theoretical luminous efficiency of an MC LED equals the luminous efficacy of the spectral distribution, provided that the chips are of 100% quantum efficiency and the absorption of the radiation of one chip by another is avoided (Mueller-Mach and Mueller 2000).

6.3.1. Dichromatic Multichip LEDs

Dichromatic LED system features the highest efficacy of all white solid-state sources. Table 6.3.1 presents data for dichromatic spectral distributions metameric with three standard CIE sources (A, B, and 065). The estimate assumed having two Gaussian lines with a width of 30 nm, which is the average value for colored LEDs. The SPOs were globally optimized for the highest efficiency at a particular color temperature. However, the values of the general CRI are hovering close to zero. Although an increase in CRI by a few points is possible at the expense of the efficacy, the dichromatic white MC LED is unable to produce white light of reasonable quality. In the area of general lighting, the dichromatic white MC LED might be able to compete only with the low-pressure sodium lamp for applications that do not require high color rendering (see, e.g., Dixon 2000, Roberts 200 I).

Table 6.3.1. Data for white two-chip LED (30·nm lines) globally optimized for the highest efficacy

Color Wavelength (nm)fpower
temperature K (lm/W) Ra
(K) )·I/¢el ,1.21 <!l e2
2856 450/0.157 580/0.843 492 -13
4870 450/0.325 57210.675 430 3.0
6504 450/0.399 569/0.601 393 9.5 130


However. up to now, both AllnGaN and t\IG .

cient LEDs in the range 570 to 580 . .aTnP technologIes do not offer effi-

. ' nrn required for a high-perfo di I

mane system (Haitz et al. ?OOO) F rth d I . rmance IC lro-

. . - . u er eve opment of this ki d f hi . .

In progress. For instance, Damilano et 01 (?OO I) d no,,: I~e LED IS

matic LED composed of two stacks of In'G;N/G ~eport,e. on a monol.lthlc dichroferent thickness. ( a mu tiple quantum wells of di[.

An example of an SPD for a dichromatic MC LED .. .

luminous efficacy at a color temperature of 4870 K' h optlmlz~d for the highest IS s OW11 In FIg. 6.3.1a.

6.3.2. Polychromatic Multichip LEOs

Polychromatic white AllnGaN- and AIGa]nP-ba .

since AllnGaN and A/GalnP tech I . sed MC LEOs are .qllJte plausible,

chromaticities for LED chips that ~~g~7Yb~sa::~~~ ~n?ugh to, provide a variety of

ing three or more different chips t " e into white emitters. Employ-

. no on Y Improves color rendering of white

K=430 ImlW, Ra=3









400 450 500 550 600 650

Wavelength (nm)

M.odel spectra ~f white emission from rnultichip LED lamps (CT = 4870 K

WIth 30-nm line Widths of " LE" )

(b) trichromatic lam: . prun,:ry Ds: (a) dichromanc lamp:

(After Zukauskas el:; 2~~ (,~adfiChIOmaIIC lamp; (d) qUlIlllchroma!ic lamp.



light but also allows for the emission of any chromaticity within the extended color gamut (see Figs. 23.2 and 2.3 .3). This might be achieved by varying the average current in each individual chip or by adjusting the number of different color chips in the assembly.

The simplest implementation of a trichromatic emitter is a package with three single LEOs (red, green, and blue), such as the emitter used for full-color video displays (Bogner et al. 1999) (see Section 7.3.2). Further improvement in white Me LEOs might be achieved using the heterointegration of light-emitting devices into a single epitaxial structure, The potential of such heterointegration was demonstrated by using wafer bonding (Floyd et al. 1999) and selective-area metalorganic chemical vapor deposition (MOCYO) (Yang et al. 2000),

However, display applications require a wide color gamut, whereas lighting applications call for high efficiency and good color rendering. Mueller-Mach and Mueller (2000) investigated the trade-off between the efficiency, color rendering, and junction temperature in a trichromatic MC LED for Lorentzian spectral distributions of the individual chips. At room temperature, the efficacies of 400 Im/W were shown to be plausible with the general CRI, reaching 80 points, With increasingjunction temperature, the emission lines broaden and the color rendering might improve at the expense of efficacy.

As can be seen in Fig. 6.1.2, the trichromatic and quadrichromatic MC LEOs have the highest efficacies for 5 < Ra < 85 and 85 < Ra .:0:: 98 , respectively. These solid-state lamps can meet requirements for most lighting applications, One additional CRI point can be gained by introduction of the fifth chip. Although this improvement is marginal, quintichromatic LED lamps and lamps with a larger number of primary LEDs are capable of providing supreme-quality white light with quasi-continuous spectra. Such lamps might be important for special lighting applications (e.g., for people with color vision properties that deviate strongly from those of the 1931 eIE Standard Observer).

Figure 6.3.1 depicts model spectra of white solid-state lamps (4780 K) composed of two to five primary LEOs with line widths of 30 nm. The spectra refer to the highest reasonable CRr values for each number of primary LEOs (points marked by crosses in Fig. 6.1.2).

Figure 6'),2 presents the dependences of the LEOs' peak wavelengths on the general CRI for lamps that fit the optimal boundary of the (K,Ra) phase distribution shown in Fig. 6.1.2. As shown in Fig. 6.3,2, the dependences contain almost continuous regions that branch abruptly to larger numbers of primary LEOs.

Data presented in Fig. 6.3.2 suggest that current LED technologies can provide primary LEOs required for a high-efficiency trichromatic white lamp (50 < Ra < 85 ). This lamp can be composed of an AIGalnP-based LED with peak wavelength around 600 nm and AllnGaN-based LEDs with the peak wavelengths of 450 and 540 nm. Equation (6.3.1) implies that by using AIGalnP and AlInGaN LEDs with radiant efficiencies of 50% (Krarnes et at 1999) and 20% (Koike et al. 2000, Wierer et al. 200 I b), respectively, trichromatic lamps with a luminous efficiency of 90 to 100 Im/W can be assembled. (Even higher efficiency could be obtained if the luminous efficiency were optimized instead of the efflcacy.) Such






0 2 LEOs
• 3 LEOs
6 4 LEOs
• 5 LEOs ... P
8 " te~ ~
~ . • 6~ ~ IiII
l> w
""l!Ii ~~
"'"IllII!! ~ ~ ~
~ ~
.~ .-
"1'1"1 65






.FIG. 6.3.2.

5 10 20 3040506070 80 90 95 9899

Ra Variation of peak wavelength with th

l~ps (CT '= 4870 K) composed of r:~e~eral C~ value for while solid-state wah 30-nm line widths (Afte Z· k k' hree, tour, and five prrmary LEDs

. r u aus as et al. 2001.)

lamps can surpass fluorescent tubes that hav' .

80 ImJW (see Table 3.7. I). 'e luminous efficIencies around

As for quadrichromatic lamps whi h .

lighting (85 < R < 98) th I n'fi .: IC are SUItable for providing "deluxe"

a - , e ow e icrency of current LED .

around 570 nm (?O/". Kr I s, with wavelengths

- /0, ames et a . 2000) do '.

~ous efficiency far above 20 ]mJW Alth~u es n~t make .It P?sslble to boos! lumi-

Inc~ndescent lamp, this drawback ~a hi gh !.hls value IS higher than that of the mane lamp. n mder ImplementatIOn of the quadrichro-

Nevertheless, AIGalnP and AlInGaN t .

development and furnishing highl ffici echnologles have. potential for further

. . .. ye IClent colored LED f I

ncuy, Therefore the primary chi ber I so a most any chroma-

creased in the future. This wi I-' ~~k:ui~ ~r ~n Me LEI? white. lamps can be insion spectra that satisfY any desired re PUi ssible to obtain .qua~l-continuous emis-

long run, Me LEDs will id q. rernents for while light quality. In the

. proVl e supenor and ec . I .

sometimes referred to as the next li hti . onormca sohd-state lighting

Ig mg revolution. '



Commercial use of visible LEOs began in the late 1960s (Bergh and Dean 1976, Williams and Hall 1978, Gillessen and Schairer 1987). The first LEOs were known as small colored indicator lamps used in all kinds of electric and electronic devices. Soon, compact LED-based alphanumeric displays composed of dots and bars appeared in the first calculators and electronic watches and in measuring and entertainment equipment. One-dimensional arrays of LEOs served as flying-dot or bar-graph displays and image projection systems. Later, as soon as most instrumental displays underwent conversion to liquid-crystal technology, the application sphere of LEOs began shrinking back to indicators. However, starting with the introduction of first transparent-substrate A1GaAs LEOs in the 1980s, highbrightness electrolurninescence devices are gradually moving into all applications related to artificial-light production.

In this chapter we present new LED applications that emerged based on highbrightness technology. This technology is encroaching on niches that are occupied not only by conventional incandescent bulbs and fluorescent tubes but also by such sources as lasers. After considering LED driving circuits (Section 7.1), we describe well-established large-volume applications of high-brightness LEOs in power signals (Section 7.2) and displays (Section 7.3). Then new applications of bright solid-state lamps in medicine (Section 7.4), photosynthesis (Section 7.5), and optical measurements (Section 7.6) are discussed. Section 7.7 is devoted to illumination, which is the ultimate goal of solid-state lighting technology.




We do not emphasize less challenging applications, such as LED applications for automotive instrument cluster warning lights or backlighting dashboards and liquid-crystal displays (see, e.g., Hodapp 1997, Brown et al. 2000). Backlighting does not require extreme efficiency, brightness, and longevity and can be provided by a variety of alternative technologies, for instance, by white organic eiectrolun-], nescenr devices (Kido et al. 1995) or by ZnSe LEOs (Katayama et al. 2000).


7.1.1. LED Strings

LEDs have exponential I-V characteristics (see Section 4.3.3) that exhibit large variations of current with small variations of voltage. Roughly, the forward voltage across an LED is V F '" Eg / q , where Eg is the bandgap energy of the semicon-

ductor and q is the elementary charge. For instance, VF is approximately 1.9 V for red, 2.1 V for yellow, and 3.0 to 4.0 V for green and blue LEDs (see Fig. 4.3. J 0). To stabilize the output, LEOs are driven at a constant forward current IF (typically, 20 rnA). Typically, the LED output characteristic is almost linear in the vicinity of the nominal current (see Fig. 4.3.9).

In lighting appl ications, the circuit can contain any number of LEOs, which are arranged in single or multiple strings. Generally, one should consider n LED strings (n == I, 2, ... ) that are connected in parallel for n » I, as shown in

Fig. 7.1. I. Each string can represent a serial connection of HI LEOs ( m == I, 2, ... )

~ n Strings

~ n Strings


n Strings

FJG.7.1.1. Circuit diagrams for driving LED strings: (a) unregUlated; (b) with voltage regulator; (c) with current sources. Vs is the unregulated de voltage. Exam. pies of LED strings are shown in Fig. 7.1.2, (After Hodapp 1997.)



. 7 I ?) The total number of d contain k parallel substrings (k '" 1,2, ... ; see Fig. . ._ .

an . kl

EDs is nmk and the overall drive current IS n F. .

L . h re three ways to maintain a constant current across stnngs of

Baslc~~l~, t ~;~;. The simplest circuit, which utilizes .an unregulated" de

LEOs (H pp conta{ns ballast resistors connected in series With each LED strmg

voltage source, . .

" 7 1 1 ) The estimated value ofthe ballast resistance IS (Fig ... a.

R",V-mVF (7.1.1)


where V is the voltage applied to the strings (VS for the unregulated and V C for

ththe r:~i~lt~re~hc~~~~i~:~:~;~~i;:~~)~I:oS~ltl:b~~z~:h~o~:;~t~~~t:;~I~:;~a~~: ~cf;~~des

e r . .V 2 14 V) the typical number 0 re

LEOs In automotive applications ( = I to, . d .

. .. - 4 A more stable operation can be achieve using a

LEOs per stnng IS m - . .

regulator that provides constant voltage V C (Fig. 7.1.1 b). .

However, for outdoor applications, the variation of VI" With temperat~re ight be significant and high values of the voltage drop across the ballast .reslS~ :ce are required. F~r instance, in the temperature range of 10~ K, "" vanauon 0

. hi h 0 5 V To maintain stable current In this temperature

VF might be as rg as. .

range the ballast resistor should dissipate the power comp~rable to that consumed by· th'e LED string. To remove this drawback, electronic current sources are




2iF ¥}SLEDS tt t ~~men'


tt t'


m LEOs

k Substrings

k Substrings

FIG. 7.1.2.

I.EO strin s driven by constant current: (a) single ~lring. of n: J .EOs;

g trl of k parallel substrings: (c) composite strmg With cross

(b) composite s nng . ,

connections that form rungs of s LEOs. (After Hodapp 1997.)



introduced into each string as shown in Fig. 7.1.1 c. These current SOurces rnav feature a simple circuit (Gillessen and Schairer 1987) or contain more sophisticated means of thermal stabilization of the LED output (see, e.g., Mroczka and Parol 1994). Integrated circuits designed specifically for LED applications (LED constant-current drivers) are also being introduced.

The design of LED strings depends on several factors. To reduce the number of . components in a~ LED array (resistors or electronic current sources), a simple stnng (FIg. 7.1.2a) IS replaced by a composite string (Fig. 7.1.2b) that contains parallel substrings. However, this solution requires LED chips that have closely matched forward voltage characteristics. Otherwise, the current distribution betw~en the substrings becomes uneven. To improve the viability of long composite strings such as those driven directly by rectified ac power, the strings can be crossconnected as shown in Fig. 7.1.2c. Here, random fail ure of a sol e chi p does not break the entire substring but affects only a segment of sLEDs.

. Electrical Protection. When LEOs are driven within electric systems that contam rapidly interrupted inductive loads (e.g., ignition coils or alternators in automotive systems), high-voltage transients can occur. Forward overloads result in reduced LED lifetime, and large reverse voltages can cause device failure. The electric robustness of LEOs is being improved (Osinski et at. 1998, Harle et a/. 1999, Osinski and Barton 2000), and some LEDs contain integral transient suppressors. Nevertheless, some external means preventing both forward overloads ~nd reverse breakdowns might be necessary. A typical protection circuit is shown III FIg. 7.1.3. The high-voltage silicon diode impedes the reverse current and the

transient suppressor Ii m its di rect voltage across the LED stri ngs, '

Dimming. In simple applications, LEOs can be dimmed up to 10% of the maximum output by reducing the drive current. However a further reduction of the drive current can cause a noticeable colorimetric shift (see Mukai et al. 1999 Harle e/ al. I ?99) and uneven luminous intensity of LEOs within an array. Also: the output might vary nonlinearly with the current. All these drawbacks can be removed by using pulse-width modulation (PWM) at the constant driving current B~ changing the duty cycle of the current pulses, the average output may be varied within three to four orders of magnitude. Typically, repetition frequencies above 100 Hz are used to prevent visual inconvenience.



High-Voltage Silicon Diode

Transient Suppressor

FIG. 7.1.3. Circuit for protection of LED strings from electrical transients. (After Hodapp 1997.)



Digital Control. Low-voltage operation of LEOs offers benefits in digital control of large LED arrays and lighting fixtures. Intelligent fixtures, which comprise built-in microprocessors or are driven by external computers, can be devised. pWM, individual remote control of LED strings, and sensor-based intensity variatjoll are easily implemented by using digital control approach. An example of LED-based digital lighting network was described by Ducharme and Morgan (2001 ).

7.1.2. Battery-Powered LEOs

Solid-state lighting involves many battery-powered applications. Driving of visible LEDs requires sources with at least 2 V for red and 3 to 3.5 V for blue colors, Meanwhile, most battery cells provide voltages in the 1.2 to 1.5 V range. This means that two or three cells are required. However, owing to the high radiant efficiency of high-brightness LEOs, the capacity of a single cell might be sufficient for device operation. To reduce the size, weight, and maintenance cost of a battery-based LED power source, single-cell drivers are being devised.

A simple circuit that is capable of driving a red LED with voltage as low as 0.85 V is depicted in Fig. 7.1.4. The circuit features a self-starting sinusoidal oscillator operating in the 1 ~O-kHz range with the peak amplitude of a sinusoidal output of2 V. A similar circuit with voltage multipliers has been used to drive a blue LED (Liao and Chang 2000b).

7.1.3. High-Power Pulsed Drive

Some spectroscopic applications, such as time-resolved fluorescence measurements (see Section 6.2), require short, intense pulses of light. Araki and Misawa (1995) and Araki el al. (1997) designed current drivers for nanosecond pulsing of LEOs at currents in excess of 1 A. The current drivers were based on rapid dis-


FlG. 7.1.4. Circuit diagram of a low-voltage LED driver. (After Liao and Chang 2000a.)



50 ns -. ._'_~nn~~~~

5 V if1__

FIG. 7.1.5.

Circuit diagram of a current driver for generation of nano . :t li

lIsmg an LED. (After Araki et al. 1997.) seconc ight pulses

charge of a capacitor through the LED by switchin I h .

7 I 5 . - - g ava anc e transistors F-

ure . . presents a CIrcuit th t . . Ig-

. a contains a cascade of two av I· h -

Inductor is connected in parallel with th LED a anc e transistors. All

injection Current. Depending On the ca aci to shorten the pulse duration of the as ShOl1 as 1.2 ns. p tance value, the driver produces pulses


~righ~ lights, usually distinguished by color, are used wid I' d

claus indoor environments for carrying information The I ~_Yhm out oor and spapower signals Fo .' ese Ig ts are referred to as signage. Filteri'n ~;~~ny years, filtered Incandesc~nt lamps dominated power tribution b g P . ches nar:o:,band spectral distribution from broadband dis-

y converting t e radiation of undesirabl \ I _ .

pov:er-dissipating process. Typicall , filtered . e 'lave engths to heat. This IS a

exhibit luminous efficiency in the ral;ge of 3 Im/~c~ndes~ent and halogen lamps (amber) colors. Also, power signals re uire hi o.r re " and IO.'mlW for.yellow conventional bulbs operating in ~ hIgh. rehabl.hty th.at IS not typical for

ciently convert electrical energy ~~ ~~~~:;:::~glln~ .. I--Ilg~-bnghtness LEDs etlilongevity Hence LED . radiatIOn and have unsurpassabJe

. , s are a superror alternar] . f .

AIGalnP and AllnGaN technologie ff hi ive or power signage. Mature

more efficient than filtered inca dis 0 Jer llgh-bnghtness LEOs that a. re much

. n escent amps for the ti f

luminance of colored SOurces assembled of c d en In: ran.ge 0 colors" The

resolve signals even when the are ex an e~a-cJass L.EOs IS high enough 10 advantages, high-brightness L~Os are Pd~:e~a t? ~Irect sunl~ght. Because of these numerous applications of pow' P crng filtered Iflcandescenl lamps in

er srgnage such as traffi I' h ( .

a~tomotive exterior Jighting (Section 7.2.2) and 11 k- IC I.g ts Section 7.2.1), SIgnals (Section 7.2.3), ' a rnds of safety and emergency



7.2.1. Traffic Lights

Traffic lights were the first mass application of high-brightness LEOs. This appl ication gave a tremendous boost to the solid-state lighting industry. In the early 1990s, when only long-wavelength LEOs were available for power signage, these devices were approved for red traffic "stop" lights and orange "don't walk" pedestrian signs (Lewi n e/ at. 1997). Despite the relativel y high price, red LED lights offered economical benefits, si nee they operated at 65% duty cycle. The first LEDbased traffic lights contained hundreds of chips arranged into assemblies with visually dotted patterns, Advanced lights might contain just a dozen efficient LEDs equipped with secondary plastic optics to provide a smooth radiation field (Krames et al. 2000). Although green "go" light heads are still more expensive than the red ones, and yellow (amber) "caution" lights operate at only 3% duty cycle, all-solid-state traffic lanterns are being actively promoted and installed.

The main advantages of LED-based traffic lights follow from economic considerations based on reduced power consumption and extended lifetime. Conventional traffic lights are equipped with either 135- or 70-W incandescent lamps in comparison with a 1 0- W or even lower wattage LED module (arrow I ights can be run by just a few watts). In industrial countries such as the United States, 1000 signaled intersections are operating on average per I million population, with about 50 traffic lights per intersection (among which 20 are lit simultaneously). The replacement of all traffic lights in the United States with LEDs would reduce electricity consumption by more than OA GW. Again, to preserve traffic safety from abrupt failure of conventional heads, the bulbs are replaced two or three times per year. Keeping in mind that LED heads can serve for 10 to 15 years, this makes replacement of bulbs by LEOs economical even if the LED module is much more expensive than the bulb. Finally, solid-state light longevity results in reduced maintenance and emergency repair costs. Consequently, the payback period for the replacement is less than one year, with further annual savings of$1 000 to 2000 per intersection plus no-maintenance thrift.

Improved traffic safety is even more important. Compared with incandescent bulbs, LEOs have a very low probability of abrupt failure. Even if a random failure occurs or the signal is attacked by vandals, only some portion of the illuminance is lost. Also, owing to the long lifetime of LEDs, the traffic is less disturbed by maintenance and emergency service. Another advantage is the absence of a large reflector, which is often a cause of accidents by drivers unable to distinguish between a lit-on red signal and an extinguished green signal exposed to a low sun (the sun phantom problem). Although high-brightness LEOs contain a built-in reflector cup, its area is negligibly small and the unlit signal is always recognized as dark. Furthermore, even if some light is retlected from an LED signal, the reflection is not colored, since LEOs with transparent encapsulants contain no colored parts.

Traffic signals must meet strict standards that are subject to regulation by national and international institutions (see, e.g., Hodapp 1997), and separate standards are being developed for LED-based traffic lights. For instance, in the United States, the Institute of Transportation Engineers offers an interim purchase specifi-



cati~n within, the VehiCle Traffic Control Signal Heads standard (see ITE 1998). In particular, th~s standard requires that the color coordinates of the signals fit predete~mn~d regions o! the ~IE I ~3 I chromaticity diagram. The modules must maint~m rnuurnum luminous lI~tenslt~ values specified for both horizontal and vertical VIew angles, and the luminous intensity should not exceed maximum values of 800. cd for. red, 3700 cd ~or yellow: and 1600 cd for green light. The drive circuitry must provide ~t I east 8~ Yo output I f a catastrophic fail ure of one ch i p occurs. The standard contam~ technical ?otes that des~ribe optional dimming, compatibility of LEI? m.odules w.tth conventional load switches, operating temperature range, and testing m an environmental chamber.

Typically, in the United States and Europe, AIGalnP or AIGaAs LEOs with an emISSIOn peak above 620 nm are suitable for red, AIGaInP 590-nm LEDs are s.ultable for amber, and 505-nm AllnGaN LEOs are suitable for green traffic lights. AIGalnP 600- to 605-nm LEDs meet color requirements for U.S. pedestrian "do~ 't walk" signals (Portland Orange). In Japan, the wavelength of the "go" si _ nal IS somewhat shifted to blue. g

Figure 7.2.1 shows the schematic cross section of a typical retrofit replacement SIgnal L~O head. The module contains a printed circuit with soldered LEOs, seco~da~ OP~ICS to enhanc~ optical performance, and an electronics unit. Usually, the ~Ircultry mv.olves multiple cross-connected strings to increase viability (see Section :.1). A lightly colored polycar~onate COver lens is sealed to the housing to protect internal components from moisture. The outer rain-prote.ctive gasket IS fastened to the door of a standard traffic light lantern.

_The l~w-power. e~ectronics unit can contain additional compensation circuits for mcr~as~ng the driving current at elevated temperatures (early LED-based heads were crincized for ?utput v.ariatton with temperature). Such functions as dimming at dark, compensation of line voltage vananons, and wireless control can easily

Housing Gasket

Housing --+-[

Drive Electronics _--..... ....

Secondary Optics


Standard Connectors

LED Assembly

FlG.7.2.1. Schematic cross section of a retrofit replacement signal LED head. (After Evans 1997.)



be implemented. Because of the relatively Jow power consump~ion of LED ?eads, drive units can be equipped with backup batteries for us~ dunng. po,:er failures, AU-battery and solar powering of the signals became feasible, which IS Important

for traffic management in remote areas. .. . .

LED-based power signals are superseding incandescent lamps In railroad signage (Evans 1997). In this a~plication, high .relia.bility is of primary importance. Railroad power signage compnses grade crossing lights and several types of wayside lights that utilize red, yellow, green, and blue, as well as lunar white colors (see Rea 2000), all available using solid-state lighting technology.

7.2.2. Automotive Signage

Automotive signage is a promising large-volume application of solid-state lighting. Performance of high-brightness LEOs is high enough to supersede conventional light sources in all automobile signal lights [stop, tail, rear and front tum, side markers, and center high-mounted stoplight (CHMSL»). This means that potentially, several hundred LEOs can be installed in a vehicle. As annual worldwide production of cars and trucks is about 60 million units, automotive si~nag.e offers a tremendous market for high-brightness LEDs. The process of replacing incandescent lamps by LEOs in automotive signage has already started. However, its acceleration depends strongly on a decrease in LED cost, since most automobile manufacturers demand inexpensive technical solutions.

Advantages of automotive LED signage include longevity, low power consumption, durability, and short switching time. Actually, solid-state lamps are "perpetual," since their lifetime exceeds the effective lifetime of vehicles. Therefore, replacement costs are excluded, no maintenance is required, and the space and weight required are minimized. Lower energy consumption requires a smaller alternator, and the cross section and weight of wiring can be reduced as well. The power conversion efficiency is even more important in vehicles that are completely or partially run by electric power (hybrid-, fuel-cell-, or rechargeablebattery-powered vehicles). There are safety benefits as well. First, abrupt signal failure is almost excluded. Second, LED stoplights feature instantaneous switchon, compared with -0.1 s for a bulb filament to light up. This improves driver response time, when the vehicle in front of a car brakes, and several meters of additional stopping space is saved.

All colored automotive signals (red and amber) are compatible with AIGalnP technology. However, like traffic lights, automotive power signals must satisfy numerous requirements of national and international organizations. The specifications involve color, luminous intensity, angular distribution of radiation pattern, illuminated area, mounting position, and geometric visibility. Hodapp (1997) reviewed many aspects of automotive signage (color, luminous intensity, and flux requirements, as well as optical, electrical, and thermal design considerations). Typically, red lights (stop, tail, rear tum) require central wavelengths above 610 run, and amber lights (rear and front turn) require wavelengths around 590 run. Minimum luminous intensities vary from 1 cd for side tum and marker



la~n?s to 200 cd for turn lights. Minimum luminous flux required is 0.23 1m for tailtights and around 231m for amber turn lights. This means that signals might co~tam from 1 to 75 chips, depending on LED output (Decker 2000). Such organi_ zauons as the Society of Automotive Engineers (U.S.) has standards that refer t LED-based automotive lighting devices. SAE (I 999) provides methods for coloo and thermal testi.ng, performance and design requirements, and guidelines applica~ ble to motor vehicle slgnalmg and marking lights that contain LEDs.

The high vibration and shock resistance of LEDs, as well as the reduced size offer new design possibilities. For instance, solid-state signals can be mounted 0 ' a trunk lid, since in contrast to incandescent lamps, they withstand slamming when SWitched o~. ~mce LEOs produce little heat, plastic optical systems can be placed 111 the proxinuty ?f the s~urce. This allows the use of efficient, compact optical fixtures and the introduction of new design styles (Decker 2000) For instance LE,?s are suitable for coupling into light conductors. Figure 7.2.2 depicts a sche~ mane cross section of an automotive signal that is based on a plastic optical conductor c?upled with an LED. The reverse side of the conductor is shaped to outcouple light. The optical conductor can be designed in cyl indrical, thin-sheet or oth~r g~ometries that tit the vehicle contours. Conn and Bannet (2001) consid~red design Issues related to a smooth lighted appearance of an LED-based automotive signal lamp.

The first automotive application of high-brightness LEOs was the CHMSL mandatory in the United States since 1982. This signal must emit at least 25 cd of luminous intensity at a flux of 2.9 1m. The 1994-1996 Ford Thunderbirds were equipped with a CHMSL composed of red AIGaAs transparent-substrate LEDs. Later, CHMSLs that contain 626-nm red AIGaTnP LEOs were designed. Performance of the present LEDs allows one to reduce to six the number of chips in a CHMSL.

AIGalnP LEDs penetrated rapidly into the truck and bus taillight market where design and installation of the fixtures are relatively simple and cost considerations are less important. For cars, the entire LED-based tailli zht section was intr~du~ed by General Motors on the 2000 Cadillac DeVille (Buchholz 1999). The section IS composed of 72 red AIGalnP LEOs. Four chips are used for the side marker lights, and the remaining 68 chips serve as the brake light. Of the latter, 16 LEDs can handle the taillight function when the brake signal is not lit. Front

FIG. 7.2.2.

Schematic design of a vehicle light signal that features a light conductor optical system with a coupled LED. (Afler Decker 2000.)



flasher lights based on orange-LED clusters were designed for the VW Golf. T.he Dew BMW 5 series features rear lights based on optical waveguides coupled With LEDs.

7.2.3. Miscellaneous Signage

In addition to traffic and automotive lights, diverse high-brightness LED-based signage is evolving rapidly. Belo~, we describe typica~ appli~ations.

Exit Signs. Among the first Wide applications of high-brightness LEDs are the exit signs that are mandatory in large public buildings. These signs. are perm~nently lit to assist evacuation during power failures. ConventIOnal. signs contam incandescent, compact fluorescent, or tritium lamps. Compared With other light sources LEDs offer lower power consumption and a longer lifetime. Long-lasting signs c~ntaining radioactive tritium are more expensive than LED-based u~its. Also, the public opposes radioactive signs (Peralta and Ruda 1998). In the Un~ted States, where the exit signs are red, the use of LEOs in exit signs became practical since development of the first AIGaAs high-brightness chips. Typically, I - to 5- W LED fixtures substitute for 12- to 50-W fluorescent or incandescent signs. LEDbased red signs save up to 90% of power and require a much smaller backup battery. In Europe, where green exit signs are required, the penetr~tion of AlinGaNbased LEOs into this signage application started later and IS stili less costeffective.

Safety Beacons. A variety of beacons are widely used for ,:vamin~ pu~pos~: in vehicles (off-road vehicles, trucks, etc.), aircrafts, towers, industrial facilities, roadway zone safety lights, marine lights, and in potentially dangerous environments. Beacons are usually operated in a flashing mode that causes no LED deterioration, in contrast to incandescent light sources. Owing to power and maintenance savings, high reliability, longevity, and durability, high-brightness LEOs are rapidly penetrating this sphere. An example of beneficial replacement of conventional lamps are the solar-powered marine lights used in various kinds of buoys and navigation devices and as fishery markers. Such mai~tenance-free lights are available for all relevant colors (red, yellow, green, and white), LED-based marrne Lights can provide luminous intensities of hundreds of candelas that ensure visibility in the range 4 to 6 nautical miles. Since smaller batteries a~d solar panels are required, buoys maintain stable positions because of reduced wind loadings and a lower center of gravity. Other examples are handheld safety flashlights and flashing safety vests, which provide the wearer with high visibility in dangerous work environments such as highway construction (Dixon 2000).

Airport Runway and Taxiway Lighting. Airport lighting systems consist of low-intensity runway and taxiway markers embedded in the pavement. Typically, blue 2-cd and green 20-cd lamps are used for taxiway edge and centerline lights (Peralta and Ruda 1998). Solid-state lighting technology offers highly rei iable low-maintenance markers based on single green and blue AllnGaN chips or small LED clusters. A promising solution is a light strip system that encapsulates a series of closely spaced LEDs to provide a continuous marking light.



Roadway Crosswalk Warning Systems. Many crosswalks that are not at intersection areas are not equipped with signals and require safety improvement. Lewin and O'Farrell (2000) described an LED-based pedestrian crossing warning system that utilizes highly directional in-pavement roadway markers. The system contains amber LEDs that start flashing when a pedestrian is present in the crosswalk. Flashing yellow LEDs can also be integrated into a supplementing pedestrian warning sign. Evaluation of test installations has shown that the system has a positive effect in increasing driver awareness, especially under conditions of low visibility.


High-brightness LEDs provide luminous intensities sufficient to display information for large viewing distances and/or in the direct sunlight. Therefore, mature AIGarnP and AlInGaN technologies made feasible various kinds of LED-based displays (Hodapp 1997). Of those, the most important are alphanumeric displays (see Section 7.3, I) and full-color video displays (Section 7.3,2). Typically, LEDbased displays employ multiplex driving that reduces the number of driving units (Gillessen and Schairer 1987, Hodapp 1997).

7.3.1. Alphanumeric Displays

LED-based bright variable message signs (VMSs), the descendants of alphanumeric displays in watches and calculators, are widely used for advertising, communicating news, and displaying scores in sport arenas and travel information in airport, rail, and bus terminals, Such signs show temperature and time and provide motorists with roadway and parking information,

Solid-state lighting technology is competing for alphanumeric display appiications with a variety of alternative technologies (Hodapp 1997). These involve incandescent lamps, fiber-optic bundles, and reflective electromechanical pixels (flip disks or rotating prisms). Colored LEDs are rapidly superseding both filtered and unfiltered incandescent lamps owing to low power use and low maintenance costs. Fiber-optic bundles, which deliver light to alphanumeric pixels from a remote central source, are also expected to convert to solid-state technology (LEOs or semiconductor lasers). In this application, solid-state technology offers higher efficiency and simplicity of operation. Flip disks contain mechanical parts that are wearing out, produce noise, and require supplemental lighting in the dark. However, they have an advantage over LEOs in power use since energy is consumed only during the switching. Hybrid LEDlflip-disk displays that combine the advantages of solid-state and electromechanical technologies are being developed.

Typically, alphanumeric displays contain orange or amber AIGalnP LEDs.

Red AIGalnP LEOs are also used, mostly for indoor applications. Characters are composed of pixels (e.g., a 5 x 7 matrix). Depending on viewing distance and LED brightness, each pixel can contain from one to several dozen LEDs. The dimen-



. s of the characters also depend on the viewing distance and can range from

sian . I·" .

~ "or a 30-m distance to 150 cm for a 1000-m distance. In genera -information

y crn l' d I··

d. lays horizontal viewing angles can reach 170". For roa way app icanons,

ISP , .

reduced viewing angles (15 to 30°) are sufficient. . .. .

Owing to low-voltage control circuitry, high reliability, and energy efficiency,

LED-based VMSs are easily integrated. into intelligent transportauon systems (ITSs), which are being deployed along highways to mana~e traffic flows (Evans ] 997). ITSs utilize LED displays to alert motorists when incidents occur and to provide real-time information on alternate routes and on road and weather condi-

tions. . 1

Portable, intelligent trailer-mounted LED-based VMSs be~ome practica as

well. For instance, they are indispensable for ensuring safety 111 roadwa~' work zone areas (Evans 1997). Typical signs are capable of displaying messages 111 multiple lines of characters that are visi~le fr?m a distance .of 4?0 m 24 hours a. day. The character arrays measure 45 ern 111 height by 30 cm 111 width and a:e deslg~ed as a matrix of amber AIGalnP LEDs. The VMS operates over the entire ambient temperature range, exhibits high corrosi?n resistan~e, and withstands w~ather elements (wind, rain, snow, sand, and sunlight) and high levels. of m~ch~nlcal s~ock and vibration. The sign is driven by a solar-powered electncal circuit combined with a rechargeable battery, which is able to operate for three weeks WIthout sunlight. The VMS features a controller that accommodates a lot of preprogrammed and user-defined messages. A cellular pho~e dial~up opti?n pr_ovides the user with convenient remote control. A speed detection option, which displays the speed of an approaching vehicle, is also available.

7.3.2. Full-Color Video Displays

At present, large full-color video displays are the major application of highbrightness LEDs. The first installations occurred in downtown Tokyo and Manhattan soon after the appearance of AlinGaN green and blue LEDs. Later, huge LED screens found many applications in advertising an? are quickly becoming an integral feature of entertainment arenas and sports stadiums.

Because of the fast switching required, only cathode-fay-tube (CRT) techno!ogy was formerly suitable for large video displays,. How~ver, this technology IS becoming obsolete, especially in outdoor installations, sll1ce. LED-?as~d VIdeo displays have a higher brightness, require only low-voltage driving circurtry, and

exhibit low power consumption, small weight, longevity, and rO?Ust~1ess. .

The basic element of large video displays is a trichromatic pixel that emits light of a required chromaticity using additive color mixing. The pixel is composed of red, green, and blue LEOs. Small pixels contain one LED of each color, whereas in larger pixels, from two to six or more LEOs of the same color can be used (see Fig. 7.3.1). Typically, horizontal viewing angles must be as large as pos-, sible (up to 170°) and vertical view angles vary from 30 to 900• Each w:oup ~t colored LEDs in a pixel is driven by 10- to 12-bit drivers, ~at 1024 to 4~96 I~te.nslties). This makes from 1.0 to 68 million combinations ot mtensiues, WIth millions



0®G 00 800

FH;.7.3.1. Possible pixel configurations used in large-area video displays.

of chrornaticities and thousands of brightness levels possible. The pixel size depends on the viewing distance and varies from 4 to 15 mrn for indoor applications and from IS to 40 rnm for outdoor products. Usually, a display is formed ofmodular panels that contain from 4 to 512 pixels. The panels are arranged side by side to provide for up to several thousand pixels width and height, Large displays can measure up to tens of meters by tens of meters, with the total number of LEDs varying from ~I06 to _107. For instance, a giant LED screen owned by the National Association of Securities Dealers in Times Square, New York, contains 8200 modular panels with 16 x 16 (256) pixels in each (Schweber 2000). Each pixel of 30 x 30 mm size comprises eight LEDs (three red, three green, and two blue). The display measures 36.6 by 27.4 m and contains 2.1 million pixels and 16.8 million LEDs. The peak lum'inance is 5000 cdJn/.

Large video displays are managed using a remote desktop-computer control with standard video and SVGA input. Processing includes digitizing of video signal, image resizing, video mixing, and color correction. The control is connected to the electrical driving circuitry via a GBPS-range serial interface using a coax or fiber-optic cable. The driving circuitry features multiplex operation and pulsewidth modulation. A typical refresh rate is 300 Hz (IO times higher than that used in CRT screens).

In most applications, LED-based video displays produce images that are consistent with TV color standards, which were developed for phosphor-based CRT displays. Since LED chromaticities do not match those of CRT phosphors, the video signal is transformed by the processor. However, AIGaTnP and AllnGaN LEDs can emit light within a variety of chromaticities, and more hues can be obtained by means of color mixing than can be obtained with present TV standards Enhancement of the color gamut of LED-based video displays has been discussed (Evans 1997, Ponce and Bour 1997).

To estimate properly the benefits of such enhancement, the color gamut should be considered within the elE 1960 UCS diagram, which exhibits more uniform distribution of hues than the CIE 1931 diagram (see Section 2.4). Figure 7.3.2 depicts the crE 1960 diagram with the color gamuts offered by different trichromatic systems. Chromaticities that can be acquired within NTSC (U.S. TV) and EBU standards (European TV) are contained within the dashed- and dotted-



540 560



.- 0.3
- 0.2
> 0.1 FlC.7.3.2.

0.0 L_~___...J~~___l-_'__~=-----,

0.0 0.2 0.4 0.6

u Chromaticity Coordinate

1960 CIE uniform chromaticity scale. c~lor dia~ram. DaShed-N~~~d-E:~d solid-line triangles embrace chromallcilies obtainable within (d" t'

d . I LED (450 52' and 626 nm) systems, respectively. SO) potn S

an tnco or- .""

mark the (II,V) chromaticity coordinates of primary LEOs.

. I· . I embraces chromaticities that can

line triangles, respectively. The solid- me tnang e 1 th 450 nm full

be obtained by mixing emissions from a b]lue [pea~5~~v~r~~:0 nm) and red width at half magnitude (FWHM) 30 nm , grtehen rea that exceeds ;hat of the

( ru'16 nm) LEDs The LED system covers ea.. .

N6T26SCnnbY 60% and th~t of the EBU by 80%. Tn particular, the .g~lIl In numhber of c 1 d d ink regions Furt er enobtainable hues is seento r~fer to ~lue-pu~h~ft~~ ;~e -r~D wavele~gths toward

bancement of these reglO~s r IS :osslb~e b~oweve~ this will result in considerable the red and blue corn.ers a t ie orses oe. ecause of reduced sensitivity of reduction of the luminous efficacy of the system b I . f th 1960 CIE the eye at the boundaries of the visible spectrum. °furh ana lYSIS or t e,n the blue-

dditi I 150/, enhancement 0 t e co or gamu

diagram suggests that a mona 0 ri h . t m with the

. d b rti g to a quadric rornatic sys e

green region can be obtame Y conve In

fourth LED at approximately 490 nrn.






Conventional L~Ds are ~i.del~ used in medicine as miniature light Sources for motion control, Image gUIding ~n surgery, production of visual stimuli for electro_ encepha.lography an.d electroretinography, capillaroscopy, and optical-transmission rnorutoring of arterial oxygen saturatiol~. Considerable radiant flux generated by hIgh-brightness LEDs enables new applications that are reviewed in this seen Recent advances in solid-state lighting in medicine involve phototherapy ofne on.

I" . ona-

La Jaun.dlce (Sectl?n ~.4.I), photodynamic therapy (Section 7.4.2), and dental.

~omposl!e cunng (Section 7.4.3). In Sec.tion ?.4.4, the potential of solid-state lightIng ~or phototherapy of seasonal affective disorder is discussed. Other promising applicatIons of LED sources, such as wound healing by means of photostimulation (Somm~relal 2001), are emerging as well. (An example of surgical illumination system IS presented in Section 7.7.1.)

7.4.1. Phototherapy of Neonatal Jaundice

JOli'!dice, yellow coloration of the skin and white of the eye, OCCurs in more than halt of newborns. It is caused by the accumulation of pigment bilirubin in the ~loodstrealll. ~onnally, neonatal hyperbili rubinemia disappears by the end of the h.r:t w~ek ofh~e, th~ per!od required by the hepatic cells to adjust to excrete all the bilirubin (phYSIologIcal Jaundice). However, high levels of bilirubin are toxic to the brains of neonates. Also, the persistence of jaundice after the second week of life indicates some pathology.

Ph~totherapy is widely used to control neonatal jaundice. Light with wavelengths in the range 420 to 480 nm is absorbed in the skin and converts the bilirubin to water-soluble isomers that are easily excreted via thc liver and kidneys. The most cO~lmonly used light Sources are fluorescent tubes, halogen spotlights, and fiberoptic blankets .. These devices have a number of disadvantages, including broadband ~pectra (resulting in heat dissipation, UV radiation, and ozone generation), flickering, h~gh voltage .supply requirements, low reliability (short lifetime, poor shock and vibration resistance), fragility with hazards from glass parts, and low efficrency, Therefore, high-brightness LEOs have been proposed as possible light sources for the phototherapy of hiperbilirubinemic neonates (Vreman et aI. 1998).

A prototype phototherapy device (Vrernan et al. 1998) consisted of three units each comprising 100 blue AlinGaN LEOs with a peak wavelength of 459 nm and a FWHM val.ue of 22 nm, The efficiency of the LED phototherapy device was compared wl.th th~t. of ~onventional devices by measuring the in vitro photodegradatlOn ot bilirubin in human serum albumin. Although the overall output power was less than r W, the irradiance in the relevant spectral range ge~erated by the LED was greater and the rate of bilirubin degradation was higher. ThIS showed that the LED light should result in a more effective treatment for neonate jaundice than light from more conventional sources.

Seidman et al. (2000) performed cl inical investigations in order to evaluate whether the LED light has the same therapeutic effect as conventional sources.



They examined 69 jaundiced, but otherwi~e healthy, newborns. Two groups of

articipants received either LED or conventional (halogen-quartz bulb bas~d) phoiotherapy with similar levels of irradiance. The LED p~ototherapy device consisted of six arrays each comprising 100 blue AlInGaN chips. !he effe~t of p~ototherapy was estimated from the measurements of seru~ bilirubin ttl capillary blood samples. Statistical analysis showed that the duration of phototherapy and the rate of decrease of bi Ii rub i n were not di fferent in these two groups. Under L~D treatment, no side effects were noted in any of the newborns. Also, the nursing staff experienced no discomfort. Based on these results, the authors conclude~ that at comparable light intensity, LEO phototherapy is as efficient as conventlon~1 phototherapy. LEDs can provide even higher irradiance with gre~ter therap~uttc effect. Tn addition, LEDs have many advantages, such as small Size and weight, low heat generation, low de voltage supply, long lifetime, durabilit~, and la~k of glass parts. This makes LED sources especially suitable for use JIl safe, lightweight home phototherapy devices.

7.4.2. Photodynamic Therapy

Photodynamic therapy employs phototoxic substances (photosensiti.zing agent~) that destroy cells when exposed to light. In particular, photodynamic therapy IS used for treatment for some types of cancer, including cancers of the bladder, brain, larynx, skin, and oral cavity. The photosensitizing agent (pho(o~ensitizer) is administrated into the bloodstream and localizes in the mitochondria of tumor cells. Under irradiation by particular wavelengths, the agent produces active forms of oxygen that is deadly for the cells. The widely used photosensitize~s are forms of porphyrin (e.g., hematoporphyrin derivative, porfimer sodium) that IS a component of red blood cells. These agents are activated by red light (630 nm) usually

produced by lasers. . .

Schmidt et at (I 996) introduced LEDs as a new cost-effective light source for photodynamic therapy of brain tumors. They hav,e. investigated t~e use of L~D arrays for excitation of porfimer sodium photosensitizer.. The LED light was delivered to the tissue by an applicator, similar to a conventional laser probe. The applicator consisted of a hollow steel tube with 144 LED chips arranged in acyl inder at the tip. AIGaAs chips (677-nm peak wavelength, 25-nm FWHM) ~ere u~ed. The temperature of LEDs was controlled by circulating water..The enttre. em~tt~r was placed into a 2-cm-diameter latex balloon inflated with a dl~fus~r h~UId (l~pId solution). The LED probe and laser adapter were compared by In Vitro mvesngation of brain stem toxicity in canines. The LED adapter was shown to be SUitable

for photodynamic therapy of brain tumors with porfimer sodiu.m. .. .

Since porphyrin-based agents have several drawbacks (skin Se~Sltl~atJOn, slow excretion, not the optimal excitation wavelength for deep penetrat~on mto tissue), new photosensitizers are being developed with simultaneous sel.ectlOn of appropr.late LED sources. Colasanti et at. (1997) demonstrated the In uuro phototoxic properties of two derivatives of chlorin, green natural pigment. The excitation Source was an array of 260 AIGaAs LEOs emitting in the wavelength region



around 675 nm (22 nm FWHM), where tissues are relatively transparent. Schmidt et al. (1999) investigated in vivo a second-generation photosensitizer benzoporphyrin derivative excited by an LED source for photodynamic inhibition of brain tumor (glioma) growth, They used an LED probe similar to that described by Schmidt et al. (1996). The emission at 688 nm was shown to result in significant glioma toxicity, whereas the normal brain tissue encountered only limited toxicity under appropriate conditions.

Ignatius and Ignatius (1998) pointed out the potential of LED-based SOurces for photochemical purging of autologous bone marrow grafts. In an autologous bone-marrow transplant, a portion of the patient's own bone marrow is removed prior to radiochernotherapeutic treatment and preserved at cryogenic temperatures. After treatment, the stored marrow' is reinfused to facilitate recovery of the damaged bone marrow. However, the preserved marrow may contain live tumor cells. To decontaminate the marrow grafts, a photosensitizer merocyanine 540, which is effective against a broad range of leukemias and lymphomas, can be used (Kubo and Sieber 1997). Excitation of rnerocyanine 540 by 525-nm LEDs was shown to he more effective than by using fluorescent lamp sources.

Ben-Bur et at. (1999) applied an LED source for photodynamic inactivation of viruses in red blood cell concentrates. Decontamination of blood components is required for increasing the safety of the blood supply. For this purpose, a photosensitizing agent, the sil icon phthalocianine Pc 4, was introduced. Pc 4 targets the lipid envelope of pathogens; red cells can be protected by means of quenchers. An LED array emitting 670-nm light was employed to excite the photosensitizer at the peak of the absorption band. This technique was shown to be capable of inacti vating even the most resistant viruses, such as human immunodeficiency virus (HIV).

7.4.3. Photopolymerization of Dental Composites

Light-activated dental composites are now widely used. Polymerization is usually performed by excitation of a photoinitiator, camphorquinone, that is present in a composite. This photoinitiator absorbs I ight in the 410 to 500 nm range, with the peak of the absorption spectrum at 468 nm. The photocxcited camphorquinone facilitates the production of free radicals that invoke polymerization of monomers contained in the restorative resin. The most commonly used light sources are halogen bulbs in combination with colored glass filters. However, halogen-bulb-based I ight curing un its (LC lJ s) have low effecti ve 1 i fet i me « I 00 hours) because of high-temperature-induced degradation of the bulb, reflector, and filter. Often, these units are not maintained properly and therefore have a reduced power output. This may result in reduced monomer conversion and increases the risk of premature fai lure of restorations.

To sustain curing quality, Mills (1995) proposed to replace halogen lamps with high-brightness blue LEOs that undergo little degradation for many thousands of hours. An example of an LF.D-based LCU unit, which consists of 25 blue LEDs and a polymer optical taper, was described by Mills et 01. (1999). The LEOs illuminate the wider end of the taper (15 mm). Due to internal reflection



. . concentrated at its narrow end (6 mm). In contrast to the

in the taper, the hght IS , rna fan The performance of the LED

halogen LCU, the LED uni~ re;~I~e~t~ ~~~v~~ion~1 halogen-bulb unit. Although LCU unit was compared With 64!1< of the irradiance of the halogen unit, it exhibthe LED LCU produced only . o. St I I et al (2000) attributed this effect to

. d hi her depth of polymerIZatIOn. a 1 . f

)te a g.. D emission spectrum to the absorption spectrum 0 cam-

better matching of the LE , . d ble part of the filtered halogen-bulb emis-

horquinone. In companson, a cons: era

~ion is wasted because of the broa~er spe:tr~;:i flexural properties of three differ-

Stahl et al. (2000) mvest~gate mehc;m h' h were photopolymerized with

com osites with three different s a es, w IC 2. .

ent P , b Ib LCU (755 mW/cm irradiance] Of a blue-LED

·ther a commercial halogen- u .' diff

~CU (350 mW/cm\ respectively. In most case\:~ pS~:~~~;i~~dSI;;lt~~c:~\der:~~

. echanical properties between compost d 1

ences In m I h that compact uncooled, an ong-

LCUs were observed. These resu ts s ow. er;ede conventional halolifetime LED light curing units have the potential to sup

gen-bulb LCUs.

7.4.4. Phototherapy of Seasonal Affective Disorder

. lications of phototherapy is in cou~teracting

One of the mo:t recognlz~d ap6) SAD is a mood disorder that strikes some seasonal affectwe disorder (SA I '. d (mostly women and younger people).

I h live 1ll the northern atitu es m d d

peop e w 0 . h I ith SAD experience depresse moo ,

During the fall and winter mont s, ~eop e WI d I . CI weight gain social

k f ti f on excess ive eating an s eept ng, '

fatigue, lac 0 mo Iva I , f SAD are not known. The

withdrawal, and decreased libid,o. The exact cat~ses oOf melatonin a sleep-related

., th t SAD IS due to secre Ion '

widespread opinIon ~ . d f d k s bv the pineal gland, is challenged

h duced during peno s ° ar nes . . . SAD'

ormone, pro , (Lee et al 1997a), Despite this ambIgLllty, IS

by a variety of other theones .' idi t ated successfully by exposure to

. d t al light vanauon an IS re . f

att.nbut~ 0 season Common treatment involves application 0 a

bright lIght (Rosenthal et al, I ~89~ t tubes mounted on a metal reflector. light box, which is a set of white uorehscen s the residual UV component of The box is shielded by a plastic screen t at remove

the emission. .' . h b " Ilt 'S red LEOs for SAD treatment.

Levitt et al. (1994) applied hig - . r~g ~es d AIGaAs LEDs (660 nm). Two

They used head-mounted units com~r~slllg -c . f subiects. A notice-

. • .' I admlIllstered to two groups a J

very different dlummances w ere I h h significant difference in response

hi t effect was achieved, a t aug no ., im li h

a e treatmen . , " br I t 1 ight and those recervmg dun ig t.

was observed between patients receiving Ilg 1. . onse depends only weakly on

These results implied thatei:her the the~apeutIC :~~~c treatment (placebo). illuminance or the red LEI? light ~~~uilt~~~:s~n~DS seem to exhibit a somewhat

Green, b~ue, and white hig r g 8 sed on metaanalysis of published and

higher potential for SAD phototherapy. La t I (1997b) concluded that light of unpublished data on SAD Photothl era;py, nJee ~IO:) is essential for the therapeutic short to medium wavelengths (b ue gree ye



e.ffect, whereas red light and UV radiation are relatively ineffective Th

srons suggest rapid penetration of AIGalnP and AllnGaN LED .' hese conclu, ment of compterety solid-state light boxes and head-mounted U~i:;t~n tc e deve.lop_ with convennonal phototherapeutic devices, LED-based tech nolo' f~m~anson ad.vantages, such as variable chromaticity and color temperature dg~ 0 .ers ~nany rmttent operati I . . ' Imming, mter-

ron, ongevlty, safety, low heat generation, and high efficiency.


PrhofosynthesiS in green plants and Some microorganisms is a photon,invoked ~ ocess by which oxygen and organic compounds are produced from carbon dio Ide, water, and minerals. The primary conversion of the photo . xel troni " . n energy mto the

ec ro~1C eXCitatIOns required for photosynthetic reactions occurs in chlorophyll

green pigment. Chlorophyll absorbs fight in the broad wavelength range fr ' a proximately 400 to 700 nm. Figure 7.5.1 depicts a schematic spectrum re~~tt:from the photosynt?etlc quantum action of plants. This spectrum is c1ose'to thg chlorophyll absorption spectrum (McCree 1972) Bula et at (1991) die LED h . . suggeste t rat

s w ose spectral output peaks within the red maximum of the cbloro h II

quantum action spectrum offer improved photosynthetl' ffici . p Y

id bl . . c e icrency, smce a con

SI era e pornon of the short-wavelength photon energy is converted to heat-

Since that urne, man? attempts have been made to develop LED-based . hotos -nthet~~ systems. In this section, the use of high-brightness LEDs in Plan~ urow;n

(Section 7.5.1) and photoblOreactors (Section 7 5 7) I' d ib d '" g

.. _ s eSCTl e .

. '. " . ,

: '

• •• , .

, ,

• " It

" " " " • •

" • •

· , , ,

• I I ,

• I I








Wavelength (nrn)

FIG. 7.5.1. S~hem~ti~ spectrum of p~oros)'nlhelic quantum action of plarus (bold line: a er c fee 1972). ElTlISSlO1l spectra of red AIGaAs (660 nm) and bl

AlInGaN (450 nm) LEOs arc also indicated. c ue



7.5.1. Plant Growing

Bula et al. (I 991) introduced high-brightness red LEOs as a radiation Source for plant growth. An array of 540 double-heterostructure (DH) transparent-substrate (TS) AIGaAs LEOs with a peak wavelength of 660 nrn and FWHM of 25 nm was used to cultivate lettuce. The radiation output of LEDs was supplemented with -10% flux of a blue fluorescent lamp in the spectral region 400 to 500 nm (the use of blue LEDs was predicted, however). The 2! -day-old plants exhibited characteristics similar to those of plants grown under similar radiative flux from cool-white fluorescent and incandescent lamps, The electrical power conversion efficiency of the LED system was compared with that of fluorescent systems. It was estimated that the LED source was at least twice as efficient as the fluorescent-lamp system. At the moment of writing, advanced AIGalnP LEOs exhibit 2.5 to 5 times higher efficiency in the relevant wavelength range (Krames et at. 1999) than that ofTS AIGaAs LEOs used in 1991.

Barta et al. (1992) proposed space-based plant irradiation source. They employed high-brightness DH AIGaAs LEDs assembled into arrays containing 17,200 chips per square meter. Spectral features and the efficiency of OH-TS A1GaAs LEDs with the emission spectrum peaked at 660 nm (20-nm FWHlvt) were compared with the characteristics of conventional light sources (highpressure sodium, metal halide, and cool-white fluorescent lamps). The OH-TS AIGaAs LEOs exhibited the efficiency of power conversion into photosynthetically significant radiation comparable to that of conventional sources. In addition, LEOs offer many advantages, such as low mass and volume .. long lifetime, and safety that are important for space-based applications. However, once again, it was noted that plants require some irradiation by blue photons (supplied by supplementary blue-phosphor fluorescent lamp). The LED-based irradiation source developed was incorporated into a complete space-based plant growth facility that included a water and nutrient delivery subsystem, a nutrient composition control subsystem, and an automation robotics subsystem (Bula et at. 1992).

The effect of the narrow emission spectrum of LED on the morphology of plants was studied in detail. Hoenecke et at. (1992) examined the growth of lettuce under different sources of light. When red LEOs were used as the sole source of irradiance, seedlings exhibited some structural abnormalities. The addition of blue light maintained healthy plant growth. Similar results were obtained for pepper plants (Brown et ai. 1995). Tripathy and Brown (1995) compared wheat plants cultivated under continuous illumination by red-LED light of varying intensity and under while light (fluorescent and incandescent). Pure red light was found to inhibit abnormally chlorophyll accumulation in the leaves when the roots of the seedlings were exposed. However, no abnormalities were observed when the red light was supplemented with 6% blue light. Again, Goins et al. (1997) discovered that larger wheat plants and greater amounts of seed were produced when the red LEOs were supplemented with 10% blue light from a fluorescent lamp.

Tennessen et al. (1994) reported on the thorough evaluation of photosynthesis process under red LED illumination of kudzu leaves. They compared narrowband LED emission with broadband white light for photosynthetic rate, the ability of




light to open stomata (tiny pores in the epidermis of a leaf through which gases and water vapor pass), and isoprene production (indicator of adenosine triphosphate status). No significant difference in net photosynthetic response and isoprene production was observed. However, pure red light resulted in some decrease in stomatal conductance that can limit the rate of photosynthesis. Addition of blue photons was suggested to remove this drawback.

Photosynthetic activity of plants may be affected by shading the light by canopy. Canopy that fills up space above the leaves alters the spectrum of the incident light by reduction of blue and red and relative enhancement of green and far-red wavebands. To simulate the effect of shading, LEDs that emit far-red light in the 7JO-nm range are used (see, e.g., Heraut-Bron et at. 1999).

Soon after the appearance of high-brightness blue LEOs, Okamoto et al. (1996) developed various kinds of completely solid-state apparatus using red AIGaAs (660 nm) and blue AllnGaN (450 nm) chips. In their first experiments, 88 red and 88 blue LEOs were assembled on a printed wiring board with equal distribution. The resulting flux contained 33% of blue photons and 67% of red photons and was demonstrated to result in healthy growth of lettuce seeding. An advanced integral apparatus consisted of a metal chamber, ventilation fans, an LED board attached to the inside ceiling, and a de current source that provided individually adjustable current for different groups of LEDs. Up to 1120 LED chips were employed (in some experiments, 730-nm LEDs were also used). Plant cultivation under pure red, pure blue, and mixed light with an adjustable ratio of blue and red was investigated, A computer-controlled multichamber apparatus was also introduced, To facilitate plant growth experiments, Takita et at. (1996) developed a computer simulation system that visualized angular and spectral distribution of the photon flux irradiated by a two-component LED array.

Plant growth under illumination by a dichromatic 660/450-nm LED system was investigated in detail. Yanagi et al. (l996a) studied the effect of photosynthetic photon flux produced by red and blue LEDs on growth and morphogenesis of lettuce plants. Lettuce plants grown hydrophonically for 20 days exhibited higher leaf development under blue/red LED light than under pure red or pure blue. Yanagi et af. (1996b) examined the net photosynthetic rate of mature strawberry cultivated under eight different combinations of blue and red illumination. The highest rates were obtained for an approximate 1:2 ratio of blue to red power. Also, pure red flux exhibited 2.5 times higher photosynthetic efficiency than did pure blue tlux. Tanaka et of. (1998) demonstrated growth of orchid plantlets under blue and red LED illumination.

In summary, the photosynthetic red/blue LED system matches well the requirements for healthy growth of plants. In addition, the present performance of high-brightness blue and red LEOs makes this system more power efficient than that of any conventional lamp system. A prototype ail-solid-state source for future commercial plant growth has already been demonstrated.



7.5.2. Photobioreactors

II " rforrned by phototrophic

. t f bi otechnologies rely on photosyn leSIS pe -. .

A. vane Y ~ 1 I . rti ular algal cultures such as Chlorella vulgariS are. SUI.tmicroorgalllsms. n pa IC , .' d ox > regeneration 111 able for wastewater treatment, agriculture applicatIOns, an oxygen

ecological life-support systems,. ' I Lee and Palsson

T loit the commercial potential of algae efficieru y, . .

o e~~si ned, constructed, and tested an LED-based. hlgh~dens~ty

(1994) g (PHn) (FI'g 75?) Thev utilized arrays of hlgh-bnghtness

hofObwreactor "- .' .- . J .", ,'d PBR

P . t 680 rm to provide a power density ot light 1I1S1 e a

AIGaAs LEOs el11lttlllg a I. .' . . ,

f23 mW/cn}. Simple delivery of light by placmg chips inside the react~r,~~~

;hoton loss, low heat,generat:on, and absence of harmful wavelengths make

'llumination optimal tor PHRs. . h b

1 , LED-based algal PBR for regeneration of atmosphere 111 I e a -

A compact '. .,' (L'e and

sence of gravity was developed for long-term manned mlsslO~s 111 spaceadd~tional Palsson 1995). Small size and weight, as well as long lifetime, were

d ta es of space applications of LED-PBR.. ' d

a vaLn g d p. I n (1996) studied the effects of dlfferent-\.vavelength LED an

ee an a sso . LEO··tb the

uorescent lamp irradiation on cultivation of Chlordfa vulgariS. .r- s ~I.

f1 I' h 660 10 700 nrn supported the growth of the algae,

peak wavelenat 1 111 t e range .' 1-

The introduction of blue LEOs did not improve the gro\,,:h PCI formance I n com

. SOil with the tluorescenl lamp, LED irradiation was to.und t? reduce t .le aver-

pan! f the cells' however this invoked no alteration ot the biomass pro-

age vo ume 0 , ,

ductiTonkrate. f I (1995) devised an LED-based PHR for cultivating cyanobactc-

a ano e a, bil: . uch as

rium strains. Cyanobacteria are an important source of ,phY~o I Ipr~t~l~n~;aS nosti~

phycocyanin, that have unique Ouorescence properties and are usef, sog I' h-

d ti 'S An LED lamp unit was constructed rom 11g

reagents an cosmc IC .

LED Illumination Chamber

Ultrafiltration Unit

Gas Control


Medium In

Circulation Pump

FIG. 7.5.2.

'. ' . f L'-'O-based pholobioreaclor. (After Lee and pulsson

Scnemanc design 0 an L




brightn~ss AIGaAs ~EDs. ~ed light was found to be essential for phycocyanin synthesis, In comparison with a broad-spectrum fluorescent lamp, red LEOs resulted in higher productivity of both phycocyanin and biomass.

Peculiarities of photosynthetic reactions imply that time-modulated lizht can result in higher efficiencies of chlorophyll cycles. Therefore, LEOs are an ideal source, since they are easi Iy switched on a nanosecond time scale. Nedbal et at (1996) exploited a simple means of LED modulation for comparison of intermittent and equivalent continuous light regimes in PBRs. Both algae and cyanobacteria exhibited peak growth rate at the intermittent light periods in the lO-ms range. Matthijs et al. (1996) used 5-~ls flashes from 659-nm LEOs at different repetition frequencies to optimize light-energy conversion in an algal strain. Enhancement of photosynthesis by flashing in the kilohertz domain was observed.


Conventional LEOs are widely used in various instruments as compact and stable light sources for optical transmittance and reflectivity measurements, as well as for photodetector cal ibration. High-brightness LEOs generate fluxes of light that are sufficient for the excitation of detectable fluorescence levels or for altering optical properties in a variety of objects, In this section the instrumental applications of high-brightness LEOs are reviewed.

In comparison with conventional optical excitation sources, such as arc lamps and .Iasers, LEOs offer advantages of a highly stable and low-noise response, elcctrornc modulation, low heat production, small dimensions, longevity, durability, and. low cost. Although LED-based fluorirneters were developed years ago (see SmIth. et af. 1988 and references therein), the development of high-brightness and, especially, blue and UV chips caused a dramatic increase of LED-based fluorimetry appl ications, especially in the fields of biochemistry and bioengineermg.

In the simplest LED-based fluorimeters, the intensity and spectral distribution of the fluorescence are analyzed in a continuous regime. This kind of devices is used for detection and spectral identification of various molecular compounds (see Section 7.6.1). Another kind of fluorimeters employ LEDs that operate in pulsed an.d high-frequency modulation regimes. Owing to simple means of generation of microsecond and nanosecond light pulses and modulation in the I DO-MHz range, low-cost tluorescence li fetime measurements became feasible. The lifetime can be measured in either the time or frequency domain, as described in Section 7.6.2. Other applications, such as anti-Stokes luminescence stimulation for sediment dating and photoreflectance spectroscopy, are described in Section 7.6.3,

7.6.1. Fluorescence Sensors

Compact and inexpensive LED-based f1uorimeters are used for fluorescence analysis of a variety of objects in physics, chemistry, biochemistry, bioengineer-



ing and other spheres. Typically, a fluorim:ter comprises ~ high-brightness L.ED that illuminates the object; a spectral distinguisher, ~hlch fluorescenc~ passes through; and a detector [photomultiplier (PMT) or semlconduc~or .photodw~eJ. In low-cost devices, the detector is spectrally isolated from the excrtation ermssion by using various kinds of optical tilters. The simplest geometry involves an LED and a photodiode placed close to a lransp.arent cell (Fig. 7 .6.1 ~). Remote configurations employ optical fibers that deliver both excmng l~ght and fluorcscenc.e (Fig. 7.6. I b). Chemical sensors use fluorophores that are elth~r quenched or activated by the substance measured. The tluorophore can be admixed rnto the sample

or immobilized in the fiber cladding as shown in Fig. 7.6.lc.. _ .'

Simple geometries (Fig. 7.6.1 a) can be utilized in a variety ot apphcatlons, such as for chlorophyll fluorimetry in biotechnology. An example of matured LED-based tluorimetric technique is the sensitive detection of growth of ph~totrophic microorganisms, cyanobacteria (Karsten et 01. 1996). The method relies on the detection of in uivG fluorescence of chlorophyll a and bactenochlorophyll a and monitoring growth through variation of the signal. TIle excitation was performed by blue (450-nm) LEDs that were mounted around a cuve~e. To separate the fluorescence signal from the ambient light, LEDs were pulse.d With a frcq~ency of 870 Hz and the photodiode detector was combined WIth an amplitudemodulation detection setup. Different long-pass filters were used to separate longwavelength tluorescence from chlorophyll a and bacteriochlorophyll a. The fluorimeter exhibited high sensitivity, superior to that of commercial Instruments







FIG. 7.6.1.

Examples of LED-based fluorescence sensors: (a) direct measurement system; (b) remote sensor with bifurcated optical fiber bundle; (c) chemical sensor utilizing optical fiber with fluorescent cladding. Key: PO, ?hotodt:teclOr; EI', excitation filter; FF, fluorescence tiller: L, lens; C, cell: fC, l1uorescent




and comparable with state-of-the-ali systems that employ bulky xenon lamps. Sensing of chlorophyll a and chlorophyll b was demonstrated by using blue-LED excitation and a detection system comprising an acousto-optic tunable filter (AOTF) and a PMT (Alexander et at. 1997). Both fluorescence and fluorescence excitation spectra were measured. In the latter case, the AOTF was placed before the sample and broadband emission from the LED was utilized. Bennoun and Beal (1997) introduced an imaging instrument for monitoring algal colonies through LED-excited fluorescence. They used a matrix of 18 x 18 yellow (594-nm) LEOs for excitation and a charge-coupled device (CCO) camera for detection.

Compact and simple UV-LEO-based f1uorimeters can be useful for detection of various organic and inorganic compounds. Stroka and Anklam (2000) described a battery-powered device, which comprised a 370-nm AllnOaN LED and a photodiode. The fluorimeter was applied for thin-layer chromatography to monitor aflatoxins, toxic metabolites offungi that may appear in food.

An example of remote fluorescence measurement via fiber is an LED-based fluorimeter adopted for genetic engineering technology (Randers-Eichhorn et at, 1997). The sensor is used for on-line monitoring of production of green fluorescent protein (OFP), an important marker for protein expression. The device consists of a bifurcated optical fiber bundle with the common end inserted into the growth cell (see Fig. 7.6.1 b). One fiber is used for excitation by means of a highbrightness 470-nm LED. Another fiber delivers the fluorescence to a photomultiplier. Bandpass filters are used to separate the excitation and emission spectrally.

Chemical fluorescence sensors with- LED excitation have been developed. A prototype LED-based chemical sensor is an oxygen sensor reported by Mac Craith et al. (J994; see Fig. 7.6.lc). The chemistry of the device relies on fluorescence quenching in ruthenium complex RU(lI) tris(4,7-diphenylI, J O-phenantroline), abbreviated as [Ru(ddp )3]2+, by molecular 02. The ruthenium complex is immobilized within a porous sol-gel-processed glass film that replaces a portion of original plastic cladding of an optical fiber. The output light of a low-power blue LED is launched into the fiber via lenses. The fluorescence that is produced due to evanescent wave absorption is detected by a silicon photodiode. To improve sensitivity, the LED current is modulated and a lock-in amplifier is used for registration. High-brightness LEOs enabled the development of simple and cost-effective chemical sensors. Watkins et al. (1998) designed an oxygen sensor with a fluorescent film containing [Ru( ddp )3]2 + that is deposited directly on the epoxy housing of a high-brightness LED. Li and Dasgupta (2000) used transverse illumination of a liquid-core waveguide for measurements of atmospheric hydrogen peroxide and hydromethyl hydroperoxide. The chemistry of this sensor is based on the production of fluorescent trichrome through the oxidation of non fluorescent thiamine.

Of most sophisticated applications of high-brightness-LED fluorimetry, chemical sensing of nucleic acids is of high interest. Burns et at. (1998) developed a compact device for nanoliter DNA analysis that can be used for rapid, low-cost medical diagnostics. In addition to rnicrofabricated fluidic channels, heaters, and



. .' ntegrated fluorescence detectors and an

ors the device compnses I A f

temperature sensors. .' The LED excites the emission of DN reac Ion

external blue LED for excltatlOn. Belgrader el al (1999) described bac-

Products tha. t are identified by the detec:ors"d Iyzer (M,jAA) equipped with an

. . advanced nucleic aCI ana f

reria detection using an . . blue LED excitation source or

fl . t The device contams a

LED-based uonme er. . h . cti Ion of nucleic acid with a target-

. fl produced in a C am rea

rnonitonng uorescence . t f a green "reporter" dye and an

fl . probe The probe consis SOd d

specific uorogenlc. '. . the dye molecules are displace an

orange "quencher" dye. During th.e reacltlOn, ched exhibits increased green

separated. The reporter ~ye. that ,lhS no . onger. q~:pntured by a photodiode detector

.' nder LED excitation- 1 e errussron IS

emISSIOn u . I' ftware tbat generates a signal for the ana yzmg so .

7 _6.2_ Time- and Frequency-Domain Spectroscopy .

. . rovide useful information on structure, envi-

Fluorescence lifetime measurements p . ds In the time-domain

. lution of molecular compoun .

ronment, and transient evo .. after excitation with a short pulse is traced '. The

measurements, flu?rescenc~ dec~y sed to extract the decay time. Typically, tirnefluorescence transient obtained IS u . f rrned using repetitive short pulses of resolved fluorescence measurements a~e :e~k 0 lasers or pulsed arc lamps. Highlight that are produced by costly land u. y offer an attractive alternative. AI-

LED t d in a pu se regime ..

brightness s opera e . atively small they gain in stability,

though the energy of the LED pulses IS com par ffici '

.' d 1 mall size and cost-e rciency.

simplicity ofblasmg an con.tro, s , d imple LED-based arrange-

. (\995) have demonstrate a Sl

Araki and Mlsawa. . nts on the nanosecond time scale. By

ment for .fluorescence lifetime .mea_s:;:~ecurrent driver (see Section 7.1.3), they using a simple av~lanche-translsto~ LED b 4-ns pulses with a peak current of pumped a high-bnghtness AlInOa . ~ tl e LED "enerates 4-ns LN light

. . t of 10 kHz In this regime, 1 '" f h

2 A at a repetition fa e . B I I 1997) The peak value 0 t e

I gth f380 nrn (see asrur ea. .

pulses at the wave en 0 .. as sufficient to record fluorescence

pulsed LED output was 40 mW .. T~lsu~~we~ ;onventional time-correlated singlekinetics in quinme sulfate solutio g lif ti e of 19 5 ns was determined from

photon counting system. The fluorescence I e 1111 •

the transient profile of the fluorescence ~ecaJ" CCO camera to capture two-

Bennoun and Beal (1997) emp oye a from algal colonies with 40-ms

. f LED excited fluorescence .

dimensional pictures 0 - . d time-resolved two-dimenSIOnal

resolution. Liebsch et al. (2000) reporte d on a lsed LED array and a gateable

. . . . system base on a pu

fluorescence lifetime imagmg d . t 8 V and 1 2 A to produce

bl 470 LEDs were nven a .'

CCD camera. The ue -nrn t sients from chemically quenched

submicrosecond light pulses. Fluorescence ,ran ) were examined on the subrni-

. d I tinum camp exes . .

fluorophores (rut~enJum an . P a scales The technique was capable of momtonng

crosecond and microsecond time d b d'oxide concentrations, as well as pH spatial distributions of oxygen an car on I

and temperature. d '1 bilit of different-color LEOs allow

. I" f dri . g LEDs an avai a I I Y .

The Simp .city 0 .fl:lfl ts of fluorescence transients usmg the

one to implement sophIstIcated measuremen



series of excitation and sam lin ulse f .

pendent control of intensfty g d~r t" s 0 dlffcr~nt wavelengths with an inde-

(Trtilek et al. 1997). Kolber el;{ (I9;;on, d ~d mterval between the pulses a technique, called fast-repetitio~-rate (~~) fl orbunov et al. (l9~9) applied such photosynthetic energy conversion and phot huonmetry, to the ~nvestigation of

Although the time-domain measureme~?nt et~c pe~formance In algal cells. cence transient time-consuming process! s fProvlde direct msight into a fluores_ for extraction of the lifetime valu r cFessrtlOhg 0 numerous data points is necessary

, es. u ermore a pulsed itati

wlde-frequency-band pulsed el troni ' exci ation Source and

. ec rOfllCS or optoelectronics . d

In thejrequency domain the lifetim _ b ' ' are requrrea, However

I . ' e can e measured rapidly ith t d '

ana YSIS of transient decay profiles a d th ..' WI ou a etailed

tude modulation. Provided that th ,n : e~cltatl~n SOurce requires only amph.

e source IS slIlusoldali dulated

fre~uency, ca, the fluorescence is modulated at the sa~e~~ u ate at an angular

finite fluorescence lifetime T leads t . equency. However, a the fluorescence, and the modulan 0 a phase shIft .between the excitation and

1999). For a single-exponential dt;~: d~~:h of the s~gnal decreascs (Lakowicz depth, m, are given by y, phase Shift, (), and the modulation

tan& = WI',

(7.6. I)


~Y measuring the phase shift and modulation dent '.

In two independent w Thi epth, the lifetime can be extracted

. ays, IS method referred t th h

technique, requires less expensi ' 0 as e P ase-modulation

standard lock-in amplifiers; see l~e narrow/-frequenCY-band electronics (typically

.' -iarms et a. 1999). '

The emergence of high-brightness LED . .

complexity and in the cost reduct' f f s resulted I.n a ~ubstantJal decrease in are inexpensive and can be d lOIn ad p l~e-modulatlOn IIlstrumentation. LEOs mo u ate easily up It·

megahertz (Sipior et al 1996a 1997) h'l 0 requenCles of hundreds of

fluorescence lifetimes ~n the n' ord I? ier, This enables the measurement of

. anosecon nrne seal S" I

tenzed high-brightness blue and g e LED fi e. rpior et a . (1996a) charac-

lifetime measurements The LED r en b s or phase-modulation fluorescence

. " s were lased at 5 rnA .

lated. with a radio-frequency power of 4 m W B cl:rrent, which was modu-

cay tune in a standard fluoropho fl : y measunng the fluorescence de-

. re uorescern they h h h

technique is comparable to that based ' . ave SOwn t at LED-based

Pockels cell. Similar results were obtaine~n f an argon-Ion lase~ ~odulated with a Using a similar approach Murtagh t I (190;6~~orescentpH indicator SNAFL-2. sor for oxygen gas and dissolved ea. Th eveloped a highly sensitive senurernent frequency (75 kHz) a d doxyge~. dey have found the optimum rneas-

n etermme calib at'

oxygen levels and the phase shift of th '. r IOn curves that related the

Ie emISSIOn from a fJ .

com, p ex. Subsequently a variety of h . I uorescent ruthemum

LE ' c emrca sensors d 1 .

D-based phase-modulation techni S.. were eve oped using an

dioxide sensor with a blue LED q~e .. ipior et al. (I 996b) reported on a carbon

fl - excitarroe The sen I' .

uorescence lifetime of tluorophore s It; h' d . sor re res on change III the u or 0 amrne 101 (SRIOI) quenched by a

7.6.3. Other Optical Applications



pH-sensitive nonfluorescent m-cresol purple (MCP). Chang et al. (1997c) developed a similar methanol sensor that utilized a ruthenium-complex-based fluorophore. Murtaza et at. (1997) have demonstrated a blue-LEO-based pH probe that employs a ruthenium complex with pH-dependent fluorescence decay time.

Modulated UV emission from LEDs offers more possibilities, since most Ofganic compounds exhibit excitation spectra in the UV range. Sipior et al. (1997) have demonstrated phase-modulation measurements of fluorescence decay time in standard fluorophores (9-cianoanthraccne and green fluorescent protein) by driving a blue LED at a high bias current that invokes UV emission. Szrnacinski and Chang (2000) adopted original 373-nm UV LEDs for frequency-domain measurements. Efficient intensity modulation of the l-mA biased LED was shown to cover the frequency range from ! 0 kHz to 330 MHz. Decay times from 0.15 ns to 10.2 j..tS were measured in different fluorophores.

An LED-based phase-modulation technique allows one to assemble sensors into arrays to monitor several chemical or biological species simultaneously. Rabinovich et al. (2000) reported on a multichannel detection system that is able to monitor changes in the fluorescence lifetimes of many samples in real time. The system detects LED-excited fluorescence at different wavelengths by means of a multi anode photomultiplier and resolves lifetime changes of different fluorophores using phase meter software and a computer.

The high output of modem LEOs can substitute for lasers in many applications dealing with nondestructive alteration of optical properties of various objects. The alteration occurs through electronic excitations that result in detectable response. In this section, two examples of such applications are presented. One such application is sediment dating by stimulation of anti-Stokes luminescence by intense light; another application concerns the light-induced modulation of optical constants that synchronously modify the reflectance signal (photoreflectance).

Optical dating of sediments relies on freeing the electrons produced by radiation of natural radioisotopes (Huntley et al. 1985). Some of these electrons are trapped at defect sites that do not relax on a geological time scale unless excited by light. Provided that the sediment is concealed from sunlight, the number of trapped electrons is proportional to the radiation dose accumulated since the last exposure to light. "V hen illuminated, the sediment may exhibit characteristic anti-Stokes luminescence that is due to radiative recombination of carriers optically excited from the trap states. Under appropriate calibration, the amount of photons emitted is a measure of time since sediment deposition. Initially, 514·nm green emission from an argon-ion laser was used for stimulation of blue luminescence in quartz. As an alternative, Galloway et al. (1997) developed compact and economical setup using high-brightness 525-nl11 LEOs. The measurement system, which comprised an array of 16 LEOs, provided sensitivity comparable to that of the Ar+.laserbased system. Pulsed stimulation, which offers an improved signal-to-background ratio and a deeper insight into the relevant phenomena via time-resolved rneas-



ure~lents, was implemented by drivin~ the LEOs with 100-mA current pulses (Chitharnbo .and Galloway 2000). Dating of quartz and feldspars was demonstrated. Again, the performance of the LED-based sediment dating system was comparable to that of much more expensive systems using a pulsed laser, such as

an Nj-laser pumped organic-dye converter.

Ph()/orejlecta.nce. is a. contactless variation of electroreflectance spectroscopy that pro~t~ces deflvatlv.e-!t~e features In the reflectance spectra. It is widely used for s~nsltlve characterization of band structure, surface properties, and built-in electric fi.elds 1Il semicon~uctors. The reflectance is modulated by carriers that are ph~toexclted by an additional s~urce of light. Typically, mechanically chopped e.mlssl~n of lasers or. arc lamps IS used as a pump. In comparison with conventional ligh~ sO~lrces, ~Igh-brightness LEOs offer many advantages for photoreflecranee apph~atl0n~. First, the LED-pump waveform (intensity, modulation depth, and. shape) IS easier to control. Second, LEOs offer inexpensive solutions with a variety of pump waveleng:hs, high stability, broad frequency range, and absence of mechanical parts. Gasktll et al. (2000) were first to point out these advantages and to apply an LED-based technique for photoreflectance measurements. Blue m.ld green AlInGaN LEDs with output in the milliwatt range were driven with no bias at 5 kHz frequ~n.cy and] 00% modulation depth using a function generator. The LEOs were. positioned near the samples and shielded by colored glass filters to reduce detection of stray pump light. Reliable photoreflectance spectra of GaAs structures were measured.


Creati~lg a visual environment by illumination of particular objects and extended areas IS the most important a.pplic~tion of lighting technology. The principal human needs that are served by illumination are visibility of objects/patterns and task performance. However, mood and atmosphere, visual comfort, aesthetic judgment. health. safety, a~d well-being, as well as social communication, are also very important. These lighting needs are 10 be balanced with an architectural context econOI~ics, and. environn:rent (Rea 2000). Illumination is an evolving appJicati~n of sohd-~tate, lighting ~wlth a numbe: of advantages. First, the small size of chips, arraylJ~e to.rm _of fixtures, and diverse chromaticity and directionality of LED?ased 11lul~lIlatlOn offer facile integration with architectural elements and versatility In d~slgn. Second, LED-based illumination meets high environmental standards, slll~e LEOs contain no mercury as do most discharge-based light sources and ~Xhlblt low power consumption with the potential to increase the luminous efficiency (see Chapter 5). White ph_osphor conversion (PC) LEOs have already surpassed incandescent lamps III luminous efficiency and white multichip systems have caught up with fluorescent lamps (see Chapter 6).

. The major impediment is a large contribution of chip price to the cost of white light [the first term in the right-hand side of Eq. (3.7.1)] Unlike for bulbs and tubes, the price of LED arrays is proportional to the luminous flux and this hinders high-power applications. Fortunately, the price of a chip per unit of luminous



flux is decreasing by a factor of 10 each 10 years (Krames et at. 2000), and owing to the long lifetime of LEOs, this drawback is expected to disappear in the beginning of the second decade of the twenty-first century.

For economical reasons the first practical applications of solid-state illumina-

tion are low-power orland colored installations for local lighting (Section 7.7.1). Future-oriented general lighting applications are presented in Section 7.7.2.

7.7.1- Local Illumination

Low-power and/or colored solid-state illumination can only target particular objects and therefore is referred to as local. In addition to the i.nherent robustness ~f LED lamps, small weight, and low maIntenance cost, an important property IS their low heat production, which allows designers to use low-cost low-temperature plastic materials for fixtures and secondary optics (Dixon 2000). Solid-state I~cal illumination became economically justified with the development of red highbrightness LEOs. One of the first demonstrations of cost-efficient illumination was decorative Christmas tree lighting developed by the Lighting Research Center at Rensselaer Polytechnic lnstitute in 1991 (Peralta and Ruda 1998). The installation featured four-LED clusters instead of 7- W miniature incandescents.

Since that time, colored illumination has undergone considerable development. According to Krames et 01. (2000), the price charged by the AIGalnP LED suppliers to OEM manufacturers dropped to $0.06 per lumen in 2000. For a 610-nm LED with 100 lrn/W luminous efficiency (Krames et al. 1999), this translates to the cost of light of $1.6/Mlm·h, which equals the cost of light for fluorescence triphosphor and low-pressure sodium lamps and is much lower than that f?r incandescent lamps (see Table 3.7.1). Since discharge-based tubes are not available for wattage below 5 W, AIGalnP LEDs are rapidly occupying the niche of low-wattage red-to-orange illumination. For example, such illumination is required for outlining the contours of buildings and other structures for decorative and advertising purposes. Contour lighting systems, which are made of LED chains and LED-backlighted plastic tubes, are gradually replacing neon tubes. The first consumers of LED-based contour lighting are restaurants, gas stations, hotels, and other commercial enterprises that require good visibility in the dark. Despite higher cost, green, blue, and white LEOs are also penetrating this niche for decorative needs. An example is a 70-meter-long bridge in Duisburg, Germany, illuminated with 140 vertical strips of white LEOs (Dixon 2000).

Another important application of colored LED illumination is low-wattage se-

curity and landscape lighting. All kinds of stairs, ce!lar~, passageways, _ aisles, park/garden walks, and lanes can be illuminated cost-efficiently by LED hxtures

to provide orientation and psychological safety.

Since LED output can be confined to narrow angles, inexpensive low-wattage

spotlighting can be implemented for all colors. For instance, durable LED-based pocket flashlights have much higher battery life and feat~re no shock-breakab~e filament as do their incandescent counterparts. LED spotlights may be useful JIl museum and gallery applications, causing no deterioration of artwork due to heat



and UV light, which are drawbacks of co . .

sources (Peralta and Ruda 1998)' S nvent~onaJ Incandescent and haJog

. , . ameconslderatlO J' en

109 merchandIse in retail shops. ns are app Icable for sp0tlight_

. LEJ?-based spotlighting and illumination of co

~atlOns In au~omotive interior lightill . An exam mpacr spaces find many appli. Incorporated IOta automobile rearvie g . ple IS that of map-reading lamps highly efficient white dichromatic s w :mrrors (DIxon 2000). The SOUrce features ~

two blue-green AlInGaN chi ThYS em composed of four amber AIGalnP d

o t' h IpS, e lamps are equi d . h an

P ICS t at collimate and mix the LED ,ppe WIt plastic secondary

ers gradually convert to all-LED . t ?utPJ~t umformly. Automobile manufaclur Par h d In enor ighting For' t' - . sc e mo cis are equipped with LED' . '. !ns ance, smce 200 Jail

Illumination of the central console cock 1~:e~lOr. ~nentatlon lights. LEDs pro~jde latches. ' pI , IgOltlon lock, light switch, and dOor

. Shimada et al. (200 I) reported on Sur ical i . .

white PC LEOs equipped 011 both id g f . lJummat.lOn system compOsed of

(Kawak . 1 SI es a light plastic go I Th

. . amI, et. a . 200 1) supply a total lumin gg es. e goggles

cient Jllumlllance at the gazin oint . ou~ flux ~f 200 1m that provide suffl. halogen lighting by the surgeo~': head whIch IS partially sc~eened from ceiling was successfully used in an operation. 'I~~~fiL~O-based ~urglcal lighting system LEDs used was pointed out. how . icrent rendItIOn for red color of the

H' h ' ever.

. I.g -power LED-based local illumination suc .

srderatlOn, Garcia-Botella et al. (2000) devel . h as fl~odllghting is under COnt~ates the emission of a large number of LED op.ed an optical s.ystem that concen.

51!)'. The system comprises an LED. s Into a flux of hIgh luminous inten

that collimates the light (Fig. 7. 7.1 rr;~~ a~~~centrator, and a parabolic refJecto;

array employs hexagonal close-

Parabolic Reflector

Cone Reflector

FIG. 7.7.1.


LED Array

Design of an LED-based floodli . htin _

limator. The Concentrator is co g ~ sy;!em WJlh a concentrator and a coland the parabolic reflector serv;:p~se °11,a Fresnel lens and Cone reflector 2000.) , a co lmaror. (After Garcia-Botella et al.'



packed geometry with 90% utilization of the area. A Fresnel lens and a cone reflector concentrate the light into the focus of a semi parabolic reflector. The behavior of the optical system was modeled as a function of its geometrical parameters. The model calculations of the beam divergence were in good agreement with the experimental data obtained from a fixture composed of 152 AIGalnP LEDs operating at 61O-nm wavelength. Also, the device was proposed to generate white light by polychromatic LED arrays.

7.7.2. General Lighting

The purpose of genera/lighting is to provide a substantially uniform level of illumination throughout an area. General lighting includes such fields as residential, office, retail, and industrial lighting; illumination at educational, hospitality, and health-care facilities; lighting of entertainment and sports arenas; and outdoor environment and roadway lighting (Rea 2000). All these applications require high luminous fluxes, which are still costly to generate by solid-state devices. Nevertheless, a further decrease in chip price and increase in luminous efficiency will make LED-based general lighting practical in the future. At the present time, solid-state general lighting is in the experimental phase.

Although the physical properties of light produced by LEDs are well understood, LED-based general lighting applications require a deeper insight into the physiological issues involved. In this view, an important investigation of laboratory animals under solid-state lighting was performed by Heeke et al. (1999). An LED source was compared with a cool white fluorescent (CWF) lamp as to the ability to suppress pineal gland melatonin content and maintaining normal retinal physiology in rats. An array of red, yellow, green, and blue LEDs with spectral power distribution similar to that of a CWF lamp was used. Rats were housed under either LED or CWF lighting at five illuminance levels (0.1, 1.0, 10, 40, and 100 Ix). Melatonin concentration was shown to be suppressed equally by LED and CWF light. This implies that a normal sleep-wake cycle is maintained. Also, the results obtained by electroretinography and morphological investigation of the retina revealed no statistically significant difference in maintaining normal retinal physiology. Although these results refer to animal-habitat lighting, they provide strong support for physiological grounding of mammal well-being under LEDbased general lighting.

Since applications in general lighting employ large arrays of LEDs, fixtures require new design solutions and characterization. Tamura et al. (2000) demonstrated a lighting source using lQ-cd class white AllnGaNIY AG:Ce3+ LEDs and investigated its illumination characteristics. The fixture was composed of 697 LEDs arranged on a glass epoxy substrate in 17 parallel series. The LEOs were driven by 3 to 17-mA current that was provided under full-wave rectification of the ac 100 V and 60 Hz. At an electric power of about 20 W, the maximum luminance of the source was 95,000 cdlm2 and the maximum illuminance was



10,000 Ix at a distance of 30 em from the LED array. These characteristics are sufficient for such applications as reading lights or automotive headlights. Thermal characterization revealed some deterioration of chromaticity above 50°C because of reduced yellow emission from the garnet. Since the array temperature increased by 65 K after I hour of operation, the necessity of heat management was pointed out.

Using large arrays of white LEDs, Taguchi et at. (2001) devised the first LED-based street lighting system. The prototype energy-saving street lamp Consists of two phosphor-con version-LED fixtures and a solar-cell-powered battery system. Each fixture contains 700 units of lO-cd class LEDs. The system is equipped with a control that increases illuminance from 80 to 660 Ix when a person approaches the lamp.

The first demonstration of LED-based technology for advanced room illumination was presented by an Austrian company, Bartenbach Lighting Laboratory, at the Insbruck Architecture Forum (see Eisert et al. 2000). The installation contained about 14,000 white, blue, blue-green, green, orange, and red LEDs that were integrated in the ceil ing. To obtain various color temperatures, color mixing was performed by means of a computer program. The assembly provided illuminance of 600 to 700 Ix that is adequate for normal office lighting.

Craine and Irvine-Halliday (2001) reported on the first LED-based household lighting project, which was implemented in remote communities of Nepal. White LED lamps equipped with different power generating systems (pedal dc generators and simple wind turbines) were installed in 142 households and two schools. Owing to extremely low maintenance costs ($3Ihousehold/year), LED lamps successfully superseded kerosene wick lamps and burning sap-filled pine sticks, which are still used in the area.

In general lighting applications, a combination ofLEDs with optical fibers offers attractive design solutions. Lighting via fibers is also used to deliver light for decorative purposes or into places where electric wiring is dangerous or difficult to utilize (see Coaton and Marsden 1997). Solid-state technology offers many advantages in fiber-optic lighting applications: small size, efficiency, low cost, and reduced heating. The latter property implies that 110 fan is required and that plastic fibers can be used. In addition, by using polychromatic arrays, simple means of color control can be implemented. An example of a new lighting device is an LED fiber light engine (Krames et al. 2000). A schematic cross section of the LEDbased fiber light engine is shown in Fig. 7.7.2. Another example is an innovative headlight in Ford's concept city vehicle 021 C presented in 1999. The car features a single headlight designed as a programmable fiber display with a phosphorconversion white LED source.

A breakthrough of LED-based technology in general lighting is expected to occur when the luminous efficiency of solid-state sources approaches 200 lm/W (Haitz et al. 2000). This turning-point value is more than twice as high as the luminous efficiency of fluorescent lamps. For polychromatic multichip LED systems, such performance can be attained in the near future (see Section 6.3.2). However, a breakthrough could occur at even lower efficiencies if the significance of environmental and energy-saving considerations increases.




Electronic Driver

FIG. 7.7.2.

LED Array

. s secti f an LED fiber liaht engine. (After Krarncs et or

Schcmal1c cress sec lOll 0 '"


7.7.3. The Future of Solid-State Lighting

. 1 inci Jes that can surpass radiative

What's next? Are. t~ere any ~ther physlcaOI:: in ~ ht emission efficiency? Can

recombination of II1Jected electron~ a~d htl ualitv ~nd quantity standards? How white light produced by L~Ds ~all1tall1 a h q I a'? What amount of effort is redeep will LEDs penetrat~ into h~htmg tee no ~~~I~? One answer to these quesquired to i~plement .solid-state 1~h/1Il(2~~~)P who ~tated that "it is vital 10 know tions was given by Nick Hol?nya , r. . ~'inci Ie and in practice, and that its that the LED is an ultimate [orm of lamp. In P til ~l power levels and colors are development indeed can and will contlilue un I a




coefficient, sec Materials systems losses, 89, 98,101,104

Acccptorfs), see Doping

Advertising, 144-145, 163

Algae, 155-156,158-160 Antireflection coatingfs), 102-103, 107 Arc, see Discharge

Automotive signagc, 23,141-142 See also Lighting, vehicle

Backlighting, 57, 134, 163 Backup battery, 141, 143 Ballast resistor, 135


direct, 45, 46, 60

discontinuity, 49, 58-64. See also

Heterostructures filling, 80

indirect, 45, 46, 58, 61, 82

offset, see Band, discontinu ity potential fluctuations, 39---41 See also

Materials system, AllnOaN, composition inhomogeneities in

Bandgap, 39-43, 46-64,70-77, 108, 134. See also Materials systems

bowing, 63

crossover from direct to indirect, 57-60, 72, 73, 92

direct, 41--46, 55--61. 70, 92, 95

engineering, 49--50. See also Strain energy

band engineering indirect, 41--45,55-63,71 photon ic, 1 12-1 15 renormalization, 80

Beacons, 143

Bilirubin, see Phototherapy, of neonatal jaundice Blackbody, 3,14,17,21,22.37,118. See also

Spectral power distribution, blackbody Bone marrow graft, 150

Brightness, see Luminance

Cancer treatment, see Phototherapy, photodynamic

Candela (cd), 9


Jablochkoff 4, 5

Cando luminescence, 3


excess, 37-50

leakage of, 53,72-73,80-81 majority, 43, 46, 47 minority. 42,46,47,50 mobility of, 53, 76

Ccnter(s) luminescent, 43, 48

of nonradiative recombination, 43--45, 49, 55, 58,65,80,81

Center high mounted stoplight (CHMSL), see Automotive sign age




current-voltage (I-V). 4S, 54-55 76 79-8~

134 " -,

emissive, 79-80. See also Spectral power

d istribution

output. 75, 7~-8 i. 134

temperature, 82

Chemical sensor" see Fluorescence sensors Chlorophyll, 152-158 . Chromaticity, 7, 12-19, 119, 131, 132. 140. 145

147,152,162,166 . ,

, diagram, see ClE. 1931: CIE. 1960

Chromatic adaptauon 16-17



1931,11-17,124,125 131 140 146

1951, S ",

1960,16,17,146147 1964,12,16-19 '

1974, 17

1995, 17

s tandard observer, 12. 1 3 I

standard sources, 14, I I S, 120. 129 test-color method 16-19

Circuits '

digital control, 137, 146, 154. 166

dimming, 136 .

dnvtng: 134-138, 140---141, 145, 165. See also Light-enutting diode, modulation; Pulse Width modulation

high-power pulsed, 137-13S, 159, 162 low-voltage battery, 137

multiplex, 144, 146

string, 134-137

electrical protection, 136 solar powered 141 145

Colort s), 5, 7,8, '11-19.26,30-33 13 I 138

140-147, 159, 166 '"

coordinates, see Chromaticity

gamut, 13L 146-147

metameric, 15, 16, II S, 129

mlXmg, 14-15,117, I 19,124,145-147 166

monochromatic, 11-15,29 '

prl 111 ary, 11_-15, 119-122, 127. 130-132, 147 rendermg,),7, 15-19, 2L 26-35 117-132

164 ",

trade-off wrth luminous efficacy, 1 18-122.

See also Phase d rstribution subtraction. I I

temperature (CT), 14, 17,24,35,118-[32

152,166 '

correlated (CCT), 14.35

white 14, 15,26, 119,141 Colorimetric shift, 15-16 Colorimetry, 11-15

Configuration-coordinate diagram, 24-25, 43--44 Conservation


energy, 40 momentum 41

Coruacus), 38: 39,66,74-78,84--99, 104--112 ohmic, 76, 107

patterned, 77-78, 89-91

transparent, 76, 78, 89, 92, 95. 99



in organic dyes, 121_ 127

m phosphors, 24-25, 3 I, 95, I 17, 122-1 n I n polymers, 122 . in semiconductors. 1:!2-123, 127

up-, 117

Cost of light, 34-35, 162-163

Cntical angle. 84-86, 96-97. 100 See also Law

Snell's '

Crystal structure

cubic (zinc blcnde), 57, 59. 62 wurtzite, 62---65


aperture. 89. 104. 109- 1 10 crowd ing, 78

diffusion, 46-47, 54-55 injection. see J njccnon recombination, 48. 54-55

ideality factor ~t~ 48, 81 reverse, 46----48, 55, 136 source. 134-136 154

spreading, 75-75. See also Layer.

current-spreading .

tunnel, 54-55. 81 -

Cyanobacteria, 155-156, 157


of carriers, 46-54, 65 coefrlcient 47

Discharge, 3~, 16, 22, 24-- 37 156 159 162

163 ' , , ,



alphanumeric, 133, 144--145 full-color video. 131, 145-147

Distnbuted Bragg reflector COBR), 87-88. 92 93

99,107-114 . ,.,

DNA, 158-159

Dome, 83-86, 102-103. See also Encapsulation Donorts}, ,ee Doping

Dopantrs), see Doping

Doping, 39--40_ 43, 47-52, 57-76. See also Impurities; Materials systems

E r~dive masses, see Materials systems cfflclency

external quantum, 38, 39, 89-102, 108- 1 09

115,125 '

of AIGaAs LEDs. 89-90

of AIGalnP LEOs, 94--95, 102-103, 105


Efficiency (Confi,med)

uf AllnGaN LEDs, 98 of white LEDs, 125 feeding, 38----39. 75 loiection, 38,48,53,72,75

interna' quantum, )0,38-49,70,12, R3.

89-100,98, 118 in AIGaA" 58

in AIGalnP, 72

in AllnGaN, 62. 73,80,98

I ight extraction, 38, 48. 71. 83-1 15, 1 t 8 of AIGaAs LEOs. 91

of AIGalnP LED,. 92-94 of AllnGaN LEDs, 96-99

luminous, 10, 36, 118-121 ofdischarge lamps, 29-33, 35 of fluorescent lamps, 28.35

of incandescenllamps, 22-24, 35. 103, 138

of LE1K 38, 100. 162-166

AIGaAs, 57,58,92 AIGalnP, 102-103, 163 AlInGaN,98

white, 126-132, 162

optical, see Erficiency, light extraction

radiant, 10, 118

of LED" 38, 92.9&, )04, 109, 128, 12\1,


radiative, see Efficiency, internal quanturn wall-plug, see Efficiency. radiant

Eleclrode(s), 5,27-34 See also Contacts Electroluminescence, 37-38,70---82, 133. See also Injection, luminescence

decay time of, 133 Electron affinity, 75

Electron-hnle pair, 38.39,43,44,52

Electrons, see Carriers

Encapsulation, 66,83-84, 102-103, 126, 139

Environmental issue" 29. 69, 162. 166 Epitaxtal lifl-off. 99, 113

Epitaxy, see also Substrate

chloride vapor-phase (CIYPE), 68

hydride vapor-phase (HYPE), 68, 94, 98 [iquid-phase (LPE), 58, 6 1,66-67,70,90-92 mnleeular-beam (MBE), 66, 68. 99-109 organometallic vapor-phase (OMYPE), see

Metalorganic chemical vapor deposition

vapor-phase (VPE), 66, 92

Epoxy, 83-86,97, 102-103, 122. 158, 165. See

also Plastics Escape cone(s), 84---108

Etching, 78, 90, 94, 96, 10 1,104---105, 113-115

EXCiton, see Transitions

Exit signs, 143

Fabry-Perot cavity, 106-110 Failure

of discharge lamps, 29-33


of !1uorescent lamp', 28

o[ Incandescent lamps, 22-24, 136 of LEOs, 136, 139-141

FLha optics. 109, 144-148, 157-158, 166-167 Filament. 5,21-24,31, 141, 163

Fire, I

Flashlights, 143, 163

Floodlighting, 5. 24, 33,164-165 Fluorescence sensors. 156-161 Fluorophores. see Fluorescence sensor'


luminous, 9-10,35, 142

ofLEDs, 94,102,126,142,162-165

radiant. 9,10,14,148 Forwardcurrenl,79-&0. 134-137, 140,154,


Forward voltage, 39, 47, 134--136

Fovea, 8

Fresncllosses, &6.97


color matching, I 1-12, 26

objective, 119, 121

relative tuminous efficiency, 8, 9,12,22.

55-56,58, 103

work, 27.75-76

Gap, see Bandgap Gas mantle, 4 Gel,103

General color rendering index (CRl), see Color,


Green fluorc,;cnt protein (GF?), 158. 161

Heat sink, 126

Ileterointeralace(s), 50-54, 57-58,61,65 Hcterostrllclure(s), 48-75, 89. See also Epitaxy;

Light-cmitting diodes: Materials systems:

Metalorgan ie chemical vapor deposition;

Quantum wells

double (DH), 50

multiple-well (MWH), 70-71,95

single (51-I), 49

Highly refractive glasses, 102 Iloles. see Carriers

Illuminance, 10-11, 139, 151. 164-166 lliumination, 162-167

local, 163-165

Impact iOllllJ!lion, 38,43

I mpurity, see also Doping acti vation, 65 compcnsal1on. 61

deep level, 43--47 ionization, 54,61, 81 oxygen,SS-59,61,67,72

Incande,cence. 1--4.21-24,37. S,'I< also Lamp,

i ncande,cence


Indium tin oxide (ITO), 78, 89. See also Contacts. transparent

Injection, 37-55, 57, 58, 70--75,138 bidirectional, 50

luminescence, 37-55,70--82

lntelhgent transportation systems (I'f'Ss), see Displays, alphanumeric


luminous, 9, 136, 140--144, 164 radiant, 9


heterojunction, 49-50, 64 homojuncuon, 46-48 p-n,46-50,66, 71, 76 Schottky, 75-76 temperature, 131

tunnel, 99, 109

Lamp(s) Agrand,2

compact fluorescent (CFL), 6, 27 electrodeless discharge, 33

Jluoresceru, 5,16, 19, 24-29, 35, 103, I 17, 120,

132,133,143,148-156,162-166 gas, 4

high-intensiry discharge (!-lID), 30 high-pressure mercury vapor (HPMV), 30--31 high-pressure sodium (l-IPS), 32 incandescent, 5, 6, 21-24, 27-35, 103, 117,

126,132,133-153,162-164 filtered, 57, 90, 92, 103, 138, 144, 151 general lighting serv ice (GlS), 22 induction, 33-34

kerosene, 3, 4, 166

low-pressure sodium (lPS), 29--30 mercury, 5

metal halide (MH), 32-33 Nernst,5


solid-state, see Light-emitting diode sulfur, 34

tritium, 143

tungsten halogen, 23-24 Lattice

constantrs), 50, 90, 95, 114. See also Materials systems

(misjmatch, 50, 57-{i5, 71-73,90--95,98 Law

Snell's, S4 Vegards, 57,62 I.ayer(s)

active, 38-54, 58-{i6, 70--81, 84-98, 104-109,113

co-doped, 74, 79,81

buffer, 62, 95


cladding, 50-54, 58-63, 66,71-76, 80--8J, 89-99,107, I 13-114

tensile strain barrier (TSBC), 72-73. See

also Layer, electron blocking conductive, see layer(s), cladd ing confining, see Layer(s), cladding

contact, see Layerrs), cladding current-blocking (CBL), 89, 92, ~3, 99 current-spreading (CSL), 76-78,89,92-95 electron-blocking (EBL), 53, 54, 72, 73, 8 I pscudornorphic, 50, 63, 65

window (WL), 69, 72, 86-94, 98



nonradiative, 44-46, 58,61.65 overall,45-46

radiative, 43-46, 106, 110

of discharge lamps, 29,31,33,35 fluorescence, 156-161 measurements

time-domain, 159--160 frequency-domain, 160--161 of fluorescent lamps, 28, 35

of incandescent lamps, 23-24, 35 economic, 23

ofLEDs, 94,103,108,109,126,136, 1J9--16) Light

box, see Phototherapy, or seasonal affective disorder

curing unit (LCU), See Phoropolymenzation of dental composites

engine, 166-167

extraction, 83-116. See also Efficiency, light extraction

Light-emitting diode(s) (LED), 6,37,38 absorhing-substrate (AS), 83-87, 90, 9 I, 104 AIGaAs, 57, 70-71, 76-77,87-92,100-108, 111,113,133,140--142,149-161 AIGalnP, 71-73, 76-82,87-95,102-105,109, 114, 130--132, 138. 140--146, 152-153, 163-165

AllnGaN, 73-82, 87-90, 95-99, 109, 117, 122-132,138-149,152-154,158-159, 162-165

dichromatic, 117, 120, 122-132, 154, 164 distributed Bragg reflector (DBR), 88, 92-93 dcuble-heterostructure (DH), 43, 50, 54,70--95,


flip-chip (FC), 98-99 hemispherical, 100--10 I

high-power, 75, n, 94, 98-103, 126, 164 modulation, 156-162

muluple-quantum-well (MQ\V), 50, 54, 70--75,

81,95,99,130 multiple-well-heterostructure (MWH), 71, 95 nonrectangular and nonplanar, 99-105 nonresonant cavity (NRC), 104-105


light-emitting diode (Continued) organic (OlED), 57, 1 17, 134 paIabolic,IOO--IOI

planar rectangular, 84-89

p-n junction, 43-48 16E polychromatic, 127-128, 130--132, 165- )

quadrichrOinatic, 130--132

qui n tichrorna tic, I 30--1 32

resonant cavity (RCl. 106-1 10, I 12

response nme 0 f, 46 _ .

single-heterostructure (SH), 49, )0, 90 single-quatltum-well (SQW), 50, 54, 66,

70--75 80,81,95

space appli~ations of, 153, 155 surface plasmon-enhanced, 110-1 ! 2

tapered, 10 I . 153

transparent-substrate (TS), 86-96, 98,

trichromatic, 117,127-128,130--132 -103 truncated-inverted-pyramld (TIP), 101 8 ultraviolet (UV), 37, 62, 66, 95,109, 117, 12 ,


white 95 117-132, J 34, 151,162-167 m~ltichip (MC), I 17, 128-13:, 162-165 phosphor conversion (PC), 9~, In, 122-128,162-166

Lighting airport, 144

architectural, 126, 163 " ..

automotive, see Autornouve signage. Lighting,

vehicle Christma_, tree, 163

commercial, 11,29,33, 164-165 decorative, 163, 166

deluxe, 26, 127, 132

economy, 34-36, 162-163 electric, 4-6

global power consumption of, 36

gas, 3-5

general, 165-167 _ industrial, 3, 11,29,33,16)

landscape, 163 .

medical, 148, See also lighting, surgical;

Phototnerapy officc,29,33,165-166 physioillgical issues of, 165

residential, 3,6, II, 23, 24, 36, 165-166 road, see lighting, street

security, 30, 163

surgical, 164

street, 3-6, 30--32, 165-166 vehicle, 24,33, 143, 164, 166

Limelight. 3, 5

Line width, 40, 52, 79-80, 109, 120--132

Lumen (1m), 9 46

. 7 II 109 134 138 140,144-1 ,

llimmance, -. , , ,

165 . 8-131 147

Luminous ellicacy, 10,21, 29, 38, I I ,


trade-off with color rendering, 118-121. See also Phase distribution

Lux (lx), 10

Marin~ lights, see Bcaco~s _ 76

Materials systernis), 45, )0, S5 70,


IV-IV, 55


AIGalnP, 59-61,92-95, 102-103 atomic ordermg in, 59-60

AllnGaN, 61-66, 95-99 .. 6" 73

composition lnhomogcnenv 111,63,), ,


quaternary alloys of, 62-66, 75,128 .

"~ Eoita Y' Metalorganlc chemical

growUl, see pi xy,

vapor deposiuon Melatonin, 151, 165 Mesa, 78, 96

Metallization, 76, 91. See also Conlacl' Metalorganic chemical vapor deposmon

(MOCVD), 61, 66, 68-73, 92-100, 108,

109,131 ,

Metamerism see Colors, metameric .

M· 'h" 106 110 III See also Fabry-Perot

IcrocaVI,;<, , , .


Microlenses. 101, 105

photoresist, 103

Modes, see also Photon denslty-of-states

engineering guided, 108-1 10, I 13 leaky 108-109

surfa~e plasmon-polariton (SPP), 106,


Multiquantum barrier (MQB), 72-73

N aiural Iithllgraphy, 104

Opal 115 . .

Opti~al boundary, see Phase distribuncn Oxygen regeneration, see Photobloreactors

Phase distribution, 120--122, 131 Phase-modulation technique, 160--161

Phllnons, 25,41,43--45, 79 II7-P8 146, 153.

Phosphor(s), 6, 24-31, 34, 9S, . -, ' ..

See also lamp, fluorescenl~ Light-emitting di de white phosphor converSIOn

10, , , -127 165-166

ceri u III -do ped garnets , 123 ,

halophosphate,6, 19,26--'27 PhotllbioreaclOrs, 155-! 56

Phlltometry,9-11 . 05 115

Photon density-of-states engllleerlllg, I -

Pnotonic crystals, 112-116

Photonic dot, I 10

Photon recycling, 89, 105, 108 Photopigment(s),7-8


Photopolyrnerization of dental composites,

150-151 Photoreceptors, 7-8 Photoreflectance, 161-162

Photosensitizcrs, see Phototherapy. photodynamic Photosynthesis, 152-156

quantum action of, I 52

Phototherapy, 148-152

of neonatal jaund Ice, 148-149 photodynamic, 149-150

of seasonal affective disorder (SAD), 151-152 Piezoelectric effect, 53,64-65

Pixels, 144-146

Planck ian locus, 13-14, 16, I 19, 120, 124 Planckian radiator, see Blackbody

Plant growing, 153-154

Plasmon, 41, 106, 110-111

Plastics, 102,126,139,142.151,163-166 Potential barrier, 46-49, 54,71.75-76

Power signals, see Automotive signage; Beacons; Exit signs; Lighting, airport; Trame lights Precursorrs), See Metalorgarnc chemical vapor


Primary sources, see Colors, primary

Principle of detailed equilibrium, 42 Pulse-width modulation (PWM), 136-137, 146 Purcell enhancement, 106

Pyroluminescence, I

Quantum-confined Stark effect, 53 Quantum confinement 49-52, 72. See also Quantum wells

Quantum dous), 109-110, 115

Quantum wellis), 49-54, 65, 66, 72,89, 109, III asymmetric, 54, 74

multiple, 50, 107


triangular, 53

Radiometry, 9

Randomization of photon trajectories, 91-92, 104-105

Reabsorption, 24, 30, 4S-50, 90

Reactor, see Metalorganic chemical vapor deposition


band-to-band, see Transitions

coefficient, 42-43, 52. See also Materials systems

temperature dependence or. 42, 52 electrons and holes of, 39-46 donor-acceptor, see Transitions

of ions, I

nonradiauve, 25, 43-45, 49, 54, 55, 72. 73. 80-82. 128

Auger, 43

due to deep levels, 43-44, 49, 58, 61, 65


via multiphonon emission, 43-45, 58 surface,48, I 14

radiative, 1,25,37-55,57,74,80,82, 161,

167, see also Transitions bimolecular, 43,52 impurity assisted, 39-4 1 intrinsic, 39-41 monomolecular, 43,49

in quantum well, 52-54

rate of, 41-45, 50-55, 72, 80, 82, 105 Reflector, 100-101, 107-109, 139, 164-165 built-in, 98, 103, 122, 139

Refractive index, 42, 84-88,96-98, 102. 103.

I 11-115. See also Materials systems Relative eye-sensitivity spectrum, see Function>

relative luminous efficiency Retina, 7-8, 165

Roadway crosswalk warn ing systems, 144 Ruthenium com plex, 158-161

Sawing, 91, 92, 102, 104 Sediment dating, 161-162 Series resistance, 39. 48, 75,81 Shielding, 76, 86, 9 J, 92 Signals, see Power signals Slider, see Epitaxy, liquid-phase

Spectral power distribution (SPD), 9-11, 118-121 of blackbody (Planckian), 22, 118

of colored LEOs. 79

of fluorescent lamps, 26

of high-pressure discharge lamps, 31 trimmed Planckian, 118-119

of tungsten lamps, 22

of white LEOs, 125-132, 165 Spontaneous emission, see Recombination,


Spontaneous polarization, 53, 64-65, 75 Spotlighting, 24, 148, 163-164

Stimuli, II

Stokes shift, 25,28,37, 120-l28 Strain, 53, 63-66, 72-73, 75,102 Strain energy band engineering, 66, 75 Stress, 63. See also Strain

Substrate, 66-6M, 85-101,108,109. See also Light-emitting diode, absorbing substrate; Light-emitting diode, transparent substrate AIGaA',71

GaAs, 57-59,69,71,88-95, 104-109 GaN,65

epitaxially laterally overgrown, 99 GaP, 102

lattice mismatched, 62--{)5 LiAI02,99

quartz, 99

sapphire, 62-69, 78, 95-98 silicon, 62

6H-SiC, 62, 95-98



Substrate (Colltinued) spinel, 62, 95 ZnO,62

mirror, 94, 104 .

Susceptor, see Metalorganic chemIca I vapor


System ,

dichromatic, 119-122, 124, bO, 164

polychromatic, 119,121-122 quadrichromatic, 119, 122, 147 quintichromatic, I 19, 122 trichromatic, 12, 119, 121, 146-147

Test samples, see CIE, test-color method; CIE,

1964 _ . -9 95

Thermal expansion coefllelcnt, 22-23, J., ,


TOlal internal rellection, 85-87, 97, 98 See also

Cri tica Ian glc

. . 1 24-25 39-54 123 See also

TransitIons" '.. '

Recombinalion, radiative 4

band-to-band> 39-42, 61, 65, n, 7 bound-exciton, 40-41

direct, 41-42, 65

donor-acceptor, 39-40, 74, 80 exciton, 39-40, :52, 64

via impurity levels, 39-41. 43, 58

ind irect, 4 I

inlersubband, :52-53

intrinsic, 39-41, 52,74 localized-exciton, 40-41

T ffi lizhts 23 139-141.SeealsoRoadway

ra ic I"", 0, , . I

crosswalk warning ,ystems~ Disp ays,


Trislimulus values, I 1-13

unitcrm chromaticity scale (UCS) diagram, see ClE,1960

Valley, see Band . I

Variable message signs (VMSs), see DISP ays,


Vertical cavity surface-emitting lasers (VCSELs),


Virus inactivation, \ 50

V is ible spectrum, 8 Vision, 7-11, 15, 16, 131

photopic, 8, 10, 12

scotopic, 7-9 - . S I

' 8 22)6 ee (J SO spectral sensitivity 01,. , . ."',.

Function, relative [urnmous dhClency

Visual environment, 162

visual tasks. 1 I, 162

Wafer bonding, 69, 92, 94, ~9, : 14, 131 Wavcfunction overlap. 39, )2-)3 Wick, 2- 3

Wire bonding, 77, 91

Sign up to vote on this title
UsefulNot useful