Professional Documents
Culture Documents
C1
Learning T. Delbrück & C.A. Mead
Qp Vb
Qfb
Model Adaptive
Vf element
Vo
I bg+i 3.4
Transient
C2 Qcas Vcas Responses
3.2
Peak
Vp Output
Qn (V) 3
Steady-state
Input response
2.8
2.6
2.4
-3 -2 -1 0 1 2 3 4 5
10 10 10 10 10 10 10 10 10
Irradiance (mW/m2)
Adaptive
element
Log Intensity
Signal Amplitude (dBV/√Hz)
-100 -2
Feedback
capacitor
(C )
2 -1
Amplifier 1/f2
(C ) 0
o -120
Storage Cascode
transistor 1/f Instrumentation
Capacitor
(C )
1
-140
6 µm
-1 0 1 2 3 4
10 10 10 10 10 10
Frequency (Hz)
Feedback transistor
Photodiode (C )
p
(also hole in overlying metal)
This paper describes an adaptive photorecep- which the bandwidth of the receptor is 60 Hz,
tor circuit that can be used in massively paral- is at approximately 1 lux, which is the border
lel analog VLSI silicon chips. The receptor between rod and cone vision and also the
provides a continuous-time output that has limit of current consumer video cameras.
low gain for static signals (including circuit We describe an adaptive element that is
mismatches), and high gain for transient sig- resistant to excess minority carrier diffusion.
nals that are centered around the adaptation We show measurements of the effectiveness
point. The response is logarithmic with illu- of guard structures, and of the spectral sensi-
mination, which makes the response to a tivities of devices that can be built in a BiC-
fixed image contrast invariant to absolute MOS process.
light intensity. The logarithmic transduction process
The 5-transistor receptor can be fabricated makes the time constant scale inversely with
in an area of about 70 by 70 µm2 in a 2-µm intensity. As a result, the total A.C. RMS
single-poly CMOS technology. It has a receptor noise is constant, independent of
dynamic range of 1–2 decades at a single intensity. The spectral density of the noise is
adaptation level, and a total dynamic range of within a factor of two of photon shot noise
more than 6 decades. Several technical and varies inversely with intensity. The con-
improvements in the circuit yield an addi- nection between shot and thermal noise is
tional 1–2 decades dynamic range over previ- beautifully illustrated.
ous designs without sacrificing signal quality.
The lower limit of the dynamic range,
defined arbitrarily as the illuminance at
CONTENTS
Continuous-Time vs. Sampled Receptors...............1 Total Noise in Adaptive Receptor.................. 12
Biological Motivation ...............................................1 Using Shot Noise Statistics to Compute the
Gain Control .....................................................2 Noise Power............................................... 12
Time-Constant Control .....................................2 Shot vs.Thermal Noise?................................. 13
The Goal of Phototransduction................................2 Effect of Temperature.................................... 14
Simple Logarithmic Receptors................................2 Noise Spectral Density .................................. 14
ADAPTIVE RECEPTOR CIRCUIT....................................3 The Essence.................................................. 14
The Feedback Loop ................................................3 Measurement and Theory..................................... 14
Cascode...........................................................4 Assumptions of Difficulty ............................... 14
Photoreceptor Gain .................................................4 Photodiode vs. Phototransistor:
Advantages of Active Feedback.............................5 Noise Behavior.................................................. 15
Speedup..................................................................5 Noise Advantage of Continuous-Time Receptors
Miller Capacitances ................................................5 over Sampled Detectors ................................... 15
Rise Time & Bandwidth ...........................................6 MINORITY CARRIER DIFFUSION & GUARD STRUC-
Gain–Bandwidth Product ........................................6 TURES ...................................................................... 15
photodiode area? ....................................................6 Summary ....................................................... 16
SMALL-SIGNAL ANALYSIS............................................6 SPECTRAL SENSITIVITY ............................................. 16
Second-Order Temporal Behavior ..........................7 The Spectral Responses ...................................... 18
ADAPTIVE ELEMENT .....................................................8 Absolute current level........................................... 19
Adaptation Rate ......................................................9 PREVIOUS WORK........................................................ 19
Other Adaptive Elements ........................................9 RELATION TO BIOLOGICAL PHOTOTRANSDUCTION..
PHOTODIODE VS. PHOTOTRANSISTOR ....................10 20
RECEPTOR LAYOUT....................................................10 SUMMARY ................................................................... 20
THE ILLUMINATION LIMIT (SPEED) ............................10 Technical Innovations........................................... 20
Illumination Limit: High End...................................11 CONCLUSION ............................................................. 21
THE DETECTION LIMIT (NOISE)..................................11 ACKNOWLEDGEMENTS ............................................. 21
Empirical Observations .........................................11 REFERENCES .............................................................. 21
Theory of Logarithmic Receptor Noise .................12 INDEX .......................................................................... 23
Using Equipartition to Compute the Noise
Power..........................................................12
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
Apr 2, 1996 1
ver the last few years, people have The transduction process seems mun- led to wide-spread use in machine vision
increments). The stimulus spot is 3.2 mm in diameter on the retina. Peak responses
measured from the dark-adapted resting potential (dotted line) are plotted as a -10 Steady-State
Dark
function of test illumination. The thin curves connect the measured points. The thick -3.2 Response
-4.4
curve is the steady membrane potential measured at least two minutes after -5 -2.1 -1.0
background onset. The average slope of the transient responses is 9.5 mV/decade,
Transient
and the slope of the steady-state, adapted response is about 1.8 mV/decade. The 0 Responses
ratio of transient gain to the steady-state gain is about 5. The total dynamic range is
about 15 mV. The illuminations are given as log attenuation from a baseline value. The -6 -5 -4 -3 -2 -1 0
unattenuated test stimulus (0 log) is 6.4 1015 quanta(640 nm)(cm2s)-1 on the retina, Log intensity
equivalent to an irradiance of about 20 W/m2. (Direct office fluorescent lighting is
about 1 W/m2.) The unattenuated background illumination is 9.1 1015
quanta(640 nm) (cm2s)-1. Adapted from Normann and Perlman (1979).23
Authors may be contacted by email at tobi@pcmp.caltech.edu. These circuits have been patented as T. Delbruck, C.A. Mead, "Adaptive photoreceptor including
adaptive element for long-time-constant continuous adaptation with low offset and inensitivity to light," U.S. Patent 5,376,813.This document (and other related
ones) exist in electronic form available via the World Wide Web; the URL for the Physics of Computation Group is http://www.pcmp.caltech.edu/
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
2 Apr 2, 1996
10 Vb
FIGURE 2 Time to
peak response in
Vo
response to a flash of
light, vs. gain, in toad
response (s)
rod receptors.
Time to peak
n++ p- n++
Replotted from
1 Baylor et al., (1980).1
Vb
60 mV/decade
Vo
0.1
0.01 0.1 1 10
Gain (arbitrary units) log Intensity
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
ADAPTIVE RECEPTOR CIRCUIT Apr 2, 1996 3
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
4 Apr 2, 1996 ADAPTIVE RECEPTOR CIRCUIT
FIGURE 6 Responses of the source- addition of this single cascode transistor in-
Voltage follower and adaptive receptor over 7
-6 creases the dynamic range of the receptor
decades of background. Stimulus is a by about a decade.
200 mV
square-wave variation in the intensity,
centered around a mean value. The
Figure 7 shows the effect of the cascode
-5 on the small-signal time response of the re-
numbers by each section are the log
intensity of the mean value; 0 log is ceptor. Under these operating conditions
2.9 W/m2, about the level of direct office the cascode speeds up the response by a fac-
-4 5s Time fluorescent light. The source-follower tor of about 6.
receptor begins to smooth out the 1 Hz
PHOTORECEPTOR GAIN
square wave input at the lowest
intensities, while the adaptive receptor still It’s simple to compute the steady state and
-3 responds. The 1.8 Hz square wave transient gain of the receptor if we assume
stimulus is from a red LED (635 nm). The that Aamp is large. When the input current
irradiance varies by a factor of about 2, or changes an e-fold, the gate of the feedback
-2
about 0.33 decades or 0.8 e-folds. capacitor must change by VT/κ to hold vp
clamped. That means in steady state the
Log small-signal gain is given by
-1 clearly useful in systems that care about the
Intensities vo ⁄ V T 1
contrast changes in the image, and not the ------------------- = ---
absolute intensities. i ⁄ I bg κ
0
The receptor adapts very rapidly in re- (linearized steady-state closed-loop gain) (2)
sponse to the decade changes of intensity. and for transient signals, where the output
Source- Adaptive This rapid adaptation is due to the use of an must go through the capacitive divider, the
receptor
follower adaptive element with an expansive nonlin- gain is
receptor earity. Large changes in the output adapt
rapidly, while small signals around an adap- vo ⁄ V T 1 C1 + C2
A cl ≡ ------------------ = --- --------------------
0 tation point only adapt slowly. i ⁄ I bg κ C2
CASCODE. Qcas has two effects:
(linearized transient closed-loop gain) (3)
1. It shields the drain of Qn from the large
where κ ≈ 0.7 is the back-gate coefficient
voltage swings of Vo. Because the
-1 describing the effectiveness of gate voltage
source conductance of Qcas is larger
changes on channel surface potential. We
than the drain conductance by a factor
shall call the gain for transient signals the
of approximately Aamp, the drain of Qn
-2 closed loop gain Acl from now on.
moves only about as much as vp. With-
out Qcas, the large voltage swings across Writing the gain in dimensionless form
the gate–drain capacitance of Qn load displays the logarithmic, contrast-sensitive
-3 response properties. The response to a con-
down the input node. They make fF
gate-drain capacitance appear to the stant i/Ibg is independent of the background
-4 input node to be on the pF scale, a phe- current. The ratio between transient and
nomenon called the Miller effect. steady-state gain is the capacitive-divider
ratio (C1+C2)/C2. We generally use a capac-
2. Qcas also multiplies the drain resistance itive divider ratio of about 10, which is the
-5 of Qn by a factor of approximately
ratio of poly-poly to metal-poly capaci-
Aamp, through a cascode action. This
tance. Figure 8 shows measured transfer
increase in drain resistance increases the
-6 curves. The plots illustrate the shifting of
gain of the amplifier by a factor of
the adaptation point to the ambient intensity
about 2.
and that the transient gain is much larger
Both the reduction in effective input capaci- than the steady-state gain.
objects. We varied the overall intensity level tance and the increased gain translate into
by interposing neutral density filters (i.e., speedup. The additional speed of the recep-
sunglasses) with various attenuation factors tor makes it usable at lower intensities. The
between the light source and the receptor,
while recording the receptor outputs. These (a) Input to LED
changes in the overall intensity level repre-
sent changes in the ambient lighting, as
would be caused by passing from shadow FIGURE 7 Effect of cascode on time- (b) Dim light
into sunlight or vice versa. response and noise. (a) shows the without cascode
The amplitude of the response to the small-signal input signal. (b) shows the
response of the adaptive receptor when
small contrast variation is almost invariant the cascode is shorted. (c) shows the
to the absolute intensity, owing to the loga- (c) Dim light with cascode
response with the cascode activated,
rithmic response property. The adaptation (along with averaged curve)
along with a time-averaged response.
makes the receptor have high gain for rapid- (d) shows the response of the receptor
ly varying intensities and low gain for slow- when the light is 10 times brighter. The
ly varying intensities. Hence, the response noise properties are discussed in the (d) Bright light (10x)
50 mV
to an intensity change of a decade, after ad- text (see page 11). without cascode
aptation, is almost the same as the response
to the 15 % variation. A receptor like this is 10 ms
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
ADAPTIVE RECEPTOR CIRCUIT Apr 2, 1996 5
3.4
FIGURE 8 Step response operating curves. Each s- Transient
shaped curve shows the peak value of the response to a Responses
step change of irradiance, starting at the intensity marked 3.2
with a circle. The ordinate shows the peak value of the Peak
response to the step, and the abscissa shows the Output
absolute incident irradiance. The total range we tested (V) 3
Steady-state
spans 6.5 decades. The receptor was allowed to adapt response
back to its steady-state value for 5 s before each step
2.8
stimulus. The steady-state gain is between 30 and
40 mV/decade; the transient gain is approximately
1.5 V/decade, decreasing at the lowest intensities due to 2.6
interaction between rise time and adaptation time
constants.
2.4
-3 -2 -1 0 1 2 3 4 5
10 10 10 10 10 10 10 10 10
Irradiance (mW/m2 )
Equation 2 is the small-signal equivalent line speed of the source-follower receptor. gate to the drain of Qn and Cfb from the gate
of the large signal expression for the Its time constant is to the source of Qfb have a substantial effect
steady-state output voltage, CV T on the time-response of the receptor, even
C
Vo I bg + i τ in = --- = ------------- (baseline speed) (5) though the capacitance is only 0.1 fF/µm of
1 g I bg channel width. We can compute the time
-------- = V p + --- log ---------------
VT κ I0 The adaptive receptor is shown again in constant of the receptor response including
(large-signal steady-state response) (4) Figure 9, this time with relevant capacitors, these parasitics by using the fact that the
both explicit and parasitic. In the absence of Miller effect increases the relevant
where Vp is the clamped voltage at the gate
Cfb and Cn, the speedup obtained by using gate–drain and gate–source capacitances by
of Qn and I0 is the preexponential in the
the active feedback to clamp the input node a factor of A, and the feedback increases the
subthreshold transistor law.
is given by effective conductance at the source of Qfb
ADVANTAGES OF ACTIVE by a factor of the total loop gain. (In this
A amp
FEEDBACK A loop = ------------- (speedup) (6) analysis we shall imagine that we have left
An active feedback circuit has three advan- A cl out Qcas.) The time constant including these
tages over a passive feedback circuit like The speedup is equal to the total loop gain, effects is given by
the one used in many early Mead lab obtained by following the gain all the way C
projects: The bias current in the output leg τ = ---
around the loop. However, this result is na- g
of the receptor is capable of driving arbi- ive, because it ignores the important para- A amp
trary capacitive loads. The bias control al- sitics Cfb and Cn. C p + A amp C n + ------------- C fb
lows us to low-pass filter at a chosen κA cl
frequency, which lets us filter out flicker MILLER CAPACITANCES = -----------------------------------------------------------------
The Miller effect occurs when a capacitor A amp I bg
from artificial lighting. Most important, the ------------- -------- (7)
feedback, by clamping the vp node, extends feeds back the output of an inverting, high- A cl V T with cascode
the usable dynamic range of the receptor gain amplifier back to the input. If the input
by speeding it up. The small photocurrents needs to move a certain amount, it must V T A cl 1
charge not only its own side of the capaci- -------- ------------- C + A cl C n + --- C fb
need only charge and discharge the small κ
= I bg amp p
A
changes in vp, rather than the large swings tor, but also but charge the other side of the
of vo. capacitor which moves A times as much in
the opposite direction. Hence a small capac- C eff
SPEEDUP itance C looks like a capacitance (A+1)C to We can think of this result as implying an
When we computed the closed loop gain, the input. Since A >> 1, we usually ignore effective capacitance Ceff, as shown above,
we assumed that the gain of the feedback the 1. The Miller capacitors Cn from the at the input node, which can be combined
amplifier formed from Qn and Qp is infi-
nite, which of course is only true in the
sense that it is much larger than the closed
loop gain. Our assumption meant that the
input node is perfectly clamped, which is Qfb Qp Vb
Vf C2
also not true—the input has to move a little
bit to change the output voltage by the re-
quired amount. The larger the gain Aamp of C1
the feedback amplifier, the less the input Vo
node needs to move, and the more speedup I bg+i Cfb
we obtain. On the other hand, for a given Qcas Vcas
feedback amplifier, the more closed-loop FIGURE 9 Adaptive receptor Cn
gain Acl we design in the receptor (by ad- circuit with important parasitic Vp
justing the capacitive divider ratio), the capacitors. (Adaptive element is left Qn
more the input node must move, and the out.) Cp
slower the response. We’ll compute the
speedup of the receptor relative to the base-
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
6 Apr 2, 1996 SMALL-SIGNAL ANALYSIS
n++/p-
out of the picture. With proper layout, Cfb
should be reducible to 0.6 fF, a factor of 5 p++/n
10–90 % Rise Time (s)
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
SMALL-SIGNAL ANALYSIS Apr 2, 1996 7
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
8 Apr 2, 1996 ADAPTIVE ELEMENT
Vf x 10 -8
Vo Vf Vo 10
(a)
5
Vf Vo
Bipolar
p+ p+ n+ Current 0
(b) n-well Vo > Vf (A)
p-substrate MOS
-5
Vf Vo
p+ p+ n+
(c) Vo < Vf
-10
-0.5 0 0.5
Voltage (V)
FIGURE 14 Expansive adaptive element (a), shown in two
schematic forms, along with the capacitor that stores the FIGURE 15 Measured current–voltage relationship for the new
adaptation state. expansive adaptive element shown in Figure 14. The bipolar
(b) The mode of conduction when the output voltage is higher mode conduction e-folds every 28 mV, compared with 48 mV
than the capacitor voltage: The structure acts as a diode- for the MOS mode, leading to a quantitative difference in the
connected MOS transistor. voltage at which the current rapidly increases. At any scale of
(c) The opposite case: The p+/n junction is forward-biased, current, the curves have the same appearance; the voltage
and the device as a whole acts as a bipolar transistor with two scale changes logarithmically with the current scale. This data
collectors. was taken from a p-well chip.
the maximum possible amount of ringing. element acts like a pair of diodes, in paral- back gate (the well) is driven at the same
This ringing is not very severe, because the lel, with opposite polarity. The current in- time, the current e-folds every VT/κ.
maximum Q of the circuit is generally less creases exponentially with voltage for For opposite polarity, the bipolar is
than 5. We have labeled these conditions on either sign of voltage, and there is an ex- turned on. The driven side forward-biases
the root-locus plot in Figure 13. tremely high-resistance region around the the p++/n emitter–base junction. The bipo-
Usually we turn the bias current up origin, as shown in Figure 15. lar transistor has two collectors: the driven
enough to give a response that is fast The I–V relationship of the expansive el- side and the substrate. The current e-folds
enough for the situation at hand, but slow ement means that the effective resistance of every VT volts. (The back gate of the MOS
enough to filter out flicker from artificial the element is huge for small signals and transistor is also turned on, leading to a cur-
lighting. A nice feature of this mode of op- rent that e-folds every VT /(1- κ ), but this
small for large signals. Hence, the adapta-
eration is that the speedup is only effective small component is invisible relative to the
tion is slow for small signals and fast for
at low intensities, while at higher intensi- large bipolar current.)
large signals. This behavior is useful, be-
ties, the low pass filtering reduces the noise. These characteristics may be seen in the
cause it means that the receptor can quickly
adapt to a large change in conditions—say, data shown in Figure 16. This data was tak-
en with and without light shining on a near-
moving from shadow into sunlight—while
by hole in the metal covering of the adaptive
maintaining high sensitivity to small and
element, to illustrate that the currents in the
slowly varying signals. The adaptation time capacitor node (Vf) are unaffected by mi-
ADAPTIVE ELEMENT rate is proportional (in some sense) to the nority carriers generated in the substrate.
Adaptation occurs when charge is trans- signal amplitude.
The I–V relationship for the element is
ferred onto or off the storage capacitor. This For voltage polarity Vo > Vf across the el- given by
charge transfer happens through the adap- ement, the MOS transistor is turned on and
tive element. The adaptive-element is a re- the bipolar transistor is turned off. The driv- I = I m – I b + I par
sistor-like device that has a monotonic I-V en side (Vo ) in the MOS case acts as the
relationship. For analog VLSI circuits, source of the transistor, but because the ∆V ≡ V o – V f
however, true ohmic resistors available in a κ∆V ( 1 – κ ) ( – ∆V ) (17)
plain CMOS process are much too small for I m = I 0,m[ e –e ]
adaptation on the time scale of seconds.† In- † Assume we need an RC time constant of a sec-
–∆ V
stead, we use transistors in our adaptive ele- ond, and that C = 1 pF (a 50 µm by 50 µm poly-to- I b = I 0,b[ e – 1]
ment—a sacrifice with unanticipated poly capacitor). Then we need R = 1012 Ω. Polysil-
icon has a resistance of 20 Ω/square, so we would
benefits. We have developed two novel need 5x1010 squares—a 2 µm-wide poly resistor
where I is the current flowing onto the ca-
adaptive elements with dual nonlineari- with area 0.6 m2! Some DRAM processes have an pacitor. I consists of three components, the
ties—expansive and compressive.6 Here, extremely high-resistance undoped polysilicon MOS transistor current Im, the bipolar tran-
we shall discuss only the expansive ele- with ohmic properties, but it is unstable, with large sistor current Ib, and the parasitic photocur-
ment, shown in Figure 14. The expansive variations, and very temperature dependent. rent in the emitter–base junction Ipar. The
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
ADAPTIVE ELEMENT Apr 2, 1996 9
Current (A)
10 -8 10 -8
The detailed measurements in Figure 16 indicate that I0,m Light
and I0,b are both approximately equal to 0.1–1 fA. The
parasitic photocurrent Ipar in the emitter–base junction is 10 -9 10 -9
e-fold
small but nonetheless important, because it determines 48 mV
the imbalance across the element that must be maintained 10 -10 10 -10
in an adapted condition of zero net current. Of course, Ipar
is never zero, due to scattered light and dark light (thermal
photons). In any case, the conductance near the adapted 10 -11 10 -11
Ib -1
I FIGURE 19 A bad choice
for an adaptive element.
Parasitic photodiode pulls
adaptation node towards
substrate voltage.
FIGURE 17 Theoretical plots of currents in Equation 17,
assuming κ=0.7, I0,m and I0,b both equal 1, and Ipar is zero.
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
10 Apr 2, 1996 PHOTODIODE VS. PHOTOTRANSISTOR
element has offsets of less than 100 mV in A nearby substrate contact sinks the photo-
an adapted state, owing mostly to the zero current to ground. The transistors in the p++ emitter
back-gate voltage. feedback amplifier are long, to maximize
the gain and hence the speedup. The adap- n well base
tive element is made from an isolated well
p– substrate collector
with a single MOS transistor. The capaci-
tive divider is formed from the two levels of FIGURE 20 Parasitic vertical bipolar
PHOTODIODE VS. polysilicon plus a metal plate, but a MOS transistor in an n-well process.
PHOTOTRANSISTOR capacitor may be used instead. The total
area in this conservative layout is about
Previous logarithmic photoreceptor designs 80x80 µm2 in a technology with a 2-µm the photocurrent, but the larger the capaci-
from the Mead lab all used the parasitic ver- feature size. tance. Since we are pretty much stuck with
tical bipolar transistor, shown in Figure 20, The capacitive-divider ratio determines a fixed technology, we regard this speed as a
instead of a photodiode. We used the bipo- the gain of the receptor for transient signals. constant of the problem. †
lar transistor because bipolars have much The capacitance of the adaptive element it- In the devices available in an ordinary
less 1/f noise than MOS surface channel de- self is usually not negligible compared with CMOS process, the maximum receptor
vices, and because the larger output current the explicit feedback capacitor C2 and must speed is obtained from one of the substrate-
is more capable of driving capacitive loads. be taken into account. It consists mostly of junction photodiodes, because these have
Later on, when we started to use active the active-well junction. A typical value is the largest quantum efficiency and the low-
feedback circuits, we continued using bipo- 15 fF, equivalent to a square metal–poly ca- est capacitance. We have chosen to use the
lars for these same reasons. pacitor with an edge length of about 17 µm. active–substrate diode rather than the
Once we started to characterize the noise, well–substrate diode in all of our designs
we immediately discovered that 1/f noise is because the layout is much more compact.
negligible compared with shot and thermal We expect that the well–substrate diode will
noise. In fact, there is a big disadvantage to
using bipolar phototransistors in the adap- THE ILLUMINATION LIMIT have a similar capacitance and quantum ef-
ficiency.
tive receptor. The active feedback speeds up
the response by clamping the input node.
(SPEED)
When we use a phototransistor instead of a From an evolutionary perspective, it is clear † The speed could be increased if we had access to
photodiode, the feedback clamps the emit- that animals that can see in the dark occupy PIN diodes 28, where the p and n regions are sepa-
ter node, but the base is left floating. Indeed, an important niche. The same is true for our rated by an intrinsic region that increases the vol-
the base must not be clamped if the bipolar silicon receptors. If they can only be used in ume, and hence the quantum efficiency, and
transistor is to function with its normal cur- bright sunlight, then they are not very use- simultaneously decreases the capacitance of the
ful. device—but we know of no CMOS process with
rent gain mechanism. The current available these characteristics.
to charge and discharge the base of the tran- The source follower receptor has a re- In a typical 2-µm n-well process, the active-sub-
sistor is approximately the same photocur- sponse speed that is determined primarily strate capacitance is around
rent available from the photodiode. As a by the ratio of quantum efficiency to capaci- 120 aF/µm2+200 aF/µm. In the more heavily
result, we obtain no speedup. The dynamic tance per unit area in the photodiode. The doped 1.2-µm process, these values jump to
range is at least 1–2 decades smaller with a larger we make the photodiode, the larger 500 aF/µm2 + 400 aF/µm.
phototransistor than with a photodiode. In
the discussion of receptor noise, we shall
see that the noise properties of photodiode
receptors and phototransistor receptors are
indistinguishable (page 15). In the context Adaptive
element
of an active feedback circuit, it makes no
sense to use bipolar phototransistors.
Feedback
capacitor
(C2)
Amplifier
RECEPTOR LAYOUT (Co)
Figure 21 shows the layout corresponding
to the schematic in Figure 5. The photo-
diode can be constructed from any of the pn Storage Cascode
Capacitor transistor
junctions that are part of a CMOS process,
(C1)
but the one that we use in practice is the
junction between native source–drain diffu- 6 µm
sion and substrate. In an n-well process, the
junction is between the p– substrate and the
n++ diffusion. We use this junction because
the quantum efficiency is high and the ca- Feedback transistor
pacitance per unit area is relatively small, Photodiode (Cp)
resulting in a fast response. This junction
(also hole in overlying metal)
also has the advantage that it may be con-
structed simply as an extension of the FIGURE 21 Photoreceptor layout. This layout is nonoptimal because the feedback
source of Qfb. All parts of the circuit except transistor is wider than it needs to be, leading to excessive Cfb (see page 5).
for the photodiode are covered with metal.
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
THE DETECTION LIMIT (NOISE) Apr 2, 1996 11
A reasonable definition of the lower lim- illumination the current to ground that they The important observation is that the to-
iting intensity for operation of the receptor produce may exceed the bias current sup- tal noise power, integrated over the entire
is the intensity at which the photoreceptor plied by Qp , pulling the output node to passband of the receptor, is a constant inde-
has a rise time of 15 ms—corresponding to ground. This situation may be ameliorated pendent of intensity. The lower the intensity,
the time for a single field of a video camera by shielding the native-type transistors in the smaller the bandwidth of the receptor,
that scans at 60 Hz, and approximately the the amplifier from light using a metal wir- but the larger the noise level within the
same as the cutoff frequency for human vi- ing layer, and by surrounding them with a passband. The responses of the logarithmic
sion under photopic† conditions. We can guard bar made from native diffusion that is receptor to white noise stimulation show
see from Figure 10 that the fastest photore- preferably tied to V dd . (It doesn’t make that the shape of the noise spectrum is the
ceptor circuit that we have tested has a rise much difference if you tie the guard bar to same as the shape of the receptor transfer
time of 15 ms at about 1 mW/m2 irradi- ground or Vdd.) Guard structures are dis- function, at each intensity level. But while
ance—equivalent to an illuminance of cussed further starting on page 15. the transfer function measurement shows
about 1 lux††, or approximately the light- that the receptor contrast gain is constant,
ing of the full moon. This receptor is built independent of intensity, the underlying
with a feedback amplifier with a gain of noise behaves differently, becoming larger,
several hundred, a closed loop gain of the lower the intensity.
about 10, and a photodiode with an area of THE DETECTION LIMIT We know that the noise arises from with-
20 µm by 20 µm. (NOISE) in the receptor, and is not an artifact of the
CCD detectors are integrating devices; measurement (for instance, from noise in
every frame, they dump out all the charge What is the smallest signal that can be de- the LED light source). The reason is that the
they collected since the last frame. The tected? How is this value affected by light amplitude of the noise depends on the level
sensitivity limit is determined by the elec- intensity, and by detector area? How does of intensity. If the noise arose from the
tron counting noise in the charge-sensing the performance compare with commercial steady light source, then reducing the inten-
amplifier. Current commercial amplifiers, CCD cameras and biological rods and sity by interposing neutral density filters
in consumer end-product devices, function cones? What is the physical basis for the re- would result in a set of curves like the top
at a noise level of about 100 electrons, ceptor noise? set of curves in Figure 23 showing the re-
meaning that the RMS noise in the output We’ll investigate the noise properties of sponse to a white noise source. In other
is equivalent to 100 electrons in the charge the receptors and their detection ability, em- words, if the noise arose in the supposedly
bucket. If we assume that the cameras must pirically and theoretically. We’ll start em- steady source, then the bottom curves
have at least 4 bits resolution to be accept- pirically, with measurements of noise would duplicate the top set of curves but be
able, then the number of electrons collect- properties. These measurements show that shifted down by a constant amount. Since
ed must be 16 times the noise level, or the underlying behavior is very simple. The the behavior is clearly quite different, we
1600, each 1/60 s. A typical CCD pixel simplicity of this behavior motivates an are certain that the noise arises in the detec-
area is 100 µm2. The quantum efficiency is equally simple theory that intuitively and tor itself.
about 30 %. From these numbers, we com- Another important observation is that
quantitatively describes how noise works in
pute that the irradiance is 1.4 mW/m2, or flicker noise (1/f) in the receptor is negligi-
logarithmic photoreceptors.
1 lux at 555 nm, consistent with the adver-
tised ratings††† of a few lux. EMPIRICAL OBSERVATIONS Bias Light
The borderline between rod and cone vi- The observations shown in Figure 23 were
sion occurs at an illuminance of about 1 measured from the simple source-follower
lux, which is approximately the level of (a) Output
detector shown in Figure 22. We captured voltage
bright moonlight. In Figure 10, we have la- the noise power spectra using two types of
beled the rod-cone border and the moon- Photodiode
stimuli, a steady illumination from an LED
light irradiance. and a white-noise source. We used the white
In summary, the current photoreceptor noise source as a direct measurement of the
circuit functions down to about the same receptor transfer function. We did each
intensities as consumer CCD cameras and measurement at several levels of intensity,
human cone receptors. separated by decades. It is clear that all
ILLUMINATION LIMIT: HIGH END characteristics are well above the instru- (b)
MOS transistors Qn and Qcas that form the mentation noise.
bottom of the amplifier circuit in the adap-
tive receptor contain parasitic photodiodes e–
from their drains and sources to the sub- FIGURE 22 The simple logarithmic
strate. Ordinarily, these parasitic photo- photoreceptor used in the study of
diodes are irrelevant, but under intense receptor noise. This circuit forms the
input stage to the adaptive receptor.
(a) shows the schematic form. The bias
† Photopic means cones vision, mesopic means voltage sets a reference for the source
cones and rods are both used, and scotopic means
(c)
voltage, which is the output. (b) shows
rod vision. that the compact receptor consists of a
†† At 555 nm wavelength (the peak of human sen- single MOS transistor whose source Floats to
sitivity under photopic conditions), forms the photodiode. (c) shows the necessary
1 lux = 1.4 mW/m2. For white light (uniformly electron energy diagram. The source level
distributed over visible), voltage floats to whatever level is
1 lux=4 mW/m2=104 photons/µm2s. 21 required to spill the photocurrent over
††† Whatever these ratings mean—supposedly
Electron
the channel and into the drain. energy
each manufacturer makes up their own definition,
but none of them tell you what it is!
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
12 Apr 2, 1996 THE DETECTION LIMIT (NOISE)
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
THE DETECTION LIMIT (NOISE) Apr 2, 1996 13
Amplitude (dBV/√Hz)
receptor is about 60 mV per decade. The gain of
the adaptive receptor is about 1.3 V/decade, or Adaptive receptor
27 dB more than the gain of the source follower -80
receptor. The plots show the measured power without
spectra for each detector, along with the spectrum
of the follower pad instrumentation. The DC Source follower Cascode
-100
irradiance for this measurement is 0.2 W/m2, and
the injected signal is a combination of a 150 Hz 1/f 2
sinusoidal signal and a 1.5 kHz signal each with
-120 1/f
Rayleigh contrast of about 1 %. We can see from Instrumentation
the height of the signal spike relative to the
surrounding noise that the SNR for each detector
are nearly indistinguishable. For definition of dBV -140
units, see Figure 23. 10 0 10 1 10 2 10 3 10 4
Frequency (Hz)
2
The macroscopic flow of current through energy is given by q τ , where τ is the response of the node to a disturbance. If the
the receptor circuit consists of the micro- mean length of time that the charge is disturbance consists of a sudden excess
scopic movement of discrete charges. As present. Hence, the noise energy in a single amount of charge on the node, the system
shown in Figure 25(a), single charges event is given by returns to its equilibrium level after some
cause step changes in the voltage on the ca- 2 time—the time being the average lifetime
pacitor. The charges are collected by the E≡q τ (24) of the individual charges that make up the
photodiode and leave via the channel of the The current consists of a random stream of disturbance. The decay from the excess is
feedback transistor. The time at which a single events. All the events are indepen- exponential, and the e-folding time is once
charge appears, and the time that it stays on dent, so we can compute the mean square again τ . Hence, in Equation 25 we can re-
2
the capacitor, are both random. A given charge fluctuation ∆Q (the noise power) place τ with the time constant of the sub-
charge spends a random amount of time sit- in the signal by taking the noise energy in a threshold circuit, CV T ⁄ I bg , to obtain
ting on the output node. These times are typical event given by Equation 22 and CV T
multiplying by the average number of 2
distributed according to a Poisson distribu- ∆Q = qI bg ----------- = (26)
tion. events per unit time, given by I bg ⁄ q : I bg
Each charge is independent of all the 2 I bg 2 I bg kT
∆Q = E ------- = q τ ------- = qI bg τ (25) qCV T = qC ------ = kTC
others, which means that if we can com- q q q
pute the statistics of the average charge, We can think of the preceding computation which is the same result as Equation 21, ob-
then we can easily obtain the statistics of in two ways: Either as an average of a single tained using equipartition. Notice that here
the current as a whole. We shall first com- particle over many lifetimes, or as an aver- we have suddenly slipped thermal behavior
pute the average noise energy in the step age of many particles over a single lifetime. into a discussion that was strictly statistical.
change of charge caused by a single The two viewpoints lead to identical results; Boltzman statistics determines the time
charge. Using the independence of the the former is perhaps more rigorous in scale of the integration, which in turn deter-
charge events, we shall then compute the terms of obtaining the correct multiplier, mines the total noise.
noise power in a current that consists of a while the latter offers a better intuition of SHOT VS.THERMAL NOISE? W h e n
flow of single charges. what process physically governs the noise we initially attempted to compute the noise
The amplitude of the step change is the behavior—an average over a time window. in the receptor, our theory consistently dif-
unit charge q of a single electron. The noise Formally, the equivalence of the two ap- fered from the measured noise power by a
energy contained in the step is the integral proaches arises from the ergodicity of the factor of about two—our theory always pre-
over time of the squared deviation of the system. dicted a noise power twice what we mea-
charge from the mean value. The mean val- A crucial fact is that the average lifetime sured. This distressing situation was not
ue is zero, since this single charge has finite τ of a charge is the same as the time con- resolved until Rahul Sarpeshkar pointed out
duration. During the presence of the stant of the node—a fact that is obvious but that shot and thermal noise are alternative
charge, the value is simply q, the size of the hard to prove. The reason it is true is that the aspects of a single underlying phenomenon.
charge. Hence, the mean value of the noise time constant describes the time scale of the We had been under the impression that they
τ lifetime
disappears. The distribution of times that the impulse
lasts is described by a Poisson process. The mean q
charge level is zero. The average energy in the
impulse is q2τ. Time
(a) (b) Mean level
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
14 Apr 2, 1996 THE DETECTION LIMIT (NOISE)
-80
FIGURE 26 Photodiode and
Signal -80
phototransistor noise properties are
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
MINORITY CARRIER DIFFUSION & GUARD STRUCTURES Apr 2, 1996 15
S a r p e s h k a r e t a l . 2 6 a n d R o s e ’s t wo The total noise is equivalent to a frac- tion is not good for short distances. The space
books24,25 are helpful. tional variation of 100% of the signal! constant is shorter, the closer the spot to the
PHOTODIODE VS. Under dim lighting conditions, shot sensor, probably because of geometrical ef-
PHOTOTRANSISTOR: noise dominates any detector system. fects due to the finite size of the test spot or
NOISE BEHAVIOR The limited integration time of a CCD three-dimensional effects of the interaction of
Earlier we saw that phototransistors have the detector means that shot noise is a real surface recombination with the pure expo-
disadvantage that the base cannot be limitation to the usable limit of opera- nential decay. With no guard bar, the mea-
voltage–clamped to speed up the response. tion for video imagers, where the inte- sured current is reduced by a factor of 10 in a
Figure 1 shows that the noise properties of gration time is at most 1/30 s. At distance of about 40 µm.
photodiodes and phototransistors are indis- moonlight illumination levels, the num- To test for the possibility that this relatively
tinguishable. The reason is obvious, given ber of charges collected per integration long diffusion length of 30 µm is due to scat-
that the analysis just given may be directly time is less than 1000. The shot fluctua- tering of light, and not minority carrier diffu-
applied to the base-emitter junction of the tion, 30, means that at best—with a sion, we compared the diffusion length
phototransistor. noiseless amplifier—the noise level is measured with a red and green LED. The ra-
3%, limiting the SNR to 30 dB. tio of absorption length for the red compared
NOISE ADVANTAGE OF CONTINUOUS- with the green LED is a factor of 2–3, so if
TIME RECEPTORS OVER SAMPLED light scattering is a significant effect in the
DETECTORS
measurement of diffusion length, we expect a
Continuous-time logarithmic photorecep- difference between the diffusion lengths mea-
tors have a natural advantage in noise per- MINORITY CARRIER sured with the two LEDs. There is no differ-
formance over sampled detector systems.
The advantage comes from the scaling of
DIFFUSION & GUARD ence. The decay with distance is nearly
bandwidth with intensity, so that circuit, in a STRUCTURES
sense, adapts its integration time to how
much signal is available. The less light there Turning now to an different topic, we’ll
is, the longer the integration time. The total discuss the diffusion of minority carri-
noise is held constant. ers and guard structures. We built the
simple structure shown in Figure 27 Probing light spot
In a CCD detector, the bandwidth is limit- 3 um
ed by the sampling rate, because the pixels specifically to measure lateral diffusion
cannot store state across samples. This limit- length of minority carriers, and to test 17 um
ed integration time means that the number of for the effectiveness of various guard Sensing
integrated photons goes as the intensity. The structures in blocking their diffusion. Phototransistor
total noise power goes inversely with the in- The structure consists of a central para-
tensity—it gets worse as the intensity de- sitic bipolar phototransistor surrounded
creases instead of staying constant. There is on three sides by guard bars with differ- 7 um
nothing magical about it: the CCD detector ent widths. The fourth side is left bare.
holds its bandwidth fixed while the log re- The bipolar transistor acts as a probe for
ceptor holds the number of integrated pho- minority carriers by collecting and am-
tons fixed. plifying the local concentration of ex-
cess minority carriers in the bulk. This Guard bars
If we look at the spectral density of the
chip was built in a 2 µm p-well technol- Path of light spot
noise power, we see that there is no differ-
ogy. We measured the emitter current
ence: in both types of detectors the spectrum 1Distance
9/16"
using a Keithley 617 picoammeter, with
of the noise comes from the quantal nature Phototransistor
a 2 V collector-emitter bias.
of light and charge; in both, the noise spec-
tral density goes inversely with the intensity. The guard bars consist of two rectan-
One additional complication is that a gles of ordinary p-type source–drain
CCD detector invariably uses a charge sens- diffusion and a rectangle of p-well dif- Light
ing amplifier to convert the photocharge to a fusion. All guard bars were grounded to spot Source-drain
diffusion
voltage—the signal that is actually used. the bulk potential. Well diffusion
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
16 Apr 2, 1996 SPECTRAL SENSITIVITY
Current (nA)
fective in reducing the minority carrier den- 7 µm diff
sity. The widest guard bar, a 17 µm well, 40 17 µm well
reduces the minority carrier density by up to
a factor of 10, particularly when the test 20
spot shines directly onto the guard bar. The
other two guard bars, which consist of
0
source-drain diffusion, are less effective, re- 0 100 200 300
ducing the carrier density by a factor of 2–3. Distance from center (um)
Interestingly, the 3 µm guard bar seems to
(c) Diffusion expt comparision
be more effective than the 7 µm guard bar. of Red and Green LED
We don’t know the reason for this result, but (b) Effect of guard bars 1
2
we can repeat it. no guard
SUMMARY. We measured a diffusion 3 µm diff
length of about 30 µm at distances of great- 7 µm diff Red LED
er than 70 µm from the source. A mini- 17 µm well response with
mum-size silicon retina pixel has about the superimposed
same dimension, which means that minori- 0
green
log Current (log nA)
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
SPECTRAL SENSITIVITY Apr 2, 1996 17
0
10
p++ n p–
-1
10
B G R IR
10 -2
0.4 0.6 0.8 1 1.2
Wavelength l (µm)
Light
Carrier
Concentration
concentration at the edge of the junction.
L = 5 µm (Red)
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
18 Apr 2, 1996 SPECTRAL SENSITIVITY
(charges/photon)
with discrete LEDs. The photopic visibility well-substrate
curve shows the relative visibility of
photons under photopic conditions; this
curve is normalized to 1 at its maximum.
pb
The primary colors, according to the CIE 0.1 ase
color chart, are labeled on the wavelength -we
axis. A shelf, marked with a hollow arrow, acti ll
Photoptic ve-
appears around the band edge for both wel
the vertical NPN phototransistor and for l see text
Visibility
the p base-well photodiode that forms the 0.01 act
ive
base to collector junction (page 19). The -p b
monochromator calibration curves show ase
the spectral line width of the
monochromator and the calibration
wavelengths. 0.001 Monochromator
calibration
B G R
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2
did not test the parasitic PNP that uses Wavelength λ (µm)
p–base as emitter because we forgot about
it, although almost certainly its behavior is
similar to the usual parasitic PNP that uses 3.75 2.81 2.25 1.88 1.61 1.41 1.25 1.13 1.02 .94
active as emitter. Photon energy (eV)
A scale bar in Figure 31 shows the ap-
proximate vertical dimensions of the junc- es including the gain, so Q(λ) can be larger • The shallow junctions formed inside the
tions of this 2-µm feature size BiCMOS than 1. well are most sensitive to a wavelength
process, according to the MOSIS fabrica- We used a prism monochromator, in con- around 500 nm, while the deep well-
tion service. The n++ emitter is arsenic- junction with a tungsten incandescent substrate junction sensitivity peaks at
doped, with a junction depth of about source, to produce a continuously variable, about 900 nm, well outside the visible.
0.3 µm, and a surface concentration of nearly monochromatic source of light. We • The peak absolute quantum efficiency
1020 cm –3 . The p+ base is boron-doped, had to carefully calibrate several parame- for the photodiodes varies from a high
with a junction depth of 0.45–0.5 µm, and a ters to obtain a reliable measurement 6. of about 0.8 at near-infrared photon
surface concentration of 1–2·1017 cm –3 . energy, to a low of about 0.3 for the
The n-well is phosphorous-doped with a THE SPECTRAL RESPONSES shallow, volume-limited junctions.
junction depth of approximately 3 µm, and Figure 32 shows the absolute quantum effi-
a surface concentration of 3-4·1014 -cm–3. • All the quantum efficiencies are approx-
c i e n cy v e r s u s p h o t o n w a v e l e n g t h . imately 0.3 at the blue end of the spec-
The p-type substrate has a doping of Figure 33 shows the same data on a linear
3–4·1014 cm–3.† trum.
scale. Figure 32 also shows the photopic
We distinguish between the deep, diffu- (daylight) visibility curve, the band edge for • The phototransistor spectral responses
sion-limited detectors where the carrier silicon, the monochromator bandwidth, and are very close to constant multiples of
collection volume is defined mostly by the the wavelengths of the primary colors. We the deeper-junction responses contribut-
minority carrier diffusion length, and the can make several phenomenological obser- ing to the base current for the pho-
shallow, volume-limited detectors where va t i o n s . totransistor. For example, the parasitic
the carrier collection volume is mostly de- PNP response is very similar in shape to
fined by junction edges. • There is a clear distinction between the response of the well-substrate junc-
The aim of the measurement is to obtain photodetectors that collect light-gener- tion, but is completely unlike the
the absolute quantum efficiency Q(λ) for ated carriers from a junction-limited response of the shallow active-well
each of the devices, as a function of photon volume and the photodetectors that col- junction. A similar observation can be
wavelength λ, where Q(λ) is defined as: lect light-generated carriers from a dif- made for the vertical NPN. This obser-
fusion-limited volume defined by the vation simply means that most of the
Q(λ) = (33) diffusion length of minority carriers. base current in the phototransistors
The diffusion-limited detectors are more comes across the deep junction.
# collected charges s e n s i t ive t o l o n g e r wave l e n g t h s .
----------------------------------------------------------------------------------- • The current gain is about 100 in the par-
# incident photons at wavelength λ
The phototransistors have built-in current • The response spectra are broad band. asitic bipolar phototransistor, and is
gain; for them we count the collected charg- All the detectors cover more than the about 30 in the vertical bipolar pho-
visible spectrum. Also, all of the detec- totransistor. This gain is a soft function
tors have responses that are flat within a of the current level, and decreases at
† These parameters come from the specifications factor of 2 or 3 within the visible spec- both high and low intensities. Figure 34
for the MOSIS 2–µm n-well BiCMOS process. trum. shows the current gain for the bipolar
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
PREVIOUS WORK Apr 2, 1996 19
1
FIGURE 33 Absolute quantum efficiencies plotted on a linear
Photoptic scale. Data are a subset of Figure 32.
0.8 Visibility well-substrate
Quantum Efficiency
0.6 200
0.4
active-well 150 Parasitic PNP vertical bipolar
0.2 p base-well
ß-1 100
active- p base (IE/I B)
0
0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 Vertical NPN bipolar
50
Wavelength (µm)
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
20 Apr 2, 1996 RELATION TO BIOLOGICAL PHOTOTRANSDUCTION
tion, one at the phototransducer, and the tricks like PIN photodiodes and avalanche stage, by the exponential I-V relationship at
other in the feedback loop. This design had multiplication. the source of the feedback transistor. The
a large transistor count, lacked any speedup logarithm gotten for free, courtesy of Boltz-
advantage of the active feedback, and used man, is apparently not used for this same
an inferior adaptive element. purpose in biological receptors. The obvi-
Delbrück and Mead 7 built an adaptive ous first guess, a membrane channel with
silicon retina for sensing time derivatives of RELATION TO diode rectifying characteristics, is not what
the image contrast, using the same feedback BIOLOGICAL the cones or rods use.
arrangement as described here, but lacking
the speedup advantage of using a photo- PHOTOTRANSDUCTION
diode and a cascode, and using an inferior In the introduction we discussed functional
adaptive element. This silicon-retina chip characteristics of biological photoreceptors SUMMARY
uses the scanning mechanism to reset the and how they inspired the silicon version.
pixel output every time the pixel output is We have discussed a photoreceptor circuit
Three characteristics stood out: The adap- that is useful for continuous-time pho-
sampled. tive properties, the gain control resulting in
Mead17 built an adaptive silicon retina totransduction. It consists of a logarithmic
illumination independent contrast response, input stage coupled to an adaptive feedback
that uses ultraviolet light to move charge and the invariance of the response time to il-
onto and off floating gates for storage of an- circuit. The feedback circuit produces an
lumination. output with high gain for transient—and
alog mismatch compensation. A pho-
totransistor with a drain-type load produces In the silicon receptor, the adaptation presumably interesting—signals, and has
high voltage gain at the input, and this high state is stored as a charge on a capacitor. In low gain for static signals, including circuit
gain point is used in the UV adaptation. Un- rods and cones, the adaptation state is (at offsets. The feedback also speeds up the re-
fortunately, making high gain at the input is least partially) stored as the internal calci- sponse by clamping the input node voltage.
incompatible with high speed operation, um concentration. The full story of photore- The receptor can be fabricated in an area
and inherently prevents the type of voltage- ceptor adaptation mechanisms is too of about 70 by 70 µm2 in a 2–µm single
clamping speedup we have discussed. Sub- complex for discussion here, but the idea poly CMOS process. The useful dynamic
sequent work has tended away from UV ad- that there is an independent state variable range, which we arbitrarily define as the
aptation, because of technical problems for adaptation state is similar in the silicon range over which the bandwidth is at least
with shining high-energy photons onto the and biological receptors. 60 Hz, extends down to moonlight illumi-
chip and making the circuit function cor- More germane is the invariance of re- nation levels, or at least 1.5 decades more
rectly at the same time. sponse time to illumination, because here than a plain logarithmic detector without
Mahowald12 incorporated the feedback we see some functional differences. How active feedback.
arrangement described here into a silicon does the response-time invariance come The receptor noise in a plain logarithmic
retina with interesting network feedback. about? The enormous gain generated in detector is almost exactly what is predicted
Her receptor does not have the speedup ad- rods comes from 3–4 amplification stages, by a simple noise theory that assumes that
vantage of the active feedback, and uses an each with modest gain. The gain of each the noise is purely shot noise, with a band-
inferior compressive adaptive element, stage is at most a few hundred, but the com- width set by the conductance of the feed-
leading to very unsymmetrical adaptation bination of all stages results in a maximum back and the capacitance of the input node.
rates for bright- and dark-going illumina- gain that closes 10 6 ion channels in re- The noise in the adaptive receptor is within
tion changes. Mahowald’s chip with the re- sponse to a single photon. a factor of two of this value.
ceptor improvements we have discussed The gain of the rod is inversely propor- TECHNICAL INNOVATIONS
here is in fabrication. tional to intensity. The mechanism for this The three technical improvements in the re-
Mann15 developed several adaptive pho- gain control probably lies in a modest gain ceptor are as follows:
toreceptor circuits that are more flexible reduction of each stage of the amplification. 1. Using a photodiode, instead of the pho-
than the ones described here, in that they A small change in the gain of each stage re- totransistor used in previous designs,
have an adjustable temporal passband. sults in a large change in the total gain. Each increases the dynamic range by at least
These receptors also use U.V. mismatch stage has a fixed gain–bandwidth product: a decade, without degrading signal qual-
compensation. They use a larger number of When the gain is reduced, the response time ity.
components than the receptor described decreases proportionally. The trick is that 2. Using a cascode transistor in the feed-
here. They also lack the advantage of active the total gain goes as the power of the num- back amplifier yields another 0.5–1
speedup and use inferior adaptive elements. ber of gain stages, and the total time re- decade dynamic range.
None of the previous photoreceptors sponse goes linearly with the number of 3. Using a new adaptive element increases
were satisfactorily characterized, in the gain stages. the adaptation time constant and the
sense that no one bothered to measure sim- This multistage gain mechanism con- symmetry of the receptor response to
ple engineering metrics like usable dynamic trasts with the silicon receptor, which has light- and dark-going transients, com-
range and sensitivity. only a two-stage amplifier, one of which has pared to previous designs, which used
Up until lately, continuous-time analog fixed gain. As shown earlier, this gain elements that were very susceptible to
photoreceptors have not received a great mechanism results in a fast response time, the effects of parasitic minority carriers.
deal of commercial attention. Nearly all ef- but the response time is also inversely pro-
fort is concentrated on sampled imagers portional to the intensity. We have not elect-
like CCD video cameras. Photoreceptors ed to try to build a silicon receptor with
are used in very specialized applications multistage gain, in analogy with the biolog-
like optical repeaters, where high integra- ical mechanism, because the response time
tion density and wide dynamic range are not is adequately fast already, and the log gen-
so important compared with response erating mechanism in subthreshold opera-
speed, and where special fabrication tech- tion is so convenient. All the gain control in
nologies are available that allow use of the silicon receptor is done right at the input
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
CONCLUSION Apr 2, 1996 21
We have also discussed two theoretical top- manium and silicon at 77 ˚K and 300 ˚K,” 16. C.A. Mead, “Neuromorphic Electronic
ics in which we have developed a much Physical Review, vol. 99, pp. 1151–1155, Systems,” Proceedings of the IEEE, vol.
clearer understanding: 1955. 78, pp. 1629–1636, 1990.
1. We quantitatively understand the physi- 4. D. Cohen and G. Lewicki, “MOSIS—the 17. C.A. Mead, “Adaptive Retina,” in Ana-
cal origin and dominant sources of ARPA silicon broker,” in Proceedings from log VLSI Implementation of Neural Sys-
the Second Caltech Conference on VLSI, tems, C. Mead and M. Ismail, Eds.,
electronic noise in continuous-time log- Boston: Kluwer Academic Pub., pp.
California Inst. of Tech., Pasadena CA,
arithmic receptors. 239–246, 1989.
pp. 29-44, 1981.
2. We quantitatively understand the limits 5. T. Delbrück, “Silicon retina with Correla- 18. C.A. Mead, Analog VLSI and Neural
on response speed that effectively limit tion-Based, Velocity-Tuned Pixels,” IEEE Systems. Reading, MA: Addison--Wesley,
the lower end of the receptor dynamic Transactions on Neural Networks, vol. 4, 1989.
range. no. 3, pp. 529–541, May 1993. 19. C.A. Mead, “A Sensitive Electronic
6. T. Delbrück, Investigations of Analog Photoreceptor,” In 1985 Chapel Hill Con-
VLSI Phototransduction and Visual ference on VLSI, H. Fuchs, Ed., Rockville:
Motion Processing, Ph.D. Thesis, Dept. of Computer Science Press, pp. 463–471,
Computation and Neural Systems, Cali- 1985.
CONCLUSION fornia Institute of Technology, Pasadena, 20. C.A. Mead and M.A. Mahowald, “A Sili-
Operations that require spatiotemporal CA, 91125, 1993. con Model of Early Visual Processing,”
computation on noisy input data in real 7. T. Delbrück and C.A. Mead, (1991), Sili- Neural Networks, vol. 1, pp. 91–97, 1988.
time are a natural fit to analog VLSI. We con adaptive photoreceptor array that 21. J.R. Meyer-Arendt, “Radiometry and
computes temporal intensity derivatives, Photometry: Units and conversion fac-
have shown how to build a practical photo-
in T.S. Jay Jayadev (ed.) Proc. SPIE, tors,” Applied Optics, vol. 7, pp.
receptor circuit that is suitable as a front- 2081–2084, 1968.
end transducer for analog VLSI vision sys- Infrared Sensors: Detectors, Electronics,
and Signal Processing., vol. 1541, pp 22. L. Nielson, M. Mahowald, and C.
tems. The receptor is usable under photo- Mead, “SeeHear,” in Analog VLSI and
pic vision conditions. The continuous 92–99.
Neural Systems, by C. Mead, Reading:
transduction process leads to intensity-in- 8. T. Delbrück and C.A. Mead, “An elec-
Addison-Wesley, chapter 13, pp.
tronic photoreceptor sensitive to small
variant total noise that is within a factor of 207–227, 1989. (adapted there from 1987
changes in intensity,” in Advances in
two of photon counting noise. Photorecep- Neural Information Processing Systems
International Association for Pattern Rec-
tors of this general type will undoubtedly ognition, 5th Scandinavian Conference
1, D.S. Touretzky, Ed., San Mateo: Mor-
be used in commercial vision systems, on Image Analysis.)
gan Kaufman, pp. 720–727, 1988.
where designers will not be able to resist 23. R.A. Normann and I. Perlman, “The
9. T. Horiuchi, J. Lazzaro, A. Moore and C. effects of background illumination on the
the capability to link low-power, continu- Koch, “A delay line based motion detec-
ous-time computation together with trans- photoresponses of red and green cones,”
tion chip,” in Advances in Neural Informa- J. Physiol., vol. 286, pp. 509–524, 1979.
duction. tion Processing Systems 3, R. Lippman, 24. A. Rose, Concepts in Photoconductiv-
J. Moody, D. Touretzky, Eds., San Mateo, ity and Allied Problems, InterScience
CA: Morgan Kaufmann, pp. 406–412, Publishers (div. of John Wiley), p. 104,
1991. 1963.
10. M.A. Mahowald, An Analog VLSI Sys-
ACKNOWLEDGEMENTS tem for Stereoscopic Vision, Kluwer Aca-
25. A. Rose, Vision: Human and Electronic,
Plenum Press: New York, 1973.
We acknowledge support from the ONR demic Publishers: Norwell, 26. R. Sarpeshkar, T. Delbrück, and C.A.
and by MOSIS, the ARPA silicon foundry.4 Massachusetts, ISBN 0-7923-9444-5, Mead, “White noise in MOS transistors
David Van Essen inspired this detailed 1994. and resistors,” IEEE Circuits and Devices,
study (especially of the noise properties). 11. M.A. Mahowald, VLSI Analogs of Neu- vol. 9, no. 6, pp. 23–29, 1993.
Rahul Sarpeshkar made a crucial observa- ronal Visual Processing: A Synthesis of 27. R. M. Shapley and C. Enroth-Cugell,
tion about the relationship between shot Form and Function, Ph.D. Thesis, Califor- “Visual adaptation and retinal gain con-
nia Inst. of Tech., Dept. of Computation trols,” in Progress in Retinal Research, N.
and thermal noise. David Standley and
and Neural Systems, Pasadena CA, Osborne and G. Chader, Eds., New York:
Dick Lyon provided helpful comments 1992.
about the manuscript. Micah Siegel and Pergamon Press, vol. 3, pp. 263–346,
12. M.A. Mahowald, “Silicon Retina with 1984.
Sanjoy Mahajan pointed out that κ varia- Adaptive Photoreceptor,” Proc. 28. S.M. Sze, Physics of Semiconductor
tions affect the gain vs. intensity character- SPIE/SPSE Symposium on Electronic Sci- Devices, Second Edition, New York: John
istics and their measurement discrepancies ence and Technology: from Neurons to Wiley & Sons, chapter 13 and appendix I,
inspired the discovery of the importance of Chips, vol. 1473, pp. 52–58, April 1991. 1981.
the Miller effect in Qfb. 13. M.A. Mahowald and C.A. Mead, “Sili- 29. J.E. Tanner and C. Mead, “An Inte-
con Retina,” in Analog VLSI and Neural grated Analog Optical Motion Sensor,” in
Systems, by C. Mead, Reading: Addison- VLSI Signal Processing, II, S.Y. Kung,
Wesley, pp. 257–278, 1989. Ed., New York: IEEE Press, pp. 59–76,
14. M.A. Mahowald and T. Delbrück, 1986.
REFERENCES “Cooperative stereo matching using
1. D.A. Baylor, G. Matthews, K.-W. Yau, static and dynamic image features,” in
“Two components of electrical dark Analog VLSI Implementation of Neural
noise in toad rod outer segments,” J. Systems, C. Mead and M. Ismail, Eds.,
Physiol. vol. 309, pp. 591–621, 1980. Boston: Kluwer Academic Pub., pp.
2. R. G. Benson and T. Delbrück, (1991). 213–238, 1989.
Direction-selective silicon retina that 15. J. Mann, “Implementing Early Visual
uses null inhibition, in D.S. Touretzky, Ed. Processing in Analog VLSI: Light Adapta-
Advances in Neural Information Process- tion,” Proc. SPIE/SPSE Symposium on
ing Systems 4. pp. 756–763. Electronic Science and Technology: from
3. W.C. Dash and R. Newman, “Intrinsic Neurons to Chips, vol. 1473, pp.
optical absorption in single-crystal ger- 128–136, April 1991.
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
22 Apr 2, 1996 REFERENCES
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
INDEX Apr 2, 1996 23
INDEX J S
junction capacitance, typical 10 Sarpeshkar, Rahul 13, 15
Numerics junction-limited volume 18 shelf in spectral response of phototransistor
1/f noise 11 K 19
60 Hz 11 signal to noise ratio in CCD video imager 15
κ(kappa) 3 signal to noise ratio of photodiode and pho-
A
L totransitor compared 14
absolute current level in photodiode 19 silicon retina 19
layout
absorption length of light 16, 17 single photon counting, relation to contrast 2
area of adaptive receptor 10
adaptive element 8–10 SNR, see receptor, signal to noise ratio
adaptation rate 9 minimizing Miller effect 6
of adaptive receptor 10 spectral density of noise 14
capacitance of 10 spectral sensitivity 16–19
conductance of 9 used to test minority carrier diffusion 15
lighting of the full moon 11 speedup
area of adaptive receptor 10 definition 5
artificial lighting 3, 8 lux, relation to W/m2 11
subthreshold transistor law 3
B M sunglasses 4
BiCMOS process Mahowald adaptive silicon retina 3
Mahowald, Misha 19
T
junction depths 18 thermal voltage 3
photodetectors 16 Mann, Jim 20
biological photoreceptors Miller effect 4, 5 V
adaptation state 20 minimizing 6 VT 3
gain and adaptation characteristics 2 minority carrier diffusion 15–16
gain mechanism 20 effect of diffusion on high illumination limit
invariance of response time 2, 20 11
bipolar transistor effect of lifetime on receptor response
current gain 18 speed 6
noise behavior 15 moonlight illumination level 11
structure 10 MOSIS 14, 18, 21
using in adaptive receptor 10 N
C neutral density filters 4
capacitance noise in receptor, see receptor
of adaptive element 10 noise, see receptor
of typical junction 10 P
typical values in receptor 6
capacitance to quantum efficiency ratio 10 photodiode, absolute current level 19
capacitive divider 3, 4 photometry 11
cascode photopic visibility 18
effect on time response and noise 4 PIN photodiode 10
functions in adaptive receptor 4 primary colors 17
CCD detector 11 Q
degradation by carrier diffusion 16 quantum efficiency
CMOS process, photodetectors can con- definition 18
struct in 16 discussed 18–19
D measured, plot of 18
Dash and Newman 17 peak 18
dBV units 12 R
density of states 16 ratio of photodiode quantum efficiency to
diffusion-limited volume 18 capacitance 10
doping of typical BiCMOS process 18 receptor
E effect of finite minority carrier lifetime 6
effective input capacitance 5 gain-bandwidth product 20
effective input conductance 5 logarithmic, technological context 19
e-fold 3 noise 11–15
ergodicity 13 noise in adaptive 12
F response time 6
feedback circuit second-order behavior 7–8
closed-loop gain 4 signal to noise ratio 13
total loop gain 5 signal to noise ratio degradation by feed-
voltage gain 3 back 12
feedback circuit, noise in 12 simplest logarithmic 2
flicker noise 11 source-follower 2
fluorescent lighting, typical level of 19 spectral density of noise 14
flying insects 1 speedup 7
foveal resolution, need for in imagers 1 speedup obtained by active feedback 5
full moon lighting 11 steady-state gain 4
transient gain 4
G
resistor, impracticality of making in CMOS
guard structures 15–16 process 8
I rod-cone border 11
image contrast 2 Rose, Al 15
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead
24 Apr 2, 1996 INDEX
Analog VLSI Phototransduction CNS Memo #30 T. Delbrück & C.A. Mead