You are on page 1of 137

Solar energy lecture notes

Modeling and analysis of photovoltaic and electrochemical solar


energy systems

Raymond A. Adomaitis
Department of Chemical & Biomolecular Engineering and
Institute for Systems Research
University of Maryland
College Park, MD 20742
adomaiti@umd.edu

c Copyright by
Raymond A. Adomaitis

June 29, 2016


2
Contents

1 Energy - units and basic concepts 7


1.1 Basic concepts from thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 SI units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Derived quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.1 Electrical quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.2 A look ahead . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2 The Sun 15
2.1 Blackbody radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.1 AM0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.2 Numerical integration of spectral irradiance . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Atmospheric absorption effects on spectral irradiance . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.1 AM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.2 Direct versus diffuse radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3 The projection effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.1 Optimal panel tilt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3.2 Integrating projection and absorption effects . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3.3 3D modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4 Insolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.5 Cloud cover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.5.1 Insolation data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3 Transient energy balances 30


3.1 Energy conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1.1 A simple example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1.2 Over a single sunny day . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.1.3 Battery storage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

1
CONTENTS 2

3.2 Material and energy balance review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33


3.2.1 Net work by inlet/outlet flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3 Solar water heater - closed system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4 Solar water heater - open system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.4.1 Computing the warm-up period . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.5 Pumped storage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.6 Sodium-sulfur batteries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

4 The physics of PV devices 42


4.1 Band structure of solid materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.1.1 Fermi level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.1.2 Doping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.1.3 Semiconductor electrical properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.2 Junctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2.1 Forward/reverse bias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.2.2 Diode non-ideality β . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.2.3 Other junctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.3 Photon interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

5 PV cell architecture and manufacturing methods 53


5.1 Crystalline Si cell substrates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.2 PV cell fabrication processing steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.3 Thin film deposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.3.1 Film morphology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.3.2 CVD as a thin-film manufacturing process . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.4 Amorphous Si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.4.1 Device structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.4.2 Device manufacture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.5 CIGS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.5.1 Cell fabrication steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.5.2 Replacement of CdS Layer with ZnO by ALD . . . . . . . . . . . . . . . . . . . . . . . . 62
5.6 CdTe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.6.1 Device architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.6.2 Device manufacture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
CONTENTS 3

6 PV cell power and the equivalent circuit model 66


6.1 Cell currents and the equivalent circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
6.1.1 Short circuit current, open circuit voltage . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.2 Maximum cell power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
6.2.1 Fill factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
6.3 Parasitic resistances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6.3.1 Known values of Io and Iph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6.3.2 Known values of Voc and Isc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.3.3 An iterative approach for Pmp when Rsh → ∞, Rs > 0 . . . . . . . . . . . . . . . . . . . 75
6.3.4 Maximum power with parasitic resistances - general case . . . . . . . . . . . . . . . . . . 75
6.4 Temperature effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

7 External and internal quantum efficiency 80


7.1 Photon energy and spectral response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
7.1.1 Internal quantum efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7.1.2 External quantum efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
7.1.3 Spectral responsivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
7.2 Theoretical efficiency as a function of Ebg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
7.2.1 Multijunction devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
7.2.2 A preview of concentrating PV systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
7.2.3 Splitting the spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
7.3 Antireflection coatings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
7.3.1 AR film optical properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
7.3.2 Reflectance modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
7.3.3 AR film absorbance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
7.3.4 Integrating the elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
7.4 The Shockley-Queisser limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

8 Cell and panel interconnections 96


8.1 Identical cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
8.1.1 Rectangular arrays of cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
8.1.2 Thin film commercial panels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
8.1.3 Commercial crystalline Si PV panels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
8.2 Shaded, faulty, or otherwise nonuniform cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
8.2.1 Shaded cells: parallel-cell case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
8.2.2 Shaded cells: series case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
CONTENTS 4

8.2.3 Experimental validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103


8.2.4 Bypass and blocking diodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
8.3 System integration issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

9 Dye Sensitized Cells 108


9.1 Electrochemistry review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
9.2 Operating principles of the DSSC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
9.3 Equivalent circuit model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

10 Photoelectrochemical systems 114


10.1 Water splitting fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
10.1.1 Electrolyte pH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
10.2 p-type PEC cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
10.3 n-type PEC cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
10.4 Quantitative analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
10.5 Nonidealities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

11 Photosynthesis and the efficiency of bioethanol production 123


11.1 Photosynthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
11.1.1 Chlorophyll . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
11.2 Light reactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
11.2.1 Electrons in the Z-scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
11.3 Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
11.4 Case study: energy efficiency of converting corn to ethanol . . . . . . . . . . . . . . . . . . . . . 126
11.4.1 A PV farm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
11.4.2 Corn-based ethanol production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
11.4.3 Downstream processing operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

12 Elementary numerical methods 130


12.1 Linear interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
12.2 Quadrature and the trapezoidal rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
12.3 Finite differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
12.4 The Euler integrator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
12.5 Newton’s method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
12.5.1 Quadratic convergence - numerical analysis . . . . . . . . . . . . . . . . . . . . . . . . . 133
CONTENTS 5

12.5.2 Newton-Raphson . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134


12.5.3 Approximation of Jacobian elements by FD . . . . . . . . . . . . . . . . . . . . . . . . . 135
12.6 Class software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

List of variables and symbols


Parameter Value Units Description
c 2.99792458 × 1017 nm s−1 speed of light
e or q 1.60217646 × 10−19 C elementary (positive) charge
Esc 1366 W m−2 solar constant
F 9.64853399 × 104 C mol−1 Farady constant
h 4.13566733 × 10−15 eV s Planck’s constant
6.62606896 × 10−34 Js
kB 1.3806503 × 10−23 J K−1 Boltzmann constant
8.6173324 × 10−5 eV K−1
m 9.109 × 10−31 kg electron mass
RE 6.37 × 106 m Earth radius
Rg 8.31447 m3 Pa mol−1 K−1 gas constant
8.31447 J K−1 mol−1
β 1 to 2 diode nonideality factor
0 8.8542 × 10−12 A2 s4 kg−1 m−3 vacuum permittivity
r (Si) 11.68 Si relative permittivity
σ 5.670400 × 10−8 W m−2 K−4 Stefan-Boltzmann constant
CONTENTS 6

Variable Units Description


Ea , Ec , Ed , Ev eV acceptor, CB, donor, VB energies
Ebg eV band gap
Eeλ W m−3 or W m−2 nm−1 spectral irradiance
EE λ W m−3 or W m−2 nm−1 solar spectral irradiance
ED , EF , EG W m−2 direct, diffuse, and global irradiance
Feλ photons s−1 m−2 nm−1 photon spectral flux
FE λ photons s−1 m−2 nm−1 solar photon spectral flux
I A conventional current
Io A diode model dark saturation current
Iph A diode model photo current
NA sites cm−3 acceptor dopant site number density
ND sites cm−3 donor dopant site number density
q C charge
Q J thermal energy (heat)
Rs Ω diode model series resistance
Rsh Ω diode model shunt resistance
t s time
V m3 or V volume or voltage
w J work
z m or dimensionless position or coordinate basis vector
o
δ declination
o
ζ solar zenith angle
ηL luminous efficiency
o
λ m or wavelength or longitude
ν s−1 frequency
o
φ latitude
Φ J work function
Chapter 1

Energy - units and basic concepts

Energy conversion, such as that which is concerned with converting solar radiant energy to electrical, chemical,
or kinetic energy is a key topic in these lecture notes, so starting with a review of the basics of energy, power, and
radiant intensity unit definitions makes sense. To put these issues in context, consider Fig. 1.1 where a typical
silicon solar cell is illustrated. The basic questions we seek to answer include

1. How do we relate photon energy and flux to the electrical power produced by the PV cell?
2. How do we develop mathematical models describing the internal currents and total electrical power produced
by PV cells?
3. How do we engineer the PV cell components to optimize cell performance?
4. How do we expand our fundamental knowledge of solid-state PV devices to electrochemical and biological
systems?

Let us now consider Fig. 1.2 where the electric power demand of approximately 30 million Californians [CaISO]
is plotted for a 24-hour period. When we examine the power production potential of wind, solar, and other
renewable resources, we observe a number of important phenomena:

1. The renewable energy sources can introduce a high degree of dynamic (in time) variability in the power
supply resulting from wind variations and passing clouds that reduce the intensity of the solar radiation
reaching the solar panels.
2. The temporal variations have multiple time-scales, introducing higher-frequency fluctuations with time scales
measured in seconds or minutes and diurnal variations that take place over the day.
3. There is a strong incentive to include energy storage elements in the overall design to reduce the power
fluctuations and to allow for peak power delivery at the 5 pm demand peak.

To compensate for the variations introduced by solar and wind power sources, some type of an energy storage
system is used to help match power production levels to demand; see the right plot of Fig. 1.2 where the difference
between production and demand is plotted. The increasing use of renewable power sources complicates the energy
picture and requires the integration of a number of energy generation, conversion, and storage technologies.

1.1 Basic concepts from thermodynamics


We begin our review of basic concepts with this short list of elementary definitions from thermodynamics; a more
substantial review will take place in the chapter on transient energy balances and later when the ultimate efficiency

7
CHAPTER 1. ENERGY - UNITS AND BASIC CONCEPTS 8

e-
n0 -
n1
n2 + h+ e- n-type
+ + +
- - - - emitter Figure 1.1: The architecture of a typical Si
PV cell.

h+ e- Ie,sc
h+ e-
p-type
Ih,sc base
h+ e-

+
2500 09/02/13 45000 09/02/13
geothermal demand
wind scaled renewables
2000 solar 40000

1500 35000
MW

MW

1000 30000

500 25000

0 5 10 15 20 20000 5 10 15 20
time of day time of day

Figure 1.2: Renewable energy production (left) and overall electricity demand (right) in California during a 24 hour period
in late summer 2013 [CaISO].

of PV cells is considered. The study of thermodynamics is crucial to rigorously assessing the efficiency of the solar
energy devices described later and provides a quantitative means of comparing competing technologies.
We now define the following terminology and conventions:

• A system is a component in a universe and is the subject of our thermodynamic analysis.


• The surroundings are in contact with the system and typically are involved in heat, material, or charge
transfer.
• A boundary separates the system from the universe.
• The universe is composed of the system and its surroundings.
• Closed systems do not allow matter to enter or leave.

• Open systems allow matter to enter or leave.


CHAPTER 1. ENERGY - UNITS AND BASIC CONCEPTS 9

• Isolated systems are closed and also do not allow energy transport across the system boundaries.
• Heat is energy in transit between a system and its surroundings or between two systems, driven by the
temperature difference between the two. Heat energy has units of J (Joules) and symbol Q; Q > 0 will
denote heat flowing into the system.

• Intensive properties do not depend on the total amount of material present; examples include temperature
T , pressure P, and molar quantities such as molar volume V (unless we specify by-mass).
• Extensive properties do depend on the quantity of material, such as volume V t or mass m. Note the use
of the t superscript.

1.2 SI units
There are seven base SI units; all other units are derived from combinations of these quantities. Some of these
we know and deal with on a daily basis such as

m, kg, s

that require no additional introduction. We are also familiar with the notion of temperature, though we be less
so with its actual definition. Kelvin, the unit of temperature is 1/273.16 of the thermodynamic temperature of the
triple point of water [NISTkelvin]. Recall the Gibbs phase rule in which the thermodynamic degrees of freedom
(d.o.f .) are related to the number of components c and phases π: d.o.f . = c − π + 2. Therefore, for the one
component of water with three phases, d.o.f . = 0 and so the temperature and pressure are both uniquely defined
at the triple point.
For the fifth base unit, a mole is defined as the number of atoms or molecules equal to the number of atoms in
12 g of carbon-12. Of course, this is equal to Avogadro’s number NA = 6.02214179 × 1023 .
Because of our focus on solar energy issues, we must also consider the remaining two base SI units. Electric
current has a base SI unit of amperes (A) defined by the current that produces a prescribed amount of force per
unit length between two parallel conductors at a specified separation. With n as the number of electrical charges,
current I is defined as
dn
I =
dt
and has units amperes (A) where 1 A = 1 C/s. The coulomb (C) is a unit of electrical charge and corresponds to

1 electron: e = 1.6022 × 10−19 C


1 C = 6.24 × 1018 electrons

We distinguish between conventional current which is the flow of positive charges (this unfortunate convention
is due to Benjamin Franklin) and electron current which to many makes more physical sense. If I is constant,
the total charge transferred due to the current is
Z t
q= I dt.
t0

Note that we assume I represents conventional current unless explicitly specified.


Finally, there is the candela (cd) which measures luminous intensity and is defined by a monochromatic light
source with frequency ν = 5.4 × 1014 s−1 and radiant intensity of
1 W
1 cd = .
683.002 sr
CHAPTER 1. ENERGY - UNITS AND BASIC CONCEPTS 10

Figure 1.3: A package of incandescent light bulbs, gener-


ating the total luminous power of approximately 10 typ-
ical candles.

A typical candle radiates approximately 1 cd of radiant power. Solid angles are measured with the dimensionless
unit of steradian (sr), defined by the quotient of a spherical cap area S and the square of the radius r of the
sphere defining that spherical cap
S
solid angle ω = 2 steradians.
r
To visualize the steradian, think of the angle created by a solid cone where the base consists of the spherical cap
surface S. If that area is chosen so that S = r 2 , the solid cone corresponds to 1 sr. One can also compare the
steradian to the more familiar unit of angular measurement, the radian. Given that the surface area of a sphere
is 4πr 2 , we see that a sphere measures 4π sr (compare to 2π radians for a full circle). Therefore, a light source
that isotropically radiates 1 candela (power/solid angle) produces 4π lumens of total luminous power. As such,
1 lm = 1 cd·sr = (1/683.002) W. To understand this in more concrete terms, examine a representative package
of light bulbs, e.g., Fig. 1.3 or the LCD projector brightness ratings for representative displays.
Taking into account the area over which the luminous flux falls, we define 1 lux (lx) = 1 lm m−2 where

cd· sr
1 lx = 1 .
m2
One must be careful in converting from candela to lux when the area in question is a flat surface, such as with an
optical sensor element. To summarize, the overarching concept connecting all of these units is that measure of
luminous intensity (candela, lumen) are values associated with the light source, while the flux (lux) can depend
on the distance to the luminous source, the geometry of the surface onto which it falls, and the nature of the
light (diffuse or direct).
Luminous efficiency ηL is defined by the fraction of chemical or electrical energy that is converted to visible light.
Therefore, ηL = 1 (100% luminous efficiency) corresponds to 683.002 lm/W.
We can convert the frequency ν of light to wavelength λ using the speed of light in a vacuum as c = 299, 792, 458
m/s and so
c 2.998 × 108 m/s (109 nm/m)
λ= = = 555 nm.
ν 5.4 × 1014 s−1
This corresponds to green visible light and is the wavelength to which the human eye is most sensitive. This
brings up an important point: the units of candela, lumen, luminous efficiency, and others just described are most
useful for lighting (photometry) applications, where only the visible portion of the spectrum is important. These
measures require the use of a luminosity function y (λ) to model the human eye’s spectral sensitivity to light.
Therefore, a light source with some total spectral power distribution Eλ0 (λ) in W/m (note that the unit of m
CHAPTER 1. ENERGY - UNITS AND BASIC CONCEPTS 11

corresponds to wavelength) emits visible light with an intensity defined by the inner product
Z∞
lm
683.002 y (λ)Eλ0 (λ) dλ lumens.
W
0

The luminosity function y (λ) is defined by several standards and takes the form of a Gaussian-like distribution
centered, for example, at λ = 555 nm for daylight conditions.
Because photovoltaic and solar thermal systems can make use of the infrared and ultraviolet portions of the
spectrum, we will find that these units generally will not be used in our calculations. More appropriate measures
of solar radiant intensity (in radiometry units) such as irradiance and insolation will be defined in detail in the
next chapter.

1.3 Derived quantities


With our description of base SI units complete, we turn to defining a number of important derived quantities. We
first recall that force, when applied to a mass, produces acceleration. Units of force are N or kg m s−2 . Pressure
is the normal force exerted by a fluid onto a surface per unit area
F
P= = N/m2 = kg m−1 s−2 = Pa
A
We make note of the useful quantities 1 atm = 101325 Pa and 1 bar = 105 Pa.
Work is force applied over distance; in differential form dw = F dz so
Z z1
w= F dz
z0

and w > 0 when done to a system. The unit of work is

W : N·m = (kg m/s2 )m = J.

We will use the convention that work done by a system has negative values and work done on a system as positive.
Energy has the same units as work. Let us consider Newton’s Second Law F = ma which can be rearranged to
produce
du
dw = F dz = ma dz = m dz
dt
but the velocity u is defined by
Z u1  2
u02

dz u1
= u so dw = mu du thus w = m u du = m − .
dt u0 2 2

The calculation above immediately illustrates that kinetic energy EK = w > 0 if u1 > u0 which is consistent with
our sign convention. Likewise, we have the potential energy EP associated with mass m and gravitational field g

dw = F dz = mg dz so w = EP = mg (z1 − z0 )

which also is consistent with our intuition and sign convention.


Before leaving this discussion, let us consider the hydrogen atom and Bohr’s model of the electron energy states.
The energy of the single electron in the H atom is given by

−2π 2 mq 4
En =
h2 n2
CHAPTER 1. ENERGY - UNITS AND BASIC CONCEPTS 12

voltage drop
voltage rise
A B
Figure 1.4: An electrical element.
+ -
I

where q is the unit electron charge, m is electron mass, n = 1, 2, 3, ... is the principle quantum number, and h is
Planck’s constant (in appropriate units). Given that the ground state corresponds to n = 1, the energy required
to move an electron from the ground state to some excited state nE > 1 is
"  #
2
−2π 2 mq 4 1
∆En = −1 >0
h2 nE

and so work is done on the electron. When the electron falls back to its ground state and emits a photon with
energy Ep = hν, we note ∆E < 0 and so work is done by the electron.
Finally, power is energy per unit time and so has units J/s or W.

1.3.1 Electrical quantities


Accelerating electrons in a vacuum or moving electrons through a conductor with some electrical resistance
requires an electromotive force, and so work must be done on a system of ne electrons (note the increasingly
precise usage of our definition of a system). Voltage (V) across an electrical circuit element is defined as the
work (in J) done on a unit of charge (+1 C) and so

J
1V=1 .
C

We will make frequent use of the electron volt (eV) which is a unit of work:

1 eV = 1.6022 × 10−19 J

We also define elementary charge as the electric charge of a signal proton, thus

e or q = 1.6022 × 10−19 C

and is a positive quantity. The Faraday constant F , which is commonly used in electrochemical calculations,
corresponds to the total charge of one mole NA of elementary charges so
F
e= .
NA

Voltage changes through electric circuit elements

In Fig. 1.4, node A has a potential of V volts (V) higher than node B. This corresponds to a voltage drop in the
direction A→B and a voltage rise for B→A. We can now ask if energy is supplied to or supplied by the element
(think resistor versus a battery). Following the definition of [Johnson, Hilburn, Johnson, 1978] (page 8):

“If a positive [conventional] current enters a positive terminal, then an external force must be driving
the current and is thus supplying or delivering energy to the element. The element is absorbing energy
in this case.”
CHAPTER 1. ENERGY - UNITS AND BASIC CONCEPTS 13

I I

A +

voltage drop
+
Figure 1.5: An elementary electrical circuit.
-

B -

Whether this energy accumulates or dissipates will be examined more carefully in the transient energy balance
chapter, but for now if we take one system to be the nh charge carriers, the work w done by/on the charge carriers
is determined by the voltage drop/rise through the device (the other system). As we saw before w > 0 when work
is done on a system. Because nh > 0 (positive conventional current) between A and B in Fig. 1.4 and because
the charge carriers experience a voltage difference −V
w = (−V )nh
implies w < 0 and so work is done by the charge carriers carriers at the following rate
  
dw dnh J C
= power = V = VI = watts.
dt dt C s
Therefore, if in Fig. 1.4, V = −5 V and I = 2 A, p = −10 W is the rate of work done by the charge carriers. By
the conservation of energy (to be discussed in more detail later), the charge carriers may transfer up to p = 10
W to the electrical element.
In Fig. 1.5 a voltage source connected to terminals A and B is depicted. Closing the circuit with the resistor
shown and using Ohm’s law
V = RI
we find
V V
R= : = Ω Ohms.
I A
In this case, the conventional current enters at a higher potential (voltage), and so positive charge carriers pass
from high to low potential implying w = (−V )I and so again, work is done by the charge carriers to the resistor.
The power dissipated by the resistor is
V2 J2 /C2
p = VI = (RI )I = RI 2 = = =W
R J·s/C2
We note that in Fig. 1.5 the battery cathode is where the conventional current flows out of the device, and so it
is the (+) terminal at A. In a device that produces power, the cathode is positive (+); for a device that consumes
power (such as the resistor), the cathode is the (−) terminal. The anode corresponds the the (−) terminal at B.
Finally, the conductance is defined as the inverse of resistance
1 A
G= : = 0 mho
R V
and so I = GV .

1.3.2 A look ahead


We will later encounter the Fermi level or electron chemical potential µ defined as
µ : work (in J) required to add one electron
CHAPTER 1. ENERGY - UNITS AND BASIC CONCEPTS 14

at some specific location within a circuit or circuit element. Therefore, considering Fig. 1.4, the voltage difference
between the terminals A and B are related to the difference in Fermi levels by
µA − µB
VA − VB = − .
e
In this case, we observe that it requires more energy to “insert” an electron into node B relative to node A and
that election current must flow from right to left. A voltmeter connected to nodes A and B effectively measures
the difference in electron chemical potential by its voltage reading.

HW and review problems


1. Given the following power demand schedule for a fixed population measured every four hours over a 24 hour
period
time, hr: 0 4 8 12 16 20 24
demand, MW: 219.5 204.2 255.1 270.5 264.0 276.5 223.0

(a) Compute total energy demand in MW·hr and MJ;


(b) Compute the number of standard cubic feet (SCF) of natural gas required to meet this demand if 20%
of the heat of combustion can be used to produce electricity.

2. Describe how the total energy and peak power dissipated in an earthquake is determined.
3. Create a short list of luminous efficiencies ηL for incandescent, halogen, compact fluorescent, and LED-based
light bulbs available at your local hardware or drug store.
4. Compute to 5 significant figures the number of photons/sec corresponding to 1 lumen (1 lm) of a monochro-
matic light source with wavelength λ = 555.17 nm.
5. What is the mean electrical power production rates of the US and China? Report your results in TW and
cite your sources of information.

References
[CaISO] California ISO http://www.caiso.com/outlook/SystemStatus.html
[Johnson, Hilburn, Johnson, 1978] Johnson, D. E., J. L. Hilburn, and J. R. Johnson, Basic Electric Circuit
Analysis, Prentice-Hall, 1978.
[NISTkelvin] From the NIST website http://physics.nist.gov/cuu/Units/kelvin.html
Chapter 2

The Sun

Our goal in this chapter is to investigate the solar spectrum, how one calculates the intensity of solar irradiance at
the Earth’s surface, both at peak intensity and hourly, how we use NREL (National Renewable Energy Laboratory)
spectral irradiance tables and insolation maps, and how those maps compare to values produced by our integrated
irradiance calculations.

2.1 Blackbody radiation


Recall the relationship between the frequency ν of light to its wavelength λ using the speed of light c in a vacuum
c
λ= .
ν
Also recall that the energy of a photon is defined as

Ep = hν

using Planck’s constant h. Introducing the Boltzmann constant1 kB and recalling that a blackbody is characterized
by its ability to absorb all radiation that falls on it, the spectral irradiance of the blackbody is given by Planck’s
law:
2πhc 2 W
Eeλ (λ, T ) =  with units (2.1)
m3
  
hc
λ5 exp −1
λkB T
which describes the spectral distribution of light emitted by a blackbody at temperature T . The units in the
denominator should be thought of in terms of m2 m, where the first factor m2 corresponds to the surface area
of the blackbody and the second unit of length corresponds to the wavelength of the emitted light. We will
frequently write the units of (2.1) as Wm−2 nm−1 to represent the wavelength in terms of the more commonly
used units of nm.
Before proceeding, we note that for smaller values of temperature T and for values of λ → 0, that the exponential
term in the denominator grows infinitely large and at a rate greater than λ5 , resulting in Eeλ → 0 as λ → 0.
The size of the exponential term, however, can become difficult to compute for small values of λ, and so one can
multiply the numerator and denominator by exp (−hc/λkB T ) to improve the reliability of this computation.
The effective temperature of the Sun’s photosphere is TS = 5777 K; the surface area of the sun’s photosphere
is AS = 6.0786 × 1018 m2 , and the area of a sphere corresponding to the mean radius of Earth’s orbit (one
1 Note how the Boltzmann constant is related to the gas constant Rg by Avogadro’s number NA through Rg = kB NA .

15
CHAPTER 2. THE SUN 16

ASTM G173 AM0


2.0 T = 5777 K blackbody
UVB lower limit
UVA lower limit
1.5 blue
green
red
EE W/(m2 nm)

Figure 2.1: Spectral irradi-


ance of the sun at a distance
1.0 corresponding to Earth’s or-
bit.

0.5

0.0 500 1000 1500 2000 2500


(nm)

astronomical unit, or 1.496 × 1011 m) is AAu = 2.8124 × 1023 m2 . Therefore, the effective solar spectral
irradiance at this location in the solar system is
 
AS 1m W
EE λ (λ, TS ) = Eeλ (λ, TS ) .
AAu 109 nm m2 ·nm
We plot this function in Fig. 2.1 along with the lines indicating the wavelengths of red λ = 700 nm, green λ = 555
nm, and violet λ = 400 nm, the first and last being the approximate limits of the visible spectrum.
The total power per unit area between wavelengths λ0 and λ1 defines irradiance and is computed by
Z λ1
EE (TS , λ0 , λ1 ) = EE λ (λ, TS )dλ
λ0

and so we can compute the ultraviolet, visible, infrared components as well as the total irradiance (be sure to
note the difference between irradiance and the wavelength-dependent spectral irradiance) as

EE (TS , 0, 400) = 167 W/m2


EE (TS , 400, 700) = 500
EE (TS , 700, ∞) = 685
EE (TS , 0, ∞) = 1352

the last of which is a good approximation to the measured solar constant Esc
W
Esc = 1366 , (2.2)
m2
a quantity also equivalent to the direct irradiance at the Earth’s surface if none of the radiation was absorbed
or scattered by the Earth’s atmosphere.
Computing the integral of (2.1) over λ ∈ (0, ∞) gives the Stefan-Boltzmann equation:
Z ∞
W
Eeλ (λ, T )dλ = σT 4 2
0 m
CHAPTER 2. THE SUN 17

with
2π 5 k 4 W
σ= = 5.670400 × 10−8
15c 2 h3 m2 ·K4
which gives the total power per unit area radiated by a blackbody object. We will make use of this relation in
later radiant heat transfer calculations. As a quick reality check, we compute
As W
σ(5777 K)4 = 1365.1 2
AAu m
which is very close to the correct value of the solar constant.
Finally, we observe the peak intensity of the Planck distribution for T = 5777 K occurs at a wavelength λpeak of
value slightly greater than 500 nm (in the green portion of the spectrum). The true peak value can be computed
using Wien’s displacement law
2.8978 × 106 nm K
λpeak = (2.3)
T
with T in K. Therefore, for T = 5777 K we find λpeak = 502 nm.
Of course it can be clearly observed from (2.3) that the wavelength of the peak intensity grows with decreasing
T ; a hot object glowing yellow will shift to orange and then to red as it cools. Using (2.1) we can also show
that the irradiance distribution as a whole also shrinks with decreasing T by computing the ratio of two spectral
irradiance distributions for two different temperatures T0 and T1 :
 
hc
exp −1
Eeλ (λ, T0 ) λkB T1
=   < 1 for T1 > T0 .
Eeλ (λ, T1 ) hc
exp −1
λkB T0

Therefore, the blackbody spectral irradiance curve corresponding to the hotter T1 must lie above that correspond-
ing to T0 for every value of λ, meaning that the cooler body must have a lower irradiance (obtained by integrating
the spectral irradiance over 0 < λ < ∞) and so emit less radiant energy. Likewise, when the spectral irradiance
peak shifts to the right for lower temperature values, the peak will never cross through irradiance distributions
corresponding to higher temperatures.

2.1.1 AM0
The US National Renewable Energy Laboratory (NREL) provides an on-line, experimentally measured reference
solar spectral irradiance data ASTM G-173 [NRELdata]. The data corresponding to the true spectral irradiance
(AM0) at the top of the atmosphere at mean Earth-Sun distance are plotted in Fig. 2.1.

2.1.2 Numerical integration of spectral irradiance


Before moving on to atmospheric absorption effects, we pause to consider the problem of integrating Planck’s
equation (2.1) or the data defining the AM0 or other spectra over a finite range of wavelength λ. For the cases
where the spectra are defined by a vector of data points, a numerical integration approach must be taken; one
may also prefer to use a numerical method to integrate (2.1) because finding a solution explicitly is nontrivial.
We describe the use of the trapezoidal rule for numerical integration in Chapter 12.

2.2 Atmospheric absorption effects on spectral irradiance


The atmosphere reduces the intensity of the solar energy reaching the Earth’s surface in a wavelength-dependent
manner through three major mechanisms
CHAPTER 2. THE SUN 18

AM0
2.0 AM1.5
AM1.5 direct
AM1.5 diffuse
1.5
EE W/(m2 nm)

Figure 2.2: Solar spectral


irradiance corresponding to
1.0 AM0 compared to global,
direct, and diffuse AM1.5
components.

0.5

0.0 500 1000 1500 2000 2500


(nm)

1. Absorption of portions of the spectrum by water vapor, CO2 and other green house gases (mostly in the
infrared region) and by ozone (in the visible and ultraviolet portion of the spectrum);
2. Scattering of (mainly blue) light by dust and ice particles in the atmosphere, giving rise to direct and
diffuse components;
3. Absorption and scattering by clouds.

The first two effects can be measured, and standardized tables corresponding to passage of the sun’s radiant energy
through the equivalent of 1.5 clear atmospheres (AM1.5: the equivalent of 1.5 air masses) also are available from
NREL [NRELdata]. The NREL data corresponding to AM0, AM1.5 global, and AM1.5 direct are plotted in
Fig. 2.2; the graph corresponding to AM1.5 diffuse is computed as the difference between AM1.5 global and
AM1.5 direct. We note that AM1.5 has become a standard method of characterizing spectral irradiance for the
design of photovoltaic systems.

2.2.1 AM
Before proceeding, we must make the concept and value of AM clearer. On page 6 of [Wenham, et al.(2007)],
the air mass (AM) is the Earth’s atmosphere is approximated by
1
AM = (2.4)
cos ζ
where ζ is the zenith angle which is defined as the angle made between a line segment extending between the
Earth and sun and local vertical, e.g., ζ = 0 when the sun is directly overhead and ζ = π/2 when the sun is on
the horizon. More will be said on ζ in Section 2.3.
The approximation (2.4) is justified by the small relative thickness of the Earth’s atmosphere relative to its radius:
RE = 6370 km while the internationally accepted definition of space is an altitude of LA = 100 km. Therefore, in
the limit of small ζ values, the distance L a ray of light must travel through the atmosphere can be calculated by
L cos ζ = LA
and so AM = L/LA = 1/ cos ζ.
CHAPTER 2. THE SUN 19

Therefore, we see that AM = 1.5 corresponds to 48o , a reasonable approximation to the mean value of the zenith
angle through the day, resulting in AM1.5 as the standard by which PV cell performance is measured.

Modeling altitude effects on AM

Atmospheric pressure and density are strong functions of altitude z. To approximate the effect air density has on
AM, we make use of equation (2.29) of [Rekioua and Matagne(2012)] to find
0.89z
AM = for z < 3 km (2.5)
cos ζ
where z is given in km. This correlation is useful for most locations we encounter, but alternative correlation
must be used for higher elevations, such as those encountered by solar-powered aircraft.

2.2.2 Direct versus diffuse radiation


The NREL spectral irradiance data of Fig. 2.2 is split into direct, diffuse, and global components. Before
proceeding, we must make the differences clear:

• The direct irradiance corresponds to light which travels on a straight line from the sun. When the surface
of the collector is normal to the incident direct radiation, this component is called direct normal irradiance.
Distinguishing the direct (normal or otherwise) component from the diffuse radiation is important because
the direct irradiance is the only portion that can be used in concentrating solar systems.
• Diffuse radiation is that component scattered by our atmosphere and is weakly directionally dependent. It,
for example, is the only component illuminating solar panels in the shade. There is no diffuse component
in space (but reflections from planets and other objects can be important).
• The global irradiance simply is the sum of the direct and diffuse contributions.

We can integrate the direct and diffuse spectral irradiance components shown in Fig. 2.2 to find the irradiance
values listed in the center column of Table 2.1.
NREL irradiance, W/m2 Model W/m2
EAM0 1348 ED⊥ (AM = 0) 1366
EAM1,global - EG⊥ (AM = 1) 1107
EAM1.5,global 1000 EG⊥ (AM = 1.5) 1002
EAM1.5,direct 900 ED⊥ (AM = 1.5) 903
EAM1.5,diffuse 100 EF (AM = 1.5) 99

Table 2.1: Integrated values of true irradiance versus our modeled values computed using (2.6) and (2.9).

The integrated NREL data of Table 2.1 shows that 10% of the global irradiance is contributed by the diffuse
component of the total irradiance. We also pause to take note that a good round number to keep in mind is that
at sea level under clear conditions and with a solar panel positioned normal to the Sun’s rays, one can expect a
global solar irradiance of approximately 1 kW/m2 , defining an absolute maximum solar power generation rate.
In the right columns of Table 2.1, the numerical AM values of 0, 1, and 1.5 can be used to approximate the
reduction of the direct normal irradiance ED⊥ (on a surface oriented such that its face is normal to the Sun’s
rays) by atmospheric absorbance with the model described by a slightly modified2 form of the model presented in
[Wenham, et al.(2007)]:
0.678 W
ED⊥ = Esc × 0.73(AM ) (2.6)
m2
2 We changed the model coefficient from the value 0.7 in the cited work to 0.73 to produce a better match with the NREL data.
CHAPTER 2. THE SUN 20

Day: 0 Day: 91

WDC ArcC
TCan ArcC WDC
Eq TCan
TCap ecliptic Eq ecliptic
TCap
AntC
Axis AntC
Axis

Day: 182 Day: 273

ArcC ArcC
WDC
TCan
WDC
TCan ecliptic Eq ecliptic
Eq TCap
TCap
AntC Axis AntC
Axis

Figure 2.3: Earth at winter solstice (top left), spring equinox (top right), summer solstice (bottom left), and fall equinox
(bottom right). In all cases, the Sun is to the left.

where Esc is the solar constant given by (2.2). The global normal irradiance EG⊥ includes the radiation scattered
by the Earth’s atmosphere but ultimately reaches the Earth’s surface through the diffuse radiation mode. As we
saw earlier, 10% of the global normal irradiance corresponds to the diffuse component EF (we assume for now
that EF is independent of solar panel orientation) Therefore, we approximate the diffuse irradiance as

EF = 0.1EG⊥ (2.7)
ED⊥ h 0.678
i
so EG⊥ = ED⊥ + EF = = 1.11 Esc × 0.73(AM ) (2.8)
0.9h i
0.678

and EF = 0.11ED = 0.11 Esc × 0.73(AM ) (2.9)

for a surface aligned normal to the Sun’s rays. Note that we use the parentheses in the exponent of 0.73 to
emphasize that AM is first raised to 0.678 and that 0.73 then is raised to the resulting value.

2.3 The projection effect


The axial tilt of the Earth is 23.44o . During the winter solstice (December 21, 2013), the northern hemisphere is
at its maximum tilt away from the sun, casting the region entirely within the Arctic Circle in 24-hour night, and
that within the Antarctic Circle in daylight (see Fig. 2.3). Given that the Sun’s rays are approximately parallel
with respect to the plane of ecliptic (by virtue of the large Earth-Sun distance), the sunlight falls vertically at solar
noon along the Southern Tropic (the Tropic of Capricorn), and so a flat solar panel would capture the maximum
CHAPTER 2. THE SUN 21

n θtilt
τ
φ
φ z
sun y
ζ latitude
δ
θ=0 sun
x
equator δ λ
ecliptic: z=0

Figure 2.4: Geometry of the Earth relative to the ecliptic in 2D (left) and in 3D (right). The equatorial plane defines the
ring shown at right.

City Latitude, o N Elevation, km


Beijing 39.91 0.04
Guadalajara 20.67 1.56
Hong Kong 22.30 0 to 0.96 Figure 2.5: Latitudes and approximate eleva-
Santiago -33.45 0.52 tions of representative cities.
Sydney -33.89 0 to 0.1
Toronto 43.65 0.07
Washington, DC 38.88 0.05

amount of light at this latitude if the normal to the panel pointed directly upward. Given the 23.44o tilt of our
planet, we can conclude that

1. The Tropic of Cancer and Capricorn lie 23.44o north and south, respectively, of the equator;
2. The Arctic and Antarctic Circles lie 90o − 23.44o north and south, respectively, of the equator.

Declination is defined as the angle between the sun-earth line segment and the equatorial plane. We can
approximate this tilt of the equator at solar noon relative to the ecliptic plane as
 
o 2πtd
δ(td ) = 23.44 cos (2.10)
365

where td ∈ [0, 365] is the date of interest in terms of the number of days after the winter solstice (e.g., td = 11 for
1 January 2014). This results in the winter and summer solstices and spring and fall equinoxes shown in Fig. 2.3.
Latitude φ is defined in terms of degrees north or south of the equator (lines of longitude λ are denoted as E
and W of the prime meridian and will be used later in this chapter). Defining θn as the angle made by a vector
~n normal to the earth’s surface from the earth’s center to the solar panel located at φ, at solar noon one can see
directly from the geometry portrayed in Fig. 2.4 that

ζ =φ+δ (2.11)

where we define ζ as the zenith angle, described previously as the angle between the sun-earth vector and ~n, or
in other words, the position of the sun relative to what appears as directly vertical.
Before concluding this section we make note of the sign convention used for the declination δ and and latitude
φ; in both cases, positive values denote degrees north of the equator, so δ < 0 during the late-spring, summer,
and early fall, and φ > 0 for locations in the northern hemisphere. We denote latitude of locations in the
southern hemisphere in terms of degrees S, or for the sake of our calculations, negative degrees N. A short table
of representative city latitudes is given in Table 2.5.
CHAPTER 2. THE SUN 22

1200 Daily clear/cloudy insolation = 7.64/1.84 kWh/m2 /day


global
direct
1000 cloudy
diffuse
mean global
800

irradiance W/m2
600

400

200

00 5 10 15 20
time

Figure 2.6: 1 degree-of-freedom optimal tilt of a solar panel located in Washington, DC (left). Irradiance profiles
corresponding to the optimal tilt (right).

2.3.1 Optimal panel tilt


Using Fig. 2.4, we see that the optimal tilt (for a solar panel oriented to capture as much of the sun as possible
and so oriented perpendicular to the rays of the sun) from horizontal θtilt,op (td ) is computed by

θtilt,op (td ) = φ + δ(td )

Plotting θtilt,op (td ), we observe that the optimal panel tilt value in mid-winter is θtilt,op = 62.32o and θtilt,op =
15.44o in summer in Washington, DC. We see an immediate use of the winter value as that which would be best
for a remote power application in which a minimal level of energy was required on a daily basis through the winter.
We note that the average of these two maximum values equals θtilt,op,avg = 38.88o , precisely our latitude of
interest. This gives a good general design rule for fixed solar panels.

2.3.2 Integrating projection and absorption effects


Combining equations (2.6), (2.8), and (2.9) we find for a panel flat on the Earth’s surface at solar noon that the
global solar irradiance is
h 0.678
i h 0.678
i
EG = Esc × 0.73(AM ) cos ζ + 0.11 Esc × 0.73(AM )
h 0.678
i
= Esc × 0.73(AM ) (0.11 + cos ζ) .

For a panel with θtilt degrees of tilt to the South Pole, it is not difficult to see that the angle ζtilt between a vector
n~p normal to and pointing out of the panel front surface to the sun-earth line is

ζtilt = ζ − θtilt = φ + δ − θtilt . (2.12)

Think of ζtilt as the zenith angle experienced by a person standing on the solar panel surface, immune to gravity
and standing normal to the panel surface. In this case

1. As before, AM = 0.89z / cos ζ for z < 3 km because the attenuation effect of the Earth’s atmosphere does
not depend on the solar panel tilt angle;
2. EF does not change because of our assumption that it is independent of panel orientation.
CHAPTER 2. THE SUN 23

Therefore, for a panel located at latitude φ, tilted at angle θtilt towards the equator, and at solar noon
h 0.678
i h 0.678
i
EG = Esc × 0.73(AM ) cos ζtilt + 0.11 Esc × 0.73(AM )
h 0.678
i
= Esc × 0.73(AM ) (0.11 + cos ζtilt ) (2.13)

where AM does not depend on ζtilt but on ζ.

2.3.3 3D modeling
For a more accurate method of computing the Sun’s position relative to a point on Earth we must use a fully
3-dimensional model. Based on the geometry shown in Fig. 2.4, we define

φ = latitude, λ = longitude.

If the Earth’s axis of rotation is aligned along the z vector in Fig. 2.4, we can convert from spherical to cartesian
coordinates

xo = RE sin φ cos λ
yo = RE sin φ sin λ
zo = RE cos φ

using the Earth’s radius RE and where we use the rotation matrix below to rotate the x and z points by the
Earth’s axial tilt δ(td ) (2.10) around the y axis (the negative values −δ(td ) are needed for the orientation depicted
in Fig. 2.4 because counterclockwise rotations result for positive angles):
       
x cos −δ(td ) − sin −δ(td ) xo cos δ(td ) sin δ(td ) xo
= =
z sin −δ(td ) cos −δ(td ) zo − sin δ(td ) cos δ(td ) zo
so

x = RE [cos δ(td ) sin φ cos λ + sin δ(td ) cos φ] (2.14)


y = RE sin φ sin λ (2.15)
z = RE [− sin δ(td ) sin φ cos λ + cos δ(td ) cos φ] (2.16)

As seen in Fig. 2.7, we can now plot the lines of latitude on the surface of the globe tilted in the correct manner.
If ~x , ~y , and ~z are the basis vectors in three dimensional space and x, y , and z the magnitudes along these
directions, the vector normal to a point on the surface of the Earth is

x~x + y~y + z~z


~n = .
RE
At the winter solstice (the condition depicted in Fig. 2.4), td = 0 and the vector the Sun’s rays make is ~s = ~x ;
because we use a time dependent model for declination (2.10), the Sun’s rays can be assumed fixed in orientation3

~s = 1~x + 0~y + 0~z .

Because both vectors are normalized, using the vector dot product we can compute

cos ζ = −~s · ~n. (2.17)


3 With the North Pole oriented vertically, the Earth orbits counterclockwise around the Sun; of course it also rotates counterclockwise

in this coordinate system.


CHAPTER 2. THE SUN 24

Figure 2.7: Circles of latitude marked in increments of 10o ; Washington, DC is shown in red.

600 Daily clear/cloudy insolation = 2.92/1.20 kWh/m2 /day 1200 Daily clear/cloudy insolation = 9.03/2.29 kWh/m2 /day
global global
direct direct
500 cloudy 1000 cloudy
diffuse diffuse
mean global mean global
400 800
irradiance W/m2

irradiance W/m2

300 600

200 400

100 200

00 5 10 15 20 00 5 10 15 20
time time

Figure 2.8: Mean and quarter-hourly irradiance values corresponding to Washington, DC, onto a plate tangent to the
Earth’s surface on January 15 (left) and July 15 (right).

Given the irradiance model (2.13) applicable to horizontal surfaces, we find our final model for global irradiance
with respect to a panel tangent to the Earth’s surface becomes

EG (td , φ, λ) = [ED + EF ] H(cos ζ)


h 0.678 0.678
i W
= Esc × 0.73(AM ) cos ζ + 0.11Esc × 0.73(AM ) H(cos ζ) . (2.18)
m2
where AM is calculated with 0.89z / cos θ for z < 3 km, constant with its original definition (2.5); H is the Heaviside
function and is required to prevent negative irradiance values on the night-side of the planet. Representative plots
of the 24-hour irradiance (onto a surface tangent to the Earth’s surface) at the latitude of Washington, DC are
shown in Fig. 2.8 corresponding to the summer and winter solstices. A comparison demonstrating the effect of
panel altitude is shown in Fig. 2.9.
CHAPTER 2. THE SUN 25

1200
EG - sea level
EG - corrected
1000

800
Figure 2.9: Solar irradiance curves correspond-
irradiance W/m2

600
ing to June 1 for a solar panel with zero tilt lo-
cated in Guadalajara, MX. The two curves illus-
trate irradiance calculated at sea level and the
400
correct elevation.

200

00 5 10 15 20
time

Tilted surfaces

For a surface tilted θtilt degrees to the South Pole, the normal vector of the tilted panel is computed using an
additional set of points (xtilt , ytilt , ztilt ) computed using (2.14-2.16) but with RE and φ replaced with

RE ,tilt = 2RE cos θtilt /2


φtilt = φ + θtilt /2

respectively. We compute the projection effect on the direct irradiance component using
1
cos ζtilt = −~s · ~ntilt with ~ntilt = [x − xtilt , y − ytilt , z − ztilt ].
RE
giving the following function for global irradiance that now includes the effect of θtilt :
h 0.678 0.678
i W
EG (td , φ, λ, θtilt ) = H(cos ζtilt )Esc × 0.73(AM ) cos ζtilt + 0.11Esc × 0.73(AM ) H(cos ζ) (2.19)
m2
where the case H(cos ζtilt ) < 0 and H(cos ζ) > 0 corresponds to a panel in daylight but tilted so that it is in the
shade and only illuminated by the diffuse component.

2.4 Insolation
Insolation is solar irradiance integrated over a specific time period, and so has units of energy/(unit area) such
as J/m2 , or more typically in solar energy applications kW hr m−2 day−1 ). Given our irradiance model (2.18) for
EG (td , φ, λ), we can translate points of differing longitude λ into hourly (or finer) intervals:

24λ
t=
360
and so if t = [0, 1, 2, ... , 24] represent the hours in one day and EG (td , φ, t) the irradiance at those times during
the day, the resulting insolation E is computed as
Z 24
E (td , φ) = EG (td , φ, t) dt. (2.20)
0

Numerical values corresponding to mid-January and mid-July are listed in Fig. 2.8.
CHAPTER 2. THE SUN 26

Month Sunny Partly sunny Cloudy


January 8 8 15
February 8 7 13
March 8 9 14
April 8 9 13
May 8 10 13
June 8 11 11
July 9 12 10
August 9 11 11
September 11 9 10
October 12 8 11
November 8 8 14
December 8 7 16
Total 105 109 151

Table 2.2: Monthly cloud cover data for Baltimore, MD.

2.5 Cloud cover


The final missing element of our irradiance model is the role cloud cover plays in attenuating the solar radiation.
Based on data compiled from the US National Oceanic and Atmospheric Administration [NOAAclouds], cloud
cover for Baltimore, MD is given in Table 2.2 where

• a sunny day corresponds to 30% or less of cloud cover, resulting in the full, unobstructed direct and diffuse
components of irradiance. Under these conditions, ED , EF and EG are defined by the values computed
using (2.18) or (2.19).
• cloudy or foggy days where we will assume no direct component of irradiance, only diffuse corresponding
to EG = EF = 0.2ED⊥ , in which ED⊥ is evaluated with (2.6), i.e., the direct irradiance taking AM into
account, but independent of panel tilt. This model is based on the arguments described on page 20 of
[Wenham, et al.(2007)]).
• and partly cloudy days where 40 to 70% of the sunlight is obstructed by clouds and where we assume the
global irradiance is computed as the average of full sun and fully cloudy values.

2.5.1 Insolation data


Consider the NREL solar map that can be generated at the URL

http://rredc.nrel.gov/solar/old_data/nsrdb/1961-1990/redbook/atlas/

for a flat plate collector for an average day in January and July, we find for the state of Maryland the figure for
the average daily insolation values
kW hr kW hr
2 in January, 5.5 in July
m2 day m2 day
We can compare these to our computed figures of
2.92(8 + 4) + 1.20(15 + 4) kW hr 9.03(9 + 6) + 2.29(10 + 6) kW hr
= 1.87 2 in January, = 5.55 2 in July
31 m day 31 m day
demonstrating the excellent predictive capabilities of our irradiance model insolation results depicted in Fig. 2.8
when they are combined with measured cloudiness data listed in Table 2.2.
CHAPTER 2. THE SUN 27

Alaska
Average Daily Solar Radiation Per Month
JANUARY

Hawaii
3.69
3.93
4.03

3.75
Hawaii, Puerto Rico, and
Guam are not shaded.

San Juan, PR Guam, PI

4.28 4.43

Horizontal Flat Plate

Collector Orientation This map shows the general trends in the amount of solar radiation received in the
United States and its territories. It is a spatial interpolation of solar radiation values
derived from the 1961-1990 National Solar Radiation Data Base (NSRDB). The dots kWh/m2/day
on the map represent the 239 sites of the NSRDB.
10 to 14
Flat-plate collector facing south on a Maps of average values are produced by averaging all 30 years of data for each site.
horizontal surface: This map shows Maps of maximum and minimum values are composites of specific months and years 8 to 10
how much solar radiation is received for which each site achieved its maximum or minimum amounts of solar radiation. 7 to 8
by a horizontal surface such as a Though useful for identifying general trends, this map should be used with caution for 6 to 7
solar pond. site-specific resource evaluations because variations in solar radiation not reflected in
the maps can exist, introducing uncertainty into resource estimates. 5 to 6
Maps are not drawn to scale. 4 to 5
3 to 4

*NREL 2 to 3
0 to 2
none
National Renewable Energy Laboratory
Resource Assessment Program FT00A01-313

Figure 2.10: NREL insolation map, January.

HW and review problems


1. Plot the spectral irradiance curves Eeλ for λ ∈ (0, 10000) nm corresponding to a halogen bulb filament
temperature of TH = 3100 K and an incandescent bulb temperature of TI = 2800 K. Then, determine the
percent of total power that is radiated in the visible, IR, and UV portions of the spectrum for each case.
2. The projected area of the Earth is equal to the area of the shadow it casts when the Sun’s rays are assumed
to be parallel. Compute an estimate of the solar irradiance over this area, assuming AM=1.5. Report your
results in TW. Assuming 10% of the irradiance can be converted to electricity, how does that figure compare
to current nominal electrical power production of the US and China?

3. Using the ASTM AM1.5 global spectral irradiance data provided by NREL and splitting the spectrum at
λ = 1127 nm, numerically integrate each portion of the spectrum and report the irradiance for each in
W/m2 .
4. 10 m2 of PV panels operating at 12% efficiency are mounted on a south-facing roof with a pitch of 6 in
12 (i.e., 6 m of vertical rise for 12 m of horizontal distance). Determine the peak potential electrical power
production (in W) for clear and cloudy days during the spring/fall equinoxes and summer/winter solstices
if the house is located in Beijing.
5. Compute the maximum power produced by a 1 m2 solar panel tilted 10o south located at a latitude of 20o
N during the spring equinox and summer solstice on cloudy and clear days. Be sure to show your explicit
calculations for AM and the direct and diffuse irradiances.

6. You have purchased a solar simulator to test solar cells in the lab. The lamp temperature is rated as
CHAPTER 2. THE SUN 28

Alaska
Average Daily Solar Radiation Per Month

JULY

Hawaii
6.00
6.62
6.67

5.20
Hawaii, Puerto Rico, and
Guam are not shaded.

San Juan, PR Guam, PI

6.09 5.09

Horizontal Flat Plate

Collector Orientation This map shows the general trends in the amount of solar radiation received in the
United States and its territories. It is a spatial interpolation of solar radiation values
derived from the 1961-1990 National Solar Radiation Data Base (NSRDB). The dots kWh/m2/day
on the map represent the 239 sites of the NSRDB.
10 to 14
Flat-plate collector facing south on a Maps of average values are produced by averaging all 30 years of data for each site.
horizontal surface: This map shows Maps of maximum and minimum values are composites of specific months and years 8 to 10
how much solar radiation is received for which each site achieved its maximum or minimum amounts of solar radiation. 7 to 8
by a horizontal surface such as a Though useful for identifying general trends, this map should be used with caution for 6 to 7
solar pond. site-specific resource evaluations because variations in solar radiation not reflected in
the maps can exist, introducing uncertainty into resource estimates. 5 to 6
Maps are not drawn to scale. 4 to 5
3 to 4

*NREL 2 to 3
0 to 2
none
National Renewable Energy Laboratory
Resource Assessment Program FT00A07-319

Figure 2.11: NREL insolation map, July.

T = 5000 K. Compute the fraction of the spectral irradiance in the visible portion of the spectrum (400 ≤
λ ≤ 700 nm).
7. For the system described in the previous problem, compute the number of photons within the visible portion
of the spectrum out of a total of 100 photons distributed over the Planck distribution.
8. Describe in two to three sentences why the factor AS /AAU is used in conjunction with the Planck distribution
to approximate EE λ at AM0 conditions.
9. As we will see in Chapter 11, plants absorb light in the wavelength ranges of 550 < λ < 700 nm and
λ < 480 nm. If eight photons are required to produce one molecule of O2 , estimate the molar production
rate of O2 by 1 m2 of leaf surface area under AM1.5 conditions. Report your result in mol/hr.
10. Compute a plot of insolation (in units kW hr m−2 day−1 ) vs. latitude valid on the summer solstice for (a) a
panel mounted flat to the ground, and (b) a panel mounted with tilt optimized for solar noon at the panel’s
particular latitude. Describe the differences you find between the two cases.

References
[NRELdata] NREL spectral irradiance data http://rredc.nrel.gov/solar/spectra/am1.5/

[Rekioua and Matagne(2012)] Rekioua, D. and E. Matagne, Optimization of Photovoltaic Power Systems,
Springer 2012.
CHAPTER 2. THE SUN 29

[NOAAclouds] US National Oceanic and Atmospheric Administration,


http://www.currentresults.com/Weather/Maryland/sunshine-by-month.php
[Wenham, et al.(2007)] Wenham, S. R., M. A. Green, M. E. Watt, and R. Corkish, Applied Photovoltaics, 2nd
Ed., Earthscan, UK, 2007.
Chapter 3

Transient energy balances

Recall our previous thermodynamically precise definitions of a system, its surroundings, and the differences between
open and closed systems. At this point we would like to develop the energy balance modeling concepts needed to
understand the interconnection of solar and conventional energy production systems, the consumers of electrical
power, and various means of storing excess energy.

load
(lamp)

power
controller

PV panel storage
(battery)

Figure 3.1: A high-level view of a representative solar energy balance for an integrated energy production and storage
system.

A simple, but nontrivial, example of these concepts is illustrated in Fig. 3.1 where a stand alone PV-powered
street lamp is shown. What makes this system nontrivial is the dynamic interconnection of the components and
the solar source itself. The basic characteristics of this system are listed in Table 3.1 and will be used in a number
of the following example calculations. For example, given the definition of luminous efficiency ηL in Section 1.2,
we can immediate compute the luminous output of the streetlamp as
lm
683 (30 W)(0.148) = 3030 lm
W
which demonstrates the considerable advantage (approximately 10×) of solid-state over incandescent lighting (see
Fig. 1.3).

30
CHAPTER 3. TRANSIENT ENERGY BALANCES 31

Component Specifications
PV panel poly-Si cells; η = 0.15 efficiency; A = 1 m2 total panel PV cell area; θtilt = 50o S
Battery voltage: 12 V; capacity: 120 Ah
Lamp assembly LED lamps with ηL = 0.148; total power consumption: 30 W

Table 3.1: Components and their specifications for the commercial PV street lamp.

3.1 Energy conservation


The first law of thermodynamics is based on the law of energy conservation (total energy is constant); it allows
us to write energy balances such as

∆(energy of system A) + ∆(energy of system B) = 0


∆(system energy) + ∆(energy of surroundings) = 0

At this point we provide the following definitions

• A system is at steady state when it no longer changes with time; in many cases it can be determined by
t → ∞ or by setting the time-derivative terms d/dt to zero;
• Equilibrium is the condition found as t → ∞; we note that equilibrium can be a steady or non-steady state.
• Instantaneous conditions correspond to a ”snapshot” in time;

• Transient states depend on time and can have derivatives or integrals with respect to time; these systems
sometimes are divided into forced and autonomous systems;
• Spatially distributed states are a function of position, generally have derivatives or integrals with respect
to position, and can be transient in nature.

3.1.1 A simple example


Consider a highly simplified situation where System A consists of the PV panel described in Table 3.1 and located
at latitude φ = 38.88o N. We wish to determine the power produced at solar noon in mid-winter by the PV panel
and then assess the relative rates of power consumption by the lamp and battery charging operation at that
instant of time.
Using the models and calculation procedures developed in the previous chapter, on td = 0 the declination δ and
zenith angles ζ and ζtilt are

δ = 23.44o
ζ = 62.32o = 1.0877 radians
ζtilt = 12.32o = 0.2150 radians

are computed using (2.10), (2.11), and (2.12), respectively. The atmosphere equivalents AM is evaluated using
ζ and (2.4) gives
1
AM = = 2.15
cos(1.0877)
giving a peak global irradiance computed using (2.13) as
h 0.678
i
EG = Esc × 0.73(AM ) (0.11 + cos ζtilt ) = 875 W/m2
CHAPTER 3. TRANSIENT ENERGY BALANCES 32

which, of course, gives the total power of the solar flux incident to the panel surface as PI = 875 W. Therefore,
with an efficiency of 15%, the peak power produced by this panel is PA = −131 W.
System B consists of the combination of the streetlamp (the load) and the battery storage system. If the lamp is
on at solar noon, we can write our instantaneous power balance as

power produced by A + power consumed by B = 0


−131 W + [30 W consumed by the lamp + PB,storage ] = 0

implies that PB,storage = 101 W to the battery storage system if the lamp is operating during the peak power
produced during the day.

3.1.2 Over a single sunny day

50o tilt, winter solstice


900 hour global irradiance (W/m2 )
800 7 0.0
700 8 208.5
600
9 508.3
10 710.9
500
EG W/m2

11 833.4
400 12 874.6
300 13 833.4
14 710.9
200
15 508.3
100 16 208.5
00 5 10 15 20 17 0.0
time, hrs

Figure 3.2: Global effective irradiance for a PV panel tilted at 50o over the day 21 December 2014 (left); daytime global
irradiance values EG (t) (right).

We use of Earth irradiance model (2.18) on the date td = 0 to compute the hourly global irradiance EG (t) for
the tilted PV panel described above; a plot of the data and the hourly values themselves are shown and tabulated
in Fig. 3.2. Using the trapezoidal quadrature rule described in Section 12.2 the total energy that can be collected
over the day is the daily insolation E , defined by (2.20) and can be simply approximated by the trapezoidal rule:

E = 208.5 + 508.3 + 710.9 + · · · + 508.3 + 208.5 = 5.397 kWh/(m2 day).

Accounting for the panel efficiency, the total energy EA generated over the day by the panel

EA = −0.15(5.397 kWh·m−2 day−1 )(1 m2 )(1 day) = −810 W·h = −2.91 MJ

The streetlamp will consume EB =(24 h)(3600 s/h)(30 J/s) = 2.59 MJ of energy/day if continuously lit. There-
fore, we can power the lamp continuously with our 1 m2 tilted panel during the winter solstice with a bit of extra
energy left at the end of the day if we store the excess energy produced during the day to power the lamp at
night.

3.1.3 Battery storage


We would like to store excess energy produced during daylight portion of the day to keep our lamp lit 24 hours. In
this scenario, our objective is to determine what electrical storage capacity is needed to power the lights through
CHAPTER 3. TRANSIENT ENERGY BALANCES 33

the night. Additionally, we would like plot the battery charge level over the course of a day to determine the
effectiveness with which the battery storage is used.
To begin this analysis we define C as the level of electrical energy charge in the battery system (in J), and a new
system is created to account for the charge stored in the battery. We note that the true capacity of the battery
system Ccap is found from Table 3.1 by integrating 120 A × 12 V over one hour:

Ccap = (120 Ah)(12 V)(3600 s/hr) = 5.2 MJ.

The rate of change of charge of charge in this new battery system is determined by the energy balance
dC
= power received from the panel − power to the street lamp
dt
subject to the initial state of the battery charge C (t = 0) = C0 . For this example, we consider the initial state to
be the battery charge at midnight (t = 0) of the day (td ) we are modeling. Under these assumptions,

dC
= 0.15EG (t) − 30 W lamp
dt
Z t
or after integrating C (t) = C0 + 0.15 EG (t 0 )dt 0 − (30 W)t.
0

We know 0 < C0 < Ccap and so we consider the case where C0 = 1 MJ. Using an explicit Euler integration
numerical method (see Section 12.4) we find the results plotted in Fig. 3.3. Note how the battery is charged
when solar energy production rates climb above the power consumption of the bulbs. To implement the Euler
method, we first rewrite the battery charge model as

dC Ci+1 − Ci EG (ti+1 ) + EG (ti )


≈ = 0.15 − 30 W lamp
dt ∆t 2
EG (ti+1 ) + EG (ti )
so Ci+1 = Ci + 0.15∆t − ∆t (30 W)
2
with i = 0, 1, 2, ... and C0 corresponding to the initial condition and with ∆t = 1 h. We note how the mean
value (over each time interval ∆t ) of the panel power EG (T ) is used in the integration procedure.

Discussion of our findings so far

The minimum battery capacity of 1.94 MJ that we computed is nearly three times the actual capacity of this
system. Of course it is not unreasonable to over-specify the storage system as this will allow for the build up of
sufficient charge to compensate for a string of cloudy days. Other issues to consider before moving on to other
examples of modeling energy system dynamics is how quickly this simple system can turn into a complex dynamic
optimization problem. For example, one can size the components and design the power system controller to adjust
the duration of the periods during which the street lamp is lit with input from the state of the battery charge and
the power generation rate of the panel.

3.2 Material and energy balance review


While less frequent in PV applications, there are situations where solar thermal and other energy conversion
processes must be described as open systems, where the time-rate of change of the system’s mass is coupled to
the energy balance. If
moles kg
m = moles or kg, ṁ = or
s s
CHAPTER 3. TRANSIENT ENERGY BALANCES 34

0
20
40
Panel power, W

60
80
100
120
1400 5 10 15 20 Figure 3.3: State of battery
2.5 charge over the course of the
day.
2.0
Battery charge, MJ

1.5
1.0 min battery capacity
0.5
0.00 5 10 15 20
time, hrs

and accumulation rate = rate in - rate out


dm
= ṁin − ṁout = net influx
dt

The molar and total internal energies U and U t , respectively, of a system are the energy associated with
the kinetic, vibrational, and chemical bond energy of the molecules making up the system. It is important to
distinguish internal energy from the kinetic and potential energy associated with the entirety of the system itself.
We note that for an ideal gas the internal energy is only a function of the gas temperature.
The general energy balance for a system consisting of some fixed volume Vst with a boundary into which a stream
flows from the surroundings with properties

Vin , Pin , Uin , uin , zin , and, ṁin

and an outlet stream with properties

Vout , Pout , Uout , uout , zout , and, ṁout

results in the power balance equation


dU t
= inlet stream energy − outlet stream energy
dt
+net work done by flows through boundary + Q̇ t + Ẇst

where Q̇ t and Ẇst denote total heat input and shaft work/unit time.

3.2.1 Net work by inlet/outlet flows


Consider an inlet pipe of cross sectional area A; a volume of fluid V t (our system for this example) passing though
this pipe will form a cylinder of length ∆x = x1 − x0 = V t /A and so the total work (force × distance) done to
the fluid (system) is
W t = PA∆x = PV t
CHAPTER 3. TRANSIENT ENERGY BALANCES 35

Returning to our control volume as the system, we see that because this fluid can be thought to enter the control
volume, the W t above is the same as the work done on the control volume system. This means

net work rate done by flows through boundary = [ṁin Pin Vin ] − [ṁout Pout Vout ].

Defining

Stream energy = internal, kinetic, potential


= U + u 2 /2 + zg

results in
dmU
= ṁin U + u 2 /2 + zg in − ṁout U + u 2 /2 + zg out + [ṁPV ]in − [ṁPV ]out + Q̇ t + Ẇst
   
dt
Because specific enthalpy H is defined by
H = U + PV
dmU
= ṁin H + u 2 /2 + zg in − ṁout H + u 2 /2 + zg out + Q̇ t + Ẇst
   
(3.1)
dt
subject to initial condition U(t = 0) = U o and, potentially, a time-dependent total mass m.

3.3 Solar water heater - closed system


Let us now develop a highly simplified thermal system model to assess the feasibility of developing inexpensive
solar water heaters. The specifications call for the heater to boil 8 fluid ounces of water in under 2 minutes at a
latitude of 19o N at solar noon on a cloudless day.
The primary component of this solar heating system is a reflector used to concentrate the sun’s rays; we assume
the reflector can be pointed towards the sun. It is important to realize that only the direct component of irradiance
can be focused on the container of water to be heated. Therefore, we compute the minimum value of the direct
irradiance over the year as
0.678 W
min ED (td ) = min Esc × 0.73(AM ) = 928 2
td td m
with AM = 1/ cos ζ and ζ = 19o + 23.44o cos(2πtd /365). The minimum ED value occurs on day td = 0.
We consider the system to be the water to be boiled; the container forms its boundary and we assume it absorbs
no solar energy. Given the mass of water m = 0.24 kg (8 oz at room temperature), the energy balance equation
reduces to
dU
m = Q̇ t
dt
We recall the definitions of heat capacity
   
∂U ∂H
Cv ≡ and Cp ≡
∂T V ∂T P

Because the water is essentially incompressible Cp ≈ Cv ; we further assume the heat capacity is not a function of
temperature and take Cp = Cv = 4.2 kJ/kg/K. These definitions lead to

dT
mCp = A(928 W/m2 )
dt
A(928 W/m2 )
T = T (0) + t
mCp
CHAPTER 3. TRANSIENT ENERGY BALANCES 36

so given the initial condition T = 20o C, we can compute the reflector projected1 area as

(100 − 20 K)(0.24 kg)(4.2 kJ/kg/K)


A= = 0.724 m2
0.928 kW/m2 (120 s)

or a circular reflector of radius r = 0.48 m, a reasonable size. Of course, the system should be over-designed to
account for reflector nonidealities.
Question: What elements are missing from this model? Answer: A partial list includes

1. Heat loss to the environment, including reflection from the water container walls;
2. Energy absorption/imperfect focusing of the reflector;

3. Temperature dependent heat capacity.

Adding these model features would improve the model accuracy, but also would make it nonlinear, and so most
likely require an iterative numerical solution procedures (e.g., Newton’s method).

3.4 Solar water heater - open system

EG

.
m, Hout

.
m, Hin

Figure 3.4: A rooftop solar thermosiphon water heating system (left) and a diagram illustrating its operation (right).

Consider an energy balance on the thermosiphon water heater shown in Fig. 3.4. The system in this case is the
water in the tank and solar collector, and the inflow/outflow pipes are the boundary for mass flow. Assuming the
total mass m in the tank is constant, ṁin = ṁout = ṁ.
The operating principle of the system is simple: because water at 20o C expands as a function of temperature
according to  
−4 3 3 1 ∂V
α = 2.07 × 10 m /m /K =
V ∂T P
the heated water rises in the solar collector into the hot-water reservoir (note that α, the thermal expansion
coefficient, becomes negative between 0 and 3o C). To model the water temperature dynamics, we first observe
1 We define the projected area as the area of a shadow cast by the reflector onto a surface normal to the sun’s rays; this area will

be smaller than the true area of the reflector.


CHAPTER 3. TRANSIENT ENERGY BALANCES 37

.
Qt
.
m, Hin, Tin
T, m
Figure 3.5: The lumped approximation of the thermosiphon hot water
heater system.
.
m, Hout, Tout = T

that no work is done on the tank and so Ẇst = 0; we also can neglect changes in fluid kinetic and potential
energy. These simplifications result in
dU
m = ṁHin − ṁHout + Q̇ t = ṁ[Hin − Hout ] + Q̇ t
dt
The inlet, outlet, and tank temperatures are Tin , Tout , and T , respectively. If we consider T to represent the
mean water temperature, we can approximate Tout = T . This effectively reduces the problem to the lumped
formulation illustrated in Fig. 3.5.
Given all the assumptions of this example, we can now write dU = Cv dT and
Z T
Hout − Hin = Cp dT = Cp (T − Tin )
Tin

resulting in the linear ordinary differential equation


dT
mCv = ṁCp (Tin − T ) + Q̇ t
dt
subject to initial condition T (t = 0) = T o . Let us now consider solving the system for Tin = 10o C, m = 200
kg, Cv = Cp = 4.2 kJ/kg/K, and

Q̇ t = EG A = (900 W/m2 )(4 m2 ) = 3600 W

3.4.1 Computing the warm-up period


We now examine the question: starting at T o = Tin and with no hot water mass flow to the shower, how long
does it take for T to reach 50o C? To find the solution, we first rewrite the modeling equation as

dT ṁ Q̇ t
= (Tin − T ) +
dt m mCp

For the case ṁ = 0,

dT Q̇ t
=
dt mCp
mCp
t = (T − T o )
Q̇ t
(200 kg)(4200 J/kg/K)
= (40 K)
3600 W
= 9333 s = 2 hr 35 min

After reaching 50o C, how long does it take to reach 95% of its steady state value with ṁ = 10 kg/min?
CHAPTER 3. TRANSIENT ENERGY BALANCES 38

At steady state
ṁ Q̇ t
0 = (Tin − Tstst ) +
m mCp
Q̇ t
Tstst = Tin +
ṁCp
(3600 W)(60 s/min)
= 10o C +
(10 kg/min)(4200 J/kg/K)
= 15.1o C

For the full, dynamic solution, we use the transformation


ṁ Q̇ t
S(t) = (Tin − T ) +
m mCp
using which we can find
S(t) = S(0)e −ṁt/m
and after a bit of algebra
T = Tstst − [Tstst − T o ] e −ṁt/m .
We wish to find
To − T
0.95 = .
T o − Tstst
Solving for T in the expression above and using it the solution gives

e −ṁt/m = 0.05
m
t = − ln(0.05)

= 60 min
or 14 min for the system to reach 50% of equilibrium.

3.5 Pumped storage


Consider the large hydroelectric pumped-storage facility located just south of Ludington, Michigan shown in
Fig. 3.6. The reservoir volume is approximately 1 × 108 m3 = 1011 kg; it is located 110 m above the surface
of Lake Michigan and is filled by six reversible turbines, each of which can generate 312 MW of electricity. The
reservoir is filled at night using excess capacity from regional coal-fired power plants.
Let us consider the scenario where the reservoir is full to its maximum level at t = 0; how long will it take to
empty it? To solve this problem, we define the system to be the water contained in the pipes connecting the
reservoir to the turbines plus the water inside the turbines. Because of the incompressibility of water, we argue
that
dmU
=0
dt
and assume the system is designed to minimize losses due to the kinetic energy of the water re-entering Lake
Michigan. Under these conditions, the energy balance equation reduces to
0 = ṁin gzin − ṁout gzout + Ẇst
0 = ṁg (zin − zout ) + Ẇst
0 = ṁ(9.8 m/s2 )(110 m) + 6(−312 MW)
6(312 × 106 J/s)
ṁ = = 1.74 × 106 kg/s
(9.8 m/s2 )(110 m)
CHAPTER 3. TRANSIENT ENERGY BALANCES 39

Figure 3.6: Aerial view of the


Ludington Pumped Storage fa-
cility.

Given the volume of the reservoir corresponds to 1011 kg of water, we see that it would be emptied in 16 h, not
unreasonable for this application.

3.6 Sodium-sulfur batteries


Sodium sulfur batteries are relatively simple, efficient, and are manufactured from earth-abundant materials making
then suitable for large-scale energy storage. Given that the sulfur and sodium must be kept in their molten states
(see the physical properties in Table 3.2), the batteries operate at approximately 300o ; however, the heat released
during the charge/discharge cycles is sufficient to maintain this temperature.

Phase Density g/cm3 Atomic mass g/mol Molar density mol/m3 Melting pt o C
Na 0.927 22.990 40322 98
S 1.82 32.066 56758 115

Table 3.2: NaS battery elemental component properties.

The batteries generally have a cylindrical geometry; as seen in Fig. 3.7 a key feature is the solid alumina electrolyte
that separates the two molten phases and conducts the Na+ ions. During discharge, the following overall reaction
takes place
1
Na + xS → Na2 S2x Ecell = 2 V
2
and so one electron travels through the external circuit for every Na atom involved in the reaction above. The
question we would like to address in this section is

How much S and Na are needed to match the Ludington pumped storage facility, a facility with an
energy storage capacity of 30 GWh?
CHAPTER 3. TRANSIENT ENERGY BALANCES 40

anode

molten Na

e-
solid
alumina Na+ load Figure 3.7: Simplified schematic of a sodium-sulfur battery
during the discharge phase.
electrolyte

S and Na2Sx
I

cathode

To start, we convert the capacity to J:

30 GWh(3600 s/h)(109 ) = 1.08 × 1014 J.

Recalling that voltage V corresponds to J/C and with Vcell = 2 V, and the Faraday constant F = 9.64853 × 104
C mol−1 , the number of moles of Na required can be found as

1.08 × 1014 J 1
nNa = = 5.597 × 108 mol Na
2V F
t
Using the densities of Table 3.2 and taking x = 2, we find a minimum total battery volume Vbatt of the Na and
S phases as
t
Vbatt = 33601 m3 or a cube with side length 32.2 m
which while large, is significantly smaller than the pumped-water facility.

HW and review problems


1. On page 10 of [Wenham, et al. (2007)] it is stated that water vapor in our atmosphere absorbs infrared light
strongly in the wavelengths between 4000 and 7000 nm and CO2 between 13000 and 19000 nm. Assuming
the Earth absorbs and emits radiation as a blackbody and that its rotation results in a uniform surface
temperature, develop a planetary energy balance and calculate the mean Earth surface temperature (in K)
for the following three cases: 1) zero atmospheric CO2 , and 2) a sufficiently large amount of H2 O such that
all radiation between 4000 and 7000 nm is absorbed by the atmosphere and half is re-emitted towards the
Earth, and 3) the same assumption regarding water as in case 2 plus a sufficiently large amount of CO2
such that all radiation between 13000 and 19000 nm is absorbed by the atmosphere and half is re-emitted
towards the Earth.

2. Consider the stand-alone streetlamp system described in the beginning of this chapter. Repeat the calcula-
tions, but use a three-day basis in which the first day is perfectly clear while the second is cloudy and the
third is perfectly clear again. What conclusions can we reach regarding the battery storage capabilities of
this system?
CHAPTER 3. TRANSIENT ENERGY BALANCES 41

3. The state of Maryland has set a goal of generating 20% of its electricity using in-state renewable resources
by 2022 [MDrenewables]. Write a short summary of the stated objectives taking care to note the solar
component of the renewable energy portfolio. Estimate the magnitude of the solar energy infrastructure
necessary to achieve this goal.

4. The enthalpy of combustion of methane is given as −802.36 kJ/mol CH4 at 25o C and 1 atm. Compute
the rate of methane consumed by a power plant for each 100 W of electrical power produced. Assume the
plant is 35% efficient and a loss of 20% in transmission of electrical power to the end user.
5. A single photon of wavelength λ = 550 nm is absorbed by 1 gm of water. What is the temperature rise
assuming CP = 4.2 J/(g·K) and the all of the photon energy is dissipated as heat within the water?

6. I estimate the lights in my campus office consume a total of 45 W of power. How much energy is saved
when I turn off the lights for 1 hour assuming
(a) it is a hot summer day with an outside temperature of 35 C, an indoor temperature of 25 C
(b) the minimum work required by an air conditioning (AC) system is

TH − TC
AC work = [heat to be removed]
TC

(c) power plant efficiency = 30%


(d) we assume 25% loss in the power line transmission between the power plant and my office
Report your solution in Joules.

7. The potential for a large-scale PV installation in the lower peninsula of Michigan is to be examined. The PV
panels are 17% efficient and will be mounted with a fixed tilt corresponding to their latitude. The Ludington
pumped-storage facility will be used so that the PV power can be supplied at a roughly continuous rate
over the 24hr day. What is the maximum area (in m2 ) of solar panels that can be used in this situation so
that the limits of the storage facility are not exceeded?

References
[Geankopolis (1978)] Geankopolis, C. J. Transport Processes and Unit Operations, Allyn and Bacon, Inc., 1978.

[MDrenewables] http://www.statestat.maryland.gov/GDUenergy.asp
[Smith, Van Ness, Abbott (2005)] Smith, J. M., H. C. Van Ness, and M. M. Abbott Introduction to Chemical
Engineering Thermodynamics, 7th Ed., McGraw-Hill’s Chemical Engineering Series, 2005.
[Wenham, et al. (2007)] Wenham, S. R., M. A. Green, M. E. Watt, and R. Corkish, Applied Photovoltaics, 2nd
Ed., Earthscan, UK, 2007.
Chapter 4

The physics of PV devices

4.1 Band structure of solid materials


It is instructive to consider some of the elements that make up solar cells we plan to study. The most common solar
cells are primarily composed of silicon (S) that is commonly doped with low levels of boron (B) and phosphorous
(P) impurities to alter the electrical performance of the Si material. Aluminum (Al) and silver (Ag) also are used
to create the cell electrical contacts. More exotic materials are found in thin-film cells including cadmium (Cd),
tellurium (Te), indium (In), gallium (Ga), selenium (Se), arsenic (As), as well as the common copper (Cu) and
nitrogen (N).
While some of these elements are toxic and rare and those implications will be considered later, let us first focus
on where they are found within the periodic table of elements (Fig. 4.1.)

IB IIB IIIA IVA VA VIA


B C N O
Al Si P S
Cu Zn Ga Ge As Se
Ag Cd In Sn Sb Te

Figure 4.1: A portion of the periodic table of elements with semiconductors shown shaded in gray; metals are to the left
of the semiconductor range and non-metals to the right.

What is interesting is that virtually all these materials are clustered in the region of the periodic table containing
semiconductors, with nearby metals and electrically insulating materials rounding out the list. Therefore, to
understand what takes place electronically within a solar cell, we begin with a closer look at band structure of
solid materials.
Recall that electrons can only inhabit discrete energy levels; these may be filled, partially full, or empty (Fig. 4.2).
As we bring atoms together, the valence electrons interact to form bonds and the energies associated with these
bonding orbitals take on a discrete structure as a result of the Pauli exclusion principle, which dictates no two
electrons can have identical sets of quantum numbers1 As atoms bond to form a crystal, the number of molecular
orbitals that can be created grows, forming a band structure when a large number of atoms (e.g., 1023 ) make up
the crystal.
1 For electrons we have n the electron shell (energy level), l the subshell (the s, p, d, etc. orbital), m the orbital (e.g., which of
l
the 3 p orbitals), and ms the spin (always one of two states).

42
CHAPTER 4. THE PHYSICS OF PV DEVICES 43

These bands are characterized by band edges and whether they are filled or empty. The regions between are
the forbidden gaps. In cases where the topmost band containing electrons is completely filled, the material is
nonconductive unless an electron is promoted to a higher-energy conduction band, leaving a hole in the valence
band. Electric current can flow as a result of either, or both, charge carriers. Charge cannot be transferred within
a filled band - there simply are no unoccupied energy levels within the band for charges to go to. However, when
bands are only partially filled, as in the case of metals, even a small applied electric field is sufficient to raise an
electron to a nearby unoccupied state, resulting in an electrical current.
We note that in metals, semiconductors, and insulators, the distinction between the last two is somewhat artificial
- the commonly accepted limit for semiconductor band gap is 4 eV.
Finally, we note that compound semiconductors can be formed by combination of elements, such as the III-V
compounds AlN, GaN, and InP.
electron energy

band edges

1 atom n atoms conductor insulator

Figure 4.2: Energy levels of an individual atom (left), finite number of discrete states for a small number n of atoms
(center left) band structure of a conductor (center right) and of an insulator (right).

4.1.1 Fermi level


We begin our quantitative discussion of that band structure and energy levels of a semiconductor by recalling the
Maxwell-Boltzmann distribution which defines the probability f of finding a particle with energy greater than E
1
f (E ) =
Ae E /kB T
where A is a normalization constant. The energy level for which the probability of finding an electron is 1/2 is
called the Fermi level. This also corresponds to the maximal electron energy level at T = 0 K. We see that
the Fermi level defines the median (not mean) value for electron energy; the specific form of the Fermi-Dirac
distribution function is
1
f (E ) =
1 + exp ((E − Ef )/kB T )
and so it is easy to see that when E = Ef , F (E ) = 1/2. The Fermi level also is equivalent to electron chemical
potential. Recall that

∆G = ∆H − T ∆S
 
∂(nG )
µ=
∂n P,T

and so because spontaneous processes require a reduction in ∆G , variations in Fermi level can drive electrons
from energetically favorable states to those that are less so.
CHAPTER 4. THE PHYSICS OF PV DEVICES 44

1.0
300 K
500 K
0.8

Figure 4.3: Fermi-Dirac distribution function


0.6 values about the Fermi energy (marked in red)
f(E)

between the valence and conduction bands edges


0.4 at 0 and 1.1 eV, respectively at ambient (300 K,
blue solid) and an elevated temperature (500 K,
green dashed).
0.2

0.00.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2


E, (eV)

For silicon, the band gap is 1.1 eV; we can plot the function to observe how electrons are promoted to the
conduction band at higher temperatures in a semiconductor (Fig. 4.1.1).
For a semiconductor with Ef , Ec , and Ev marking the Fermi level, conducting band bottom, and valence band
top edges, respectively, and with me∗ and mh∗ as the effective electron and hole masses, we can compute the
number density of the electrons ne with energies above Ec and holes nh with energies below Ev in an intrinsic
semiconductor by
3/2
2πme∗ kB T
  
Ef − Ec
ne = 2 exp
h2 kB T
3/2
2πmh∗ kB T
   
Ev − Ef
nh = 2 exp .
h2 kB T
Setting ne = nh in an uncharged semiconductor, we can then compute
Ec + Ev 3kB T Ec + Ev
Ef = + ln (mh∗ /me∗ ) ≈ = Ei
2 4 2
The intrinsic carrier concentration ni then is defined by the geometrical mean ni2 = ne nh . For silicon, ni =
1.45 × 1010 charges/cm3 at 300 K [Middleman and Hochberg (1993)]. We observe that the intrinsic carrier
concentration increases with temperature, resulting in electrical conductivity for semiconductors that increases
with temperature:  
3/2 VG
ni = no T exp −
2VT
where VG is the band gap and VT = kB T /q the thermal voltage.
Before we move on from this discussion, we list the following Fermi-level facts:

• The Fermi level corresponds to the thermodynamic work necessary to add an electron to a system; it thus
defines the electron chemical potential;
• As seen in the calculation above, the Fermi level for intrinsic semiconductors is approximately mid-way
through the forbidden gap;
• The previous fact does not imply there are electrons within the gap because there are no available states
within the forbidden gap of a defect-free intrinsic semiconductor;
• The “spreading” of the Fermi distribution explains why the conductivity of a semiconductor increases with
temperature;
• The Fermi level for a metal is found where the conduction and valence bands overlap.
Si Si Si Si Si
Si Si Si Si Si
CHAPTER 4. THE PHYSICS OF PV DEVICES 45

4.1.2 Doping
Si Si Si Si Si Si Si Si Si Si
The band structure of intrinsic semiconductors can be modified by doping, a means of introducing new energy
Si Si Si Si Si Si Si P Si Si
levels into the forbidden gap. In the discussion that follows, we will focus on substitutional doping, as opposed to
Si toSicases
interstitial doping and will also limit our analysis Siwhere
Si the
Sidopant concentration
Si Si Si level Si is Si
relatively small
(typically true) so that the dopant atoms do not interact.

Si Si Si Si Si Si Si Si Si Si Si Si Si Si Si
Si Si Si Si Si Si Si P Si Si Si Si B Si Si
Si Si Si Si Si Si Si Si Si Si Si Si Si Si Si

Figure 4.4: Lewis structure of crystalline Si (left); n-type Si resulting from phosphorous doping (center), and p-type from
Si Si
boron doping (right). Si Si Si Si Si Si Si Si
The location Si SinewPstates
of the Si relative
Si to theSi band
Si edges
B of Si theSiforbidden gap depends on the nature of the
dopant atoms. Consider, for example, the Si crystal shown in Fig. 4.4. Si is a column IV element meaning it
Si electrons.
has four valence Si Si PSi Si
(phosphorous) Si Si SiV element;
is a column Si Si four of its valence electrons can bond with
the Si neighbors, leaving an unbound electron (while still remaining neutral). This single electron inhabits an
energy level close to the bottom edge of the conduction band (see Fig. 4.5 and the ionization value tabulated
in Fig. 4.1.2), requiring little extra energy to excite the extra electron of the new state to the conduction band.
Si Si Si Si Si
Therefore, addition of dopant atoms results in an increase in electrical conductivity of the material, in this case an
n-type semiconductor
Si Si where B Si electrons
Si are the majority carriers. In this context, the P atoms act an an electron
donor; after ionization, the positively charged P remains fixed within the Si crystal while the e− charge is free
Si way,
to move. In this Si weSican Si Si why electrons are the majority carrier and are found in the conduction
understand
band of n-type semiconductors.
The number density of P atoms within the Si is given by ND atoms/cm3 or sites/cm3 . The number density of
atoms in a pure Si crystal is approximately 5 × 1022 atoms/cm3 . We compare this density to typical dopant
densities of 1010 to 1020 atoms/cm3 [Middleman and Hochberg (1993)].

CB CB CB
band gap
electron energy

hole energy

Ef Efn Efp

VB VB VB

intrinsic n-type p-type


semiconductor

Figure 4.5: Energy levels and band structures for an intrinsic semiconductor (left), n-type (center) and p-type (right).

We are familiar with electrons, but may be uncomfortable with the concept of holes. Roughly

1. Holes mark the position in an atom lattice normally occupied by an electron and so has positive charge e;
CHAPTER 4. THE PHYSICS OF PV DEVICES 46

2. Holes can move in response to an electric field, resulting in conduction through the em valence band of a
material;
3. Holes can be assigned an effective mass mh∗ , useful for quantifying the mobility of these charge carriers.

With these concepts in mind, consider doping the Si with B (boron), a column III element. In this case, three
valence electrons are available to bond. This gives rise to a broken bond in the crystal, creating a new energy level
slightly above the valence band. Analogous to the previous doping case, a small amount of energy is necessary
to excite an electron from the valence band to this new energy level. Another way to look at this is a small
increase in hole energy (as seen in Fig. 4.5, oriented in the direction opposite of electron energy) is required to
promote it to the valence band. This is a p-type semiconductor, in which holes are the majority carrier found
within the valence band (c.f. electrons in the CB for n-type materials). We note that boron is widely used as
a p-type dopant in electronic devices instead of silicon’s neighbor aluminum because of the lower diffusivity and
higher solubility of B in Si relative to Al [Middleman and Hochberg (1993)]. As with n-type dopant, the number
density of acceptor sites in the Si is given as NA sites/cm3 .
The important notion here is that for the n- and p-type semiconductors described, the Fermi level of the doped
material is shifted from the approximate center of the forbidden gap towards the conduction band of the majority
carrier (see Fig. 4.5). For an extrinsic n-type semiconductor

ND
Ef = Ei + kB T ln
ni

a relationship valid for with 1011 < ND < 1018 cm−3 for Si. Of course a similar relationship can be derived for
p-type materials [Leck (1967)].

dopant donor Ec − Ed acceptor Ea − Ev


Al - 0.057 to 0.067 eV Figure 4.6: Ionization energies of important dopants of Si
B - 0.045 eV taken from Table 2.3 of [Leck (1967)].
P 0.045 eV -

4.1.3 Semiconductor electrical properties


Resistivity ρ has units ohm·m and is defined as specific electrical resistance; with
V
R= = Ω ohms
A
we have
L
R=ρ
A
where L is the length of the material specimen for which we are measuring the resistance and A the cross-sectional
area.
Conductance G is simply 1/R with units mho or siemens. Specific conductivity σ = 1/ρ therefore has units
siemens/m or 1/(ohm·m). For a semiconductor

σ = ne qµe + nh qµh (4.1)

where n are the charge carrier number densities (commonly in cm−3 for semiconductors), q is unit charge in C,
and µ is charge carrier mobility. For Si

µe = 1500 cm2 /(V·s)


µh = 475 cm2 /(V·s).
CHAPTER 4. THE PHYSICS OF PV DEVICES 47

As a quick check of units


cm2
   
charges C 1
nqµ : =
cm3 charge V·s ohm·cm
which is correct. In an electrically neutral material

nneg = npos
ne + NA = nh + ND .

In an n-type semiconductor, NA = 0, therefore

ne = nh + ND ≈ ND
ni2
ne nh = ni2 and so nh =
ND
For a p-type semiconductor, ND = 0, therefore

nh = ne + NA ≈ NA
ni2
ne nh = ni2 and so ne = .
NA
Therefore, given the charge carrier number densities ne and nh and the charge carrier mobilities, µe and µh listed
above, the specific conductivity σ can be computed using (4.1).

4.2 Junctions
Recall p- and n-type semiconductors, their band edges and Fermi levels; these are depicted in Fig. 4.7. A crucial
point is that both pieces of semiconductor are electrically neutral when they are separate. As a “thought”
experiment, we bring them in contact to form a double layer inside the depletion region.
When the doped semiconductors are separate, the n-type and p-type have different Fermi levels. When joined,
the materials must come to equilibrium, forcing the Fermi levels to become equal.

p-type n-type E
holes electrons

depletion
region
CB
CB CB
qVbi
electron energy

- CB
hole energy

Ef -+
- +
VB +
VB VB
VB

Figure 4.7: p- and n-type semiconductors before (left) and after (right) they are brought into contact.

The electric field is electric force per unit charge E = F /q. The direction of the field is taken as the direction
of force acting on a positive charge. Electric field has units N/C or V/m.
CHAPTER 4. THE PHYSICS OF PV DEVICES 48

The effect of this is band bending, resulting in the built-in potential Vb . Recall 1 V=1 J/C, so energy is (unit
charge)×(Voltage). When the p-type and n-type semiconductors are brought into contact, the conduction and
valence bands (which were identical because both correspond to Si) of the p-type material effectively shifts upward
in electron energy qVb relative to the n-type. This seems counterintuitive, in that the electrons, when migrating
across from the n-type to p-type material during the formation of the space-charge region, are moving uphill in
terms of energy. While energetically unfavorable, this process is entropically favorable. The process is analogous
to the normal diffusion of gases when a partition separating two gas species is removed.
The work function Φ is defined by the energy required to remove an electron from the Fermi level of a material
to vacuum. After the n- and p-type semiconductors with work functions Φn and Φp are brought into contact and
their Fermi levels equilibrate, the built-in voltage of the junction is simply

qVbi = Φp − Φn .

The built-in voltage can be explicitly computed as [Middleman and Hochberg (1993)]
 
kB T NA ND
Vbi = ln + ln
q ni ni

and the depletion zone width as equation (3.90) of [Gray (2003)]


s  
2 NA + ND
WD = (Vbi − V )
q NA ND

where V is the applied voltage (bias) and  is the electric permittivity of the semiconductor.
When the junction is at equilibrium, no net current flows. However, there is quite a bit still going on. Thermally
generated or drift currents Ihdrift and Iedrift result from excitation the minority carriers: electrons are thermally
excited into the conduction band on the p-side of the junction and holes are promoted into the valence band on
the n-side. These minority carriers diffuse through their respective quasi-neutral regions of the junction, and if
they reach the junction before they recombine, are swept through the junction by the electric field, generating
currents Iedrift and Ihdrift , respectively.
The recombination or diffusion currents Iediff and Ihdiff , are due to the majority carriers (holes on the p-side,
electrons on the n-side). This process, as described earlier, is due to the relatively large concentration of these
carriers on their respective sides of the junction, and can cross the barrier due to entropic effects.
Overall, this leaves us with the saturation current across the junction:

I = Ihdiff + Iediff + Ihdrift + Iedrift = 0


   

 
because Ihdiff + Iediff is positive and
 diff
Ih + Iediff = − Ihdrift + Iedrift
  
at zero bias.

We note that in this notation positive conventional current flows from left to right, and positive electron current
flows right to left.
In summary,

• Iedrift : created on the p-side by thermal excitation, and forced to the n-side by the electric field;
• Iediff : majority carriers of the n-side, driven by a concentration gradient across the barrier - similar to what
happens when the materials are first brought into contact;
• The degree to which the CB and VB are shifted, qVb , is typically less than the band gap of the undoped
semiconductor.
CHAPTER 4. THE PHYSICS OF PV DEVICES 49

E E E
- - ++ -- + -+
p --- --- ++ ++ n p -- +
+ n p -+ n
++ +
e- - + + - e-

reverse bias forward bias

Iediff
Iediff
Ie drift Iedrift Iedrift
electron energy

hole energy
Ihdrift
Ihdrift Ihdrift
Ih diff Ihdiff

Figure 4.8: Generation and recombination currents in an unbiased, forward-biased, and reverse-biased p-n junction.
Parameters for the diode equation are Io = 1 mA, β = 2, and T = 300 K.

4.2.1 Forward/reverse bias


A p-n junction forms a diode, an electronic circuit element that allows electricity to flow easily in one direction
(when forward biased) and not so much in the opposite (reverse bias). We previously considered the case where
the junction was in equilibrium.

Forward bias

A forward bias (attaching a positive electrode to the p-side of the junction) of magnitude V reduces the barrier
height by qV

1. This has little effect on Iedrift and Ihdrift ;


2. Iediff and Ihdiff are greatly enhanced because of the reduced electric field that resists the recombination
current. With sufficiently large bias, |Iediff + Ihdiff |  |Iedrift + Ihdrift |

In this case, the total current has a large positive value (see Fig. 4.8)

I ≈ Ihdiff + Iediff .
CHAPTER 4. THE PHYSICS OF PV DEVICES 50

Reverse bias

A reverse bias (attaching a positive electrode to the n-side of the junction) of magnitude |V | increases the barrier
height by −qV

1. Again, this has little effect on Iedrift and Ihdrift


2. Iediff and Ihdiff are greatly reduced because of the increased resistance to the recombination current.

In this case, the total dark current is consists mainly of the thermally generated current and so

I = Ihdrift + Iedrift = −Io ,

a smaller, negative value.


The overall response (the dark current) of the diode to forward (positive V ) and negative (negative V ) bias can
be described by the empirical diode equation:

I (V ) = Ihdiff + Iediff + Ihdrift + Iedrift


   
 
qV
= Io exp − Io
βkB T
   
qV
= Io exp −1 (4.2)
βkB T

where Io is the limiting value of current described earlier for high negative bias values, kB is the Boltzmann
constant, and β is the non-ideality factor, described in the next section. The electric circuit symbol for a diode is
shown in Fig. 4.9; the diode equation is given by (4.2).

E
-- +
p -- +
+ n Figure 4.9: The p-n junction and corresponding symbol for a
+ diode. We recall that electron current flows out of the anode

anode I cathode under forward bias (when a positive electrode is attached to


the p-side).

+ -

4.2.2 Diode non-ideality β


The factor β in the diode equation is a measure of diode behavior ideality. For the ideal case, β = 1, the p-n
junction is idealized as an infinite plane and no recombination takes place inside the depletion layer (space-charge
region). In this situation, recombination takes place only in the quasi-neutral regions - when current flows under
positive bias, for example, electrons enter though the n-type material side, diffuse through the junction, and
combine with holes on the p-type material side of the junction. Hole transport and recombination takes place in
and analogous (but opposite) manner.
For non-ideal junctions, β > 1 and for extreme cases where recombination takes place primarily in the depletion
region, β → 2. Because of the recombination taking place in the depletion region, it makes sense that the effect
of β > 1 is to lower the electric field barrier height. Overall, β ∈ [1, 2] for the PV cells we study; there are
situations in photoelectrochemical systems where β > 2 is possible.
CHAPTER 4. THE PHYSICS OF PV DEVICES 51

4.2.3 Other junctions


Other important junctions, such as ohmic, Schottky, p-i-n, and heterojunctions will be described where applicable.

4.3 Photon interactions


Photons with energy greater than the band gap of a material have the potential to be absorbed by that material,
generating an electron/hole pair. When this happens, the electron is promoted to the conduction band of the
material while the hole is is created and remains in the conduction band. This means the crystalline material is
essentially transparent to wavelengths of light longer than a critical value. We note that many materials with
which we are familiar have large band gap values, such as sapphire (Al2 O3 , band gap = 9.9 eV) or diamond (band
gap = 5.5 eV).
For silicon the band gap is E = 1.1 eV. Recall E = hν where h is Planck’s constant (h = 6.626 × 10−34 J·s or
4.136 × 10−15 eV·s) and that the speed of light is 2.998 × 1017 nm/s and so

E c ch
ν= , λ= = = 1127 nm
h ν E
The value above is consistent with Fig. 2.8 of [Wenham, et al. (2007)].

Transport of charged species across the junction

The key idea is simple: photogenerated (minority) carriers in both the the n- and p-type sections behave much
like the thermally generated Ihdrift , Iedrift versions of these carriers.

E E
-- +
+ - ++ n
p -- +n p
+ -

short circuit open circuit


Iediff
electron energy

Iediff
hole energy

hv Ieph Ieph
hv

Ihdiff
Ihph Ihdrift
Ihdiff

Figure 4.10: PV cell currents under short circuit (left) and open circuit conditions (center). These points are marked by
the circle and square markers, respectively, in the plot of the PV diode equation (right). For this example, Iph = −10Io
taken from Fig. 4.8, as are the other model parameters.

As seen in Fig. 4.10, the total photocurrent the sum of the electron and conventional photocurrents is

−Iph = Ihph + Ieph ≤ 0.

Based on the sign convention described earlier, for the short circuit case, the conventional current is negative
because it flows to the right of the cell. This means under illumination, electrons flow out from the n-type end of
the junction, travel around the external circuit, and recombine with holes on the p-side of the cell.
CHAPTER 4. THE PHYSICS OF PV DEVICES 52

The photo and dark currents add linearly in our idealized analysis, and so the voltage-current equation for the
solar cell now is a simple modification of the diode equation
   
qV
I = −Iph + Io exp −1 . (4.3)
βkB T

when considered in the sign convention shown in Fig. 4.10.


A representative plot is shown in Fig. 4.10 (right) illustrating the short-circuit current (Fig. 4.10, left) marked by
the circle, and the square the open-circuit voltage (Fig. 4.10, center). Between these limits is the range of useful I
vs V values, a range resulting from the external load resistance varying from zero (short circuit) to infinity (open
circuit).
At this point we ask, how do we find the actual power produced by the cell? This will be the focus of the following
chapter, in which the equivalent circuit model of the PV cell and external load are considered.

HW and review problems


1. n- and p-type Si samples are doped with the same concentration levels of NA = ND = 1017 cm−3 of their
respective dopants. Compute the change carrier concentrations in units cm−3 and specific conductivity in
siemens/cm for each of the two materials.
2. Characterize the p-n junction created from the materials of the previous problem by computing the built-in
voltage Vbi (in V) and depletion zone width (in nm).

3. Create a plot of the depletion zone width (in nm) as a function of NA and ND in cm−3 . Assume the external
bias is V = 0.
4. Compute the minority and majority charge carrier concentrations for n- and p-type doped Si for a) NA = 1015 ,
ND = 0 cm−3 ; b) NA = 0, ND = 1015 cm−3 ; c) NA = 1015 , ND = 1016 cm−3 .
5. Compute the built-in voltage Vbi and depletion zone width WD of a junction formed from materials in part
(a) and (b) of the previous problem.
6. Estimate the fraction of ionized P and B atoms in lightly doped Si over the temperature range 250 ≤ T ≤
350 K.

References
[Gray (2003)] Gray, J. I., Chapter 3: The physics of the solar cell. In Handbook of Photovoltaic Science and
Engineering, A. Luque and S. Hegedus (Ed.), John Wiley & Sons, Ltd, 2003.

[Leck (1967)] Leck, J. H., Theory of Semiconductor Junction Devices, Pergamon Press, 1967.
[Middleman and Hochberg (1993)] Middleman, S. and A. K. Hochberg, Process Engineering Analysis in
Semiconductor Device Fabrication, McGraw-Hill, Inc., 1993.
[Wenham, et al. (2007)] Wenham, S. R., M. A. Green, M. E. Watt, and R. Corkish, Applied Photovoltaics, 2nd
Ed., Earthscan, UK, 2007.
Chapter 5

PV cell architecture and manufacturing


methods

PV cell design is generally split between thin-film (a-Si, CIGS, CdTe) and Si-substrate cells. Let us consider the
entire manufacturing process for creating multicrystalline silicon (mc-Si) solar cells, starting with the production
of the Si substrates that form the base of the PV cell; additional details can be found in the overviews by
[Ceccaroli and Lohne (2003)] and [Ranjan et al. (2011)].

5.1 Crystalline Si cell substrates


Step 1: Reduction of SiO2

Lumpy quartz (SiO2 ) → Metallurgical-grade Si

The first step of the process is SiO2 reduction in which lumpy quartz (see Fig. 5.1) is converted to metallurgical-
grade Si. The primary reaction
SiO2 (s) + 2C(s) → Si(l) + 2CO(g)
takes place in a carbon electrode arc furnace at approximately 2200 K (compare this to 2000 K, the melting point
of quartz). Additional reactions compete with the Si production in the furnace, such as those producing SiC.

Figure 5.1: An atypical sample of lumpy quartz found outside the Geological
Museum of China, Beijing.

53
CHAPTER 5. PV CELL ARCHITECTURE AND MANUFACTURING METHODS 54

Step 2: Silicon hydrochlorination

Silicon → Chlorosilane liquids

Iron, boron, and aluminum remain as some of the impurities in the metallurgical-grade Si. Silicon hydrochlorination
is the first step to removing these impurities. The reactions take place in a fluidized-bed reactor at approximately
300o C, with the primary reaction
Si + 3HCl → H2 + SiHCl3 .
Additional chlorosilanes (e.g. SiCl4 , SiH2 Cl2 ) are produced, as well as other chlorinated impurities.
Step 3: Distillation

Chlorosilane liquids and impurities → Purified chlorosilane

Consider the boiling and melting points of the hydrochlorination products:

compound SiH4 SiH3 Cl SiH2 Cl2 SiHCl3 SiCl4 FeCl3 AlCl3 BCl3
boiling pt (o C) -112.3 -30.4 8.3 31.5 57.6 316 - 12.7
melting pt (o C) - - - - - - 190 -

Our desired product is SiHCl3 , silicon chloride hydride. Given the wide separation of boiling points, it is a relatively
simple matter to design two sequential distillation columns capable of producing the desired product stream.
Step 4: Chemical vapor deposition

Purified chlorosilane → Si

As the final purification step, electronic grade silicon (EGS) is produced in the Siemens CVD process and the
equilibrium reaction
SiHCl3 (g) + H2 (g) ↔ Si(s) + 3HCl(g)
Note this is CVD because of the gas-phase precursors and the solid film that is deposited; this reaction is the
reverse of the (exothermic) hydrochlorination reaction, and so high temperatures (1400 K) are needed to shift
the reaction above to the right.
[HCl]3 −∆G o
 
= exp = K (T )
[SiHCl3 ][H2 ] RT
200-300 hours of deposition are required to produce EG polycrystalline silicon.
Step 5-C: Single crystal growth

Polycrystalline Si → Single crystal Si

The polycrystalline Si is remelted (melting point: 1687 K) and then grown into a single crystal (and further
purified) via

1. The Czochralski method, in which an ingot (1 m long and 15 cm in diameter) is pulled from molten Si
forming a single cylindrical crystal; impurities tend to stay in the melt;

2. The float-zone process, where the cylindrical ingot is heated by a moving heater zone, allowing the Si to
melt and recrystallize; the segregation coefficients of impurities are < 1, meaning they prefer to remain in
the melted region, and so are swept from the crystal with each zone pass.
CHAPTER 5. PV CELL ARCHITECTURE AND MANUFACTURING METHODS 55

In each process, a small quantity of boron is added to melt to create the p-type substrates.
Step 5-M: Multicrystalline Si ingots

Polycrystalline Si, waste EGS → Multicrystalline Si

Multicrystalline Si ingots then are created from a combination of polycrystalline Si and waste EGS from the
semiconductor industry. Molten Si is pored into crucibles and solidification takes place from the bottom up.
Small, mostly vertical crystals (grains) are formed during the solidification process; this directionality is important
to the operation of the final solar cell. For p-type Si, a small amount of boron is added to the molten Si prior to
pouring the mixture into the crucible.
For silicon ingot formation a technique called “mono-casting” appear to be a promising development. In this
case, a monocrystalline ingot is cast (not pulled, such as in the Czochralski process). This is a much faster and
therefore less expensive process. Typically portions of the ingot are still multicrystalline, but large sections are
monocrystalline. Better casting process control will improve monocrystalline yield and subsequently will have a
significant influence on the industry.
Step 6: Wafering

(Multi)crystalline Si → Wafers

In the final wafering process, a wire saw is used to slice the Si ingots into ≈ 180 µm thick wafers. Many parallel
cutting wires and an SiC cutting solution is used in the process, where the cuts are made perpendicular to the
long crystal grains, resulting in the rectangular wafers distinctive to mc-Si cells. Kerf losses and surface roughness
are important considerations in the wafering operation [Koch et al. (2003)].

top contacts

anti-reflective coating
n-type silicon
p-type Si

back contact

Figure 5.2: Completed Si solar cell. We note that the n-type layer normally forms the top (sun-facing) surface of a solar
cell because it has a higher surface quality relative to p-type Si. The top layer is called the emitter (thickness ≈ 1 µm),
the bottom (substrate) is called the base, with current thickness of 100 µm.

At this point we have the Si substrates that will be transformed into the finished Si cells (Fig. 5.2); we recall the
basic diode equation model to make a conceptual connection between the photocurrent source term −Iph and the
junction of the model to the physical features of the complete PV cell:
   
qV
I = −Iph + Io exp −1 . (5.1)
βkB T

Given the model and final cell structure, we now examine the individual PV manufacturing steps and make note
where additional nonidealities might be introduced.

5.2 PV cell fabrication processing steps


Starting with the p-doped Si wafer described, an overview of the basic steps relevant to c-Si and m-Si PV cell
manufacturing are shown in Fig. 5.3. We now consider each step in sequence.
CHAPTER 5. PV CELL ARCHITECTURE AND MANUFACTURING METHODS 56

Starting wafer AR deposition

p-type Si

Texture etch Front contact printing

P diffusion Back contact print


n-type Si

Edge isolation Firing

Figure 5.3: Sequence of c-Si and m-Si solar cell fabrication.

Step 7: Surface finishing/texturing Chemical etches are used to smooth wire saw marks. Likewise KOH or
NaOH solutions can be used to texture the surface to decrease light reflection; the micron-sized pyramid shaped
surface results in a higher probability that a reflected photon encounters another portion of the cell surface.

Step 8: Phosphorous diffusion To create the n-type layer, a phosphorous precursor (for example, gas-phase
P2 H2 , liquid POCl3 , or solid-source precursor P2 O3 ) is used with a furnace operating at 900 o C to diffuse the P
into the Si; this diffusion can be modeled simply by

∂[P] ∂ 2 [P]
=D
∂t ∂z 2
subject to a constant concentration boundary condition at the wafer surfaces; note that all sides of the wafer
become doped.

Step 9: Edge isolation The wafer edge doping is removed to prevent short circuits through the PV device,
resulting in low shunt resistance.
Step 10: AR deposition Reflectance loss for bare Si in air can be as high as 30%. Antireflection coatings can
reduce losses below 10%; for m-Si with texturing, the combination reduces reflection to < 1%. AR coatings
typically consisting of CVD titanium dioxide or PECVD silicon nitride. The thickness of these coatings is set to
the wavelength-dependent optimal value Material refractive index is important and AR coating compositions are
chosen to give a refractive index between the values of glass and silicon.

It will be shown in a later chapter of these notes that reflectance loss for bare Si in air can be as high as
30%. Antireflection (AR) coatings reduce losses below 10%, and potentially to < 1% when combined with the
texturing described earlier [?]. AR coatings typically consist of CVD titanium dioxide or PECVD silicon nitride.
The thickness of the coatings is set to the wavelength-dependent optimal value such as
λ
d= .
4
CHAPTER 5. PV CELL ARCHITECTURE AND MANUFACTURING METHODS 57

Material refractive index is important and AR coating compositions are chosen to give a refractive index between
the values of glass (or air) and silicon.

Step 11: Front contact printing A silver paste mixture in an organic binder is used in a screen-printing process
to produce the front contacts. Line widths are ≈top
50 contacts
µm and the design [?] of the contact pattern is nontrivial.
Front contact shading can account for up to 10% loss of cell surface area.

anti-reflective
Step 12: Back contact The back contact is deposited as a metal paste containing silvercoating
and aluminum, the
latter to produce an ohmic contact with the p-type Si layer. n-type silicon
p-type Si
Step 13: Firing With the front/back contacts applied, the final assembly is fired to remove the contact organic
binders and to drive the Al contacts through the n-type back layer into the p-layer and the front contacts through
back contact
the AR film. This thermal process also can profoundly affect other characteristics of the cell performance.

Step 14: Wiring and sealing the panel A set of cells connected by series wiring is shown in Fig. 5.4; module
packaging is describe in Ch. 5 of [?]; additional details on cell wiring will be covered in later in these notes.

Figure 5.4: Cells wired in series.

5.3 Thin film deposition


In this section we consider thin film PV cells of three basic designs:

1. Amorphous silicon, including multijunction cells;


2. Cu(In,Ga)Se2 copper indium gallium (di)selenide, otherwise known as CIGS;
3. CdTe Cadmium telluride (tellurium, not the ski resort).

Crystalline versus thin-film

• Previously, Si single crystals or multicrystalline material was used both for the PV cell structural support
(substrate) and to form the junction. Typical cells were ≈ 300 µm thick;
• Now we consider thin films O(1 µm) thick deposited onto an essentially inert substrate for support - obvious
advantage: less active material required.

5.3.1 Film morphology


1. Crystalline materials, usually formed at relatively high temperatures, high concentration solutions/melts,
slow growth rates, structures of > O(10 cm);
2. Multicrystalline, structures of O(1 cm); conceptually similar to crystalline materials;
CHAPTER 5. PV CELL ARCHITECTURE AND MANUFACTURING METHODS 58

3. Polycrystalline, structures of O(1 µ m); nanocrystalline materials also can be grown with processes otherwise
used to deposit amorphous films;
4. Amorphous, faster growth rates from gas-phase precursors, lower temperature processes, can mostly preserve
stoichiometry.

Overall, the first two morphologies are relevant to c-Si and m-Si cells discussed in the previous chapter; the poly-
and nanocrystalline and amorphous materials are more relevant to thin-film cells.

5.3.2 CVD as a thin-film manufacturing process

reactor roof

reactant gas convection

reactor outlet
precursor gas phase reactions
reactor inlet

diffusion diffusion

adsorption surface processes desorption

substrate (wafer)
thermal
energy
cmcm

Figure 5.5: Twin CVD reactor system (left) and a schematic diagram of one reactor with major chemical species
transport and reaction phenomena noted (right).

• Gas phase reactants (precursors)


• Mix, react, and are transported through the gas phase
• (Hopefully) depositing on the wafer in a controlled manner

Large-scale III-V MOCVD for multijunction solar cells


Gas flows radially outward from a central feed point; susceptor contains multiple wafers. The effects of significant
depletion (gas-phase gradients) can be seen in Fig. 5.6.
Even larger substrates for thin film solar

• 5.7m2 Si PE-CVD process (appliedmaterials.com)


• Solar thin-film deposition processes adapted from flat-panel display technology

5.4 Amorphous Si
Consider the differences between the regular structure of c-Si and a-Si illustrated below. The H passivation and
amorphous structure are crucial to understanding the behavior of PV cells manufactured using these films:

• The chemical structure has much in common with crystalline Si, e.g., generally 4 bonds per Si atom;
• H atoms remove (passivate) dangling bonds which would otherwise encourage recombination;
CHAPTER 5. PV CELL ARCHITECTURE AND MANUFACTURING METHODS 59

Figure 5.6: Five-wafer planetary reactor system (left) and stalled wafer thickness profile illustrating the strong
depletion effect in these reactors (right).

Figure 5.7: Applied Materials a-Si PECVD production line (left) and an enlarged view of one reactor (right).

• Band gap depends on H fraction: Ebg = 1.6 - 1.8 eV for 5 - 20 at. % H [Archer and Hill (2001)];
• a-SiGe:H up to 60 at% Ge resulting in Ebg = 1.3 eV;
• Disorder may be one factor in much greater absorptivity.

5.4.1 Device structure


Typical a-Si:H devices are shown next. We see there are two major approaches to PV cell design:

1. Superstrate designs where the sunlight passes through the glass substrate first; this is a good design for
building-integrated PV applications where the glass is exposed to the outdoors environment;
2. Substrate designs where the glass substrate or other substrate is at the bottom of the cell; this is a good
design when opaque substrates are to be used (e.g., stainless steel in roll-to-roll processes).

Both cases are characterized by a p-i-n junction (p-intrinsic-n), seen in Fig. 5.9. The large undoped (intrinsic)
i a-Si:H layer is where most photons are absorbed; the electrons and holes then migrate to the n and p-layers,
respectively. This device architecture is used because the heavily doped n- and p-type layers result in high
recombination rates, and so they are made as thin as possible to minimize charge carrier residence times in these
regions.
CHAPTER 5. PV CELL ARCHITECTURE AND MANUFACTURING METHODS 60

Figure 5.8: Amorphous Si film structure illustrat-


ing dangling bonds (connected at one end to an Si
atom, denoted by the larger circles) and hydrogen
atoms (smaller circles).

- e- +
TCO, SnO2
+
glass - +
+ p a-Si:H
TCO, SnO2 - hv
+
p a-Si:H
i a-Si:H

i a-Si:H
n a-Si:H - +
contact
-
n a-Si:H - +
- glass
h+
contact - +
Superstrate design Substrate design p a-Si:H i a-Si:H n a-Si:H

Figure 5.9: Amorphous Si PV cell designs (left and center); band structure for single junction a-Si:H cell (right).

5.4.2 Device manufacture


The basic steps in manufacturing a-Si:H cells:

1. Starting with the glass superstrate...


2. Sputter/CVD SnO2 as the front contact; note the relatively high SnO2 band gap of 3.6 eV [Salehi et al. (2010)];
3. Plasma-enhanced chemical vapor deposition (PECVD) or hot-wire CVD using silane (SiH4 ) as the gaseous
precursor with H2 dilution;
4. B2 H6 , BF3 , and trimethyl boron (TMB) as dopant precursors for the p-type layer;
5. PH3 for n-type doping; note that in both doping cases, doping mechanisms can be much different from that
of crystalline Si;
6. Sputter Al back contact/reflector.

In [Adomaitis (2010)], an approach to empirical model development for a plasma-enhanced chemical vapor depo-
sition (PECVD) reactor used to deposit amorphous hydrogenated silicon (a-Si:H) films of transparent conducting
oxide-coated glass substrates in an industrial manufacturing system is described. Experiment design, response
surface modeling, analysis of variance, model use for process optimization, and inversion of the models to compute
process operating conditions corresponding to a specified set of film qualities are all described in that work. In the
industrial PECVD system considered, three process operating parameters were identified as potentially important
CHAPTER 5. PV CELL ARCHITECTURE AND MANUFACTURING METHODS 61

to influencing film quality: the H2 /SiH4 dilution ratio (D), reactor total pressure (P), and RF discharge power
(W ).

RT ET

1 1

Figure 5.10: Optimization of


0.5 0.5
W

W
PECVD i-layer deposition rate
(left) and final a-Si:H PV cell ef-
0 0
1 1 ficiency (right). Higher values are
0.5 0.5 indicated in red.
0 −1 0 −1
−0.5 −0.5
0 0
−0.5 0.5 −0.5 0.5
D P D P

RB EB
As seen in Fig. 5.10, the disposition rate (left plot) and final solar cell efficiency (right) are found to be strongly
1 1
influenced by the three operating parameters; in these plots, red corresponds to higher values of the growth rate
and cell efficiency, while blue denotes lower values. Of course, our objective it to maximize both. In these plots
we 0.5
spot a number of important trends.0.5In terms of deposition rate, we see that the film grows more quickly
W

with decreasing dilution ratio D and with increasing pressure P; while more difficult to see, film growth rate also
0 0
increases,
1 but at a lesser rate, with RF power 1 W . These observations all make physical sense in terms of the
chemical
0.5 vapor deposition process. Regarding 0.5 cell efficiency, that characteristic also increases with decreasing
0 −1 0 −1
dilution rate D; pressure0 P has−0.5 no effect and cell efficiency increases
0
−0.5 with decreasing RF power, the opposite of
−0.5 0.5 −0.5 0.5
growth rate.
D ThereforeP the optimal operating conditions
D areP low dilution ratio, high pressure, and a power level
that balances the overall manufacturing process economics with final product value.

5.5 CIGS
CuInSe2 is inherently a p-type semiconductor; partial substitution of Ga for In increases the CuInx Ga1−x Se2 band
gap from 1.02 eV to 1.1 to 1.2 eV, allowing the CIGS films to be made into efficient PV devices as seen in
Fig. 5.11. The band structure has a unique conduction-band offset that varies with processing conditions and
with the composition of the this buffer layer; regardless, this n-type film generally is depleted of its majority charge
carriers and so is vital to forming the p-n junction of the cell. Currently (as of December 2013) Samsung holds
the large-substrate commercial CIGS panel efficiency record of 15.7%. Cells fabricated on smaller substrates and
research-only (c.f., commercial panels) CIGS cells have reached a record efficiency of 20.8% [Wesoff(2013)].

5.5.1 Cell fabrication steps


For the substrate design shown in Fig. 5.11 (sunlight enters the top of the cell in this design), the cell components
and fabrication steps are briefly described as follows.

Substrate: The PV devices can be constructed on a wide range of flexible and rigid substrates. In
many cases, a soda-lime glass substrate is used because of its low cost and similar thermal expansion
coefficient as the CIGS film; borosilicate glass has lower coefficient of expansion and so leads to
cracks when cooling. The Na, K, and Ca oxide impurities in the substrate contribute positively to
the performance of the PV cell [Shafarman and Stolt(2003)]. However, all subsequent processing is
limited by the Tsoftening ≈ 550o C for soda-lime glass.
Back contact: The Mo back contact is created by DC sputtering. This step is relatively straightfor-
ward, though adhesion of the contact layer to the substrate can be an issue.
Absorber: Co-evaporation is a widely used method in manufacturing the CuInx Ga1−x Se2 absorber
layer. In this process, the elements, or element-containing precursors, are evaporated in a parallel
CHAPTER 5. PV CELL ARCHITECTURE AND MANUFACTURING METHODS 62

e-
ZnO:Al
-
n-CdS - +
hv
p-Cu(In,Ga)Se2 - +
- +
Mo contact
+ h+
+
n ZnO:Al
soda glass -
- +
Substrate design n CdS
p Cu(In,Ga)Sn2

Figure 5.11: Representative CIGS cell design (left), corresponding electrostatic band structure (center), and a
monolithically interconnected CIGS cell module (right)

fashion and are ballistically transported to the substrates for deposition. [Zachmann et al.(2009)]
used co-evaporation of Cu, In, and Ga as a part of the process to create a flexible CIGS film sup-
ported on polyimide. [Xu et al.(2012)] used co-evaporation from a single source with powders of the
elemental components in nominal composition (Cu-25, In-12.5, Ga-12.5, and Se-50). The powders
were evaporated using an electron beam and the composition of the film was controlled by varying
the beam current. [Jung et al.(2010)] deposited CIGS thin films on Mo/soda-lime glass in a three
stage method to study the effects of Ga composition on the films properties.
Alternatively, a two-step process can be used where the Cu, In, and Ga are first deposited, potentially
as thin separate layers, followed by a high-temperature annealing step in a Se atmosphere.
Buffer layer: A chemical bath deposition process is used to produce the undoped n-type CdS and
makes use of an alkaline aqueous solution containing the Cd and S precursors as well as a complexing
agent. Deposition times typically are several minutes. Pseudoepitaxial growth at the interface suggests
that the Cu(InGa)Se2 surface promotes the initial reaction and results from the small lattice mismatch
between the two layers.
Top contact: A transparent conducting oxide (TCO) such as Al-doped ZnO is deposited by an ALD
or atmospheric CVD process using water and diethylzinc as precursors for both processes.

5.5.2 Replacement of CdS Layer with ZnO by ALD


A number of factors have prompted the investigation of alternatives to cadmium sulfide (CdS) as the buffer layer
in CIGS thin film solar cells; in terms of device performance, the relatively low band gap of CdG (≈ 2.4 eV) results
in significant attenuation of photons in the near-UV and UV range. From a manufacturing perspective, a dry
gas-phase deposition process would be preferable to the liquid-phase chemical bath, and elimination of toxic Cd
would improve the overall sustainability of this PV technology.
Motivated by these limitations, a number of researchers have examined the potential of ALD ZnO and doped
ZnO as an alternative [Holmqvist(2013), van Delft et al.(2012)]. In the latter review, the use of ALD-prepared
(Zn,Mg)O buffer layers deposited by alternating ZnO and MgO cycles is described as one approach. Diethyzinc
Zn(C2 H5 )2 (DEZn) and MgCp2 are used as precursors with water as the oxidant source. Doping concentration
is determined by the ratio between ZnO and MgO cycles. The band gap of the (Zn,Mg)O films increases from
3.3 eV towards 3.8 eV as the concentration of Mg in the ZnO structure increases, resulting in a two-fold increase
in cell efficiency over pure ZnO. Increasing the MgO:ZnO ALD cycle ratio after this point, however, results in a
sudden drop off in device performance.
CHAPTER 5. PV CELL ARCHITECTURE AND MANUFACTURING METHODS 63

5.6 CdTe

5.6.1 Device architecture


A typical superstrate device and the corresponding heterojunction band structure diagram is shown in Fig. 5.12.
The latter provides valuable insight into the operation and design of these devices: most of the charge carriers are
produced in the CdTe layer. By the electric field found in that part of the device, electrons drift to the junction
while holes travel in the opposite direction. We see that because the CdS region is fully depleted, no field exists to
transport the charged species; its presence is to generate the field inside the CdTe film. Therefore, it is desirable
to make this film as thin as possible (see page 250 of [Archer and Hill (2001)]).

e-

glass - +
-
TCO, SnO2
hv
+
n-CdS
- +
p-CdTe - +
- + SnO2 | ITO
back contact
+ h+
- +
Superstrate
(preferred) design p CdTe n CdS

Figure 5.12: Representative CdTe cell design (left) and corresponding band structure (right).

We see that in this design, light enters the cell from the top, through the glass and TCO layer. This configuration
is consistent with the general design rule concerning the arrangement of the device semiconductor layers according
to decreasing bandgap in the direction light takes through the device: in this case, the topmost layer is SnO2 with
Ebg = 2.5 to 3 eV, CdS with Ebg = 2.42 eV, and CdTe with Ebg = 1.44 eV.

5.6.2 Device manufacture


The basic steps in manufacturing CdTe cells are described as follows.

1. Manufacturing begins with the glass superstrate;


2. SnO2 or tin-doped indium tin oxide (ITO) is sputtered or deposited by thermal chemical vapor deposition
to form the front contact;

3. Deposition of CdTe from


(a) physical vapor deposition (PVD) from elemental sources;
(b) close-space sublimation or vapor transport deposition (similar to PVD, but at higher pressure);
(c) electrodeposition in an aqueous electrolyte
(d) screen printing from a (Cd, Te)-containing paste
4. CdS by sputtering, chemical bath deposition (described earlier), and PVD
5. Cu back contact by PVD, electrodeposition, or screen printing.
CHAPTER 5. PV CELL ARCHITECTURE AND MANUFACTURING METHODS 64

5.7 Summary
Thin-film PV cells are developing rapidly and have great potential; taken from [Archer and Hill (2001)], the
following efficiencies have been reported for commercial modules:

• Single junction efficiency: a-Si: up to 8% for commercial modules


• Tandem junctions up to 9.5%
• Triple junctions over 10% efficiency
• CdTe over 10% for commercial modules

• CIGS: nearly 15% commercial module efficiency

HW and review problems


1. Describe a representative chemical process by which the pyramidal patterns are created on an Si surface.
Be sure to include reactants and etching operating conditions (e.g., temperature, etch time, etc.).
2. One of the earliest working solar cells used copper oxide as its semiconductor. Briefly describe the architec-
ture of this type of cell (the materials and configuration of the cell components) and draw a band diagram
illustrating the nature of the junction of this cell.

3. Write a sentence or two describing the role these junctions play in thin-film and crystalline Si solar cells:
(1) ohmic junction; (2) Schottky junction; (3) homojunction; (4) heterojunction; (5) tunnel junction; (6)
rectifying junction; (7) non-rectifying junction.
4. The absorption coefficient of crystalline Si is α = 105 m−1 for light with wavelength λ = 800 nm. What
fraction of light at this wavelength is absorbed within the first 1 µm of the top surface of an Si solar cell if
absorbance is modeled by αe −αz where z = 0 represents the top surface of the PV cell.
5. Given the P diffusion coefficient 6.5 × 10−13 cm2 /s, a surface equilibrium P concentration [P]o = 1020
cm−3 and a boron doped Si substrate with NA = 1015 cm−3 , compute the diffusion time (in s) required to
create a junction with depth 1 µm. Plot the resulting [P] profile.
6. List typical thicknesses for the base layers making up an a-Si, CdTe, and CIGS PV cell. Be sure to cite
sources.

References
[Adomaitis (2010)] Adomaitis, R. A., Notes on response surface modeling for a thin-film solar cell
manufacturing application. Unpublished technical document.
[Archer and Hill (2001)] Archer, M. D. and R. Hill, Ed. Clean Energy from Photovoltatics 2001, Imperial
College Press, pp. 202-203.

[Ceccaroli and Lohne (2003)] Ceccaroli, B. and O. Lohne, Chapter 5: Solar grade silicon feedstock. In Handbook
of Photovoltaic Science and Engineering, A. Luque and S. Hegedus (Ed.), John Wiley & Sons, Ltd (2003).
[Holmqvist(2013)] Holmqvist, A., Model-based Analysis and design of Atomic Layer Deposition Processes, PhD
Thesis, Lund University (2013).
CHAPTER 5. PV CELL ARCHITECTURE AND MANUFACTURING METHODS 65

[Jung et al.(2010)] Jung, S., S. J. Ahn, J. H. Yun, J. Gwak, D. Kim, and K. Yoon, Effects of Ga contents on
properties of CIGS thin films and solar cells fabricated by co-evaporation technique, Current Applied
Physics, 10 990-996 (2010).
[Koch et al. (2003)] Koch, W., A. L. Endrös, D. Franke, C Haßler, J. P. Kalejs, and H. J. Möller, Chapter 6:
Bulk crystal growth and wafering. In Handbook of Photovoltaic Science and Engineering, A. Luque and S.
Hegedus (Ed.), John Wiley & Sons, Ltd (2003).
[Ranjan et al. (2011)] Ranjan, S., S. Balaji, R. A. Panella, and B. E. Ydstie, Silicon solar cell production. Comp.
& Chem. Enging 35 14391453 (2011) .

[Salehi et al. (2010)] Salehi, H., M. Aryadoust, and M. Farbod, Electronic and structural properties of tin
dioxide in cubic phase. Iranian J. Sci. Tech., Trans. A 34 131-138 (2010).
[Shafarman and Stolt(2003)] Shafarman, W. N and L. Stolt, Chapter 13: Cu(inGa)Se2 Solar Cells, In Handbook
of Photovoltaic Science and Engineering, A. Luque and S. Hegedus (Ed.), John Wiley & Sons, Ltd (2003).
[van Delft et al.(2012)] van Delft J. A., D. Garcia-Alonso, and W. M. M. Kessels, Atomic layer deposition for
photovoltaics: Applications and prospects for solar cell manufacturing, Semicond. Sci. Tech. 27 074002
(2012).
[Wesoff(2013)] Wesoff, E., Samsung now the new CIGS solar module efficiency record holder, Greentech Media,
December (2013).

[Xu et al.(2012)] Xu, C., H. Zhang, J. Parry, S. Perera, G. Long, and H. Zeng, A single source three-stage
evaporation approach to CIGS absorber layer for thin film solar cells, Solar Energy Materials and Solar
Cells, 117 357-362 (2012).
[Zachmann et al.(2009)] Zachmann, H., S. Heinker, A. Brauna, A. V. Mudryi, V. F. Gremenok, A. V.
Ivaniukovich, and M. V. Yakushev, Characterisation of Cu(In,Ga)Se2 -based thin film solar cells on
polyimide, Thin Solid Films, 517 2209-2212 (2009).
Chapter 6

PV cell power and the equivalent circuit


model

As part of the Gemstone program, a group of undergraduate students at the University of Maryland has worked on
a number of solar energy concepts, constructing panels and preparing to test concepts in concentrating systems.
Consider, for example, the panel they constructed (Fig. 6.1, left) consisting of 30, 8 × 15 cm2 multicrystalline Si
cells, wired in series. This gives a total active cell area of Ac = 0.012 m2 per cell and Ap = 0.36 m2 for the panel.

external load
A
-
voltage drop
R V

+
B
I

Figure 6.1: Gemstone team-built solar panel (left) and the panel and external test circuit (right).

The students conducted a sequence of experiments from about noon to 1pm on September 22, 2010, in which
the voltage (V ) across a selection of resistors R forming an external circuit to the panel (Fig. 6.1, right) was
measured across the A-B terminals shown. A subset of the data1 are listed in Table 6.1.
The results are shown in the top left plot of Fig. 6.2. The current I flowing through the external circuit was
1 Scaled by dividing by the total number of cells N = 30 that are wired in series.

Rex (Ohms) 0.033 0.102 0.185 0.233 0.333 0.733 3.333


V (V) 0.075 0.220 0.315 0.347 0.399 0.448 0.483

Table 6.1: Experimentally measured voltage versus external load resistance, scaled in such a way to represent a
single-junction PV cell.

66
CHAPTER 6. PV CELL POWER AND THE EQUIVALENT CIRCUIT MODEL 67

computed using Ohm’s law:


V
I =
Rex
and the results are displayed in Fig. 6.2, bottom left. The positive value indicates conventional current is flowing
out of the bottom of the cell. Of course, power P dissipated by the resistor is simply

V2
P = IV =
Rex
and as displayed in Fig. 6.2, top right, it is important to to keep in mind that the net power produced by the solar
cell is equivalent in magnitude but opposite in sign to that which is dissipated by the resistor.

0.5
0.45
0.4
0.4
0.35

P (W)
Vex (V)

0.3
0.3
0.25
0.2 0.2
0.15
0.1 0.1
0.5 1 1.5 2 2.5 3 0.1 0.2 0.3 0.4
R (Ohms) Vex
ex

800
2
global irradiance EG, W/m2

600
1.5
Iex (A)

400
1

200
0.5

0
0.1 0.2 0.3 0.4 0 5 10 15 20
Vex (V) hour

Figure 6.2: Measured solar cell voltage versus external circuit resistance (top left), current vs. voltage (bottom left),
power versus voltage (top right) and irradiance for a horizontal solar panel on Sept 22, 2010.

Finally, in Fig. 6.2 bottom right, we plot the global irradiance for the day of the experiment (at zero degrees tilt).
Given that the experiments took place at about the time of maximum irradiance (818 W/m2 × 0.012 m2 = 9.74
W for the active cell area), the maximum cell efficiency η is

0.54 W
η= ≈ 5.5%
9.74 W
The main question we now try to answer is: how does this panel perform relative to the maximum theoretical
efficiency, and how do we begin to understand inherent versus specific device inefficiencies?
CHAPTER 6. PV CELL POWER AND THE EQUIVALENT CIRCUIT MODEL 68

h!
(a)
e-
n0 (b) -
n1
n2 h+ e-
Ie,ph
n-type
+ + + + (c) +
- - - - h+ e- -
h+ e- p-type
(b)
Ih,ph h+ e-

+
Figure 6.3: PV cell photocurrent and recombination sites.

6.1 Cell currents and the equivalent circuit


Recall the currents flowing through the illuminated solar cell shown in Fig. 6.3. In the dark, the solar cell acts as
a diode, corresponding to the electric circuit element to the right in Fig. 6.4. We recall that the current under
these conditions depends on the bias voltage V where positive bias reduces the barrier height, increasing the
recombination current flow, resulting in a large recombination current. Negative bias (V < 0) results in I → Io
Electron/hole pairs created by the photons result in 1) minority carriers swept through the junction from both
sides of the junction; 2) additional majority carriers, some of which are collected at the electrodes, with the
remainder undergoing recombination.

actual e- A
-
V = voltage drop

n-type

I R
Iph p-type Idiode
+
I<0 B
Figure 6.4: PV cell and external load equivalent circuit.

As described in the previous chapter, the photocurrent consists of the flow of electrons and holes created by
photons interacting with the semiconductor making up the PV cell; this additional current Iph flows parallel to
but opposite in direction of the diode (Fig. 6.4). Recall the PV cell single-diode model (4.3), we adopt the
conventional current sign convention illustrated in Fig. 6.4 to find
   
qV
I = Idiode − Iph = Io exp − 1 − Iph
βkB T

with Iph , Io ≥ 0.
CHAPTER 6. PV CELL POWER AND THE EQUIVALENT CIRCUIT MODEL 69

6.1.1 Short circuit current, open circuit voltage


In the context of our complete circuit shown in Fig. 6.4, in the case of a short circuit, R = 0, V = 0 and so
I = −Iph = Isc
and for the data of Table 6.1, this is approximately Isc = −2.25 A. Therefore, the short circuit current Isc is the
maximum current magnitude and corresponds to the PV cell operating at its minimum voltage.
In the opposite extreme, for the case of an open circuit, I = 0 and we read off the open-circuit voltage as
V = Voc ≈ 0.5 V, therefore   
qVoc
0 = Iph + Io 1 − exp
kB βT
and so for now, assuming we have ideal diode behavior (β = 1) we find
Iph
Io =   = 8.97 × 10−9 A.
qVoc
exp −1
βkB T
We conclude that the open circuit voltage Voc is the maximum voltage drop and corresponds to the minimum
current that can be produced by a solar cell. Given Isc and Io computed from the equation above, we write the
final form of our single-junction, single-diode PV cell equivalent circuit model
   
qV
I = −Iph + Io exp −1 (6.1)
βkB T
with Iph = −Isc for this system.

0.0 PV cell perspective 1.0 Load perspective


=1 =1
0.5 0.8
=2 =2
1.0 data 0.6 data
P (W)
I (A)

1.5 0.4
2.0 0.2
2.50.0 0.1 0.2 0.3 0.4 0.5 0.00.0 0.1 0.2 0.3 0.4 0.5
2.5 V V

2.0
current (A)

1.5
1.0
Iph
0.5 Idiode
0.00.0 0.1 0.2 0.3 0.4 0.5
V

Figure 6.5: Diode plus photocurrent model for the Gemstone data; note how Isc = Idiode at V = Voc .

Plots of the cases for β = 1 and β = 2 are shown in Fig. 6.5. As we would expect, the fit of the model is good
at the open- and short-circuit cases; the model-data mismatch in the mid-section of the data is poor, however,
indicating our diode+photocurrent model (6.1) may be missing some elements; these will be discussed next and
in later chapters.
CHAPTER 6. PV CELL POWER AND THE EQUIVALENT CIRCUIT MODEL 70

6.2 Maximum cell power


For our model (6.1), the maximum power Pmp that is dissipated by the load is computed using

Pmp = −Vmp Imp


   
qV
= max V Iph + Io 1 − exp (6.2)
V βkB T
6= Voc (−Isc )

Differentiating the power equation above with respect to V and setting the result equal to zero results in a
nonlinear equation for Vmp
    
qVmp qVmp q
g (Vmp ) = Iph + Io 1 − exp − Vmp Io exp =0 (6.3)
βkB T βkB T βkB T
which must be solved numerically for Vmp . We can simplify equation (6.3) using
 
Iph qVoc
= exp −1
Io βkB T
to find equation (3.4) of [Wenham, et al. (2007)]:
     
qVoc qVmp qVmp q
0 = exp − exp − Vmp exp
βkB T βkB T βkB T βkB T
 
qVoc − qVmp q
= exp − 1 − Vmp
βkB T βkB T
 
βkB T q
therefore Vmp = Voc − ln 1 + Vmp . (6.4)
q βkB T
It is clear from the equation above that Vmp < Voc . For T = 300 K we find that
kB T
= 0.0259 V
q
and so equation (6.4) can be safely used in an iterative manner to find Vmp , starting with an initial guess
0
Vmp ≈ Voc where superscript i = 0, 1, 2, ... indicates iteration index:
 
i+1 βkB T i q
Vmp = Voc − ln 1 + Vmp . (6.5)
q βkB T

There are other approaches to computing solutions to equation (6.3). One of the most efficient is to use Newton’s
method, which consists of the following steps:

0 0
1. Estimate a solution to (6.3) by choosing a value 0 ≤ Vmp ≤ Voc (such as before Vmp = Voc or the mean
0
value of the expected voltage range Vmp = Voc /2);

2. Compute the value of the residual g i of (6.3) as

g i = g (Vmp
i
)

where i is the iteration number (i = 0 for the initial guess, i = 1 for the first iteration, etc.)
3. The objective of the Newton procedure is to compute solutions to the problem

∂g
0 = gi + i+1 i

Vmp − Vmp
∂Vmp V i

mp
CHAPTER 6. PV CELL POWER AND THE EQUIVALENT CIRCUIT MODEL 71

Iteration number i 0 1 2 3 4 5 6 7
Iterated equation (6.5) 0.5 0.4221 0.4263 0.4260 0.4260 0.4260 0.4260 0.4260
Newton procedure 0.5 0.4766 0.4551 0.4381 0.4286 0.4262 0.4260 0.4260

Table 6.2: Computations of Vmp (in Volts) using the iterative (eqn 6.5) and Newton-based methods.

i+1
and so a more refined solution estimate Vmp is computed using

i+1 i gi
Vmp = Vmp −
[∂g /∂Vmp ]V i
mp

with     2
∂g qVmp q qVmp q
= −2Io exp − Vmp Io exp
∂Vmp βkB T βkB T βkB T βkB T
i+1
4. We increment i = i + 1 and return to step 2 and continue to iterate until the difference between Vmp and
i
Vmp is infinitesimally small (with a sufficiently good initial solution estimate, the procedure should converge
in as few as six iterations). This difference is called the update u i and is defined as
gi
u i = Vmp
i+1 i
− Vmp =−
[∂g /∂Vmp ]V i
mp

We compare the two techniques in Table 6.2 for the case Isc = 2.25 A and Voc = 0.5 V to see that the iterative
technique actually can outperform the Newton procedure for this choice of parameter values and initial estimate
for Vmp . This does not render the Newton procedure useless, however, as we will see in the later section on series
and shunt resistances.
Of course, we use Vmp to then find Imp and Pmp directly from the diode model and the definition of power.

6.2.1 Fill factor


A commonly used term to compute Pmp from the open- and short-circuit data is the fill factor FF
Vmp Imp
FF = .
Voc Isc
Noting the relatively rapid convergence of (6.5), using a single iteration we find the approximate value of Vmp as
 
βkB T q
Vmp ≈ Voc − ln 1 + Voc
q βkB T
so following the notation of Green we define νoc = qVoc /βkB T and νmp = qVmp /βkB T
Isc exp (νoc ) − exp (νmp )
Imp = Isc + Io [exp (νmp ) − 1] = Isc + [exp (νmp ) − 1] = Isc
1 − exp (νoc ) exp (νoc ) − 1
and so because
exp (νoc )
exp (νmp ) ≈ .
1 + νoc
q/βkB T
Imp ≈ Isc Voc
[1 − exp (−νoc )] (1 + νoc )
giving
Vmp Imp νoc − ln (1 + νoc )
FFapprox = = .
Voc Isc [1 − exp(−νoc )] (1 + νoc )
Again, using the PV cell parameters Isc = −2.25 A and Voc = 0.5 V, an accurate calculation of the fill factor
gives FF = 0.8033 while the approximation gives a very good estimate of FFapprox = 0.8027
CHAPTER 6. PV CELL POWER AND THE EQUIVALENT CIRCUIT MODEL 72

6.3 Parasitic resistances


Consider the new equivalent circuit model in Fig. 6.6. In this Figure, we observe two primary differences with
Fig. 6.4: that the usable voltage drop V ≥ 0 is reduced by the addition of a series resistance Rs and the usable
current I ≥ 0 is reduced by finite shunt resistance values 0 < Rsh < ∞. Physically, these additions correspond to

• Series resistance: results from the inherent resistance of the PV cell itself to charge carrier flow, from the
resistance of interfaces between the cell and metal contacts, and from the resistance of the metal contacts
themselves. Denoted RS , the optimal value of this resistance is zero.
• Shunt resistance: results primarily from manufacturing defects, allowing current to flow through the PV
cell or around its edges (recall the edge isolation step after P diffusion in crystalline PV cell manufacturing).
Denoted Rsh the ideal value approaches positive infinity.

V = voltage drop
Iph n-type Idiode Rs

p-type Rsh Ish


+
I<0
Figure 6.6: Series and shunt resistances in an equivalent circuit.

Translating this to an equivalent circuit model, we find


   
V − IRs V − IRs
I = −Iph + Io exp q −1 + (6.6)
βkB T Rsh
We note that the problem of plotting current I versus voltage drop V across the external load has now become
more complicated. To plot the I and V performance curves, we start by by considering two conceptually different
situations:

1. We know Iph and Io and wish to compute the resulting Isc , Voc and points on the I versus V curve for
Rs > 0 and/or finite Rsh . This situation can be thought of as having a pristine cell with known performance
undergo some transformation introducing parasitic resistances, degrading the performance of the cell.
2. We know Isc , Voc and either the values of the parasitic resistances, or the values of I for one or more values
of V ∈ (0, Voc ).

6.3.1 Known values of Io and Iph


Determining the open-circuit voltage for this situation requires solving the nonlinear equation
   
Voc Voc
0 = −Iph + Io exp q −1 +
βkB T Rsh
for Voc , which we can rearrange to find
 
βkB T Iph Voc
Voc = ln − +1 (6.7)
q Io Io Rsh
CHAPTER 6. PV CELL POWER AND THE EQUIVALENT CIRCUIT MODEL 73

with the initial guess for Voc the value found for ideal conditions. From this equation we observe that Rs has
no effect on the value of Voc because no current flows through the resistor corresponding to Rs . Likewise, under
short circuit conditions we find an equation that can be used immediately in an iterative manner and that should
perform well as Rs → 0:    
−Isc Rs Isc Rs
Isc = −Iph + Io exp q −1 − (6.8)
βkB T Rsh
with an initial guess of Isc = Iph . It is important to stress, however, that this iterative equation works only in the
limit of small values of Rs ; for mid-size and large values, a Newton procedure must be employed with
   
−Isc Rs Isc Rs
g (Isc ) = −Isc − Iph + Io exp q −1 −
βkB T Rsh
 
−Isc Rs Rs Rs
dg /dIsc = −1 − Io exp q q −
βkB T βkB T Rsh

The current I that flows through the external circuit is plotted as a function of V ∈ [0, Voc ] using (6.6). Again, a
nonlinear procedure (e.g., Newton’s method) must be used to solve for I as a function of V , where for this case
we wish to find the root of h(I ) = 0 where
   
V + IRs V − IRs
h(I ) = −I − Iph + Io exp q −1 +
βkB T Rsh
 
V − IRs Rs Rs
dh/dI = −1 − Io exp q q −
βkB T βkB T Rsh
given a vector of values V ∈ [0, Voc ] where Voc is found using (6.7) or another Newton procedure.

Plotting I versus Vdiode

An alternative, non-iterative approach for plotting I versus V is to recall the diode equation
   
qV
Idiode = Io exp −1
βkB T
and employ Kirchhoff’s current law to find
I = −Iph + Idiode + Ish
where Ish corresponds to the current flowing through the shunt resistor. We observe that the voltage across both
the diode and the shunt resistor are equal with value
Vdiode = V − IRs . (6.9)
The definition of Vdiode then can be used in (6.6) to obtain the explicit equation for I as a function of Vdiode
   
Vdiode Vdiode
I = −Iph + Io exp q −1 + (6.10)
βkB T Rsh
Vdiode
= −Iph + Idiode +
Rsh
= −Iph + Idiode + Ish which checks.
Therefore, the following procedure can be developed to compute I versus V given values of Iph and Io :

1. Choose a value of Vdiode ∈ [−Isc Rs , Voc ];


2. Compute a corresponding value for I using (6.10);
3. Compute the corresponding value of V = Vdiode + IRs .
CHAPTER 6. PV CELL POWER AND THE EQUIVALENT CIRCUIT MODEL 74

6.3.2 Known values of Voc and Isc


If Voc and Isc are known values, for given (or assumed, for now) values of Rs , Rsh , and β we can write
   
−Isc Rs Isc Rs
short circuit: Isc = −Iph + Io exp q −1 −
βkB T Rsh
   
Voc Voc
open circuit: 0 = −Iph + Io exp q −1 +
βkB T Rsh

so adding the two conditions gives an equation that can be solved for Io :
    
−Isc Rs qVoc Isc Rs − Voc
Isc = Io exp q − exp −
βkB T βkB T Rsh

and Iph from either of the open-circuit or short-circuit equations.


To illustrate a situation where a parasitic resistance value can be estimated from solar cell I versus V data,
consider our experimental data and recall our measured PV cell parameters Isc = −2.25 A and Voc = 0.5 V. We
will investigate the potential of explaining the mismatch between the diode model without parasitic resistances
and the experimentally measured data might be attributable solely to the series resistance, that is, that Rsh → ∞
and Rs is finite.
Using the procedure we developed above, we solve for Io = 8.97 × 10−9 A and use that to find Iph = 2.25 A,
values that are essentially unchanged from the previously computed figures (the differences cannot be seen due to
the number of significant digits we report). These values of Iph and Io are used in (6.6) together with the series
resistance Rs = 0.08 Ω and shunt resistance Rsh → ∞. As seen in Fig. 6.7, we clearly see how much better the
I versus V curve fits the experimental data, indicating a significant amount of series resistance in the PV cell
array used to obtain these performance data. One can speculate as to the source of the Rs - for example the
resistance may be attributable to resistance introduced by the soldering operations used to create the string of
series-connected solar cells in the PV panel. Likewise, environmental effects may have contributed to corrosion
and reduced conductance in the soldered connections.

0.0
Rs =0.08
Rs =0
0.5 data

1.0
I (A)

Figure 6.7: I versus V for a PV cell with


parasitic series resistance Rs = 0.08 Ω.
1.5

2.0

2.50.0 0.1 0.2 0.3 0.4 0.5


V (V)
CHAPTER 6. PV CELL POWER AND THE EQUIVALENT CIRCUIT MODEL 75

6.3.3 An iterative approach for Pmp when Rsh → ∞, Rs > 0


We now present a relatively simple (relative to the Newton-Raphson procedure described later in this section)
approach to computing Pmp for cases when Rsh → ∞. For this case where only Rs is considered
   
V − IRs
I = −Iph + Io exp q −1
βkB T

and so the power dissipated by the load is P = IV . As before, we differentiate P with respect to V to find the
peak value
∂P ∂I
=I +V (6.11)
∂V ∂V
∂I qIo /βkB T
with = . (6.12)
∂V exp [−q(V − IRs )/βkB T ] + qIo Rs /βkB T

Substituting (6.12) into (6.11), setting the result equal to zero, and making use of the open-circuit conditions
 
Iph qVoc
= exp −1
Io βkB T

and the diode voltage at maximum power as

Vmp,diode = Vmp − Imp Rs

we ultimately find the iterated relations


" ! #
i
i+1
Vmp,diode
Imp = −Iph + Io exp q −1 (6.13)
βkB T
 
i
βkB T  Vmp,diode + I i+1 Rs
i+1
Vmp,diode = Voc − ln 1 +  mp  . (6.14)
q βk T /q + I R exp qV i
B o s mp,diode /βkB T

0
Using Vmp,diode = Voc we observe the following excellent convergence behavior of the algorithm
i 0 1 2 3 4 5
i
Vmp,diode 0.5 0.4681 0.4567 0.4528 0.4516 0.4513
i
Imp - 0.0 -1.5938 -1.8293 -1.8879 -1.9041
to find for the Gemstone data the results shown in Fig. 6.8.
Finally, we compute the photocurrent produced per unit area as
2.25 A
= 18.75 mA/cm2
120 cm3
and compare that to [Krogstrup, et al. (2013)] where a state-of-the-art research nanostructured solar cell produced
a full order of magnitude greater current density.

6.3.4 Maximum power with parasitic resistances - general case


Using either the somewhat-direct method described earlier (where I is plotted as a function of Vdiode and the
abscissa is then translated back to V ) or the Newton procedure, we then can plot the power supplied to the
external load as P = (−I )V . This calculation is trivial once the values of I are computed by one of the two
procedures described. Computing Pmp and the corresponding values of Vmp and Imp explicitly, however, is another
story.
CHAPTER 6. PV CELL POWER AND THE EQUIVALENT CIRCUIT MODEL 76

1.0
Rs =0.08
Rs =0
0.8 data
Pmp
parameter value
0.6 Io 8.97 × 10−9 A
Iph 2.25 A
P (W)

Vmp 0.299 V
0.4 Imp -1.90 A
Pmp 0.57 W

0.2

0.00.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7


Vex (V)

Figure 6.8: Power dissipated by the external load (P) versus V for a PV cell with parasitic series resistance Rs = 0.08 Ω.

To derive a generally applicable method for computing Pmp , we recall that the maximum power point is defined
by (6.11) and so
   
dP dI
= −Imp − Vmp =0
dV mp dV mp
   
Rs 1
0 = −Imp 1 + ΓRs + − Vmp Γ +
Rsh Rsh

where we must differentiate (6.6) and solve the result for dI /dV to find
1
  Γ+
dI Rsh
=
dV Imp ,Vmp
Rs
1 + ΓRs +
Rsh
with  
Vmp − Imp Rs q
Γ = Io exp q .
βkB T βkB T
We note that dI /dV > 0, which makes sense given all of the I versus V curves we have seen so far. To find the
maximum power point we must find the solution vector
 
Vmp
x=
Imp

that satisfies the following pair of equations consisting of our zero-derivative condition and the diode equation
(6.6) itself
     
Rs 1
−I mp 1 + ΓR s + − V mp Γ +
   Rsh Rsh 
g (x)  
f= =  .
h(x)     
 Vmp − Imp Rs Vmp − Imp Rs 
−Imp − Iph + Io exp q −1 +
βkB T Rsh
CHAPTER 6. PV CELL POWER AND THE EQUIVALENT CIRCUIT MODEL 77

A Newton-Raphson procedure then is used to solve the coupled set of nonlinear equations such that
xi+1 = xi − [J(xi )]−1 f(xi )
where the iteration number now is denoted using the subscript i and with Jacobian array elements
1
Γ+
∂g Rsh (1 + ΓRs + Rs /Rsh )(−Γq/βkB T ) + (Γ + 1/Rsh )(ΓRs q/βkB T )
J1,1 = = + Vmp
∂Vmp Rs (1 + ΓRs + Rs /Rsh )2
1 + ΓRs +
Rsh
∂g (1 + ΓRs + Rs /Rsh )(−Γq/βkB T ) + (Γ + 1/Rsh )(ΓRs q/βkB T )
J1,2 = = 1 + Rs Vmp
∂Imp (1 + ΓRs + Rs /Rsh )2
∂h
J2,1 = = −Γ − 1/Rsh
∂Vmp
∂h
J2,2 = = −1 − Rs Γ − Rs /Rsh
∂Imp

While somewhat complicated, the procedure converges quickly to accurate values of Vmp and Imp .

6.4 Temperature effects


In the spirit of examining PV cell nonidealities that are not associated with Iph (which will be covered in the next
chapter), let us consider the question of how Pmp changes with T . [Wenham, et al. (2007)] claim that
dVoc Vg 0 − Voc + γkB T /q
=−
dT T
by differentiating  
βkB T Iph
Voc = ln +1
q Io
with  
γ −Ebg 0
Io = BT exp ,
kB T
B as a material-specific constant, γ ∈ [1, 4] [Möller (1993)], and Ebg 0 the semiconductor band gap at 0 K; for Si
Ebg 0 = 1.17 eV [Kano (1998)]. For the PV cells used by the students described at the beginning of this chapter,
Voc = 0.5 V and so by choosing γ = 2 we find
dVoc 1.17 − 0.5 V + 2(8.6173 × 10−5 eV/K)(298 K) V
=− = −0.0024 .
dT 298 K K
We note that two simplifications are possible: (1) the relatively minor temperature dependence of Ebg means
Ebg (T = 298 K) can be used when Ebg 0 is unavailable; and (2) because kB T /q = 0.0257 eV at T = 298 K, this
term can be neglected when Vbg 0 − Voc  kB T /q.

HW and review problems


1. Given the following table of Voc and Isc values

case Voc (V) Isc (A) Pmp (W) Vmp (V) Imp (A) FF
1 0.505 -2.6
2 0.645 -3.65
3 0.623 -3.15
4 0.775 -3.37
CHAPTER 6. PV CELL POWER AND THE EQUIVALENT CIRCUIT MODEL 78

(a) Plot the resulting I versus V curves all for β = 1;


(b) Complete the table by determining the maximum power point Pmp and the corresponding values of
Vmp , Imp and FF .

2. We collect the following data from a solar cell using a sequence of resistors with relatively low resistance
values:
Rex (Ohms) 0.033 0.067 0.102
V (V) 0.075 0.149 0.220
Estimate the value of Isc by fitting a linear function of the form I = aV +b to the data using linear regression
and extrapolating to V = 0.
3. We collect the following data from a solar cell using a sequence of resistors with relatively high resistance
values:
Rex (Ohms) 0.667 0.733 1.000 3.333
V (V) 0.445 0.448 0.457 0.483
Estimate the value of Voc by fitting a linear function of the form V = aI + b to the data using linear
regression and extrapolating to I = 0.
4. Using the values of Voc and Isc determined in two previous problems, plot the resulting I versus V curves
all for β = 1. Then determine the maximum power point Pmp and the corresponding values of Vmp , Imp
and FF .

5. For single-diode PV cell model, the maximum power Pmp is computed using (6.2). Differentiating this
equation above with respect to V and setting the result equal to zero results in a nonlinear equation
for Vmp given by (6.3) which must be solved numerically. For this HW assignment, develop a numerical
procedure (in Excel, MATLAB, Python or any other computing environment/language) that allows you to
(a) Take as input values of Voc (in V), Isc (in A), β and T (K) and compute Io ;
(b) Plot the I versus V curve and the P (power) versus V curve;
(c) Compute Pmp , Vmp and Imp , marking these points on the current and power curves.
6. Given the ideal (β = 1) solar cell diode equation
   
qV
I = −Iph + Io exp −1
kB T

for open circuit voltage Voc = 0.6 V and short circuit current Isc = 10 A,
(a) Determine the photocurrent and dark saturation current parameter values for the model;
(b) Estimate to two significant figures the maximum power (in W) and corresponding voltage Vmp .

7. A PV cell with Voc = 0.5 V and Isc = −5 A under conditions of Rsh → ∞ and Rs = 0 is mounted in an
outdoor location. Weathering of the electrodes causes the series resistance to increase from 0 on the day
the cell is installed to Rs = 3 × 10−5 t Ω where t is the number of days after the installation. Plot graphs
of the peak cell power and corresponding Vmp for 0 ≤ t ≤ 1000 days.
8. A PV cell with Voc = 0.6 V and Isc = −10 A has been manufactured with the perfect characteristics
Rsh → ∞, Rs = 0, and β = 1. The cell over time develops faults corresponding to Rsh = 10 and Rs = 0.02
Ω. Assuming β remains unchanged,
(a) Plot I and power as a function of V making use of the procedure described in class whereby I first is
computed as a function of Vdiode and then is transformed into a function of V .
(b) What are the relevant ranges of V and Vdiode for this problem?
(c) Report the maximum power to two significant digits.
CHAPTER 6. PV CELL POWER AND THE EQUIVALENT CIRCUIT MODEL 79

9. Given the student PV panel measured performance of Voc = 0.5 V and Isc = −2.25 A, what is the efficiency
at T = 298 K if the panel measurements were actually taken at 50o C?
10. A PV panel made up of c-Si cells is found to have Io = 4.2 × 10−10 A m−2 , Iph = 30 A m−2 , Rs = 0,
Rsh → ∞, and β = 1.

(a) Plot I and P versus V ; determine Voc , Isc , Vmp , Imp , and Pmp at T = 298 K.
(b) Compute panel temperature T when the panel is mounted horizontally with a significant air gap
between the panel and the roof. For your thermal energy balance only, assume that EG = 850 W/m2 ,
that 35% of the incoming irradiance is reflected, and that 85% of what remains is absorbed by the
PV cell as heat. The panel/air heat transfer coefficient is h = 20 Wm−2 K−1 . Take the ambient
temperature to be 300 K.
Major clue: Heat transfer to each of the panel sides is Q = Ah(Tamb − T ) W where A is the area of
one side of the panel, T is the panel temperature, and Tamb is ambient temperature.
(c) Evaluate Pmp at your calculated PV cell temperature Tcell in K.

References
[Kano (1998)] Kano, K., Semiconductor Devices, Prentice Hall (1998).
[Krogstrup, et al. (2013)] Krogstrup, P, Jorgensen, H. I., Heiss, M., Demichel, O., Holm, J. V., Aagesen, M.,
Nygard, J. and Fontcuberta i Morral, A. Single-nanowire solar cells beyond the ShockleyQueisser limit,
Nature Photonics, DOI: 10.1038/NPHOTON.2013.32 (2013).

[Möller (1993)] Möller, H. J., Semiconductors for Solar Cells, Artech House, (1993).
[Wenham, et al. (2007)] Wenham, S. R., M. A. Green, M. E. Watt, and R. Corkish, Applied Photovoltaics, 2nd
Ed., Earthscan, UK, (2007).
Chapter 7

External and internal quantum efficiency

7.1 Photon energy and spectral response


Recalling the cell current summary presented in Fig. 6.3, the primary source of the electrical current produced
in a Si solar cell are the minority carriers (electrons in the bottom p-type base, holes in the top n-type emitter)
generated when photons promote an electron from the Si valence band to the Si conduction band - note the
emphasis on Si and not the dopants. The minority carriers produced in this fashion can be swept across the
barrier layer in the junction and out each electrode of the solar cell, producing the photocurrent Iph .
The majority carriers of the electron/hole pairs also can pass through the junction in the form of the recombination
current Irec = Ih,rec + Ie,rec . While they can also make their way out of the electrodes, contributing to the total
current produced by the cell, the primary effect of the recombination current is to generate the diode current
Idiode that counteracts the useful current generated by the photons.
To begin our discussion of the photoprocesses at work in a solar cell, we recall the AM1.5 solar spectrum normalized
for a global irradiance of EG = 1000 W/m2 giving the scaled spectral irradiance EE λ
 
 1000 W/m2 
EE λ (λ) = 
Z ∞
 AM1.5(λ)

AM1.5(λ) dλ
λ=0

These data are plotted in Fig. 7.1.

Figure 7.1: Scaled (for a global irradiance of 1000 W/m2 ) solar flux intensity spectrum. Only photons with
wavelength shorter than λ = 1107 nm (indicated by the vertical line) are absorbed by the Si solar cell.

80
CHAPTER 7. EXTERNAL AND INTERNAL QUANTUM EFFICIENCY 81

We recall the silicon the band gap of Ebg = 1.12 eV and that the energy of a single photon is Ep = hν where
h is Planck’s constant (h = 4.136 × 10−15 eV·s) and that the speed of light is c = 2.998 × 1017 nm/s and so
converting Si band gap to the corresponding photon wavelength:
E c hc
ν= , λ= = = 1107 nm
h ν E
which gives the cutoff point in the spectrum, above which photons have insufficient energy to promote an electron
from the valance to conduction band.
Therefore, photons with wavelength λ = 1107 nm have precisely the right amount of energy to create an elec-
tron/hole pair in the Si solar cell. Photons corresponding to shorter wavelengths have more than sufficient energy
to do this, as seen in Fig. 7.2.

Ep - 1.12 eV

CB
Ep = h!
1.12 eV

VB

Figure 7.2: A closer look at how electrons are promoted from the valence to conduction band in Si, illustrating
the energy loss (hν − 1.12 eV) by photons with wavelength less than 1107 nm (left). The resulting usable spectral
irradiance EE λ,u versus the spectral irradiance EE λ scaled to 1000 W/m2 is shown at right.

Thinking in terms of individual photons, we know that

energy hc 1239.8 eV·nm 1.9864 × 10−16 J·nm


= = =
photon λ λ λ

so if we consider a given value of global irradiance EG in W/m2 at some fixed wavelength λ, provided

hc (eV·s)(nm/s)
λ ≤ λbg = =
Ebg eV

we will have
photons EG power flux W/m2
= = = .
s·m2 hc/λ energy/photon (J·s)(nm/s)/(nm·photon)

7.1.1 Internal quantum efficiency


We define internal quantum efficiency IQE (λ) as the number of electron/hole pairs produced per photon
absorbed by the semiconductor of the PV cell. The IQE is wavelength dependent, with
0 ≤ IQE (λ) ≤ 1 for λ ≤ λbg
IQE (λ) = 0 λ > λbg
For the case where IQE = 1 for values of λ < λbg , we can represent the IQE (λ) by adjusting the upper limit of
integration over the spectral irradiance to correspond to λ = λbg .
CHAPTER 7. EXTERNAL AND INTERNAL QUANTUM EFFICIENCY 82

7.1.2 External quantum efficiency


We define external quantum efficiency EQE (λ) as a modification to IQE (λ):
electron-hole pairs
EQE = (1 − s) [1 − a(λ)] [1 − r (λ)] IQE (λ)
photon
that takes into account the photons reflected from the PV cell top surface r (λ) ∈ [0, 1], absorption of photons by
any coating or cover above the PV cell a(λ) ∈ [0, 1], as well as the shading effect of the front surface contacts
s ∈ [0, 1]. Regarding the latter, up to 10% of the front surface may be obscured by the contact lines; for bare Si,
up to 30% of the incident light may be reflected. Anti-reflection coatings will be studied in more detail later in
this chapter.

7.1.3 Spectral responsivity


The photon energy in excess of 1.12 eV shed by the electron is absorbed by the Si solar cell in the form of
waste heat - only the 1.12 eV of photon energy goes into the useful work of creating the electron/hole pair. To
take into account the fraction of photon energy converted to useful electrical work, we now scale the spectral
irradiance EE λ (λ) spectrum of Fig. 7.1 in the following, wavelength-dependent manner to obtain the maximum
usable spectral irradiance EE λ,u for this solar cell. The key concept of this section is to find a relationship between
the incident global irradiance EG of the solar cell and the photoelectrical current it produces Iph . To accomplish
this, we define the spectral responsivity SR by first computing the maximum usable power we can obtain from
the integrated whole of EE λ :
Z ∞  
Ebg W
EG ,u = EQE (λ) × EE λ (λ)dλ 2 (7.1)
λ=0 hc/λ m
Z λbg  
Ebg
= EE λ (λ)dλ under ideal conditions.
λ=0 hc/λ
Another way to look at this is to compute the maximum photocurrent possible Igen
Z∞
EE λ (λ)
Igen = (q × EQE ) dλ
hc/λ
λ=0
W/(m2 nm)
  
C e-h pair
= 1 (nm)
electron photon (J·s)(nm/s)/(nm·photon)
A
= 2
m
where we emphasize that h has units of J·s. Following [Gray (2003)], we relate this current to Iph through ηcint ,
the internal collection efficiency of the cell
Iph = ηcint Igen .
Unless otherwise noted, we will chose ηcint = 1, thereby rendering Iph and Igen equivalent.
Spectral responsivity SR(λ) is defined by
 
q ne qλ e-h pairs qλ A
SR(λ) = = = EQE with units .
(hc/λ) nph hc photon hc W
Given this definition of SR we can write the usable portion of the spectral irradiance as
Ebg
EE λ,u (λ) = SR(λ) × EE λ (λ)
q
 
Ebg
= EQE (λ) × λ × EE λ (λ) (see equation 7.1)
hc
= EQE (λ) × SR ∗ (λ) × EE λ (λ)
CHAPTER 7. EXTERNAL AND INTERNAL QUANTUM EFFICIENCY 83

with a “dimensionless spectral responsivity” defined by


Ebg λ
SR ∗ (λ) = =
hc/λ λbg

where, of course, h has units consistent with Ebg . We summarize our results as follows
Z∞ Z∞ Z∞
Ebg ∗
EG ,u = SR(λ)EE λ (λ) dλ = SR (λ)EQE (λ)EE λ (λ) dλ = EE λ,u (λ) dλ (W/m2 )
q
λ=0 λ=0 λ=0
Z∞
Igen = SR(λ)EE λ (λ) dλ (A/m2 )
λ=0

These results illustrating EE λ,u are plotted in Fig. 7.2. The conclusion we reach from this analysis is that we can
compute the true theoretical maximum power EG ,u that can be generated by a 1.12 eV band gap Si solar cell by
integrating EE λ,u over the usable portion of the spectrum:
Z 1107
EG ,u = EE λ,u (λ) dλ = 490.7 W/m2
λ=0

noting that the upper limit of integration can be either λbg = 1107 nm or λbg = ∞ because EE λ,u = 0 for
λ > λbg by definition. This gives the theoretical maximum efficiency of a Si solar cell as

490.7 W/m2
ηSi,max = = 49.1%
1000 W/m2

This provides us with another, and potentially better, way of assessing the efficiency of the Gemstone team panel:
by using the ratio of the observed maximum panel power and the theoretically usable flux, we can state that the
panel achieves the following percentage of its maximum theoretical efficiency for a single junction cell:
0.54 W
ηof theo = = 11.2%.
(490.7 W/m )(818/1000)(0.012 m2 )
2

7.2 Theoretical efficiency as a function of Ebg


At this point we can derive a generalized relationship between maximum theoretical efficiency of a single-junction
PV cell and the semiconductor band gap energy value for that cell. In a semiconductor with band gap Ebg eV,
photons below this energy cannot excite a minority carrier across the forbidden gap, and only Ebg of photon
energy goes into the useful work of creating the electron/hole pair. This fact is embodied in our definition of
spectral responsivity; therefore, for some semiconductor with band gap Ebg , the maximum theoretical efficiency
η is computed as
Z hc/Ebg
EG ,u 1
ηmax (Ebg ) = = SR ∗ (λ)EE λ dλ
EG 1000 W/m2 λ=0
where it is important to keep in mind that this limit assumes no refection, shading, or absorption by a cell surface
layer. Furthermore, the impact of the diode current in the actual PV cell is not taken into account.
These results are plotted in Fig. 7.3. We note that optimal band gap values are those between about 0.9 and
1.4 eV; plotting some commonly used semiconductors for PV and other solar energy applications (such as for
photoelectrochemical water splitting reactions), we obtain a good idea of the potential relative effectiveness of
each.
CHAPTER 7. EXTERNAL AND INTERNAL QUANTUM EFFICIENCY 84

Figure 7.3: Maximum theoretical efficiency as a function of semiconductor band gap under AM1.5 conditions.

7.2.1 Multijunction devices


Because of the (at least) linear drop in λ characteristic of the spectral responsivity, the high-frequency portion of
the spectral irradiance is underutilized. To partly make up for this, we can create a PV cell by stacking multiple
layers where the top-most layer (the first to interact with incident light) corresponds to the layer with the highest
band gap. Take for example the tandem cell configuration shown in Fig. 7.4. In this cell, where an amorphous
Si thin film layer is deposited before (and so above with respect to the incident light) an a-SiGe film in this
superstrate design. The band gap of each layer is as follows:

Ebg ,aSi = 1.75 eV Ebg ,aSiGe = 1.3 eV


λbg ,aSi = 708 nm λbg ,aSiGe = 954 nm

1.6 SnO2 a-Si a-SiGe c-Si


1.4
EE W/(m2 nm)

glass 1.2
+ 1.0
TCO (SnO2) 0.8
p a-Si:H 0.6
0.4
0.2
i a-Si:H
200 400 600 800 1000 1200 1400

n/p tunnel junction 1.6 EE


1.4
EE ,EE ,u W/(m2 nm)

1.2 EE ,u,top
i a-SiGe:H 1.0 EE ,u,bot
0.8
0.6
n a-Si:H 0.4
contact
- 0.2
200 400 600 800 1000 1200 1400
Tandem device (nm)

Figure 7.4: Multi (tandem) junction a-Si cell design (left) and spectral irradiance curves (right) indicating the
maximum usable spectral irradiance for the upper and lower layers shown in green and red, respectively.

We also note that a transparent conducting oxide (TCO) layer of tin oxide (SnO2 ) has been deposited above
CHAPTER 7. EXTERNAL AND INTERNAL QUANTUM EFFICIENCY 85

(before) the a-Si layer; the characteristics of that layer are

Ebg ,TCO = 3.5 eV, λbg ,TCO = 354 nm

Given the AM1.5 spectral irradiance EE λ normalized to 1 kW/m2 , we can compute the fraction of each portion
of the spectrum that is absorbed by each of the three layers. If we assume the TCO to be transparent to all
wavelengths above λbg ,TCO
Z λbg ,TCO
ITCO = EE λ dλ
λ=0
= 16.1 W/m2

therefore, less than 2% of the total incident power is absorbed by the TCO film. We can then move on to
computing the fraction of light remaining that can be effectively converted to electron-hole pairs by
Z λbg ,aSi
1 ∗
ηmax,aSi = SRaSi (λ)EE λ dλ
1000 W/m2 λbg ,TCO
= 0.361
Z λbg ,aSiGe
1 ∗
ηmax,aSiGe = SRaSiGe (λ)EE λ dλ
1000 W/m2 λbg ,aSi

= 0.192

and so the maximum total potential efficiency of this tandem junction cell is

ηmax = 36.1 + 19.2 = 55.3%

a substantial improvement over the single-junction a-Si and a-SiGe cells (where the absorption of the SnO2 layer
remains for the sake of direct comparison) of

ηaSi,max = 36.1% ηaSiGe,max = 46.0%

respectively. While the improvement in potential performance is notable, the improvement is partially offset by
increased resistance across the tunnel junction and through the cell itself. However, we note that the most efficient
PV cells that currently exist are multijunction devices; a summary is shown in Fig. 7.5
We take note of two important features in this diagram

1. c-Si solar cells have improved from 13 to 25% over 25 years, but are a long way from the 49.1% potential
efficiency we computed (note that mc-Si cells are limited to slightly over 20%);
2. Both in multijunction and c-Si solar cells, significant improvement is observed when concentrated solar
radiation is used.

7.2.2 A preview of concentrating PV systems


Let us consider a system that concentrates sunlight by a factor of X ; given the diode model parameters Iph = 2.25
A and Io = 8.97 × 10−9 :    
qV
I = −XIph + Io exp −1
βkB T
As before, differentiating the power equation P = (−I )V with respect to V and setting the result equal to zero
results in a nonlinear equation for Vmp
    
qVmp qVmp q
g (Vmp ) = XIph + Io 1 − exp − Vmp Io exp =0 (7.2)
βkB T βkB T βkB T
CHAPTER 7. EXTERNAL AND INTERNAL QUANTUM EFFICIENCY 86

Figure 7.5: NREL reported PV cell efficiencies.

which must be solved numerically for Vmp . Again, we can simplify equation (7.2) using
   
XIph qVoc βkB T XIph
= exp −1 and Voc = ln +1
Io βkB T q Io

to find the same iterated equation for Vmp (equation 6.5)


 
i+1 βkB T i q
Vmp = Voc − ln 1 + Vmp
q βkB T

but with Voc corresponding to the concentrating system. Computing the maximum power point for our system
we find
X 0.2 1 2
Vmp (V) 0.387 0.426 0.443
Imp (A) -0.422 -2.121 -4.252
|Pmp | (W) 0.163 0.904 1.884
|Pmp |/X 0.816 0.904 0.942
As expected, the power increases with solar concentrating ratio, be we also observe the efficiency likewise increases.

7.2.3 Splitting the spectrum


Let us consider a semiconductor such as a-Si:H that can be deposited as a thin-film on non-planar surfaces. We
take the band gap of the amorphous Si to be Ebg = 1.75 eV resulting in λcrit = 708.5 nm. As we see in Fig. 7.6,
the portion of AM1.5 corresponding to energies too low to be absorbed by the semiconductor (hν < hνcrit )
constitutes over half of the total solar power. Of that which remains, at most 37% can be potentially used by a
PV cell and at least 12% will ultimately end up as waste (low-grade) heat.
CHAPTER 7. EXTERNAL AND INTERNAL QUANTUM EFFICIENCY 87

1.6 AM1.5
AM1.5u
1.4 h < h crit
crit
1.2

1.0 PV 368 W/m2


EE W/(m2 nm)

thermal 513 W/m2


0.8 low-grade 119 W/m2
total 1 kW/m2
0.6

0.4

0.2

0.0 500 1000 1500 2000 2500


(nm)

Figure 7.6: Splitting the spectrum into its thermally and PV-useful components.

To address the loss of the thermal portion of the irradiance spectrum (λ > λcrit ), we propose the hybrid thin-film
PV/concentrating thermal solar system illustrated in Fig. 7.7. The key idea behind this approach is to coat the
reflecting surface with a thin-film PV layer; this PV cell will generate electricity directly from the λ ≤ λcrit portion
of the spectrum and will be transparent to the longer-wavelength portion of the spectrum. The IR portion of the
spectrum then could be focused on a thermal collector for solar heating or generating electricity, such as by using
a Stirling engine.

Figure 7.7: A potential PV-thermal con-


centrating system.

waste
heat I load

The challenges posed by such as system include engineering the thin-film PV/reflector (Ebg ), systems integration,
optimization, and (static) concentrator design. However, the potential benefits of this technology include efficient
combined power + heating or power + water desalination applications. More information on these technologies
can be found in [Charalambousa, et al., (2007)] and [Mittelman, Kribus, and Dayan (2007)].
CHAPTER 7. EXTERNAL AND INTERNAL QUANTUM EFFICIENCY 88

x = 1.33 x =0
Si N Si N
N H H Si N H
Si Si Si Si
H
N Si Si Si H Figure 7.8: SiAmorphousSi
SiNx :H
H Si Si H N
N corresponding to a stoichiometric
N
Si Si Si N H Si Si
Si Si Si N
N H H Si Si Si films (left) and
N the limitingH case of
H N N H Si H N N H
Si Si N Si Si Si Si Si a pure amorphous Si film
N (right).
Si
N Si Si H Si
N Si
H N N HH N
H N
N H H
N

7.3 Antireflection coatings


Anti-reflection (AR) films are deposited on the top-most surface of the PV cell relative to the incident solar
radiation (see Fig. 6.3). While AR films can be created from a number of transparent materials, we will limit our
focus to silicon nitride, a tough and inert film readily deposited by plasma-enhanced chemical vapor deposition
[Adomaitis and Schwarm (2013)]. Crystalline silicon nitride is written with the chemical formula Si3 N4 ; because
the anti-reflection films are deposited in amorphous form and contain a significant amount of hydrogen, the films
are denoted as SiNx :H with x = 1.33 corresponding to a stoichiometric N/Si ratio. A depiction of the film
molecular structure as a function of composition is given in Fig. 7.8.

7.3.1 AR film optical properties


AR film optical and electrical properties vary strongly with film composition x. In [Adomaitis and Schwarm (2013)],
models correlating film composition to refractive index n1 (x) and band gap Ebg (x) were developed from pub-
lished experimental data. To summarize those results, AR film refractive index n1 was shown to display a linear
compositional dependence (proportional to 1/x = Si/N) giving the empirical relationship.

0.65
n1 = + 1.3 (7.3)
x
Of course (7.3) becomes invalid in the limit of pure amorphous Si (x → 0) where the true index of refraction is
between 3.5 and 4 [Janai, et al., (1979)].
AR film band gap has been shown to decrease with increasing Si content from 5.3 eV for a stoichiometric film to
1.8 eV for amorphous Si. Assuming a linear relationship, [Adomaitis and Schwarm (2013)] found

Ebg ,1 = 1.8 + 2.63x (eV). (7.4)

7.3.2 Reflectance modeling


Given a film thickness w , a range of wavelengths λ, and n0 , n1 , and n2 as the refractive indices of the surrounding
medium, AR film, and Si, respectively, we begin our discussion of reflectance by recalling that the light incident
to the solar cell surface generally will have components corresponding to both the S and P polarizations. The P
component of the incident light is a plane wave perpendicular to the reflecting surface and the S component is
parallel. The Fresnel equations describe the reflectances of each component as a function of the angle of incidence
θi ; for the case of a single reflecting surface where the incident beam originates in the medium corresponding to
CHAPTER 7. EXTERNAL AND INTERNAL QUANTUM EFFICIENCY 89

I
h!
r01

(a) r12
e-
n0 n0
n1
(b) -
n1
n2 n2 h+ e-
w
Ie,ph
n-type
+ + + + (c) +T
- - - - h+ e- -
h+filmeregion
Figure 7.9: An enlargement of the AR - p-type
(left) and thin-film reflectance (right)
(b)
Ih,ph h+ e-
n0 and reflects off the surface of material with n1 ,
2 2
+
2
 s   s  2
n0 n0
 n0 cos θi − n1 1 − sin θi  n0 1 − − n1 cos θi 
sin θi
n1 n1

   
rs = 
 s  2  ,
 rp =  s

 2


 n0   n0 
n0 cos θi + n1 1 − sin θi n0 1 − sin θi + n1 cos θi
n1 n1

where the total reflectance r is simply the mean of the two components
rp + rs
r= .
2
For θi = 0 the Fresnel equations reduce to
 2
n1 − n0
rs = rp = .
n1 + n0

Reflection for the case of a thin film

Returning to the AR film of Fig. 7.9, we can write the single-interface reflectances [Melles Griot (2009)] as
 2  2
n1 − n0 n2 − n1
r01 = , r12 =
n1 + n0 n2 + n1

(see Fig. 7.9 for an enlargement of the AR film region). At this point it is interesting to consider the (idealized)
case where no film exists; under these conditions n0 = 1 and n2 = 3.85 (for Si) and so
 2  2
n2 − n0 3.85 − 1
r02 = = = 35% reflectance.
n2 + n0 4.85

While this, at first, may appear to be a high value, we note that this case corresponds to a polished Si surface
with no surface oxide coating and no P doping.
To determine the overall reflectivity, the first and second-surface reflected wave phase difference is defined by

φd = n1 (2w )
λ
CHAPTER 7. EXTERNAL AND INTERNAL QUANTUM EFFICIENCY 90

Figure 7.10: Reflectance r as a function of λ for the case x = 1 and w = 200 nm.

where we incorporate the factor 2 to account for the length of the path light incident to the AR film of thickness w
takes through the film and back after being reflected at the AR-Si interface1 . Setting φd = π results in maximum
destructive interference of light at wavelength λ corresponding to a film thickness equivalent to one quarter of
the wavelength λ divided by the refractive index of the AR film n1 :
λ
w= .
4n1 (x)
The final form of total reflectance is
2 2
r̂01 + r̂12 + 2r̂01 r̂12 cos φd
r (w , x, λ) = 2 2 + 2r̂ r̂ cos φ (7.5)
1 + r̂01 r̂12 01 12 d

with
n1 − n0 n2 − n1
r̂01 = , r̂12 =
n1 + n0 n2 + n1
a model for reflectance that is valid in the limit of w → 0 (proof is left to the reader). For our silicon nitride AR
film
n0 = 1 for air
n1 = 0.65/x + 1.3 for SiNx :H
n2 = 3.85 for Si
An example of the reflectance as a function of wavelength for a film with thickness w = 200 nm and composition
x = 1 (n1 = 1.95) is shown in Fig. 7.10. For this relatively thick film, we observe the 1/4 wavelength minimum
at λ = 1560 nm. Within the visible and UV portions of the spectrum we find minima at λ = 520, 312, 223 nm
and more values at higher frequencies.

7.3.3 AR film absorbance


Light absorption in thin films generally is modeled as an exponential function of film thickness with a wavelength-
dependent absorption coefficient. We will model absorption in a highly simplified manner as a high-frequency
filter, with the cut-off wavelength λa defined by the x-dependent band gap:
hc
λa (x) = .
Ebg ,1 (x)
1 This and (7.5) which follows are slightly different from the form of the equations in [Melles Griot (2009)] but ultimately produce

the same result.


CHAPTER 7. EXTERNAL AND INTERNAL QUANTUM EFFICIENCY 91

Therefore
1 − a(λ) = 1 − H(λ − λa ) = 1 − IQE (Ebg ,1 )
where H is the Heaviside function.
Contact shading is proportional to the area covered by the front surface contact; we will use a constant value
of s = 0.1 (10%) unless otherwise noted.

7.3.4 Integrating the elements


Consider an AR film with x = 1 and w = 200 nm and AM1.5 normalized to 1 kW/m2 . We recall that there are
minimum reflectance values at λ = 310 and 520 nm and that the dimensionless spectral responsivity is defined
by SR ∗ = (Ebg /hc)λ. For this example we use the band gap of silicon Ebg = 1.12 eV.
Results demonstrating the reduction of the AM1.5 spectrum as a result of SR ∗ , reflectance, AR film absorbance,
and contact shading are shown in Fig. 7.11. In this figure, we clearly see the increasing effect of SR ∗ with
decreasing λ and the two minima in the reflectance dependency on λ. Integrating under the green curve of the
figure gives the maximum value of Iph possible:
Z ∞
Iph,max = SR(λ)EE λ (λ) dλ (7.6)
0
Z 1107
1
= (1 − s) [1 − a(λ)] [1 − r (λ)] EE λ (λ)SR ∗ (λ) dλ
Ebg 280
1
= 339.5 W/m2 Ebg in eV
Ebg
= 303 A/m2

where the Ebg preceding the integral has units of eV and SR ∗ is dimensionless. To determine cell peak power
requires knowing Voc ; an approximation is described in the following section.
We note that a number of other issues remain to be discussed, including (1) reduction of reflections using surface
roughening by etching; (2) modeling the AR film over these roughened surfaces; (3) front surface contact layout
optimization; (4) buried and laser-etched contact designs; and (5) back-surface roughening.

7.4 The Shockley-Queisser limit


In the previous section, we showed that a single-junction Si solar cell should be able to convert up to 49.1% of
the incident AM1.5 radiation to useful electrical power. An important factor missing, however, from that analysis
is the voltage-current characteristics of the p-n junction of the PV cell.
In our previous discussions, we have seen that Voc for an Si cell typically is limited to approximately 0.6 eV, a
substantial difference from the 1.12 eV band gap of Si. If we recall our discussion of the unbiased p-n junction, a
thermally-generated current through the junction is offset by the recombination current; we see the reduction of
Voc can be partly attributed to operating the junction at room-or-above temperature, instead of 0 K.
[Schockley and Queisser (1961)] accounted for this effect in their analysis of the efficiency limit of solar cells;
a detailed discussion of how they achieve a maximum efficiency of η ≈ 0.3 for a Si solar cell is beyond the
scope of these notes. However, as a simplified means of comparison to their analysis, we start by computing
the maximum value possible of the PV cell photocurrent Iph from (7.6). We know that for an ideal PV cell the
dark saturation current Io then can be computed using Voc ; [Krogstrup, et al., (2013)] presented a simpler to
understand interpretation:

4n2
   
T kB T Ωemit
Voc = Ebg 1 − − ln + ln 2 − ln(QE )
Tsun q Ωsun X
CHAPTER 7. EXTERNAL AND INTERNAL QUANTUM EFFICIENCY 92

1.6
EE
EE ,u
1.4 (1-r)EE ,u
(1-s)(1-r)EE ,u
1.2
power W/(m2 nm)

1.0

0.8

0.6

0.4

0.2

200 400 600 800 1000 1200 1400


(nm)

Figure 7.11: Spectral irradiance and the effects of spectral responsivity SR ∗ , reflectance, AR film absorbance,
and contact shading. We note that [1 − a(λ)] does not appreciably affect the useful power because the lower
spectral cut-off wavelength λbg = 280 nm for the AR film is essentially completely below the power spectrum in
this example.

where in our interpretation Ebg is in eV and kB in J/K, Ωemit and Ωsun are the solid angles of photons radiated
and absorbed (in steradians), n is the PV semiconductor refractive index, X is the solar concentration ratio and
QE is the emission quantum efficiency. After simplifying to
kB T 4n22
 
T
Voc = Ebg 1 − − ln ,
Tsun q X
using the refractive index of Si, and setting X = 1, the effective Voc is computed and plotted in Fig. 7.12.
While the reduction of Voc relative to Ebg appears small in Fig. 7.12, when combined with the diode model, the
effect of maximum efficiency is significant. Computing
   −1
qVoc
Io = Iph,max exp −1
kB T
and then using the (ideal) diode model to determine Pmp , the resulting maximum theoretical single-junction PV
cell efficiency for one sun under AM1.5 conditions is shown in Fig. 7.13. We see that the maximum efficacy now
is 36.8% and that there is a slight shift to the higher Ebg materials, making the maximum theoretical efficiency
of CdTe cells competitive with c-Si.
CHAPTER 7. EXTERNAL AND INTERNAL QUANTUM EFFICIENCY 93

Figure 7.12: Prediction of maximum theoreti-


htb
cal Voc using an approximation to the Schockley-
Queisser limit

Figure 7.13: Maximum theoretical efficiency as a function of semiconductor band gap under AM1.5 conditions; curve
corresponding to Pmp is shown in red.

HW and review problems


1. How many photons/sec fall on 1 cm2 of a surface illuminated by a monochromatic 1 kW/m2 radiation
source of wavelength 500 nm?
2. For the system above, what is the answer when the radiation is received at an angle 15o from vertical?
3. A solar cell made from a semiconductor with 2.0 eV band gap is exposed to monochromatic radiation of
wavelength 400 nm; what fraction of the radiation can be converted to useful electrical energy?
4. Repeat the calculation above if the wavelengths are uniformly distributed between 400 and 700 nm.
5. A PV cell created from a semiconductor with band gap 1.4 eV is coated on its top (light receiving) surface
with another semiconductor with band gap 2.5 eV. Compute the maximum potential fraction of incident
radiation converted to useful power if this cell is illuminated by light with wavelengths uniformly distributed
between 400 and 700 nm. Assume the top semiconductor layer does not contribute to electrical power
production.
6. Consider the multijunction solar cell described in this chapter. For a cell with the same TCO and upper
a-Si junction (with band gap Ebg ,Si = 1.75 eV), determine the optimal band gap of the lower junction in
terms of maximum electrical power production.
CHAPTER 7. EXTERNAL AND INTERNAL QUANTUM EFFICIENCY 94

7. For a single interface with n0 = 1 and n1 = 3.85, determine the total reflectance when the incident radiation
is normal and 45o to the surface.
8. Using the single-film reflectance equation (7.5) and n0 = 1, n2 = 4

(a) Determine the optimal values for n1 and w for an Si cell using the AM1.5 spectrum.
(b) What is the limit of (7.5) for w → 0?
(c) For n1 = 2 and w = 100 nm, determine the location of the reflectance minima for 0 < λ < 2000 nm.
What is the physical interpretation of the minima?
9. Recall the section in this chapter where a concentrating solar system design was discussed which effectively
split the AM1.5 spectrum into the IR and visible/UV components and then used the spectrum fractions in
thermal and PV systems, respectively. If the thermal portion is to be used as the heat source for a 25%
efficient Stirling engine, what is the optimal PV band gap? To solve the problem, create a plot of overall
efficiency vs. Ebg .
10. The star Betelgeuse in the constellation Orion has a surface temperature of 3500 K and therefore is cool
relative to our Sun. Someone claiming to be from that star’s planetary system would like to evaluate the
potential for generating solar energy using a single-junction PV cell under conditions of zero atmospheric
absorption. For this problem,
(a) Assume the spectral irradiance is approximated by Planck’s distribution normalized to 1 kW/m2 over
λ ∈ [200, 4000] nm; plot the spectral irradiance and estimate the wavelength corresponding to peak
irradiance.
(b) Assuming ideal internal and external quantum efficiencies, compute the maximum fraction of the 1
kW/m2 irradiance that can be converted to useful power for PV cell semiconductor band gap values
Ebg = 0.75, 1, 1.25, and 1.5 eV.
(c) Estimate to two significant figures the optimal band gap value (in eV) and the peak theoretical efficiency
at that value; do not simply interpolate between two values of (b).

References
[Adomaitis and Schwarm (2013)] Adomaitis, R. A. and A. Schwarm, Systems and control challenges in
photovoltaic manufacturing processes – A modeling strategy for passivation and anti-reflection films,
Computers & Chemical Engineering 51 65-76 (2013).
[Charalambousa, et al., (2007)] Charalambousa, P. G., G. G. Maidmenta, S. A. Kalogiroub, and K.
Yiakoumettia, Photovoltaic thermal (PV/T) collectors: A review, Applied Thermal Engineering 27 275-286
(2007).
[Janai, et al., (1979)] Janai, M., D. D. Allred, D. C. Booth, and B. O. Seraphin, Optical properties and
structure of amorphous silicon films prepared by CVD, Solar Energy Materials 1 11-27 (1979).
[Mittelman, Kribus, and Dayan (2007)] Mittelman, G., A. Kribus, and A. Dayan, Solar cooling with
concentrating photovoltaic/thermal (CPVT) systems, Energy Conversion and Management 48 2481-2490
(2007).
[Gray (2003)] Gray, J. I., Chapter 3: The physics of the solar cell. In Handbook of Photovoltaic Science and
Engineering, A. Luque and S. Hegedus (Ed.), John Wiley & Sons, Ltd (2003).
[Krogstrup, et al., (2013)] Krogstrup, P. and coworkers, Single-nanowire solar cells beyond the
Shockley-Queisser limit, Nature Photonics DOI: 10.1038/NPHOTON.2013.32 (2013).
CHAPTER 7. EXTERNAL AND INTERNAL QUANTUM EFFICIENCY 95

[Schockley and Queisser (1961)] Shockley, W. and H. J. Queisser Detailed balance limit of efficiency of p-n
junction solar cells, J. Appl. Phys. 32 510-519 (1961).
[Melles Griot (2009)] Melles Griot, Technical Guide, Vol. 2, Ch 5 (2009).
Chapter 8

Cell and panel interconnections

8.1 Identical cells


Consider the single solar cell equivalent circuit shown in Fig. 8.1 and the simplified notion used to represent the
entire circuit. Each of these cells can be described by the diode equation (6.6):
   
Vi − Ii Rs Vi − Ii Rs
Ii = −Iph + Io exp q −1 + .
βkB T Rsh

In our previous discussions, V denoted both the voltage rise across the PV cell as well as the voltage drop though
the external load. Because we now will consider cases of multiple cells wired together, a modified interpretation
of Vi on a per-cell basis will be required and is shown in Fig. 8.1.

A
- -
Vi = voltage drop

Iph n-type Idiode Rs

Vi Ii
p-type Rsh Ish
+ +
B
Ii < 0
Figure 8.1: Notation used to represent a single solar cell.

The two simplest multi-cell configurations are shown in Fig. 8.1. Before we proceed with the analysis, we recall
Kirchhoff’s Current and Voltage Laws [Johnson et. al (1978)]:

1. In a circuit, the algebraic sum of currents entering any node is zero;


2. The algebraic sum of the voltages around any closed path is zero.

Let us begin with the case of two identical cells mounted in parallel, connected to an external circuit consisting
of a resistor with resistance R. We denote the currents I1 , I2 and voltages V1 and V2 as those corresponding to
cells 1 and 2, respectively, and V as the voltage difference across the external load of Fig. 8.1. Making use of
Kirchhoff’s Laws we find:

96
CHAPTER 8. CELL AND PANEL INTERCONNECTIONS 97

-
V/N 1 I
Figure 8.2: Two identi-
V1 1 V2 2 V V cal cells arranged in par-
I1 I2 R R allel (M = 2, left) and
series (N = 2, right).
2
+ I/M I

+ I

1. Given that the conventional current of each cell is denoted downward (hence negative in our sign convention),
at the lower node where the two cells connect we have I = I1 + I2 , therefore the total current I to the
external load is the sum of the individual cells.
2. There are three unique closed circuit paths in this case: it is easy to see under these circumstances that the
voltage rise (from the top to bottom) of each cell V −!, V2 and the voltage drop across the external load
V all are equal.

Therefore, for M identical cells in parallel, V = V1 = V2 = · · · = VM and


     
V − (I /M)Rs V − (I /M)Rs
I = M −Iph + Io exp q −1 + .
βkB T Rsh

Now consider the case of two identical cells mounted in series (Fig. 8.1, right), again connected to an external
circuit consisting of a resistor of resistance R Ω. In this case, Kirchhoff’s Laws give the following relationships
between the individual cell characteristics and the two-cell string performance:

1. Because any point in the loop is effectively a node, the current must be constant through the circuit, and
so I = I1 = I2 ;
2. There is only one loop in this circuits and so V1 + V2 = V .

Therefore, for N identical cells in series, V = V1 + V2 + · · · + VN = NV1 and


   
V /N − IRs V /N − IRs
I = −Iph + Io exp q −1 + .
βkB T Rsh

8.1.1 Rectangular arrays of cells


Let us consider an array of M parallel rows of PV cells, with each row containing a string of N cells connected
in series. This matrix arrangement of PV cells constitutes a basic cell array – whether we actually would won’t
to wire PV cells in this manner will be examined more closely later in this chapter. A representative M × N cell
array is shown in Fig. 8.1.1. Note the manner in which the individual cells are soldered, particularly how the top
contact of one cell is connected to the back-side contact of the next cell in a series string. If each cell in the array
performs identically and according to (6.6), the array performance model becomes
     
V /N − (I /M)Rs V /N − (I /M)Rs
I = M −Iph + Io exp q −1 + (8.1)
βkB T Rsh
CHAPTER 8. CELL AND PANEL INTERCONNECTIONS 98

N
M

A Figure 8.3: An M × N solar cell array where


B each row of cells is wired in series and the rows
+ are connected in parallel.
-
I

where it is important to stress that I and V correspond to those values measured for the entire array, not a
representative cell in the array.
For the case Rs = 0 and Rsh → ∞, we can immediately write the array open circuit voltage Voc and array
short-circuit current Isc as:
 
NβkB T Iph
Voc = ln +1
q Io
Isc = −MIph .

For cases where Rs 6= 0 and Rsh finite, we must use a Newton or other iterative procedure to compute the solutions
to the nonlinear equations defining Voc , Isc , and I (V ).

8.1.2 Thin film commercial panels


Recall the discussion of thin-film a-Si:H PV cell design and fabrication Chapter 5 of these notes. A manufacturing
advantage thin-film cells have over their crystalline counterparts is that large-scale thin-film cells used in com-
mercial panels can be broken into segmented strips - see Fig. 8.4. This integrated manufacturing method makes
possible single panels that are rated for 12, 24, and 48 V for domestic and smaller-scale applications, and greater
than 10 V for larger-scale applications.

glass superstrate

TCO
a-Si:H
metal

Figure 8.4: Integral interconnected panel design for a superstrate thin-film PV cell array.

Another appealing aspect of a-Si:H panels is their utility in Building Integrated Photovoltaics (BIPV). An excellent
overview is given by [Schade (2004)] but will not be described further in these notes.
CHAPTER 8. CELL AND PANEL INTERCONNECTIONS 99

8.1.3 Commercial crystalline Si PV panels

Figure 8.5: A commercial c-Si solar panel (left) and panel performance specifications (right).

Consider the commercial c-Si PV panel shown in Fig. 8.5. The panel consists of 36 monocrystalline Si solar cells
mounted on an aluminum backing plate and the overall panel is sealed for outdoor use. We read off the following
performance data from the c-Si PV panel:

Vmp = 18 V Imp = −1.67 A Voc = 21.6 V Isc = −1.78 A

We note Pmp = 30 W is given, but this data point is redundant because Vmp and Imp are listed on the panel
specifications.
Because Voc = 21.6 V for the 36 cells of this panel and because we know that individual junction Voc values are
normally between 0.5 and 1 V, we can assume (and verify by visual inspection) that all 36 cells in this panel are
wired into a single series string. Recalling the PV array modeling equation (8.1) this gives N = 36 cells mounted
in series and M = 1. Given these are high-quality commercial cells, we will initially assume Rs to be small but
nonzero (on a per-cell basis), Rsh is large but finite, and β = 1. We note the panel specifications were determined
under AM1.5 conditions at T = 298 K.
These assumptions and observations lead to the panel modeling equation:
   
V /N − IRs V /N − IRs
I = −Iph + Io exp q −1 + . (8.2)
kB T Rsh

The model has the four unknowns: Iph , Io , Rs , and Rsh having fixed β = 1. We use the panel open- and
short-circuit data as well as the maximum power data to define these parameters in (8.2):
   
qIsc Rs Isc Rs
Isc = −Iph + Io exp −1 −
kB T Rsh
   
qVoc Voc
0 = −Iph + Io exp −1 +
NkB T NRsh
which gives
Isc + Isc Rs /Rsh + Voc /(NRsh )
Io =    
qIsc Rs qVoc
exp − exp
kB T NkB T
CHAPTER 8. CELL AND PANEL INTERCONNECTIONS 100

and Iph from one of the previous two equations.


Taking Rsh = 10 Ω (sufficiently large to make no real difference in the solutions) and Rs = 0.015 Ω, we compute
Iph = 1.783 A and Io = 1.227×10−10 A. The I versus V curve is computed much in the same manner as was done
when previously describing parasitic resistances and Vdiode (equation 6.9): defining Vdiode on a per-cell (junction)
basis
V
Vdiode = − IRs
N
so    
Vdiode Vdiode
I = −Iph + Io exp q −1 +
kB T Rsh
and so we can choose a range of values Vdiode ∈ [−Isc Rs , Voc /N], compute I , and then compute the corresponding
V from
V = (Vdiode + IRs )N.
We plot the results for I and panel power and mark the Imp , Vmp , and resulting Pmp for the model predictions as
well as those specified by the panel manufacturer in Fig. 8.6. Note how full panel voltage, current, and power are
used for the plots rather than presenting the results on a per-cell basis. As can be seen in the plots, the match
between the specified and computed maximum power points is nearly perfect, as are Isc and Voc .

0.0
computed
0.5 mp specs
mp computed
Voc specs
I (A)

1.0
Isc specs
1.5

0 5 10 15 20
30
25
computed
mp specs
20 mp computed
|P| (W)

15
10
5
00 5 10 15 20
V (Volts)

Figure 8.6: c-Si PV panel current (top) and power (bottom) as a function of external voltage V .

8.2 Shaded, faulty, or otherwise nonuniform cells


Now we consider the arrangements described in Fig. 8.1 but now where one of the cells is damaged or shaded.
The underperforming cells are shown as red in Fig. 8.2.
CHAPTER 8. CELL AND PANEL INTERCONNECTIONS 101

-
- V1 1 I Figure 8.7: Two iden-
V1 1 V2 2 V + V tical cells arranged in
I1 I2 R R parallel (left) and se-
ries (right) where the
+ V2 2 damage/shaded cells are
I shown as red.

8.2.1 Shaded cells: parallel-cell case


Let us start the analysis by considering the cases where we have two cells wired in parallel and in series. If cells
labeled “2” in Fig. 8.2 are in the shade while the other of each circuit is fully illuminated, Iph,1  Iph,2 and so we
consider the case Iph,2 = 0 as a simplification. With two identical cells mounted in parallel, Kirchhoff’s Voltage
Law again results in V = V1 = V2 . Therefore, summing the currents at the (lower) node where the two PV cells
are joined,
   
V − I 1 Rs V − I1 Rs
I1 = −Iph,1 + Io exp q −1 +
βkB T Rsh
   
V − I2 Rs V − I2 Rs
I2 = Io exp q −1 +
βkB T Rsh
I = I1 + I2 .

To compute I1 and I2 we will use a Newton procedure to compute the two currents directly as a function of V
instead of the procedure based on using Vdiode ; we chose this approach to simplify the numerical calculation of
I = I1 + I2 in that the Newton-based procedure guarantees the values of V used to compute I1 and I2 will match.
Of course, the alternative is to define Vdiode,1 = V − I1 Rs and Vdiode,2 = V − I2 Rs to compute I1 and I2 followed
by interpolation of one set of current values to the voltage values of the other prior to computing the sum.
Results are shown in Fig. 8.8, where the four subplots illustrate current I (top plots) and power P (bottom plots)
as a function of external voltage V . The plots in the left column of the figure correspond to two cells wired in
parallel. We observe that the fully illuminated pair of cells produce approximately twice the current (as expected)
and twice the overall power. However, examining the results for the full 36 cell panel with all cells wired in
parallel, we see that the overall impact of a single shaded cell is diminished as the number of illuminated cells
grows relative to the shaded cells. Therefore, an important conclusion can be drawn from this result: for a long
string of parallel cells or solar panels, one dark cell/panel in a string of many illuminated cells/panels results in
little performance degradation.

8.2.2 Shaded cells: series case


In the case of cells arranged in series, Kirchhoff’s Laws give V = V1 + V2 and I1 = I2 = I , and so
   
V1 − IRs V1 − IRs
I = −Iph,1 + Io exp q −1 +
βkB T Rsh
       
V2 − IRs V2 − IRs V − V1 − IRs V − V1 − IRs
= Io exp q −1 + = Io exp q −1 +
βkB T Rsh βkB T Rsh
CHAPTER 8. CELL AND PANEL INTERCONNECTIONS 102

0.0 0
0.5 10
1.0 20
1.5 30
2.0
I (A)

2.5 40
3.0 50
3.5 60
4.00.0 0.1 0.2 0.3 0.4 0.5 700.0 0.1 0.2 0.3 0.4 0.5
2.0 30
25
1.5
20
P (W)

1.0 15
10
0.5
5
0.00.0 0.1 0.2 0.3 0.4 0.5 00.0 0.1 0.2 0.3 0.4 0.5
V (Volts) V (Volts)

Figure 8.8: Current (top) and power (bottom) versus voltage for two cells arranged in parallel (left) and 36 cells in parallel
(right). Cases where all cells are illuminated are shown as the dotted blue curves; cases with a single shaded cell is shown
in red.

For the open-circuit case


   
Voc,1 Voc,1
0 = −Iph,1 + Io exp q −1 +
βkB T Rsh
   
Voc,2 Voc,2
= Io exp q −1 +
βkB T Rsh

results in Voc,2 = 0 and our familiar iterative relationship for Voc,1 :


 
βkB T Iph,1 Voc,1
Voc,1 = ln 1 + −
q Io Io Rsh

For the short-circuit limit


   
Vsc,1 − Isc Rs Vsc,1 − Isc Rs
Isc = −Iph,1 + Io exp q −1 +
βkB T Rsh
   
−Vsc,1 − Isc Rs −Vsc,1 − Isc Rs
= Io exp q −1 +
βkB T Rsh

For cases where |Isc Rs | is small, Vsc,1 can be found using a Newton procedure (as opposed to the Newton-Raphson
necessary to compute Vsc,1 and Isc simultaneously) to find that Isc is exceedingly small in this situation and that
Vsc,2 ≈ −Voc,1 meaning that what little power is produced by cell 1 is dissipated in the darkened cell in this
situation.
CHAPTER 8. CELL AND PANEL INTERCONNECTIONS 103

To compute I versus V for the two-cell system for V ∈ [0, Voc ], Kirchhoff’s Laws give V = V1 +V2 and I1 = I2 = I ,
and so
   
V1 − IRs V1 − IRs
I = −Iph,1 + Io exp q −1 + (8.3)
βkB T Rsh
   
V2 − IRs V2 − IRs
= Io exp q −1 + (8.4)
βkB T Rsh
   
V − V1 − IRs V − V1 − IRs
= Io exp q −1 + (8.5)
βkB T Rsh
This give two equations (8.3,8.5) and 2 unknowns, I and V1 , the latter which can be determined using a Newton-
Raphson technique.

0.0 0.0

0.5 0.5

1.0 1.0
I (A)

1.5 1.5

2.00.0 0.2 0.4 0.6 0.8 1.0 2.00 5 10 15 20


0.10 30
0.08 25
20
0.06
P (W)

15
0.04
10
0.02 5
0.00 00
0.0 0.2 0.4 0.6 0.8 1.0 5 10 15 20
V (Volts) V (Volts)

Figure 8.9: Current (top) and power (bottom) versus voltage for two cells arranged in parallel (left) and 36 cells in parallel
(right). Cases where all cells are illuminated are shown as the dotted blue curves; cases with a single shaded cell is shown
in red.

A plot of power and current versus voltage is given in Fig. 8.9 for the series-wired case, comparing the fully
illuminated and single-shaded cell performance curves. What is striking is how severe the performance loss is with
a single shaded cell: nearly all the two-cell PV string power is dissipated in the shaded cell, and for the case of 36
cells with one shaded, we observe that the total power output is reduced to 10 W, one-third the power of the fully
illuminated panel. We note that because so much power is potentially dissipated by the shaded cell, significant
heating and ultimate damage of the shaded cell can result.

8.2.3 Experimental validation


As seen in Fig. 8.9, it almost seems remarkable that shading a single cell out of 36 cells wired in series can reduce
the overall performance of the string to approximately one-third the full-sun performance. To examine the validity
CHAPTER 8. CELL AND PANEL INTERCONNECTIONS 104

0.0 Full sun, no shading

0.5

I (A)
1.0

1.5

0 5 10 15 20

30
25
20

P (W)
15
10
5
00 5 10 15 20
V (V)

0.0 1 cell dark

0.5

I (A)
1.0

1.5

Figure 8.10: Experimental setup illustrating how one 0 5 10 15 20


single cell was shaded; resistors and multimeter used 10
to determine measured V and I values also are shown 8
(above). Full sun (top right) and single shaded cell 6
P (W)

(bottom right) current and power versus voltage plots. 4


Experimentally measured data are shown with filled cir- 2
cles.
00 5 10 15 20
V (V)

of this model prediction, a set of experiments was performed to assess the performance of the panel shown in
Fig. 8.5. With the panel positioned at optimal tilt on a day when the global irradiance at solar noon corresponded
to EG ≈ 1000 W/m2 , the fully exposed panel data plotted in Fig. 8.10 was recorded and shown to be in good
agreement with our model predictions. Likewise, data corresponding to a single shaded cell showed the remarkable
agreement between the observations and the performance loss predictions.

8.2.4 Bypass and blocking diodes


As seen in Fig. 8.11, bypass diodes are used to prevent hot spot formation by allowing current to flow around
shaded cells or panels [Wenham and co-authors (2007)]. Bypass diodes are placed in parallel to strings of PV
cells wired in series. Under normal operation, the PV cells reverse-bias the diode and so no current flows through
the diode. However, when one or more of the cells in the string become shaded or defective, the high resistance
forward-biases the diode allowing current to flow, bypassing the shaded/damaged string of cells. Generally, groups
of 10-15 cells are protected by a single bypass diode; in many circumstances, the diodes are built into the solar
panels.
Blocking diodes are used to prevent the reverse flow of current through a dark solar panel and are particularly
important in battery charging applications to prevent battery discharge at night. Because of the resistance of the
diode ([Wenham and co-authors (2007)] cite a 0.6 V drop across the diode), these diodes should only be used
when necessary.
CHAPTER 8. CELL AND PANEL INTERCONNECTIONS 105

- I

V 1 I 2
Vex
+ Vex
Rex
Rex
I 2 V 3 I

Iex Iex
4

Figure 8.11: Bypass diodes.

8.3 System integration issues

battery DC
bank load
charge
controller
= AC
~ load
inverter

building AC
= load
~
inverter utility
grid

Figure 8.12: Stand-alone (top) and and grid-connected PV (bottom) systems. After Figs 15.8 and 15.9 of
[Archer and Hill (2001)].

The components of stand-alone and grid-connected PV systems are shown in Fig. 8.12. There are a number of
excellent sources of information on the practical aspects of PV system design, some of which are listed in the
references of this chapter. We will only briefly discuss the elements of systems design and several general design
rules.

• As a general design rule, peak installed power should be 5 times the nominal load; this rule is taken from
from page 124 of [Wenham and co-authors (2007)].
• Panel tilt should be set at the local latitude for grid-connected systems and latitude plus 15 degrees for
stand-alone systems if possible [Balfour, Shaw, and Jarosek (2013)]. Of course, panels in the northern
CHAPTER 8. CELL AND PANEL INTERCONNECTIONS 106

hemisphere are tilted south and are tilted north in the southern hemisphere.
• The DC bus of a typical domestic PV installation is designed for 12-48 VDC, with 48 VDC as the most
common [Balfour, Shaw, and Jarosek (2013)]. We compare that voltage to utility scale PV production
production where the DC bus operates at 200-600 VDC (page 184 of [Markvart (1994)]).
• Inverters are a critical component to an integrated system. Inverters convert the DC current produced by
the PV panels and/or battery storage to AC. Inverters are characterized by the AC current they produce:
whether they produce AC as a sine wave or a modified square wave, the former being more desirable. The
operation of a simple inverter capable of producing a square wave is shown in Fig. 8.13.

Iph - +
n-type

p-type Id
+ +
I
-

Figure 8.13: Highly simplified view of an inverter.

• Power controllers are crucial to the operation of the overall PV installation. Power controllers mainly
function to regulate the minimum and maximum voltage produced by the solar panels. Certain controllers
include maximum power tracking algorithm, effectively adjusting the voltage drop of the panel load to allow
the panels to operate as close as possible to Pmp ; the controller basically acts as a DC to DC transformer
in this operation. Power system controllers also are required to allow operation parallel to the grid in grid-
connected systems; in domestic applications, this operation is called net metering whereby the PV owner
sells excess power to the electrical utility.
• Battery storage is sized such that 2 to 15 days of nominal power can be provided under cloudy conditions.
Battery capacity is measured in A·hr; battery voltages commonly are 12 and 24 V. We note that solar panels
generally are designed with Vmp in increments of 18 V; this is to account for the voltage drop across the
blocking diode, the 2.8 V loss due to operation to 60o C, and the 1 V loss through the power regulator
[Wenham and co-authors (2007)], and that at least 2 V of overvoltage is required to charge a 12 V battery.
• The balance of system (BOC) refers to the remaining wiring, metering, grounding, and other elements
completing the PV installation.

A very readable article describing the home installation of a complete PV system can be found in [Murphy (2008)].

HW and review problems


1. 20 solar panels tilted 10o south with the following characteristics are mounted on the roof of the AV Williams
building on the University of Maryland campus

Vmp = 35 V Imp = 4.0 A Voc = 44.5 V Isc = 4.6 A


CHAPTER 8. CELL AND PANEL INTERCONNECTIONS 107

Compute and plot the hourly total power production on July 20 of this year assuming (a) a totally clear
day; (b) a totally cloudy day. Each panel has 72 cells wired in series.
2. We have the responsibility of designing a PV system for a remote medical clinic. The following define the
design specifications:

(a) the system is to continuously power a 150 W refrigerator, 50 W of security lighting through the night,
and 1200 W·hrs of miscellaneous load throughout the day
(b) all loads require 220 V AC power provided by a 48VDC to 220 VAC inverter which is 80% efficient
(c) 12V DC, 200 amp-hr batteries are to be purchased
(d) panels similar to those of the previous problem are to be used
(e) the country is located 8.5o N latitude and is mostly sunny
(f) diesel generator backup for cloudy days already exists
What is your recommended design in terms of the number and wiring arrangement of the panels, batteries,
and inverter? Draw a diagram illustrating your design.

3. Consider three identical PV cells with the following characteristics under standard AM1.5 conditions: Iph =
2.5 A, Io = 9 × 10−9 A, β = 1, Rs = 0, Rsh = 10 Ω, and T = 300 K. If the cells are wired in series,
(a) Compute Vmp , Imp , and Pmp for the string of three;
(b) Repeat the calculation when the middle cell receives 800 W/m2 compared to the two others which
continue to receive 1000 W/m2 .
4. Consider a tandem (2-junction) thin film PV cell with total surface area of 10 × 10 cm2 . If one layer
of the cell is composed of a-Si:H, what is the optimal band gap of the other semiconductor? For your
calculations, assume all Rs = 0 and Rsh = 10 Ω for each junction. Assume the tunnel junction separating
the semiconductors has zero resistance. Plot I and P versus V for your optimized design.

References
[Archer and Hill (2001)] Archer, M. D. and R. Hill, Clean Electricity from Photovoltaics, Imperial College Press,
2001.
[Balfour, Shaw, and Jarosek (2013)] Balfour, J. M. Shaw, and S. Jarosek, Introduction to Photovoltaics, Jones
& Bartlett Learning, 2013.
[Johnson et. al (1978)] Johnson, D. E., J. L. Hilburn, and J. R. Johnson, Basic Electric Circuit Analysis,
Prentice Hall, NJ, 1978.

[Markvart (1994)] Markvart, T., Ed. Solar Electricity, J. Wiley, 1994.


[Murphy (2008)] Murphy, T. W. Home photovoltaic systems for physicists, Physics Today, 42-47, July (2008)
[Schade (2004)] Schade, H., PV for buildings, reFOCUS, 22-27, Nov/Dec 2004.

[Wenham and co-authors (2007)] Wenham, S. R., M. A. Green, M. E. Watt, and R. Corkish, Applied
Photovoltaics, 2nd Ed., Earthscan, UK, 2007.
[Worder and Zuercher-Martinson(2009)] Worder, J. and M. Zuercher-Martinson, How inverters work,
SOLARPRO, 68-85, April-May 2009.
Chapter 9

Dye Sensitized Cells

In all PV cell architectures considered to this point, photons are absorbed by a semiconductor creating an electron-
hole pair from the promotion of a valence electron the conduction band. To prevent loss of the potential energy in-
crease produced in this process, charges are separated by a semiconductor/semiconductor or semiconductor/metal
junction where minority carriers are swept through the junction by the built-in electric field. In this chapter, we
begin to explore other types of charge carrier generation and separation mechanisms. Specifically, we consider
the thin-film dye-sensitized solar cell (DSSC), often referred to as the Grätzel cell [O’Regan and Grätzel (1991)].
This PV cell is based on very different charge generation and separation mechanisms:

1. Photons are absorbed by a light-sensitive dye, promoting an electron from the dye molecule highest occupied
molecular orbital (HOMO) to the lowest unoccupied molecule orbital (LUMO)

2. The excited electron then is injected into the conduction band of a large-band gap semiconductor such as
TIO2 as the charge separation mechanism

TiO2 dye
anode

cathode

electrolyte

Figure 9.1: The basic elements of a dye-sensitized


PV cell.

load
e-

These dye-sensitized solar cells have a number of advantages [Hardin et. al (2012)]; for example, they can be
constructed from relatively inexpensive raw materials, can be made into flexible cells with plastic substrates, have

108
CHAPTER 9. DYE SENSITIZED CELLS 109

demonstrated efficiencies of up to 12.8%, and they perform well in low-light conditions. Do-it-yourself kits have
even been published illustrating how working cells can be made using ingredients of sun-tan lotion and blackberry
juice [Appleyard (2006)].
A schematic of a typic cell is shown in Fig. 9.1 and the major cell components are described as follows. We
first recall that conventional current flows into the anode of an electrical device. The anode of a DSSC is
transparent to most of the incoming solar radiation and consists of a TCO (transparent conducting oxide) coated
glass or plastic substrate onto which a 10 µm coating of nanostructured TiO2 is deposited. The TiO2 particles
are approximately 10 nm in diameter and so form a high surface area layer (1000× compared to a flat surface)
onto which a monolayer of light absorbing dye is deposited.
The electrolyte consists typically of iodide (I− , and I−
3 ) (note that iodine is I and I2 ) in an organic solvent. The
purpose of the electrolyte is to provide a mechanism by which the dye is regenerated and to create an electric
contact with the counter electrode (the cathode). The cathode, also generally made to be transparent, supports
a platinum or carbon conducting layer.

9.1 Electrochemistry review


To understand how the overall circuit of the DSSC works, we begin with a short review of electrochemistry
concepts. Oxidation reactions involve the loss of electrons and occur at the anode of an electrochemical cell.
Conventional current flows into an anode, so electron current flows out of the anode making the anode the
negative terminal of a galvanic cell. The oxidation number of the reaction product increases in an oxidation
reaction. An example is
Zn(s) → Zn2+ (aq) + 2e− E o = 0.76 V
Note that when written in the form Zn2+ (aq) + 2e− → Zn(s), the standard potential is written as E o = -0.76 V.
Reduction involves reactions in which electrons are gained and so occur at the cathode. The oxidation number
of the reaction product decreases in a reduction reaction. An example is
Cu2+ (aq) + 2e− → Cu(s) E o = 0.34 V
A redox couple denotes the reducing species and its oxidized form, such as for the copper reaction above
Cu2+ /Cu
noting that the oxidized specie is written first [Wang (2014)]. The redox couple represents ions or molecules in
solution that can be reduced or oxidized solely by an electron transfer process; both species are present in the
solution and electrons are transferred to/from the redox couple from/to an electrode [Memming (2001)].
Reduction potential E o is a measure of the tendency of a chemical species to gain electrons (be reduced). The
more positive the value of E o , the more easily the species is reduced. The o superscript indicates standard
pressure and 1 M solutions, and the reduction potential is measured relative to the standard potential of 0 V for
the reaction
1
H+ + e − → H2 (g) E o = 0 V.
2
Comparing the reduction potentials of Cu and Zn, the potentials indicate valence band electrons in Zn have
higher energy relative to those in Cu and so an electron e− available to reduce Cu+2 or Zn+2 would find the Cu2+
reduction energetically favorable1 .

The galvanic cell

To help make these concepts clear, consider the classic copper-zinc electrochemical experiment – the Daniell
galvanic cell – consisting of the two half-reactions described above. Adding the oxidation and reduction reactions
1 In the sense of Gibb’s free energy at standard solution conditions.
CHAPTER 9. DYE SENSITIZED CELLS 110

gives
o
Zn(s) + Cu2+ (aq) → Zn2+ (aq) + Cu(s) Ecell = 1.1 V (9.1)
o o o o o
where Ecell = Ecathode + Eanode = 1.1 V gives the voltage of the battery cell. Because ECu 2+ > EZn2+ , the forward

reaction above is favored over the reverse, resulting in the metal Zn electrode dissolving (oxidizing) from the Zn
anode (anode = conventional current in). Another way to view (9.1) is that it is spontaneous in the direction
o
indicated because Ecell > 0.

Zn anode Cu cathode

Zn(s) ! Zn2+(aq) + 2e-


Zn2+/Zn
EF,Zn Eo = 0.76 V
e- Figure 9.2: Band di-
agram of the electro-
chemical reactions in
e- EF,Cu
the Daniell galvanic
Cu2+/Cu
Eo = -0.34 V cell relative to the
Fermi levels of the two
electrolyte electrodes.
Cu2+(aq) + 2e- ! Cu(s)

load
e-

An alternative view of the charge transport through this electrochemical system is presented in Fig. 9.2 from
the perspective of the electrons involved in the reactions. As stated earlier, the valence electrons of Zn have a
higher energy relative to those of Cu, and so the overall reaction cycle can be thought of as starting with the
Zn oxidation reaction, releasing Zn2+ to the electrolyte solution and 2e− to the Zn anode. In this diagram, we
represent the Fermi level of the anode as being lower in energy relative to the Zn2+ /Zn redox couple, indicating
the driving potential for this forward reaction.
The electrons produced by this reaction travel through the external circuit and the external load which results in
a voltage drop in the electrons. The electrons enter the Cu cathode at a reduced voltage and then are consumed
in the Cu2+ reduction reaction, resulting in metallic Cu deposition on the cathode. Of course there must be a
corresponding positive ion current in the electrolyte to maintain electrical neutrality of each electrode. Note that
the standard potential for the Cu2+ /Cu redox couple is written in terms of the Cu oxidation reaction to maintain
consistency with the potential of the Zn2+ /Zn redox couple – recall that the energy levels represent the relative
energies required to extract the electrons from each metal.
Before returning to the DSSC, we propose the following thought experiment involving varying the resistance R
of the external load: as R → 0, the Fermi levels of each electrode should approach a common value between the
potentials of the two redox couples. This should result in maximum driving force for each half-reaction, resulting
in a high reaction rate and zero useful external work. Likewise, increasing R → ∞ should move each electrode
Fermi level towards the corresponding redox couple potential. Because each half-reaction approaches equilibrium
under these conditions, the overall reaction stops and again, no useful external work is produced. Therefore, much
like PV cells, the optimal power output of a battery appears to be governed by the external load.

9.2 Operating principles of the DSSC


To understand the basic operation of the dye-sensitized cell, we turn to the band and redox potential diagram of
Fig. 9.3. The process begins with absorption of a sufficiently high-energy photon, exciting the dye molecule from
CHAPTER 9. DYE SENSITIZED CELLS 111

its ground state D to D∗ . The dyes are typically composed of a ruthenium complex; the dye concentration on the
TiO2 nanoparticles is approximately 0.5-1 dye molecules/nm2 and the dye absorbs from λ = 780 nm and lower
wavelengths [Hardin et. al (2012)].

D *, D + e- injection
ECB
overpotential
EF
V
I3 -/I-
dye regen Figure 9.3: Band diagram
D overpotential of the DSSC illustrating the
I3- useful voltage V produced
by this PV cell.
EVB I-

load
e-
Rather the returning to its ground state by fluorescence, the dye molecule can transfer the excited electron to
the conduction band of its TiO2 substrate. Recall that TiO2 is an n-type semiconductor with Ebg = 3 eV. The
electron then can migrate through the TiO2 nanoparticles to the TCO anode, after which it can travel through
the external circuit and load to perform useful work, finally returning to the cell at the counter electrode (the
cathode).
The dye molecule, having transferred its charge to the TiO2 is regenerated by driving the following oxidation
reaction at the interface of the dye and electrolyte

3I− → I−
3 + 2e

The I−
3 species then diffuses through the electrolyte and when it reaches the counter electrode, undergoes the
reduction reaction
I− −
3 + 2e → 3I

Overall the iodide reactions do no net chemical work (as opposed to the photoelectrochemical cells described
in Chapter 10) and simply act as charge transfer mechanisms. Just as with the galvanic cell, the overpotential
required to inject electrons into the TiO2 conduction band and to drive the dye regeneration reaction at finite
rates results in a useful external voltage V being less than the energy required to excite the dye molecule. We
can write the overall chemical reactions as

D → D∗
D∗ → e − + D +
2D+ + 3I− → 2D + I−
3
I− −
3 + 2e → 3I

where the first three reactions take place at the anode and the final reaction at the cathode.
CHAPTER 9. DYE SENSITIZED CELLS 112

V = voltage drop
Iph n-type Idiode R3 R1 Rh

p-type Rsh Ish


+
I
Figure 9.4: Equivalent circuit diagram of the DSSC.

9.3 Equivalent circuit model


As seen in Fig. 9.4, the equivalent circuit model of the dye sensitized solar cell is functionally equivalent to the PV
cell circuit with parasitic shunt and series resistances. One of the two major differences between the dye-sensitized
and solid-state PV cell models is in the definition of the series resistance: because of the electrochemical reactions
and the resistance of the electrolyte to iodide ion diffusion, the series resistance Rs is made up of three components

Rh : ITO resistance
R1 : charge transfer at Pt electrode
R3 : electrolyte resistance
and so Rs = R3 + R1 + Rh

The second major difference is in the diode nonideality factor β: for the dye sensitized cell, values of β ≥ 2 can
result from recombination of e − and I−
3 from the TiO2 surface.

Finally, we note that this model applies only to DC operating conditions; this is important because dye-sensitized
cell parameters are frequently determined by applying an alternating external potential.

HW and review problems


1. Consider a DSSC which makes use of a dye that absorbs all photons of λ ≤ 780 nm wavelength. If the e−
injection and dye regeneration overpotentials each correspond to 15% of the energy required for the D → D ∗
transition, compute the maximum theoretical efficiency possible for this cell under AM1.5 conditions.

References
[Appleyard (2006)] Appleyard, S. J., Simple photovoltaic cells for exploring solar energy concepts. Physics
Education 41 409-419 (2006).
[Hardin et. al (2012)] Hardin, B. E. , H. J. Snaith, and M. D. McGehee, The renaissance of dye-sensitized solar
cells. Nature Photonics 6 162-169 (2012).

[Memming (2001)] Memming, R., Semiconductor Electrochemistry, Wiley-VCH, (2001).


[O’Regan and Grätzel (1991)] O’Regan, B. and M. Grätzel, A low-cost, high-efficiency solar cell based on
dye-sensitized colloidal TiO2 films. Nature 353 737-740 (1991).
CHAPTER 9. DYE SENSITIZED CELLS 113

[Wang (2014)] Wang, C. S. Personal communication December (2014)


Chapter 10

Photoelectrochemical systems

hc/λ"
V
I
counter electrode

Figure 10.1: The basic elements of a pho-


semiconductor

toelectrochemical cell. Note the direction


of the external current I is not yet specified.

ions

Consider the electrochemical system illustrated in Fig. 10.1. Sunlight enters through the top of the cell and falls
on a semiconductor electrode. As with the solid-state PV cells we are familiar with, an electron/hole pair is
formed and subsequently separated by a junction. However, rather than doing external work, the photon energy is
used to drive an electrochemical reaction in the electrolyte solution. In this chapter, we will study the similarities
between this system and the PV cells we considered to this point, and will address the following questions:

1. What is the nature of the junction that separates the e− /h+ charges?
2. How do p- and n-type working electrodes differ in terms of system behavior?

3. What are the chemical reactions that take place and how are they driven by the sunlight?
4. What gases are produced at the electrodes?
5. What is the nature of the electrolyte, e.g., is it an acid or base, and does the pH matter?

6. How do we extract the most useful energy from this system?

114
CHAPTER 10. PHOTOELECTROCHEMICAL SYSTEMS 115

10.1 Water splitting fundamentals


In this chapter we focus exclusively on using solar energy to split water into hydrogen and oxygen:

2H2 O → 2H2 + O2

To start, we consider doing so not by solar energy but by an electrolysis process where direct current from an
external source is used to drive the reaction shown above. An externally applied voltage bias provides the driving
force to pull electrons from the anode through the external circuit, and then to return them to the cell through
the cathode to enable the reactions there. The product of that bias voltage and external current determines the
total applied power.
o
Later in this chapter we will show that the minimum cell potential required for this reaction is Ecell = −1.23 V,
noting the large difference from the positive cell voltage of 1.1 V for the Daniell cell (Fig. 9.2) described previously.
With ∆G o as the change in Gibbs free energy under standard conditions, we write

∆G o = −nFEcell
o

n = moles of electrons
F = Faraday constant = 96485 C/e− mole
o
Ecell = cell potential, V

so for water electrolysis we will show n = 2 and so obtain

∆G o = −(2 mol e− /mol H2 O)(96485 C/e− mole)(-1.23 V)


= +237, 350 J/mol H2 O.

Positive ∆G o indicates nonspontaneous reaction and the minimum work required to split one mole of water;
again, we compare this to the Daniell cell which produces electrical current spontaneously.

Relationship to the solar irradiance spectrum

o
Given the minimum cell potential of Ecell = −1.23 V, we can argue that in the context of solar energy-driven
reactions that photons with energy less than 1.23 eV cannot be used for water splitting reactions. This energy
corresponds to photons with wavelength
hc 1239.8eV nm
λ= o = = 1008 nm
Ecell 1.23 eV

indicating that (a) most of the visible spectrum of sunlight has the theoretical potential be used to drive the
PEC water splitting reactions, and (b) a semiconductor with Ebg ≥ 1.23 eV is necessary for the PEC reactions to
proceed in a single-junction device.

10.1.1 Electrolyte pH
One of the questions posed at the outset of this chapter relates to the role electrolyte pH plays in the overall water
splitting process. As a quick review, recall that pure water partly dissociates into H+ and OH− ions according to
the equilibrium relationship
[H+ ][OH− ]
= Keq
[H2 O]
where the equilibrium constant Keq has units mol/l (molar, or M). Under normal laboratory conditions [H2 O] =
55.5 mol/l and so
Kw = 55.5Keq = [H+ ][OH− ]
CHAPTER 10. PHOTOELECTROCHEMICAL SYSTEMS 116

where Kw = 10−14 mol2 /l2 . This means [H+ ]=[OH− ] = 10−7 mol/l in pure water, resulting in a pH of

pH = − log10 [H+ ] = +7 for pure water.

Given that HCl is an acid (proton donor) and a strong one at that (because it essentially fully dissociates in
water), a 1 M solution of HCl will have pH=0. For a 1 M NaOH (a strong base and proton acceptor) solution in
water
Kw
[H+ ] = = 10−14 mol/l
[OH− ]
resulting in pH=14.

Electrolyte conductivity

Pure water is a relatively poor conductor of electricity, and one important aspect of electrolyte pH is its role
in determining its charge carrriers, the electrolyte conduction mechanisms, and the mobility of those carriers
[Memming (2001)]. For example, one factor necessary to compute the overall conductivity of an electrolyte is the
combination of charged species mobility and the concentrations of the species themselves. Consider, for example,
the two cases illustrated in Fig. 10.2. For a basic solution, negative charges are transported through a hopping
mechanism where a hydroxyl ion OH− removes a proton from a nearby water molecule, transferring its negative
charge to that molecule. In an analogous manner, hydronium ions (H3 O− ) transfer a proton to a neighboring
water molecule, thus transporting the positive charge though the electrolyte.

e-

O2 H2
H H H
Figure 10.2: The charge jumping mech-
OH- O H O H O H O H O- anism at work in an aqueous electrolyte,
shown for the basic (top) and acidic (bot-
H H tom) cases.
base
acid
H H H H
H O H O H O H O H H+
+

To qualify this conduction mechanism, electric ion mobility data below (from page 460 of [Levine (1978)]) corre-
spond to 25o C, 1 atm of pressure, and are extrapolated to infinite dilution.

Ion H 3 O+ Na+ OH−



u cm2 /(V s) 3.63 × 10−3 5.19 × 10−4 2.05 × 10−3

Multiplying each value by a representative electric field value, e.g., 10 V/cm, shows the relatively slow pace by
which ion charge is transferred through the electrolyte. From this we conclude that concentrated electrolyte
solutions will be required for practical PEC devices.
CHAPTER 10. PHOTOELECTROCHEMICAL SYSTEMS 117

Electrolyte Fermi level

At this point we have extensively discussed the concept of the Fermi level EF in the context of semiconductors and
metals. Otherwise known as electron chemical potential µe , it is characterized by the energy level corresponding
to a 0.5 probability of electron occupation, or the highest occupied energy state in the limit of T → 0 K.

• For a semiconductor, EF = f (EVB , ECB , T ) and, of course, also is a function of acceptor and donor concen-
trations NA and ND in the case of doped semiconductors.
• For an electrolyte, EF ,redox = f (Eredox ) = f (pH), so the pH of the electrolyte plays a critical role in
) Ox + e− the electrochemical potential
determining the solution Fermi level. For the redox reaction Re *
[Ox ]
µe,redox = µored − µoox − RT ln
[Re ]
and so the chemical potential decreases with increasing oxidized species concentration [Memming (2001)].
In the case of Ox = H− this would result in decreasing electrochemical potential with decreasing pH.
• Reminiscent of the thought experiment involving semiconductor band structures and the band bending that
results when a junction is formed (Fig. 4.7), when a semiconductor and electrolyte are brought into contact,
the Fermi levels equilibrate and a junction is formed.

10.2 p-type PEC cells


To examine the semiconductor/electrolyte junction in more detail, we first consider the case of a p-type semicon-
ductor/electrolyte junction.

When EF < EF ,redox (a basic electrolyte) When EF > EF ,redox (acidic electrolyte)

• electrons move from the electrolyte into the • holes build up at the semiconductor edge
semiconductor forming the depletion region; nearest the electrolyte;

• the EVB and ECB bend downwards inside the • anions (negatively charged species) in the
semiconductor until EF = EF ,redox . electrolyte migrate to the surface to balance
the positive charges forming the accumula-
tion region;

• the EVB and ECB bend upwards inside the


semiconductor until EF = EF ,redox .

These two situations are depicted in Fig. 10.3 where the band bending corresponding to a basic solution is shown
at left and in an acidic solution, right. In the latter situation, if an electron is promoted to the conduction band,
the hole becomes available to drive an oxidation reaction at the electrode/electrolyte interface. The electron,
however, would be required to traverse the full width of the semiconductor - being the minority carrier, it will
encounter substantial resistance and likely undergo recombination, reducing the effectiveness of the cell. This
results in the preferred combination of p-type semiconductor with a basic solution where a reduction reaction
takes place.

The PEC cell

We observe the PEC cell in operation in Fig. 10.4, where an electron is promoted to the conduction band in
the semiconductor, while the majority carriers (holes) migrate through the semiconductor to the electrode. The
reduction reaction takes place at the photocathode, resulting in OH− ions which migrate through the electrolyte
CHAPTER 10. PHOTOELECTROCHEMICAL SYSTEMS 118

depletion region accumulation region

ECB ECB
-+ +-
Figure 10.3: p-type depletion region (left) and
-+ +- accumulation (right) regions. PEC operation un-
-+ ERedox +- ERedox der conditions where a depletion region exists is
EF EF the more common case.
-+ +-
EVB EVB
electrolyte electrolyte
p-type p-type
semicond semicond

to the counter electrode where the oxidation reaction takes place. Electrons involved in the anode reactions at
the liquid/semiconductor interface then are replenished by those traveling through the external circuit. In our
archetype PV cell, we recall that some resistance across an external load is required to generate useful electrical
work from the cell shown in Fig. 6.3. In the PV cell, a short (external) circuit results in no useful work. However,
an external short circuit forces the PEC cell to maximize useful chemical work.

e-

e-
2H2O + 2e-  2OH- + H2

OH -
1.23 V Figure 10.4: p-type PEC cell operation in a basic elec-
EF,Redox trolyte.

h+ 2OH-  H2O + ½O2 + 2e-

hc/λ"

Surface reactions

The details of the surface reaction are important because of the strong dependency the reaction rates demonstrate
with different surfaces. Typically catalysts are required to produce effective reaction rates. The water-splitting
reaction details also are an important first step in understanding other photocatalyzed reactions. For the case of
an p-type semiconductor in a basic electrolyte, we can write the oxidation reactions that take place at the anode
as OXB the the reduction reactions at the cathode as REB .

OXB : OH− + S → OH + e− OH + OH− → O + H2 O + e− O+O → 2S + O2

S S S S S

REB : H2 O + S → H+ + OH− H+ + e− → H H+H → 2S + H2

S S S S S
CHAPTER 10. PHOTOELECTROCHEMICAL SYSTEMS 119

The anode and cathode reactions are summarized as follows; when the two half reaction potentials are added,
the cell potential is found be -1.23 V, the value discussed earlier in this chapter.

Anode (oxidation) Overall cell potential


4OH− (aq) → O2 (g) + 2H2 O(l) + 4e− E o = −0.40 V
o o o
Ecell = Ecathode + Eanode
Cathode (reduction) = −0.83 V + (−0.40 V)
2H2 O(l) + 2e− → H2 (g) + 2 OH− (aq) E o = −0.83 V = −1.23 V

10.3 n-type PEC cells


Following the same process as with the p-type, we now consider the case of a n-type semiconductor/electrolyte
junction.

When EF > EF ,redox (e.g., an acidic electrolyte) When EF < EF ,redox (basic electrolyte)

• electrons move from the semiconductor into • electrons build up at the semiconductor edge
the electrolyte forming the depletion region; nearest the electrolyte;
• the EVB and ECB bend upwards inside the • cations (positively charges species) in the
semiconductor until EF = EF ,redox . electrolyte migrate to the surface to balance
the negative charges;

• the EVB and ECB bend downwards inside the


semiconductor until EF = EF ,redox .

depletion region accumulation region

ECB ECB
+- -+
EF ERedox EF ERedox
+- -+ Figure 10.5: n-type depletion region (left) and
accumulation (right) regions. As with the p-type
+- -+ semiconductor, the more common configurations
result in a junction featuring a depletion region.
+- -+
EVB EVB
electrolyte electrolyte
n-type n-type
semicond semicond

These two situations are depicted in Fig. 10.5 where the band bending corresponding to an acidic solution is
shown at left and in a basic solution, right. In the latter situation, if an electron is promoted to the conduction
band, it becomes available to drive a reduction reaction at the electrode/electrolyte interface. The hole, however,
would be required to traverse the full width of the semiconductor - being the minority carrier, it will encounter
substantial resistance and likely undergo recombination, reducing the effectiveness of the cell. This results in the
preferred combination of n-type semiconductor with a acidic solution where an oxidation reaction takes place.

The PEC cell

We observe the PEC cell in operation in Fig. 10.6, where an electron is promoted to the conduction band in the
semiconductor, while the minority carriers (holes) drift through the depletion zone to semiconductor/electrolyte
CHAPTER 10. PHOTOELECTROCHEMICAL SYSTEMS 120

interface. The electrons, as majority carriers, migrate through the semiconductor to the electrode. The oxidation
reaction takes place at the photoanode, resulting in H+ ions which migrate through the electrolyte to the counter
electrode where the reduction reaction takes place. Electrons generated in the anode travel to the counter
electrode through the external circuit.

e-

e-
2H+ + 2e- ! H2

1.23 V
EF,Redox Figure 10.6: n-type PEC cell operation in an acidic
electrolyte.
H+

H2O + 2h+ ! ½O2 + 2H+


h+

hc/λ"

Surface reactions

For the case of an n-type semiconductor in an acidic electrolyte, we can write the oxidation reactions that take
place at the anode as OXA the the reduction reactions at the cathode as REA .

OXA : H2 O + S → OH + H+ + e− OH → O + H+ + e− O+O → 2S + O2

S S S S S

REA : H+ + S → H+ H+ + e− → H H+H → 2S + H2

S S S S S
The anode and cathode reactions are summarized as follows; when the two have reaction potentials are added,
the cell potential again is found be -1.23 V.

Anode (oxidation) Overall cell potential


2H2 O(l) → O2 (g) + 4H+ (aq) + 4e− E o = −1.23 V
o o o
Ecell = Ecathode + Eanode
Cathode (reduction) = 0 V + (−1.23 V)
2H+ (aq) + 2e− → H2 (g) Eo = 0 V = −1.23 V

10.4 Quantitative analysis


Our motivation for studying PEC systems is based in their potential to directly create a storable and portable
energy carrier: H2 . As a first step in developing this understanding, let us consider the feasibility of using TiO2 ,
an n-type semiconductor with Ebg = 3.0 eV. Let us determine the maximum theoretical H2 generation rate by a
fully illuminated electrode. Our choice is based on the pioneering work of [Fujishima and Honda(1972)].
CHAPTER 10. PHOTOELECTROCHEMICAL SYSTEMS 121

Because we are operating under (external) short-circuit conditions, the maximum theoretical current is deter-
mined in exactly the same manner as with Iph,max for a PV cell: first we compute λbg for the titanium dioxide
semiconductor as
λbg = 413 nm
and then
Z∞
Iph = SR(λ)EE λ dλ
λ=0
Z λbg
EE λ
= q dλ
0 hc/λ
= 18.68 A/m2

Figure 10.7: Usable portion of


the spectrum for TiO2 -based
PEC.

The number of moles of electrons produced per unit area then is


Iph
ṅ = = 1.93 × 10−4 mol m−2 s−1 .
F
From the anode/cathode reactions described earlier, for all cases we know that 2 electrons are required for every
molecule of H2 produced. Given that the ideal gas law is PV t = nRT and is a reasonable approximation at the
PEC cell conditions of atmospheric pressure and 300 K
ṅRT [1.93 × 10−4 mol m−2 s−1 ][8.314 J mol−1 K−1 ][300 K]
V̇ t = =
2P 2[101325 Pa]
= 2.4 × 10−6 m3 H2 /(m2 s)

At this point in the analysis we must ask “Is this a lot?” Or “Are we certain we have an energy balance?”

10.5 Nonidealities
We make note of some of the many nonidealities that limit the effective performance of these PEC systems. These
include

• Recombination: where electrons and holes may recombine within the semiconductor because of limited
charge carrier mobility and faults in the semiconductor, reducing the rate of the redox reactions.
• Electrolyte redox competition
– For n-type semiconductors, electrolyte cations may be reduced (instead of H+ ) and electrolyte anions
may be oxidized (instead of H2 O being split to (1/2)O2 , 2H+ , and 2e− );
CHAPTER 10. PHOTOELECTROCHEMICAL SYSTEMS 122

– For p-type semiconductors, electrolyte cations may be reduced at the photocathode in place of the
2H2 O + 2e− → H2 + 2OH− reaction, and electrolyte anions may be oxidized (instead of OH− reacting
to form O2 , water and e− ).
• Electrode degradation In some cases, the semiconductor itself may be oxidized or reduced instead of the
water or electrolyte anions or cations.

HW and review problems


1. What fraction of AM1.5 photons are absorbed by the n-type semiconductor TiO2 (Ebg = 3.0 eV)?

2. Consider the p-type semiconductor Cu2 O (Ebg = 2.1 eV) with an active electrode surface area of 10 m2 in a
PEC system operating at ambient conditions with an AM1.5 spectral irradiance normalized to 1000 W/m2 .
Assuming ideal conditions, determine the maximum possible molar flow rates (mol/sec) of the H2 and O2
produced as well as the water needed to replenish the system.

References
[Fujishima and Honda(1972)] Fujishima, A. K. Honda, Electrochemical photolysis of water at a semiconductor
electrode. Nature, 238 pp. 37-38 (1972).

[Levine (1978)] Levine, I. N., Physical Chemistry, McGraw-Hill, (1978).


[Memming (2001)] Memming, R., Semiconductor Electrochemistry, Wiley-VCH, (2001).
Chapter 11

Photosynthesis and the efficiency of


bioethanol production

The DSSC and PEC devices discussed in the two previous chapters demonstrated photon absorption and charge
separation mechanisms that were new, but not entirely different from those that govern the solid-state PV devices
making up the majority of this text. We now turn our attention to photosynthesis, with the ultimate goal of
estimating the efficiency of natural photosynthesis processes as a means of comparing natural photosynthesis and
biofuel production to the PV systems we have studied.

11.1 Photosynthesis
The overall photosynthesis reaction is referred to as carbon fixation, the conversion of carbon dioxide to sugars
such as glucose:

6CO2 (g) + 6H2 O(l) → (CH2 O)6 (aq) + 6O2 (g) (11.1)
o
The standard Gibbs energy of glucose formation by the reaction above is ∆G298 = 2880 kJ/mol (page 224 of
o
[Akins and DePaula (2002)]); for liquid water we have the Gibbs energy of formation ∆G298 = −237.1 kJ/mol
o
and for gaseous CO2 ∆G298 = −394.4 kJ/mol (pp. 686-687 of [Smith et al. (2005)]). Of course, work (energy)
can be produced by the reverse reaction; an objective of this chapter is to determine the solar energy required
to produce energy-storing molecules such as sugars and alcohols, and to give an estimate of the efficiency of
photosynthesis.
Carbon fixation is a redox reaction, and so (i) electron transfer is involved and (ii) energy is required to drive
the process because of the positive ∆G of (11.1). It is easy to show the reaction above is a redox reaction: 4
electrons are needed to reduce CO2 to the carbohydrate1 . Of course the reverse reaction is oxidation of sugar to
water and CO2 .
Before we proceed to our more quantitative analysis, we consider the following pre-history facts: photosynthesis
in early life on our planet (3.5 billion year ago) converted H2 and H2 S to bio-products; one half billion years after
that, water-splitting photosynthesis led to the O2 in our atmosphere.
Modern photosynthesis is a two-stage process

1. Light reactions capture solar energy and create ATP and NADPH, described by the Z-scheme; these
reactions are mostly unaffected by temperature, and the light reactions are responsible for splitting water
in the photosynthesis process;
1 Definition: Carbohydrates are hydrates of carbon with general formula Cm (H2 O)n .

123
CHAPTER 11. PHOTOSYNTHESIS AND THE EFFICIENCY OF BIOETHANOL PRODUCTION 124

2. Light independent reactions convert CO2 to sugar. The reaction rate of this process increases with
temperature and these reactions form the Calvin-Benson Cycle, consuming CO2 , ATP, NADPH and H+ to
generate a 3-carbon carbohydrate, ADP, NADP+ , and water.

It is interesting to note that the rate of the carbon-fixing reactions increases initially with solar irradiation (which
makes sense), reaches a plateau near the peak irradiance of 1 kW/m2 , and then decreases, though some sources
cite a 100 W/m2 plateau [Hall and Rao (1991)] (page 26).

11.1.1 Chlorophyll
There are two types of chlorophyll in green plants: chlorophyll a and chlorophyll b. Typically, the ratio of
chlorophyll a molecules to b is three to one. The function of chlorophyll a is to drive the water splitting reaction;
chlorophyll b also absorbs sunlight, but transfers its energy efficiently to chlorophyll a, expanding the spectral
range over which plants can absorb light.

2
second singlet
heat
1 green
first singlet
480 nm 550 nm 700 nm

hν"
energy available for
chemical work Figure 11.1: Chlorophyll excitation and ground states.

ground state

Plants absorb light between 550 and 700 nm, and for λ < 480 nm. It is the gap in between that results in the
green coloration of most plants. We note the sodium doublet (at 589 and 589.6 nm) falls within the absorption
range, resulting in the efficient use of sodium vapor lights for indoor plant growth (Fig. 11.1).
The molecule of chlorophyll a has at its center an Mg atom surrounded by a structure containing 10 C-C conjugated
double bonds that are delocalized, switching between resonance structures. Generally, the higher the number of
double bonds in a system, the lower the energy required to form an excited state by a photon.
In Fig. 11.1, we observe two excited chlorophyll states denoted as the first and second singlets; the second
corresponds to and energy level accessible by the higher-energy photons (λ < 480 nm). This state will relax to
the first singlet releasing heat; the first singlet also can be accessed directly from photons in the lower-energy
fraction of the absorption spectrum. The energy associated with excitation to the first singlet can drive the light
reactions of photosynthesis (described next) or, much like a direct band gap semiconductor, release the energy
via fluorescence with a peak intensity of λ = 685 nm [Hall and Rao (1991)] (page 63).

11.2 Light reactions


Plant cell chloroplasts contain chlorophyll inside light-harvesting complexes. Chlorophyll molecules absorb light
only at specific wavelengths: as seen in Fig. 11.2, light at 680 nm is absorbed by the chlorophyll of Photosystem
II, and 700 nm for Photosystem I. Photosynthesis would be extremely inefficient if plants were limited to light
absorption at these discrete wavelengths, leaving much of the visible spectrum unusable.
CHAPTER 11. PHOTOSYNTHESIS AND THE EFFICIENCY OF BIOETHANOL PRODUCTION 125

2e-
2hν
λ=700 nm
NADP+ + H+ + 2e- ! NADPH

-
e transport
2e- rxns
2hν PS I Figure 11.2: The photosyn-
reaction center
λ=680 nm thesis ”Z” scheme, so-named
because of the ”Z” shape of
the energy levels.

PS II
reaction center

H2O + 2h+ ! ½O2 + 2H+

To understand how plants have engineered their light-harvesting capabilities, recall the discrete energy levels of
molecules as compared to the band structure of crystalline materials (Fig. 4.2). The light-harvesting complexes
inside chloroplasts are tightly packed crystalline structures composed of bundles of different pigments, each capable
of absorbing photons of higher energy relative to the chlorophyll center. These additional pigments (responsible for
fall colors in trees?), as part of the antenna proteins, essentially transfer their absorbed energy to the chlorophyll,
widening the spectral range over which the light-harvesting structure can absorb sunlight.
Photosystem II is a chemical complex with a chlorophyll a molecule at its center and absorbs light of wavelength
680 nm or less, hence the notation P680 or PS II. Photosystem I can absorb light of wavelength 700 nm or less
and is denoted P700 or PS I. We note that the reaction center of PS I also has chlorophyll a at its center.
The energy-carrying molecules ATP and NADPH are generated by the light reactions and subsequently provide
the energy source for the carbon-fixing reactions. NADPH is a reducing agent; when it is produced it leaves the
chlorophyll of PS I in a positively charged (oxidized) state.

11.2.1 Electrons in the Z-scheme


In an attempt to understand the basic steps of the light reactions, we follow the charges through the Z-scheme
shown in Fig. 11.2. The chlorophyll a molecule at the center of PS II promotes an electron to a higher energy level
when a photon is absorbed by the chlorophyll molecule. As described earlier, the photon must have wavelength
λ < 680 nm. The excited system PS II∗ is unstable, and initiates a chain of redox reactions, transferring the
electron from one (electron acceptor) molecule to the next in the electron transport chain (ETC). Each of these
reactions has a lowered redox potential, hence the downward slope of the path to PS I in Fig. 11.2. The effect of
this is to separate the charges, preventing recombination at the chlorophyll site.
The positively charged PS II+ dissociates water in the following oxidation reaction
H2 O → 1/2 O2 + 2H+ + 2e−
regenerating PS II and leaving it in an uncharged state. The H+ ions eventually are used as part of the ATP
production process. We note that it is this step that is responsible for O2 generation by plants.
Ultimately, the electron produced by the PS II system is transported by the ETC to regenerate the PS I+ site
which is produced when an electron is promoted to a higher energy of PS I by a photon of λ < 700 nm or shorter,
producing excited state PS II∗ . In this ETC, the final electron acceptor is the NADP+ , where this molecule is
reduced to form NADPH.
Overall, the general path taken in the light reaction scheme is
2 photons + H2 O → 1/2 O2 + PS II → biochemical reactions of the ETC → PS I + 2 photons → NADPH
CHAPTER 11. PHOTOSYNTHESIS AND THE EFFICIENCY OF BIOETHANOL PRODUCTION 126

where the NADPH is a sufficiently strong reducing agent to reduce CO2 and to ultimately create glucose.

11.3 Efficiency
Examining the literature, the efficiency of photosynthesis is listed as being anything from 0.1 to nearly 10%. But
what does this mean?
To begin our analysis, we begin with the water splitting reaction in the first step of the Z-scheme and recall the
overall cell potential E o = −1.23 eV and so photons of wavelength λmax ≤ 1008 nm are required to split the
water molecule.

• Integrating the AM1.5 spectrum from 0 to 480 nm for both photosystems and from 550 to 680 nm for PSII
and to 700 nm for PSI, the mean maximum theoretical efficiencies are found to be

ηPSI = 20% ηPSII = 18%

where SR ∗ is computed using λ = 1008 nm as the equivalent of λbg . Thus, the limiting efficiency is taken
as the mean ηPS = 19%
• 70% quantum efficiency reduces η = 13.3%
• dark reaction losses, including 32% efficiency in conversion to glucose gives η = 4.3%
• and 40% of glucose is consumed by the plant itself

2
AM1.5
PS I
1.5 PS II
W/m2/nm

0.5

0
200 400 600 800 1000 1200 1400
nm

Figure 11.3: The AM1.5 spectrum and the portion used by photosynthesis.

Overall, we obtain a final estimate of photosynthesis efficiency of converting sunlight at AM1.5 conditions to
harvestable glucose as η = 2.6%.

11.4 Case study: energy efficiency of converting corn to ethanol


The objectives of this case study are 1) to develop an energy balance to determine if ethanol can be considered a
renewable energy source, and 2) use a material balance to determine whether this fuel source is carbon neutral.
We will use the following assumptions as a basis for our calculations

1. All calculations will be performed for a 1 hectare area (10000 m2 - the size of a cricket field, or about 2.5
US acres)
CHAPTER 11. PHOTOSYNTHESIS AND THE EFFICIENCY OF BIOETHANOL PRODUCTION 127

2. Glucose will be produced by corn grown in Iowa, assuming a latitude of 42.5o N and a growing season from
early May to mid-October (approximately 160 days)
3. We will omit energy costs due to irrigation, human work, transportation, and application of pesticides and
herbicides

4. That days during the corn growth period are a 50/50 mix of full sun and full clouds

11.4.1 A PV farm
To begin, we compute the total amount of solar energy that falls on our one hectare plot to determine an upper
limit on the energy that can be harvested during the 160 day growing cycle. To carry out this calculation, we use
our solar toolbox sodtware starting with day 120 after the winter solstice, randomly choosing whether the day is
cloudy or clear according to our assumed 50/50 distribution. A plot of the hourly global irradiance EG is shown
in Fig. 11.4.

Figure 11.4: Iowa global irradiance EG over the corn growing season.

Integrating the hourly EG data over the full 160 days, we find the total solar energy Esolar ,tot as
280
10000 m2
 Z
Esolar ,tot = EG (t)dt
hectare 120
= (10000)(2.97 × 109 J/m2 ) = 2.97 × 1013 J/hectare

We take a relatively conservative estimate that a PV farm would convert 10% of this energy to electricity, resulting
in a total PV energy production of

EPV ,tot = −2.9 × 106 MJ/hectare

11.4.2 Corn-based ethanol production


We now consider growing corn under the same conditions to assess the ultimate production rate of ethanol. Much
of the data for the following calculations will be taken from the study of [Pimental and Patzek (2005)]. First,
we consider the total energy input to a hectare of corn farm based on the previously described assumptions. The
CHAPTER 11. PHOTOSYNTHESIS AND THE EFFICIENCY OF BIOETHANOL PRODUCTION 128

major remaining energy inputs are

Machinery and diesel fuel : 8.46 × 103 MJ/hectare


Nitrogen for fertilizer : 10.24 × 103 MJ/hectare
Total : 18.70 × 103 MJ/hectare

So far this looks promising given that these energy requirements are significantly lower that the power input from
the sun. Note that for fertilizer
3H2 + N2 ↔ 2NH3
with H2 produced from methane reforming, partial oxidation, or electrolysis; all constitute energy and carbon
inputs.
The resulting annual corn production rate cited by [Pimental and Patzek (2005)] for this farming operating is

Corn yield: 8655 kg/hectare

11.4.3 Downstream processing operations


[Pimental and Patzek (2005)] give the following fermentation relationship for the production of ethanol:

15 l water + 2.7 kg corn → 1 l ethanol + 13 l wastewater

and the fermentation reaction itself


C6 H12 O6 → 2C2 H5 OH + 2CO2
We can immediately conclude that the ethanol annual production rate Retoh is
 
8655 kg/hectare
Retoh = = 3206 l ethanol/hectare
2.7 kg/l

In producing 99.5% pure ethanol, the following utility costs from [Pimental and Patzek (2005)] for ethanol distil-
lation are used; based on the rate of ethanol production Retoh , we convert the utility costs to the same basis as
the corn production energy inputs:

Steam : 3.41 × 104 MJ/hectare


Electricity : 1.36 × 104 MJ/hectare
Total : 4.77 × 104 MJ/hectare

The final step requires that we compute the energy content of the product ethanol. For this, we assume that it
is used in the most efficient form of the combustion reaction (e.g., using a fuel cell)

7
C2 H5 OH(l) + O2 → 2CO2 + 3H2 O(g )
2
and so the maximum work that can be extracted is
o
∆G298 = 2(−394000) + 3(−229000) − (−175000) = −1.3 × 106 J/mole ethanol

Given ethanol’s molecular mass of MW = 46.07 g/mol and density of ρ = 0.789 kg/l, our ethanol energy
equivalent is  
1000 o
Eetoh = ρRetoh ∆G298 = −7.14 × 104 MJ/hectare
MW
We summarize our findings in the following table
CHAPTER 11. PHOTOSYNTHESIS AND THE EFFICIENCY OF BIOETHANOL PRODUCTION 129

energy input/output MJ/hectare


corn production 1.87 × 104
ethanol production 4.77 × 104
ethanol energy value −7.14 × 104
net ethanol energy value −0.50 × 104
max theoretical photosynthesis −72.9 × 104
PV farm −290 × 104

In conclusion

• The total energy input (not including the sun!) necessary to produce fuel-grade ethanol essentially equals
the ethanol’s fuel value itself;
• Numerous energy inputs were omitted, but other factors, such as the dry distiller’s grain: 33% of corn (page
68 of [Pimental and Patzek (2005)], constitute an additional source of revenue for the overall production
process;
• Much can be done in optimizing the water/ethanol separation process;

• Because of its high oxygen content, we should consider biomass as a potentially more significant value as
a feedstock for chemical production instead of as a liquid fuel.

HW and review problems

References
[Akins and DePaula (2002)] Akins, P. W. and J. DePaula, Physical Chemistry, 7th ed., Oxford University Press,
2002.
[Hall and Rao (1991)] Hall D. O. and K. K. Rao, Photosynthesis, 6th ed., Cambridge University Press, 1991.

[Pimental and Patzek (2005)] Pimentel, D. and T. W. Patzek, Ethanol production using corn, switchgrass, and
wood; biodiesel production using soybean and sunflower, Natural Resources Research 14 (1). pp. 65-76,
2005.
[Smith et al. (2005)] Smith, J. M, H. C. Van Ness, and M. M. Abbott, Introduction to Chemical Engineering
Thermodynamics, 7th ed., McGraw-Hill Chemical Engineering Series, 2005.
Chapter 12

Elementary numerical methods

In this chapter we summarize the main numerical techniques used in our solar energy calculations.

12.1 Linear interpolation


Given two data points (xd1 , yd1 ) and (xd2 , yd2 ), we can define a line
yd2 − yd1
y= (x − xd1 ) + yd1 .
xd2 − xd1
This gives us the interpolated value y for some intermediate value x. Of course, x can fall outside the range of
the interval over which the data are defined [xd1 , xd2 ]. These cases of extrapolation should generally be treated
with caution.

12.2 Quadrature and the trapezoidal rule


To illustrate the use of the trapezoidal rule for numerical integration (also called quadrature), consider the problem
of determining the definite integral of function f (x) with limits x0 and xN
Z xN
F (x0 , xN ) = f (x) dx. (12.1)
x0

We have the following values of f (x) within the interval of integration, including the two endpoints

[f (x0 ), f (x1 ), f (x2 ), ... , f (xN )]

corresponding to the discrete variable values arranged in ascending order:

[x0 , x1 , x2 , ... , xN ].

Given these data, we can compute an approximation to the definite integral (12.1) as

f (x0 ) + f (x1 ) f (x1 ) + f (x2 ) f (xN−1 ) + f (xN )


F (x0 , xN ) ≈ (x1 − x0 ) + (x2 − x1 ) + · · · + (xN − xN−1 )
2 2 2
Note that linear interpolation must be performed when either or both of the limits of integration of (12.1) do not
match values of xn exactly.

130
CHAPTER 12. ELEMENTARY NUMERICAL METHODS 131

12.3 Finite differences


The finite difference method is one technique for approximating derivatives. If δ is a small number (but significantly
larger than the machine roundoff error), g is a vector of nonlinear equations, and z is the vector of variables, for
centered finite differences
gi (zj0 + δ) − gi (zj0 − δ)

∂gi
≈ (12.2)
∂zj z0

and for forward differences
gi (zj0 + δ) − gi (zj0 )

∂gi
≈ . (12.3)
∂zj z0 δ

12.4 The Euler integrator


Consider the single differential equation model:
dx
= g (x) = ax − b
dt
with a = −0.5 and b = 1, subject to the initial condition x(t = 0) = 5. We can immediately see that the steady
state solution is xst.st. = −2 and it is stable because of the negative eigenvalue λ = a = −0.5. The corresponding
particular solution is:
x(t) = −2 + 7e −0.5t
We wish to compare the exact solution computed over 0 ≤ t ≤ 10 to the solution approximated with the (forward)
Euler method:

dx
= g (xn ) (exact)
dt t=tn
xn+1 − xn

tn+1 − tn
and so xn+1 = xn + (tn+1 − tn )g (xn )
= xn + ∆t g (xn ).

If we consider xn = x(tn ) to be a point on the exact solution to the differential equation, we can write the Taylor’s
series approximation to the time-dependent behavior of x (the exact solution) in the neighborhood of xn as

1 d 2 x

dx
x(t) = x(tn ) + (t − t n ) + (t − tn )2 + h.o.t.
dt tn 2 dt 2 tn
1 2 d 2 x

dx
x(tn+1 ) = x(tn ) + ∆t + ∆t 2 + h.o.t.
dt tn 2 dt tn

1 2 dg (x)
= x(tn ) + ∆t g (xn ) + ∆t + h.o.t.
2 dt tn

and so subtracting the results from one step of the Euler integrator and retaining only terms of order ∆2t or less
CHAPTER 12. ELEMENTARY NUMERICAL METHODS 132

gives an estimate of the error dn+1 generated by one step of the Euler integrator:

dn+1 = x(tn+1 ) − xn+1



1 2 dg (x)
= x(tn ) + ∆t g (xn ) + ∆t − xn − ∆t g (xn )
2 dt tn

1 2 dg (x)
= ∆
2 t dt tn
1 2 2
= ∆ [a xn − ab]
2 t
1 2
∆ (−0.5)2 −2 + 7e −0.5t + 0.5
 
=
2 t
1 2
< ∆ [1.75]
2 t
so dn+1 < 0.875∆2t for our particular problem.

This means halving the step size decreases the local error by a factor of 1/4, however, twice as many steps are
required to integrate the system over the original time step and so if we assume errors accumulate in an additive
manner over short time intervals, we see the net effect of halving the step size when using the Euler method is to
halve the error. This is why the Euler method is called a first order method, even though it is accurate to order
2.
The following points also should be noted:

• The Euler method works for sets of equations by simply employing vector notation:

zn+1 = zn + ∆t g(zn ) subject to z0

• The method works with mixed algebraic/ODE systems provided the AEs are solved at each time step for
some subset y the state variables:

zn+1 = zn + ∆t g(zn , yn )
f(zn , yn ) = 0

subject to z0 ;
• Stability issues set an upper practical limit on the time step sizes for most numerical integration techniques,
because instabilities can cause “runaway” to infinitely large global errors in systems which in reality display
bounded, long-time behavior.

12.5 Newton’s method


Consider the problem of finding solution(s) to

g (x; p) = 0

We now solve for solutions xs using the iterative Newton’s method for a specified set of parameters p. If x 0 is
our guess for a solution, a more refined solution x 1 can be computed using the Taylor’s series of g evaluated at
the current solution estimate x 0 :
1 d 2 g

1 0 dg 1 0
g (x ) ≈ g (x ) + (x − x ) + (x 1 − x 0 )2 + h.o.t.
dx x 0 2! dx 2 x 0
CHAPTER 12. ELEMENTARY NUMERICAL METHODS 133

where h.o.t. refers to order three and higher derivative terms. To compute the more refined solution estimate
x 1 , a linear curve gL (x) tangent to g (x) at the current solution estimate g (x 0 ) is defined from the Taylor’s series
expansion by
0 dg
gL (x) = g (x ) + (x − x 0 ).
dx x 0
We see that gL (x) is a linear approximation to the true g (x), but one that can be solved explicitly for the value
of x where the function crosses zero; therefore, setting the right side of the equation above equal to zero and
solving for the corresponding x = x 1 gives

g (x 0 )
x1 = x0 − .
dg /dx|x 0

Produced by this computation is a (hopefully) better estimate x 1 of a true solution to g (x) = 0; the word
hopefully is used because the actual convergence behavior of the Newton procedure can be difficult to predict.
We can continue to refine the solution estimate by replacing the previous estimate x 0 with the newer estimate x 1
and then repeating the procedure until a sufficiently accurate solution is reached.

12.5.1 Quadratic convergence - numerical analysis


The convergence of Newton’s method for solving g (x) = 0 (for cases where the first g 0 and second g 00 derivatives
do not vanish and for sufficiently-good initial guesses) is said to be quadratic: if x = r is the numerical value of
the root we are attempting to find, in the neighborhood of x = r with  = x − r ,
1
g (x) ≈ g (r ) + (x − r )g 0 (r ) + (x − r )2 g 00 (r )
2
0 1 2 00
= 0 + g (r ) +  g (r )
2
Differentiating each term of the right side of the equation above with respect to x (i.e., with respect to ) gives

g 0 (x) = g 0 (r ) + g 00 (r ).

Thus we can analyze our Newton scheme to find

g (x 0 )
x1 = x0 −
g 0 (x 0 )
g (x 0 )
x1 − r = x0 − r − 0 0
g (x )
0 g 0 (r ) + 12 (0 )2 g 00 (r )
1 = 0 −
g 0 (r ) + 0 g 00 (r )
0 g 0 (r ) + (0 )2 g 00 (r ) − 0 g 0 (r ) − 21 (0 )2 g 00 (r )
=
g 0 (r ) + 0 g 00 (r )
1 0 2 00
2 ( ) g (r )
=
g (r ) + 0 g 00 (r )
0

so for sufficiently small 0 ,


g 00 (r )
1 ≈ (0 )2 = C (0 )2 .
2g 0 (r )
Thus, we see that after six iterations, the error becomes
63
6 = 0 C 0
CHAPTER 12. ELEMENTARY NUMERICAL METHODS 134

which normally will be sufficiently accurate if |C 0 | < 1 because, for example, if |C 0 | = 0.5
63
6 = 0.5 (0.5) = 5.4 × 10−20

which is less than round-off error. It is important to point out, however, that choosing initial guesses based on
|C 0 | < 1 do not necessarily have to converge to the fixed points used to calculate the values of C .
The analysis above illustrates the quadratic convergence behavior of the Newton method when the estimate of
the solution approaches the true value. Because the next value of the error 1 is proportional to the square of
the previous error 0 , the accuracy of the solution estimates improves with each iteration in such a way that the
number of accurate digits doubles with each iteration, i.e., if 0 = 0.001, then 1 = 0.00001 in the neighborhood
of the converged solution. We note that explicit computation of g 0 and g 00 may be used to gain further insight
into the true accuracy of the solution.

12.5.2 Newton-Raphson
Now consider a representative set of nonlinear algebraic equations
 
h1 (x, y )
h(z) = (12.4)
h2 (x, y )

we wish to find solutions to the problem written in vector form


 
x
0 = h(z) with z= .
y

We can write the Taylor’s series expansion of h at (x 0 , y 0 ) as

∂ 2 h1

0 0 ∂h1 0 ∂h1
h1 (x, y ) ≈ h1 (x , y ) + (x − x ) + (y − y 0 ) + (x − x 0 )(y − y 0 ) + ...
∂x x 0 ,y 0 ∂y x 0 ,y 0 ∂x∂y x 0 ,y 0
∂ 2 h2

∂h2 ∂h2
h2 (x, y ) ≈ h2 (x 0 , y 0 ) + (x − x 0
) + (y − y 0 ) + (x − x 0 )(y − y 0 ) + ...
∂x 0 0x ,y ∂y 0 0x ,y ∂x∂y 0 0
x ,y

or, neglecting the second and higher-order terms

h(z) ≈ h(z0 ) + J(z − z0 ). (12.5)

For the representative model, the Jacobian matrix J is found to be


 
J1,1 J1,2
J= .
J2,1 J2,2

This gives us the Newton-Raphson procedure,

z1 = z0 − J−1 h(z0 ). (12.6)

Of course, we do not actually invert the Jacobian J; instead, we use the Gaussian elimination (or similar) procedure
described in the linear systems chapter to solve

J0 u1 = −h(z0 ) with u1 = z1 − z0

for the update vector u, which is used to compute the refined solution estimate z1 from

z1 = z0 + u1 .
CHAPTER 12. ELEMENTARY NUMERICAL METHODS 135

12.5.3 Approximation of Jacobian elements by FD


Recalling our centered (12.2) and forward (12.3) finite differences, the Jacobian array can be constructed using
either of these approaches  
∂g ∂g
J|z0 = ···
∂z1 ∂z2

The perturbation δ used in the forward finite difference calculation actually varies according to the magnitude of
the variable zj to which the variation of the residuals are being computed; following [Kelly (2003)]:

δj =  max [|zj |, 1]

where  corresponds to machine epsilon (round-off error).

12.6 Class software


Class software can be found at http://dev.eng.umd.edu/adomaiti/mainSolar.html with userid: solar and
password PKUsolar2013

References
[Kelly (2003)] Kelly, C. T. Solving Nonlinear Equations with Newton’s Method SIAM Press (2003).

You might also like