Solar Energy Materials

and Solar Cells

ELSEVIER Solar Energy Materials and Solar Cells 38 (1995) 45-55

On the ultimate efficiency of solar cells

H. Kiess ., W. Rehwald
Paul Sche"er Institut, CH-5232 Vu/igen PSI, Switzerland

1. Introduction

The limits to the efficiency of solar cells are again under discussion in particular
due to the ideas of improving cell efficiency by using quantum well structured
i-layers in p-i-n solar cells or by increasing impact ionization by appropriate band
structure engineering. In various papers, the efficiency of quantum well solar cells
has been discussed [1-3] whereby the different authors came to significantly
different results. Impact ionization has also been discussed [4-6] but to my
knowledge up to now no experimental evidence for impact ionization in solar cells
has been found, i.e., internal quantum efficiency > 1 in the short-circuit current
has not been observed for photons contained in the solar spectrum. The question
arises why, since impact ionization as a phenomenon would occur in semiconduc-
tors independent of theoretical arguments. Since impact ionization is sensitive to
the band structure it might appear that band structure engineering might help to
increase the probability of impact ionization. This paper is, therefore, intended to
contribute to the efficiency limits of quantum well solar cells and cells prepared of
materials in which potentially impact ionization might play a role. The considera-
tions are simplified in order to emphasize the physical phenomena and will be first
outlined on homojunctions.

2. Consideration for homojunctions

The efficiency of solar cells is determined by the short-circuit current and the
open-circuit voltage achievable under illumination and by the fill factor. With
appropriate transport parameters the photogenerated charge carriers are collected
at the electrodes prior to recombination and thus the short-circuit current density

• Corresponding author.

0927-0248/95/$09.50 © 1995 Elsevier Science B.V. All rights reseIVed

SSDI 0927-0248(94)00214-2
 46 H. Kiess, W Rehwald / Solar Energy Materials and Solar Cells 38 (J995) 45-55

~",<-s."'--=--_-_ -==--- Ec

t; / -
::r---- Ev
p

Fig. 1: Band structure with the (n, and (p.

of cells is usually close to eNabSOrbed with e the electronic charge and Nabsorbed the
number of absorbed photons/cm 2 s.
The open-circuit voltage depends on the recombination of the charge carriers. It
is, therefore, reasonable for a discussion of the ultimate efficiency of solar cells, to
calculate the open-circuit voltage under the appropriate recombination processes
and to assume for the short-circuit current density the highest possible value
eNincident' with Nincident being now the number of photons incident per unit area
and second with energies hv > E g • With a fill factor of 0.85, this will give an upper
limit of the achievable efficiency. The open-circuit voltage is given by the separa-
tion of the quasi Fermi levels [7] for the electrons and holes:
(1)
with (n = kT In(n/Nc) and (p = kT In(p/N) (Fig. 1). Consequently, the open-cir-
cuit voltage obtainable under one sun illumination is given by the carrier concen-
trations taking into account all the various recombina:tion paths. For thermody-
namic reasons, radiative recombination is unavoidable. The highest voltage which
can be achieved for a given material is obtained if only radiative recombination
occurs.
Then, in the stationary state and under open-circuit conditions the incident
photon flux must be equal to the emitted photon flux due to the recombination of
the excited charge carriers:
g = gtherm + grad = R = R rad · (2)
The generation rate g is due to the excitation by the black body radiation of the
surrounding heat bath and of the incident additional radiation. g is then a sum of
gtherm and of grad' The total recombination rate R rad can be expressed by
R rad = r(n + n;)(p + n;), (3)
where the carrier concentrations were split into those excited by the additional
radiation (n,p) and those by the black body radiation of the heat bath.
Under sufficiently high additional radiation is g therm < grad and n i < n, p
R rad = grad =mp =m 2 , (4)
whereas without grad
_ 2
gtherm - mi' (5)
 H. /(jess, W Rehwald / Solar Energy Materials and Solar Cells 38 (J995) 45-55 47

It is now assumed that the absorbed flux of photons jN(hv)dv [1/cm 2 s] for
hv > Eg is equal to the photon flux emitted by radiative recombination of carriers
within kT of the band gap. This holds certainly for semiconductors for which the
relaxation to the band edges of carriers excited high up into the bands occurs
within 10 -13 to 10 -12 S by phonon emission, a time which is usually orders of
magnitudes shorter than the recombination lifetime, in our case here the radiative
recombination lifetime. Otherwise emission of light by hot carriers would have also
to be taken into account. Implicitly it is also assumed that impact ionization can be
neglected. This means that per absorbed photon only one e-h pair is generated.
Under these assumptions is:
(6)
What remains now is a calculation of the recombination coefficient r in order
to be able to obtain the carrier concentrations in the material and the open-circuit
voltage through Eqs. (1) and (4).
The recombination coefficient r can be calculated since in thermal equilibrium
detailed balance exists between the incident and the emitted radiation [8]. The
black body radiation falls on the semiconductor at a rate of
f( hv) = 21Tv 2.:1v /( c 2 [exp( hv /kT) - 11). (7)
These photons are absorbed at a depth d. Hence, an equivalent stream of
photons is emitted from the same region (if excited e/h pairs diffuse deeper into
the semiconductor and recombine radiatively there, the emitted light is reabsorbed
and does not reach the surface). The emitted photons per cm 2 s for a material
whose refractive index is b are given by m;d/2b 2• We obtain:
(8)
If we consider only the photons in a range kT of the bandgap, then is hv = Eg
and h.:1v = kT. Detailed balance gives then for the recombination coefficient r
(9)
and with n; = NcNv - exp( - Eg/kT) and with E g > kT the recombination coeffi-
cient r is found to be
(10)
The product of the effective density of states can be expressed by the effective
masses of the electrons and holes, being

(11)
The rate constant r is determined by the gap of the semiconductor, its
refractive index b, the absorption depth d and the intrinsic carrier concentration
(or its density of states or the effective masses mn and m p )' A change of mn and
mp by band structure tailoring may, therefore, induce a change of r even if the
band gap of the material remains unchanged.
48 H. Kiess, W Rehwald / Solar Energy Materials and Solar Cells 38 (J995) 45-55

We assume now that the rate constant r is independent of the excitation levels
under consideration. Without loss of general validity for the further conclusions we
assume now that the semiconductor is intrinsic. Then the excited carrier densities
n, p are large compared to those in the dark and n = p. The carrier concentration
n, if we make use of Eqs. (3) and (5) is easily calculated to be:
(12)
From Eq. (12) we conclude that the only material parameters which influence
the carrier concentration are the effective density of states Nc and Nv (or effective
masses of the carriers) and the bandgap of the material. Details of the absorption
and of the emission of the light are irrelevant. This is, of course, in agreement with
more rigorous thermodynamic considerations.
The quasi Fermi levels are readily calculated using the formula
(13)
and
(14)
Whence the open-circuit voltage is obtained using Eq. (1) and assuming that no
optical reflection or transmission losses occurs, i.e., that Nabsorbed = Nincident'
eVoc = Eg + kT In (h 3c 2/2TrkT) (Nincident/Eg
. 2) . (15)
This formula is of remarkable simplicity and shows, that Voc depends only on
the band gap and on the number of photons of the solar spectrum of energy
hv > Eg which are absorbed per unit area and time by the cell.
The efficiency 'T1 of the cell referred to the incident intensity Iincident = 100
mW/cm 2 of the solar radiation at AM 1.5 is then, with the assumption made on
the short-circuit current, given by

'T1 = [Nincident{ Eg + kT In [( h 3c 2/2TrkT)( ~ncident/E;)]} /Iincident. (16)

Results of the open-circuit voltage are given in Table 1. The open-circuit
voltages are in excellent agreement with those of Smestad and Ries [9]; (their
values are given in parenthesis in Table 1). The efficiencies calculated with Eq.
(16) have a maximum for materials with band gaps in the range of 1.2 to 1.3 eV.
They are slightly higher than those from rigorous thermodynamic theory [11],
which is mostly due to the assumption on the fill factor f.

Table 1
Open-circuit voltage of materials with various band gaps in the radiative limit
Material GaAs Si
Band gap 1.42 1.1 1.0 0.9 0.8 0.7 0.6 0.5
Efficiency 28.2 30.6 29.9 28.6 25.7 24.3 20.8 17.7
Voc [mY] 1147 (1150) 860 (860) 775 680 590 502 413 326
 H. Kiess, W. Rehwald / Solar Energy Materials and Solar Cells 38 (1995) 45-55 49

3. Open-circuit voltage of solar cells with quantum well structure

In a quantum well structure the electrons are confmed perpendicular to the well
and their kinetic energy is quantized into discrete energy levels. This leads to a
modification of the density of states compared to the three-dimensional isotropic
case (compare Figs. 2 and 3). The density of states in the jth subband is given by
[10]:
(17)
with L z the width of the well and with m~ the mass of the electrons in the jth and
subband. If we assume that the subbands are separated by more than kT and that
only the lowest subband is occupied, then the carrier densities in the electron and
hole subbands can be calculated with the Boltzmann statistics.
The detailed balance argument gives then as for the homogeneous material for
the recombination coefficient r
(18)
This is the same expression as Eq. (10), except that now the band gap has to be
replaced by Eo where Eo = Eg + the confinement energy (see Fig. 3).
In order to calculate the carrier concentration under illumination, we assume
again that the photoexcited carriers relax to the "gap energy" Eo within a time
short compared to the recombination lifetime. This leads to the expression for n
given in Eq. (9), in which Eg is replaced by Eo. Hence, the open-circuit voltage is
given by:

eVoc = Eo + kT In ( h 3c 2 /2TrkT) ( Nincident/

•
Eo2) . (19)
The open-circuit voltage of a quantum well solar cell corresponds to that of a
semiconductor with Eg = Eo [3]. Quantum well structures can be used to shift the
threshold for absorption to a more favourable position with respect to the solar
spectrum giving rise to slightly higher Voc and, hence, to slightly higher efficiency.
This conclusion is based on the assumption that carriers relax to the lowest
states by thermalization with the lattice. This assumption has been confirmed for

Kx • Ky
Fig. 2. Energy as a function of k.
50 H. Kiess, W. Rehwald / Solar Energy Materials and Solar Cells 38 (J995) 45-55

I vI
L \/

~ J7'
1 DENSITY OF STATES

Eg
I .....~ .. _

L
l f.

1\
7
Fig. 3. Density of states as function of E.

multi quantum well systems by luminescence measurements, in which emISSIon

from these states was observed [12]. Hot luminescence in the picosecond range,
however, has been observed in super structures and it had been concluded that the
efficiency of hot carrier devices with an appropriate extraction of the carriers could
principally greatly exceed that of quantum devices in which carriers equilibrate
with the lattice [13,14]. On the other side it has been pointed out that the quasi
Fermi level of hot carriers of temperature T> Tlattice is closer to the center of the
forbidden gap, hence the open-circuit voltage of a hot carrier device would be
principally lower than that with thermalized carriers [15]. It seems to be true that
in the radiative limit the efficiencies of quantum well solar cells of gap energies Eo
do not differ from those of homogeneous materials with gap energy Eg = Eo.

4. Effects of impact ionization and auger recombination

4.1. Impact ionization

The short-circuit current without impact ionization was given by

(20)
whereby the quantum efficiency q was assumed to be unity. However, if the
photon energy is sufficiently high, impact ionization may occur and the quantum
efficiency will become a function of photon energy. We assume now that the
excited carriers are subject only to two competing processes before recombination,
 H. Kiess, W Rehwald / Solar Energy Materials and Solar Cells 38 (1995) 45-55 51

namely relaxation to the band edges either by phonon emission (q = 1) or by

impact ionization (q = 2; it is not expected that photon energies hv < 4 eV suffice
to give rise to more than one additional e-h pair). Then
q(E) = [2rimp+rph]/[rimp+rph]' (21)
In general rimp as well as r ph are functions of the energy. The factor 2 takes into
account that impact ionization generates two e-h pairs. Impact ionization is only
possible above a threshold energy Eth (above the band edge) if phonon assisted
processes are negligible. The short-circuit current is therefore:

(22)
Impact ionization contributes to I sc ' if r irnp == rph' Using the theory of Keldysh
[16] one finds that rirnp is approximately
rimP=ro[(E-Eth)/Eg12, (23)
with
ro = (e 2/47T'EEo)\mnm/h3)Ie2I;/(l + 2p.)3/2,
Eth = (1 + 2p.)Eg /(1 + p.),
p. =mn/mp,
Ie' Iv overlap integrals.
r0 and Eth are functions of the effective masses, of the band gap and of the overlap
integrals Ie and Iv' ro is 5 X lOll/S for a semiconductor with effective masses
equivalent to those of Si(m n = 0.33, mp = 1.26) and with overlap integrals Ie = 1
and Iv = 0.98 (see Ref. (16]). Therefore, the impact ionization rate in a semicon-
ductor of Eg = 1.12 eV (Si) near threshold (E - E th ) '" kT is low and of the order
of 109 /s. This estimate agrees astonishingly well with numerical simulations
[16-18] which give values between lOs-10 lO /s. Therefore, the energy of the
photons using the Keldysh formula (23) has to be hv == 3 Eth if the rate of impact
ionization is to reach 2 X lO12/s, a value comparable to the phonon emission rates.
If phonon-assisted processes are ignored, the threshold energy for impact ioniza-
tion has to be at least Eth = E g. With Eth == Eg. impact ionization is expected to
playa role only in solar cells with energy gaps smaller than 1 eV since the flux of

Table 2
The effect of impact ionization on the short-circuit current for cells with energy gaps between 1.0 eV
and 0.5 eV
Band gap leV] 1.0 0.8 0.7 0.6 0.5
Relative increase of Is;: 3.3% 12% 20.8% 28.5% 42.3%
Isc with impact ionisation [mA/cm 2 ] 48.7 60.5 71.3 79.4 94.5
Isc no impact ionization [mA/cm 2 ] 47.2 54.1 59.1 61.8 66.2
52 H. Kiess, W. Rehwald I So/ar Energy Materials and Solar Cells 38 (1995) 45-55

photons of energy hv > 3 e V in the solar spectrum is small compared to that for
hv < 3 eV. Table 2 gives the expected relative increase of the short-circuit current
due to impact ionization as a function of gap energy between 0.5 and 1.0 e V
assuming that all carriers excited by photons of hv > 3Eg contribute to the
short-circuit current with quantum efficiency q = 2. Furthermore, these model
semiconductors are assumed to have all the same effective masses (those of
electrons and holes of silicon), i.e., only effects of the varying band gaps are
considered.
It is, however, true that these assumptions are ~ot generally valid: ro as well as
the threshold energies for impact ionization Eth do depend on details of the band
structure (effective masses) so that semiconductors with the same band gap but
different effective masses might in reality have lower or higher impact ionization
rates [19].
In addition one has to be aware of the fact, that the increase of the efficiency
will be less than expected: an increase of the short-circuit current due to impact
ionization might be accompanied by a decrease of the open-circuit voltage due to a
higher Auger recombination provided that Auger recombination is the dominant
recombination mechanism. Impact ionization without Auger recombination shifts
the maximum of the efficiency to materials with band gaps in the range between
0.9 and 1.1 eV, however, with Auger recombination it is between 1.1 and 1.3 eV.
The efficiencies with Auger recombination below 1.1 eV are found to be smaller
than those in the radiative limit (Fig. Sa and Sb). This will be discussed in the
following section.

4.2. Auger recombination

In order to obtain an estimate of the magnitude of Auger recombination, it

would be desirable to have a functional dependence of the Auger recombination
coefficient C on the parameters of the semiconductor. In thermal equilibrium
detailed balance requires that the impact ionization rate is compensated by the
corresponding Auger recombination rate. If this argument is applied to an intrinsic
semiconductor, then
(24)
where the concentration of holes p = n i • n is the carrier concentration at Eth
above band edge in a range kT, the density of carriers at higher energies being
assumed to be so small as to be negligible for the impact ionization. Using the
expression (23) for r from the theory of Keldysh with E - Eth ~ kT and assuming
Ec - EF = E g /2, we find:
CpNcN} exp( -3Eg/2kT) = ro(kTIEg/n, (25)

41Tro(2mn)3/2(1 + 2JL)1/2 exp( -JLEg/(1 + JL)kT)

C =--------------------~~--~--------~ (26)
p Nc2Nvh3(1+JL)1/2Ei/2 '
 H. Kiess, W. Rehwald / Solar Energy Materials and Solar Cells 38 (J995) 45-55 53

Q9

Q8
:1
<17
as
Q5
'b: Q4 I~

I
Q3
Q2
I
Q1 I
o
Q5 <16 <17 Q8 <19 1.1 1.3

Fig. 4. Open-circuit voltage in [V] as a function of band gap between 0.5 and 1.12 eV, (0) without and
(0) with Auger recombination.

The same functional dependence of Cp on the energy gap has been derived by
Ridley [16]) in a different way. From (26), Cp = 4.9 X 10- 31 (cm 6 js) for a semicon-
ductor with an energy gap and effective masses of the carriers equivalent to those

...
•• • • •
35
:J)

25
EfIIc:Iency :!)
r,q 15
I
10
5
o
a
0.5 0.7 0.9 1.1 1.3
Band gap [eV)

Band gap [eV)

0.5 0.7 0.9 1.1 1.3
35

••
30 IA
25 • -
Efficiency 20
[%) 15 •
10
5
o b
Fig. 5. a: Efficiency of solar cells as a function of band gap between 0.5 and 1.12 eV, in the radiative
limit without (0) and with (0) impact ionization. b: Efficiency of solar cells as a function of band gap
between 0.5 and 1.12 eV with impact ionization and radiative recombination (0). The lower curve
depicts the efficiencies with impact ionization and Auger recombination (0).
 54 H. Kiess, W. Rehwald / Solar Energy Materials and Solar Cells 38 (1995) 45-55

of silicon, a result which is in close agreement with experimentally determined

values of Cp for Si [20]. If we assume that the effective masses do not change with
the gap energy as before, then Cp depends only on the energy gap and the increase
in Auger recombination with decreasing gap and the corresponding open-circuit
voltages can be calculated. The open-circuit voltages in the radiative limit and
under Auger recombination are compared in Fig. 4. assuming that impact ioniza-
tion is operative. Above approximately 1.2 eV Auger recombination is smaller than
radiative recombination. The combined influence of impact ionization and of
Auger recombination on the efficiency is depicted in Fig. 5a and 5b and compared
with the radiative limit. The carrier concentrations and the open-circuit voltages
. were calculated as in section 2 for n > n i , assuming additionally that all optically
generated charge carriers are homogeneously distributed in a 200 f.Lm thick
semiconductor. Auger recombination is seen to become important for semiconduc-
tors of band gaps < 1 e V. The Auger recombination rate is expected then to be
higher than radiative recombination and to reduce the efficiencies below the
radiative limit. This is true especially at low band gaps. The Auger recombination
rates are higher if the semiconductor would have been chosen to be thinner
and/or doped. The results on open-circuit voltage and efficiency can, therefore, be
considered to present an upper limit if possible band structure effects are ne-
glected.

5. Conclusions

The considerations here show that an increase of the efficiency of solar cells
beyond presently accepted thermodynamic limits by utilizing quantum well struc-
tures or impact ionization is difficult to achieve. The physics behind is the fast
relaxation of the photoexcited charge carriers to the band edges, which prevents
Voc to be determined by the larger band gap material in quantum well structured
systems and impact ionization to become significant already for carriers excited by
photons of energies hv = Eth + Eg "'" 2Eg. In addition, the expected increase in
efficiency solely due to impact ionization is counteracted by a decrease caused by
Auger recombination which is the basic limiting recombination mechanism in small
band gap materials in which impact ionization is expected to occur in the useable
photon range of the solar spectrum. In order to overcome these limitations, the
relaxation time of carriers to the band edges by phonon emission should be
increased for impact ionization to become also significant in the photon range
2Eg < hv < 3Eg • The observation of hot luminescence in superstructures gives a
hint that phonon bottle necks exist and might be used to improve efficiencies of
solar cells.

Acknowledgements

This work has profited by discussions with various groups, in particular with
colleagues from the photovoltaic project group at PSI, at the workshop on
 H. Kiess, W. Rehwald / Solar Energy Materials and Solar Cells 38 (]995) 45-55 55

'Fundamental Research on Quantum Solar Energy Conversion' in Brigels, Switzer-

land, 1994, and finally at the workshop held in conjunction with IPS-10, Interlaken,
Switzerland, 1994.

References

[1) K.W.J. Barnham and G. Duggan, JAP 67 (1990) 3490.

[2) R. Corkish and M.A Green, Proc. 23rd IEEE Photovoltaic Specialists Conf. (1993) p. 675.
[3) G.L. Araujo, A Marti, F.W. Ragay and J.H. Wolter, Proc. 12th EPSEC, Amsterdam, 1994, p.
1481.
[4) S. Kolodinski, J.H. Werner, Th. Wittchen and H.J. Queisser, JAP 63 (1993) 2405.
[5) P.T. Landsberg, H. Nussbaumer and G. WiIleke, JAP 74 (1993) 1451.
[6) P.T. Landsberg, John K. Liakos and Alexis de Vos, Proc. 12th EPSEC, 1994, p. 1343.
[7) W. Ruppel and P. Wiirfel, IEEE Trans. Electron Dev. 27 (1980) 877.
[8) e.g. A Rose, Concepts in Photoconductivity and Allied Problems (Interscience Tracts on Physics
and Astronomy, 1963).
[9) G. Smestad and H. Ries, Sol. Energy Mater. Sol. Cells 25 (1992) 51.
[10) see e.g. G.P. Agrawal and, N.K. Dutta, Long-Wavelength Semiconductor Lasers (Van Nostrand
Reinhold, 1986).
[11) e.g. G.L. Araujo and A Marti, Proc. 11th EPSEC, 1992, p. 142.
[12) F.W. Ragay, J.H. Wolter, A Marti and G.L. Araujo, Proc. 12th EPSEC, Amsterdam, 1994, p.
1429.
[13) R.T. Ross and AJ. Nozik, JAP 53 (1982) 3813.
[14) M. Archer, in: J.S. Connolly (Ed.), Photochemical Conversion and Storage of Energy (1981).
[15) H. Ries, Private Communication.
[16) see e.g. B.K. Ridley, Quantum Processes in Semiconductors (Clarendon Press, 1993).
[17) J. Bude, K. Hess and G.J. Iafrate, Semicond. Sci. Technol. 7 (1992) B 506.
[18) W. Quade, F. Rossi and C. Jacoboni, Semicond. Sci. Technol. 7 (1992) B502.
[19) Jasprit Singh, Semicond. Sci. Technol. 7 (1992) B 509.
[20] M.A Green, High Efficiency Solar Cells (Trans Tech Publications, 1987).