1138 IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 4, NO. 4, JULY 2014 Correlation of Built-In Potential and I–V Crossover in Thin-Film Solar Cells James E. Moore, Member, IEEE, Sourabh Dongaonkar, Member, IEEE, Raghu Vamsi Krishna Chavali, Student Member, IEEE, Muhammad Ashraful Alam, Fellow, IEEE, and Mark S. Lundstrom, Fellow, IEEE Abstract—Thin-film solar cells often show a crossover between the illuminated and dark I–V characteristics. Several device specific reasons for crossover exist and have been discussed extensively. In this paper, we show that a low contact-to-contact built-in potential can produce a voltage-dependent photocurrent that leads to I–V crossover at a voltage that is almost exactly the device built-in potential. This mechanism can produce crossover in the absence of carrier trapping or recombination. It can be a contributing factor to crossover, but when an anomalously low contact-to-contact built-in potential exists, it can be the dominant factor. Using numerical simulations, we examine a variety of model solar cell structures with low contact-to-contact built-in potential and show a strong correlation of the crossover and built-in potential voltages. These simulations also suggest that a plot of the illuminated minus dark current may help identify when a low Vbi is limiting device performance. Index Terms—Built-in potential, device modeling, injection current, photogenerated current, superposition. I. INTRODUCTION HE superposition principle commonly used to describe the current in crystalline solar cells [1] defines the total current measured under illumination as T Jlight (V, G) = Jinj (V ) + Jgen (G) (1a) where Jinj is an injection current, which is typically assumed to be equal to the measured dark current, and Jgen is the photogenerated current, which is typically assumed to be voltage independent and equal to the measured short-circuit current. Electrical measurements of thin-film solar cells, however, often show a crossover point between the illuminated and dark I–V characteristics, which cannot occur according to (1a). At the crossover voltage Vx , the dark and illuminated currents are equal (see Fig. 1). Although I–V crossover is not always as dramatic as illustrated in Fig. 1, it can, in some cases, lower open-circuit voltages and fill factors [2]–[5]. Moreover, such a crossover complicates Manuscript received October 11, 2013; revised March 4, 2014; accepted March 18, 2014. Date of publication May 6, 2014; date of current version June 18, 2014. This work was supported by the SRC Energy Research Initiative (ERI) and the SunShot Program funded by the U.S. Department of Energy. The authors are with the School of Electrical and Computer Engineering, Purdue University, West Lafayette, IN 47907 USA (e-mail: moore32@purdue. edu; sourdon@gmail.com; c.raghuvamsi@gmail.com; lundstro@purdue.edu; alam@purdue.edu). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JPHOTOV.2014.2316364 Fig. 1. Experimental data showing I–V crossover in a CdTe solar cell with a Schottky back contact [4], an a-Si solar cell and a CZTSSe solar cell similar to those reported on in [26]. Solid lines show dark I–V and dashed lines show illuminated I–V. the reciprocity relation between absorption and emission in a solar cell [6]; therefore, it is important to understand the physical origin of the crossover phenomenon. Several different mechanisms can cause J–V crossover and these have been extensively discussed in the literature (e.g., [2]–[5]). The key message of this paper is that a low overall device built-in voltage Vbi,d can produce a voltage-dependent photocurrent and cause I–V crossover. Here, Vbi,d refers to the contact-to-contact total built-in potential difference of the solar cell, while Vbi,j refers to the built-in potential of the individual junction within the solar cell. For example, the presence of a Schottky contact does not change Vbi,j , but it lowers Vbi,d . Surprisingly, we will show that Vbi,d related crossover can occur in many different device structures, even in the absence of traps and bulk recombination. It provides a simple, physically transparent explanation for crossover due to a Schottky back contact and for one of the factors that are contributing to cross-over in p-i-n cells. It can be the dominant factor for cells with anomalously low Vbi,d . This paper highlights the need to consider low Vbi,d as a possible cause of crossover in solar cells. II. BACKGROUND AND APPROACH Physical mechanisms that can cause superposition failure have been discussed extensively in the literature [2]–[6], but most of these studies focus on a specific technology. Fig. 1 shows I–V crossover in three different types of solar cells. In this paper, we will describe a physical mechanism, low contactto-contact built-in potential that can produce I–V crossover and will discuss how this mechanism can contribute to the observed 2156-3381 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications standards/publications/rights/index.html for more information. MOORE et al.: CORRELATION OF BUILT-IN POTENTIAL AND I–V CROSSOVER IN THIN-FILM SOLAR CELLS 1139 I–V crossover for the three different types of cells illustrated in Fig. 1. This includes ink-based CZTSSe, for which recent work shows an experimentally measured crossover voltage that is correlated with Vbi,d and independent of light wavelength and intensity [7]. Understanding these measurements is one key motivation for the study presented in this paper. Explanations for I–V crossover due to voltage-dependent current collection have been given previously [2], [3], focusing mainly on devices with short diffusion lengths. Similarly, I–V crossover is observed in a-Si solar cells and has been found to be directly related to the built-in voltage of the p-i-n junction [5]. General equations describing the mechanism have been given [8], [9], but have only recently been noticed and applied to solar cells [10]. The presence of a Schottky barrier back contact produces two diodes in series [11]. Crossover in the presence of a Schottky barrier has been analyzed in terms of this twodiode model [4], [12]. Several papers have also noted that the photoexcited deep trap levels in the window layer can change the electrostatics of the device with illumination that lead to crossover in Cu2 S and CIGS [13]–[16]. It has been shown that this effect can be identified by a change in the I–V characteristic under long- or short-wavelength light [13], [14]. To examine the physics of I–V crossover, (1a) must be generalized to with low contact-to-contact Vbi,d demonstrate that the same effect can occur in a wide variety of solar cell structures—when the contact-to-contact Vbi,d is low. In Section V, we discuss the validity of the assumptions made in the previous sections, and we conclude in Section VI that Vbi,d related crossover should be considered as a possible cause when diagnosing crossover in solar cells. Jlight (V, G) = Jinj (V, G) + Jgen (V, G) . Consider first an n-i-p homostructure as shown in Figs. 2(a) and (b). For this structure, I–V crossover occurs because of a voltage-dependent photocurrent, which can be readily understood in terms of carrier partitioning between two contacts. As shown in Fig. 2(a), under short-circuit conditions, most photogenerated carriers in the i-layer are swept out to the correct contacts by the built-in electric field. However, at the crossover voltage Vx , the bands are nearly flat, and there is an equal probability that photogenerated carriers will diffuse to either of the two contacts [see Fig. 2(b)]. This equal partitioning of the photogenerated carriers between the two contacts causes the photogenerated current to be zero at the built-in potential where crossover occurs. For even larger biases, the electric field and the photocurrent change direction. Electrons are now collected by the p+ contact and holes by the n+ contact. These effects are well documented experimentally [7], [10]. Next, we examine an N-i-p heterostructure with a wide bandgap N+ layer, which creates a blocking barrier in the valence band [see Fig. 2(c)]. Solar cells such as CIGS and CdTe often have a wide bandgap window layer at the front contact similar to this structure [20], and therefore, this model can help us understand crossover in these types of cells. In this model cell, carrier partitioning between the two contacts occurs for electrons, but the holes can only exit from the p+ contact [see Fig. 2(c)]. The photocurrent will be voltage dependent, but it cannot reach zero unless all electrons and holes exit from the p+ contact [see Fig. 2(d)]. However, as Fig. 2(d) also shows, photogenerated holes will accumulate at the heterojunction under large forward bias. The internal electric fields will, therefore, be different under illumination and in the dark, and therefore, the diode injection current becomes generation dependent. For this structure, the crossover voltage (1b) We find it useful to plot ΔJ = Jlight − Jdark = {Jinj (V, G) − Jdark } + Jgen (V, G) (1c) which shows that I–V crossover can occur in two ways. First, the diode injection current at a given forward bias may change under illumination. In this case, the first term in brackets in (1c) is dominant. Second, the photogenerated current may have a strong voltage dependence and go to zero at V = Vx . In this case, the second term of (1c) is dominant. Both terms may play a role in crossover; depending on which one is dominant, the plot of (1c) will display different characteristic shapes. Using a specially modified version of the numerical simulation program, ADEPT [17], [18], we will perform the current decomposition of (1b) exactly and examine the relative contributions of the two terms for various solar cell structures. Additional simulations were also done using SCAPS-1D [19]. The objectives of this paper are to demonstrate a close connection between the crossover and device built-in voltages for a variety of solar cells and to discuss the physics of this close correlation. This study provides a simple, unifying description of effects that have been considered to be distinct. We examine several different model structures with detailed, numerical simulations. We begin in Section III by analyzing a simple n-i-p structure for which I–V crossover can be simply explained by a voltage-dependent photocurrent. We then examine the effect of a wide bandgap n layer with a hole blocking heterojunction (because this type of structure is common [20]). This example shows how the shape of the ΔJ characteristic is related to the internal physics of the cell. In Section IV, we examine other solar cell structures and show that simulations of model structures III. N-I-P MODEL STRUCTURES In this section, we consider n-i-p homo- and heterostructures, as it allows us to develop an analytic model that illustrates the basic physics of crossover associated with the built-in potential. For simplicity, we initially assume no trap states within the bandgap, no bulk recombination, and uniform photogeneration (these nonidealities will be addressed in Section V). We will initially assume that the injection current term in (1c) is negligible, although we will later show that it is important in the heterojunction case. Although other effects such as short diffusion lengths will most likely affect superposition failure in actual solar cells, we show that Vbi related crossover can occur, even in this ideal case. A. Physics of I–V Crossover 1140 IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 4, NO. 4, JULY 2014 Fig. 2. Numerically simulated energy bands for (a) n-i-p homojunction at short circuit, (b) n-i-p homojunction at crossover, (c) n-i-p heterojunction with valence band offset at short circuit, and (d) n-i-p heterojunction with valence band offset at crossover. will be determined by both a bias-dependent photocurrent and a generation-dependent injection current. The qualitative physics of these two model cells can be described quantitatively by analytical analysis and by full numerical simulation. B. Analytic Treatment of n-i-p Structures Analytic expressions for the two terms in (1b) have been derived for an insulator with two contacts in [9] and shown to apply to an a-Si n-i-p solar cell in [10]. These expressions are summarized in Appendix A. The resulting expressions are found to be in agreement with those presented in [8]; however, separation of injection and photocurrent provide additional insight into the physical mechanisms behind I–V crossover. At voltages significantly below Vbi,d , these equations show that the device behaves according to (1a). As the voltage approaches Vbi,d , the generation current goes to zero for the reason summarized in Figs. 2(a) and (b). In this case, Vx and Vbi,d are therefore identical. The C–V measurements, which can normally be used to estimate Vbi,j do not work for n-i-p cells, and therefore, this observation offers a novel alternative for measuring n-i-p solar cells such as a-Si. For V > Vbi,d , the Jinj equation reduces to the resistor current equation Jinj = (nqμn + pqμp )E, and Jgen changes direction, as was explained in the previous section. Fig. 3. Comparison of analytical (symbols) and numerically simulated photocurrent (lines) for n-i-p homojunction and heterojunction. The difference in ΔJ is due to additional band bending induced by photogenerated charge accumulation, which is not accounted for in the analytic expression. The symbols plotted in Fig. 3 (crosses) show the analytical result for the voltage-dependent generation current of the n-i-p cell. The generation current goes to zero at Vbi,d , and crossover occurs. If the n and p regions were highly doped, Vbi would be large, and Jinj Jgen at the crossover point. Open-circuit conditions would occur well below Vbi,d , where the generation current is approximately constant, and superposition would apply. When Vbi,d is small, however, crossover occurs at relatively low injection current, i.e., near Voc . MOORE et al.: CORRELATION OF BUILT-IN POTENTIAL AND I–V CROSSOVER IN THIN-FILM SOLAR CELLS Analytic expressions can be similarly derived for the structure of Fig. 2(c) with a hole blocking heterojunction (the equations are presented in Appendix A). The injection current is the same as for the n-i-p homojunction case, except the term for holes vanishes, since the hole current is blocked by the valence band barrier (ignoring bulk recombination). In this case, however, Jgen is not symmetric about Vbi,d but is asymptotic to zero at large forward bias (see Fig. 3 circles). This occurs because at large forward bias, all electrons and holes leave from the p+ contact, and therefore, the generation current approaches zero. Because Jgen never reaches zero, crossover should not occur, but this analysis does not account for the change in electric field due to hole accumulation, as indicated in Fig. 2(d). The added positive charge from this accumulation acts as an additional forward bias on the junction and allows crossover to occur. An analytical discussion of this accumulation effect is also presented in Appendix A. To understand why the current partitioning limited crossover is determined by the contact-to-contact potential Vbi,d (rather than the junction potential Vbj,j ), consider, for example, a heterojunction formed by two intrinsic semiconductors with an energy barrier of Δχ at the interface due to different material electron affinities, similar to the HOMO/LUMO levels of an organic photovoltaic cell. Such a device will have no Vbi,j contribution from doping, yet the energy barrier provides a means of partitioning carriers between contacts [21]. Crossover may still occur, therefore, when the device is biased above Vbi,d = Δχ. C. Numerical Simulation of n-i-p Structures The analytic calculations that are discussed in Section III-B support the qualitative picture of crossover presented in Section III-A and can be confirmed by full numerical simulations. The decomposition of (1b) can be done exactly using numerical simulation. All simulation parameters, including mobility, lifetime, layer thickness, bandgap, and doping concentration can be found in Appendix B. First, the illuminated I–V characteristic Jlight (V, G) is computed by solving the semiconductor equations numerically. At each bias, we then turn OFF the illumination and compute the injection current Jinj (V, G) by using the self-consistent electrostatic potential that was computed under illumination [22], [23]. The photogenerated current Jgen (V, G) is then computed from (1b). The numerical results can then be compared with the analytical results presented as symbols in Fig. 3. It should be noted that the efficiencies of the simulated structures with low Vbi,d tend to be below 12%. Although this would be considered quite low performance for a mature solar cell technology, it would not be unreasonable to find these efficiencies in new technologies still under development. We, therefore, note that the theory presented in this paper would be useful to characterize experimental solar cells but would not necessarily be relevant to mature technologies such as c-Si. As shown in Fig. 3, the analytic results for Jgen (symbols) and the numerical results (solid lines) compare well. In addition, shown in Fig. 3 is the numerically computed ΔJ = Jlight − Jdark . For the n-i-p homostructure, Jgen (dotted line) and ΔJ (solid line) are identical up to the crossover 1141 voltage. The crossover is, therefore, almost entirely due to the voltage-dependent photogenerated current. The crossover voltage is exactly equal to the built-in potential, as expected. For the case with the hole blocking heterojunction, charging effects make {Jinj (V, G) − Jdark } > 0 in (1c). As a result, crossover does occur, due to both terms in (1c). In this case Vx is close to, but not exactly equal to, Vbi,d . We have shown by physical arguments, with analytic calculations, and by numerical simulations, that I–V crossover in n-i-p cells is directly related to the built-in voltage. Surprisingly, as we will show next, this can also be the case in very different types of solar cell structures. IV. N-P MODEL STRUCTURES In the previous section, it was shown that the voltage at which I–V crossover occurs is equal to Vbi,d for an ideal n-i-p homostructure and at a voltage close to Vbi,d for the case with a heterojunction that blocks holes. In this section, we will look at numerical simulations of the following three additional device structures: 1) a symmetrically doped n-p junction [see Figs. 4(a) and (b)], 2) an n-p junction with a Schottky back contact similar to that reported for CdTe [4] (see Figs. 4(c) and (d)], and 3) an N-p heterojunction similar to that found in CIGS or CZTS. The structures we examine here are designed to have built-in potentials well below the bandgap so that crossover can be observed under normal operating conditions at moderate forward bias. Specific simulation parameters for all devices are listed in Appendix B. Because these structures do not have a constant electric field like the n-i-p structures, analytic expressions are more difficult to obtain, and therefore, we will examine the relationship between Vbi,d and Vx , as obtained by simulation. A. Homojunction Model Structures The first structure is a symmetric n-p junction with modest doping on both sides of the junction [see Fig. 4(a)]. We vary the doping in the n- and p-layers from 1014 to 1017 cm−3 and plot Vx versus Vbi,d . For these modest doping densities, the builtin potential is low; therefore, crossover occurs near Voc . The behavior of the n-p structure near Vx is similar to that of the n-i-p structure, as expected [see Fig. 4(b)]. Simulations show that the crossover voltage Vx is equal to Vbi,d , just as it was for the n-i-p structure (see Fig. 5 diamonds). For the second structure, we leave the doping of the n and p regions constant at 1016 cm−3 and vary the work function of the back contact, changing the built-in Schottky barrier height between 0 and 0.3 eV. Simulation results again show that I–V crossover occurs very near the contact-to-contact Vbi,d , which is given by Vbi,d = Vbi,j − Vbi,sb (2) where Vbi,j is the built-in voltage of the p-n junction, and Vbi,sb is the built-in voltage of the Schottky junction, which is given by the difference in work functions of the p-layer and the metal contact, as shown in Fig. 4(c). 1142 IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 4, NO. 4, JULY 2014 Fig. 4. Energy band diagrams at short circuit plotted using SCAPS-1D [18] for (a) symmetric n-p junction (n and p doping 101 6 ), (b) n-p junction biased at V b i , (c) n-p junction with a Schottky back contact (V b i , s b = 0.3 eV), and (d) device with the Schottky back contact biased at V b i . Fig. 5. Numerical simulation results in SCAPS-1D for crossover voltage plotted versus built in voltage for the three homojunction device structures discussed. All are closely related to V b i , but the n-p and sb results may be shifted slightly above or below due to small changes in band bending under illumination. is symmetrical [see Fig. 4(d)]. The photogenerated carriers in the bulk partition equally between the front and back contacts and Jgen = 0. Even though the bands are not perfectly flat, the symmetry in band bending leads to I–V crossover just as we saw for the simple n-p device. Again, we see that Vx is equal to Vbi,d (see Fig. 5 triangles). This approach provides a simple explanation for the enhanced electron current described in [13]. Fig. 5 shows a strong correlation of Vbi,d with Vx for a wide voltage range, indicating that the current partitioning effect is present in all devices. In practice, however, we would only expect to see this effect in devices with Vbi,d well below the bandgap (1.1 eV for this simulation). This is simply due to the fact that when a large forward bias is applied to the device, the injection current would be far too large to measure properly. This type of crossover would, therefore, probably not be visible in measurements of real devices above Vbi,d ≈ 2/3Eg . B. CIGS-Like Model Structure As explained in [4] and [11], the Schottky junction changes from forward to reverse bias as the operating point moves from the fourth to the third quadrant, but it is remarkably easy to explain the crossover voltage. At V = Vbi,d , the band bending Finally, we examine an N-p heterojunction with a valence band barrier blocking the N contact. This structure is similar to that found in thin-film solar cells like CIGS [15]. It is similar to the N-i-p heterostructure in Fig. 2(c) but without the MOORE et al.: CORRELATION OF BUILT-IN POTENTIAL AND I–V CROSSOVER IN THIN-FILM SOLAR CELLS 1143 Fig. 6. Band diagram for N-p heterojunction plotted using SCAPS-1D (a) at short circuit and (b) biased at V b i , d . The window layer was modeled with parameters similar to CdS, and the absorber was modeled with parameters similar to CZTS. ilar CdTe structure with a p-side doping of 1014 . If the n-side doping is small, less band bending will occur on the p-side and there will be a smaller accumulated charge caused by photogenerated holes, so the device will require a higher voltage to achieve crossover. In either case, we see that a cell of this type can experience I–V crossover, even in the absence of traps or bulk recombination if the device simply has a low overall Vbi,d . Because low Vbi,d is associated with a low open-circuit voltage, the type of crossover discussed in each of the cases so far is an indicator of lower efficiency. V. DISCUSSION A. Effects of Traps at a Heterojunction Interface Fig. 7. Simulation results using SCAPS-1D for a CIGS like heterojunction. The dotted line is a reference showing the built-in voltage equal to the crossover voltage. Each set of data points shows the trend for a given n-layer doping, and the p-type base doping is varied from 101 3 to 10 1 6 in each case. intrinsic layer and, therefore, behaves in a similar manner. The model structure uses an n-type doping that results in a low Vbi,d determined by the front contact work function and the p-type absorber doping. By lowering the Vbi,d in this manner, the Voc is lowered and crossover occurs near Voc . Recent experiments show that this mechanism could be the cause of crossover in CZTSSe cells [7]. The simulated band structure under short circuit is shown in Fig. 6(a) and at crossover in Fig. 6(b). The structure contains a small conduction band offset at the n-p interface, but in this case, its effect on the crossover voltage is negligible. In Section V, we will discuss a structure with a larger, more significant conduction band offset. For this simulation, we vary the n- and p-side doping individually. Fig. 7 shows a crossover voltage strongly correlated with Vbi,d , but Vx may be slightly above or below Vbi,d , depending on which side of the junction is more heavily doped. If the n-side is highly doped, most of the band bending will occur on the p-side, leading to increased photogenerated charge accumulation and a crossover voltage lower than Vbi,d , as observed in [2] for a sim- In the previous section, we showed how accumulated charge at a heterojunction interface can affect the generation-dependent injection current of the device and cause crossover, even in the absence of traps. Several papers have discussed I–V crossover caused by traps in the widow layer or interface [13], [14], [16], [23], [24]. In this section, we will show that interface traps populated by photoexcitation can have a similar effect to mobile charge accumulation at a heterojunction. We first consider the effect of traps in a heterojunction without any conduction band offset, such as that shown in Fig. 6(a). Like the accumulation charge discussed previously, the trapped charge is a sheet of charge at the interface, which increases the forward bias of the junction under illumination. Adding acceptor traps at the interface increases the band bending, thus increasing the injection current (see Fig. 8). Donor traps have the opposite effect, suppressing the injection current. Larger trap densities will have a larger effect on the injection current. There is, however, one important difference in behavior between the two types of charge. As shown in Fig. 3, the accumulation charge only increases the injection current above Vbi,j , because this is the regime where accumulation occurs. The interface traps, on the other hand, influence the current at all voltages, because a fraction of the traps are charged even below Vbi,j (see Fig. 8). 1144 Fig. 8. Simulated results for different interface trap distributions. As the trapped charge at the interface increases, the electron injection current increases under illumination. IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 4, NO. 4, JULY 2014 Fig. 10. Plot showing change in crossover voltage with carrier lifetime for three different structures: n-i-p, n-p, and n-p with 0.2-eV Schottky built-in voltage. The dotted lines show the built in potential of each device. behavior in the light J–V curve, with a second bend appearing in high forward bias [25]. The offset will affect crossover insofar as it lowers Vbi,d . There may also be some inversion charge due to the band offset, which can slightly increase Vx , even pushing it slightly above Vbi,d in the case with no traps (see Fig. 9, blue squares). B. Evaluation of Assumptions Fig. 9. Simulated results for a CIGS structure with a V b i of 1.0 V and varying deep acceptor trap density in the CdS layer. The modeled CdS/CIGS band offset is 0.4 eV. Inset: light (solid line) and dark (dashed line) I–V curves for the case with no traps. In [13], Gloeckler et al. discuss a CIGS solar cell where crossover is caused by deep acceptor traps in the CdS. The structure of interest has a relatively large Vbi,d of 1.0 V. This is fundamentally different from the previous structures, where crossover was predominantly caused by a voltage-dependent photocurrent. We simulate a CIGS structure like that discussed in [13] with a conduction band offset of 0.4 eV at the CdS/CIGS interface and plot ΔJ versus V – Vbi,d in Fig. 9. For structures with traps in CdS, we find that Vx is significantly lower than Vbi,d . Therefore, Vx is uncorrelated with Voc ; therefore, in this case, crossover is not necessarily an indicator of lower efficiency, since Voc is not limited by Vbi,d , as seen in the previous case. As the acceptor trap density in the CdS layer increases, the crossover is pushed farther below the trap free case. We also do not see symmetry in the plot of ΔJ because the photocurrent is not limited to any specific value by current partitioning. This structure also contains a significant conduction band offset, which we have included in our model to better match simulated results previously reported in the literature [15]. This offset results in a two-diode effect, which can lead to unexpected Up to this point, all simulations have assumed no bulk recombination and uniform optical generation. In this section, we will test the robustness of our model by simulating structures with finite recombination lifetimes and different generation profiles. Bulk recombination gives the photogenerated carriers an additional means of removal from the device. Fig. 10 shows the results for crossover voltage as a function of carrier lifetime for three device structures. Carriers that recombine in the bulk of the device reduce the collected photocurrent but do not change the current partitioning. For reasonably long lifetimes, we find that the crossover voltage remains invariant. At very low lifetimes, we find that the crossover voltage increases only slightly for the n-i-p and n-p structures and decreases slightly for the Schottky barrier structure. However, even with a short lifetime of 1 ns the crossover voltage remains within about 50 mV of the built-in potential. In this case, the diffusion lengths will affect the low bias properties such as fill factor, as discussed in [8]; however, at high forward bias, current partitioning remains the dominant effect, and therefore, Vx is not significantly changed. Fig. 11 shows the I–V characteristic for different generation profiles. Simulations of solar cells illuminated from either the n- or p-side give identical I–V curves, since for our model, we assume that electrons and holes have identical transport properties. The overall I–V is different for uniform generation and illumination from one side. However, when the device is illuminated from the n-side, electrons are able to exit from the n contact more easily, but holes also have an easier time escaping from the n contact. The crossover voltage, therefore, remains the same, even if the generation is nonuniform. MOORE et al.: CORRELATION OF BUILT-IN POTENTIAL AND I–V CROSSOVER IN THIN-FILM SOLAR CELLS 1145 When taking experimental measurements, it is also important to account for series resistance. The models that are discussed in this paper presumed that series-resistance (Rs ), is negligible; although, there are finite series resistances in all real devices. This resistance will increase the measured voltage at current I by I ∗ Rs . The measured crossover voltage must, therefore, be corrected in analysis by Vx = Vx,m eas – Ix Rs , where Ix is the measured current at the crossover voltage [2]. However, caution must be taken when correcting for series resistance above V = Vbi since, as mentioned in Section III, the exponential portion of the diode current disappears in this region, and the current becomes dominated by the bulk resistance. Fig. 11. Plot showing the effect of different generation profiles on the I–V characteristics of an n-i-p structure. The I–V is slightly different in each case, but the crossover point remains tied to the built-in potential, regardless of the generation profile. C. Experimental Interpretation of Crossover In this section, we will discuss possible measurement techniques that can be used on real devices to conclusively determine whether crossover is due to a low Vbi,d or some other cause such as light-induced charge trapping. We have already shown how plotting ΔJ can be useful to determine the cause of crossover. For Vbi,d limited crossover in a homojunction, the ΔJ curve will appear symmetric, as shown in Fig. 3, which is not the case for other causes of crossover. The case of a heterojunction is slightly more complicated. If the doping of both sides of the junction is known, then the crossover can be compared with a simulation such as that shown in Fig. 7. The crossover voltage can also be affected by traps in the window layer. As demonstrated in [13] and [14], this effect can be identified using wavelength-dependent light bias measurements. If the crossover voltage remains constant under low-wavelength illumination (i.e., with photon energy below the bandgap of the window layer), then photoexcitation in the window layer can be effectively ruled out as the primary cause of crossover, as demonstrated in [7]. Comparison of the crossover voltage with Vbj,j measured from the intercept of 1/C 2 versus voltage can be useful to characterize crossover. If the two values match closely, then it can be concluded that the crossover is limited by the built-in voltage of the junction Vbj,j . Unfortunately, as noted in [2], the C–V of thin-film devices is often difficult to interpret and may not give the correct intercept for Vbi,j , and therefore, care must be taken in this measurement. If the 1/C2 intercept is above the voltage predicted by crossover, a Schottky barrier at one of the contacts may be responsible. As noted previously, the crossover is dependent on the contact-to-contact Vbi,d , while C–V is dependent only on Vbi,j . The presence of a Schottky barrier can also be confirmed by other methods as well, such as temperature-dependent I–V [12]. The Vbi,d limited crossover voltage should also have the same temperature dependence as Vbi,d and will extrapolate to Eg at T = 0 K if limited by the junction or Eg – φsb if a Schottky barrier is present [7]. VI. CONCLUSION Determining the cause of I–V crossover in a solar cell can be helpful in understanding performance limitations. As discussed extensively in the literature, many different effects can cause crossover. All causes of crossover can be classified into two categories. Crossover may be caused by a voltage-dependent photocurrent, or by a generation-dependent diode injection current caused by a change in band bending under illumination. Both mechanisms often play a role, and the overall behavior of the solar cell depends on which term in (1c) is dominant. Previous studies have attributed voltage-dependent collection in thin-film solar cells to short diffusion lengths, but as shown in this paper, voltage-dependent collection can occur even in ideal devices with long diffusion lengths, if the contact-to-contact built-in potential is low. This effect can occur in a wide variety of cell structures. For example, we showed that I–V crossover caused by a Schottky back contact can be simply explained by the reduced contact-to-contact Vbi . We also showed that a simple plot of ΔJ = Jlight – Jdark can be useful in identifying the specific cause of crossover if series resistance is properly accounted for. These conclusions were supported with physical arguments, analytical calculations, and with detailed numerical simulation. In light of these findings, we believe that the existence of a low contact-to-contact built-in potential should be considered as a possible cause for I–V crossover. APPENDIX A ANALYTIC TREATMENT OF N-I-P STRUCTURES Assuming the electric field is constant throughout the absorber layer, the injection and photogenerated currents for the structure of Fig. 2(a) are as follows: Jinj = q n2 n2 μn i + μp i ND NA Jgen = qGL coth (V − Vbi ) eq (V −V b i )/k T − 1 q (V − Vbi ) 2kT − 2kT q(V − Vbi ) eq V /k T − 1 (A1) (A2) where G is the generation rate, and L is the length of the channel. We see from this equation that the photogenerated current has some voltage dependence. The full derivation of these equations 1146 IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 4, NO. 4, JULY 2014 can be found in [9], where they were originally derived for an m-i-m structure. They are shown to apply to a n-i-p solar cell in [10]. An analytic expression can be similarly derived for the structure in Fig. 2(c) with a hole blocking heterojunction by slightly modifying the derivation of (2). Setting the boundary condition for the hole current at the n-contact to Jp = 0, the equations for the injection and photogenerated currents become (V − Vbi ) n2 eq V /k T − 1 (A3) Jinj = q μn i q (V −V )/k T bi ND e −1 1 kT Jgen = qGL q (V −V )/k T − . (A4) bi q(V − Vbi ) e −1 The injection current in (A3) is the same as for the homojunction case, except the term for holes vanishes, since the hole current is blocked by the valence band barrier. The extra injection current is explained by a change in the charge distribution under illumination. At high forward bias, the electric field will try to push the holes toward the n-contact but holes cannot enter the n-contact due to the large valence band barrier. The density of photogenerated holes at the blocking contact builds up because they cannot escape through this contact [see Fig. 2(d)]. This accumulation of holes acts as an additional forward bias on the junction, increasing the electron injection current. The additional injection current under illumination increases ΔJ above zero, as shown in Fig. 3. The analytic expression does not account for this Poisson effect, which is why it differs from the numerical simulation in this case. It would be difficult to derive a full analytic expression for this accumulation driven current; however, a rough approximation can be made by treating the accumulation charge as a sheet charge at the n-p interface. This sheet charge simply adds a constant component to the electric field which can then be rewritten as E = E0 + E Qs (V, G) 2Ks ε0 APPENDIX B PARAMETERS USED IN NUMERICAL MODELING Material parameters for CIGS [15] Material parameters for CdS [15] n-i-p homostructure (A5) where E0 is the constant electric field in the intrinsic layer of the p-i-n junction given by (V − Vbi )/d, and E is the additional electric field due to the accumulation or trapped charge. The equation for this additional electric field is E = As previously mentioned, this equation is only an approximation of the accumulation current due to the estimation of the accumulated holes purely as a surface charge. In reality, the excess holes under illumination may have a more complicated distribution. We, therefore, introduce a correction factor α to account for this difference. We can evaluate the validity of this expression using our results from numerical simulation. The simulation gives us the hole concentration at the interface. We find that the relationship between current and charge accumulation is in close agreement with our prediction from (A7) for α = 0.8. This equation provides a simple way to understand the full numerical simulation shown in Fig. 3. N-i-p heterostructure (A6) where Qs is the surface charge per unit area, and Ks is the dielectric constant of the intrinsic layer. Note that Qs is also a function of both V and G. Additionally, trap states at the interface may also provide an additional source of surface charge as discussed previously. Because the electric field of a sheet of charge is uniform with position, the equation for our total electric field remains position independent. At high forward bias, the extra current caused by photogenerated accumulation or trapped charge can now be written as Qs, light − Qs, dark . (A7) Jacc = αND qμn 2Ks ε0 NP structure NP structure with Schottky Barrier Work function of back contact: 5.1559 eV (back contact of all other structures assumed to be ohmic). MOORE et al.: CORRELATION OF BUILT-IN POTENTIAL AND I–V CROSSOVER IN THIN-FILM SOLAR CELLS ACKNOWLEDGMENT The authors acknowledge the work of Prof. R. Agrawal’s group, particularly his students C. Hages and N. Carter, whose experiments in CZTSSe devices were the primary motivation for this work, as well as Prof. J. Sites, for his helpful comments and suggestions on this work. REFERENCES [1] N. G. Tarr and D. L. Pulfrey, “An investigation of dark current and photocurrent superposition in photovoltaic devices,” Solid State Electron., vol. 22, no. 3, pp. 265–270, Mar. 1979. [2] S. Hegedus, D. Desai, and C. Thompson, “Voltage dependent photocurrent collection in CdTe/CdS solar cells,” Progress Photovoltaics, Res. Appl., vol. 15, no. 7, pp. 587–602, May 2007. [3] Z. Wang, Z. Chang, A. E. Delahoy, and K. K. Chin, “A study of light sensitive ideality factor and voltage dependent carrier collection of CdTe solar cells in forward bias,” J. Photovoltaics, vol. 3, no. 2, pp. 843–850, Apr. 2013. [4] M. Burgelman, P. Nollet, S. Degrave, and J. Beier, “Modeling the crossover of the IV characteristics of thin film CdTe solar cells,” in Proc. IEEE 28th Photovoltaic Spec. Conf. Rec., Anchorage, AK, USA, 2000, pp. 551–554. [5] S. Hegedus, “Current-voltage analysis of a-Si and a-SiGe solar cells including voltage-dependent photocurrent collection,” Progress Photovoltaics, Res. Appl., vol. 5, no. 3, pp. 151–168, Dec. 1997. [6] X. Wang and M. S. Lundstrom, “On the use of rau’s reciprocity to deduce external radiative efficiency in solar cells,” IEEE J. Photovoltaics, vol. 3, no. 4, pp. 1348–1353, Oct. 2013. [7] J. E. Moore, C. J. Hages, N. Carter, R. A. Agrawal, and M. S. Lundstrom, “The physics of Vb i related IV crossover in thin film solar cells: Applications to ink deposited CZTSSe,” in Proc. IEEE 39th Photovoltaic Spec. Conf. Rec., Tampa, FL, USA, 2013, pp. 3255–3359. [8] R. S. Crandall, “Modeling of thin film solar cells: Uniform field approximation,” J. Appl. Phys., vol. 54, no. 12, pp. 7176–7186, Mar. 1983. [9] R. Sokel and R. C. Hughes, “Numerical analysis of transient photoconductivity in insulators,” J. Appl. Phys., vol. 53, no. 11, pp. 7414–7424, May 1982. [10] S. Dongaonkar, “End to end modeling of thin film solar cells,” PhD Dissertation, Dept. Elect. Eng., Purdue Univ., West Lafayette, IN, USA, 2013 [11] S. H. Demtsu and J. R. Sites, “Effect of back contact barrier on thin film CdTe solar cells,” Thin Solid Films, vol. 510, no. 1-2, pp. 320–324, Jul. 2006. [12] J. Pan, M. Gloeckler, and J. R. Sites, "Hole current impedance and electron current enhancement by back-contact barriers in CdTe thin film solar cells," J. Appl. Phys., vol. 100, no. 12, pp. 124505-1–124505-6, Dec. 2006. [13] M. Gloeckler, C. R. Jenkins, and J. R. Sites, “Explanation of light/dark superposition failure in CIGS solar cells,” in Proc. MRS Symp., Jan. 2003, vol. 763, pp. B5.20.1–6. [14] A. Rothwarf, “The CdS/Cu2S solar cell: Basic operation and anomalous effects,” Sol. Cells, vol. 2, no. 2, pp. 115–140, Oct. 1980. [15] M. Gloeckler, A. L. Fahrenbruch, and J. R. Sites, “Numerical modeling of CIGS and CdTe solar cells: Setting the baseline,” in Proc. 3rd World Conf. Photovoltaic Energy Convers., Osaka, Japan, 2003, vol. 1, p. 491-494. [16] G. Agostenelli, E. D. Dunlop, D. L. Batzner, A. N. Tiwari, P. Nollet, M. Burgelman, and M. Kontges, “Light dependent current transport mechanisms in chalcogenide solar cells,” in Proc. 3rd World Conf. Photovoltaic Energy Convers., Osaka, Japan, vol. 1, 2003, pp. 356–359. [17] J. L. Gray, “ADEPT: A general purpose device simulator for modeling solar cells in one-, two-, and three-dimensions,” in Proc. IEEE 22nd Photovoltaic Spec. Conf. Rec., Las Vegas, NV, USA, 1991, pp. 436–438. [18] J. L. Gray, X. Wang, X. Sun, and J. R. Wilcox. (2011). ADEPT 2.0. [Online]. Available: http://nanohub.org/resources/10913 [19] A. Niemegeers, S. Gillis, and M. Burgelman, “A user program for realistic simulation of polycrystalline heterojunction solar cells: SCAPS-1D,” in Proc. 2nd World Conf. Photovoltaic Energy Convers., 1998, vol. 1, pp. 672–675. [20] K. L. Chopra, P. D. Paulson, and V. Dutta, “Thin-film solar cells: An overview,” Progress Photovoltaics, Res. Appl., vol. 12, no. 2-3, pp. 69– 92, Mar. 2004. 1147 [21] S. Fonash, Solar Cell Device Physics. New York, NY, USA: Elsevier, 2010. [22] R. V. K. Chavali, J. R. Wilcox, B. Ray, J. L. Gray, and M. A. Alam, “Correlated nonideal effects of dark and light I–V characteristics in a-Si/c-Si heterojunction solar cells,” IEEE J. Photovoltaics, vol. 4, no. 3, pp. 763– 771, May 2014. [23] R. V. K. Chavali, J. L. Gray, and J. R. Wilcox, “Numerical method to separate the photo-current and contact injection current in solar cells,” in Proc. IEEE 40th Photovoltaic Spec. Conf., 2014. [24] G. Agostenelli, D. L. Batzner, and M. Burgelman, “A theoretical model for the front region of cadmium telluride solar cells,” Thin Solid Films, vol. 431, pp. 407–413, May 2003. [25] J. R. Sites, “Quantification of losses in thin-film polycrystalline solar cells,” Sol. Energy Mater. Sol. Cells, vol. 75, no. 1, pp. 243–251, Jan. 2003. [26] Q. Guo, G. M. Ford, W. C. Yang, B. C. Walker, E. A. Stach, H. W. Hillhouse, and R. Agrawal, “Fabrication of 7.2% efficient CZTSSe solar cells using CZTS nanocrystals,” J. Amer. Chem. Soc., vol. 132, no. 49, pp. 17384–17386, Nov. 2010. James E. Moore (M’11) received the B.S. degree in engineering physics from Taylor University, Upland, IN, USA, in 2009 and the M.S. degree in electrical and computer engineering from Purdue University, West Lafayette, IN, in 2011, where he is currently working toward the Ph.D. degree in electrical and computer engineering. His research interests include numerical and compact modeling of semiconductor devices, as well as experimental work involving electrical and optical characterization. Sourabh Dongaonkar (M’13) received the B.Tech. degree in electrical engineering from the Indian Institute of Technology Kanpur, Kanpur, India, in 2007 and the Ph.D. degree from the School of Electrical and Computer Engineering, Purdue University, West Lafayette, IN, USA, in 2013. He is currently a Compact Device Modeling Engineer with Intel Corporation, Hillsboro, OR, USA. From 2007 to 2008, he was with the Global Market Center, Deutsche Bank, Mumbai, India, as a Quantitative Analyst. His research interests include the analysis and design for variability and reliability in solar cells, modeling of novel semiconductor devices, and solid-state storage technologies. Dr. Dongaonkar received the 2008 Ross Fellowship from Purdue University Graduate School, as well as the Best Student Presentation and the Best Poster Awards at the 37th and 38th IEEE Photovoltaic Specialists Conference, respectively. Raghu Vamsi Krishna Chavali (S’13) received the B.E (Hons.) degree in electrical and electronics engineering from the Birla Institute of Technology and Science, Pilani, India, in 2009 and the M.S degree in electrical and computer engineering from Purdue University, West Lafayette, IN, USA, in 2011, where since 2011, he has been working toward the Ph.D. degree with the School of Electrical and Computer Engineering. His research interests include modeling, simulation, and characterization of semiconductor devices. 1148 Muhammad Ashraful Alam (M’96–SM’01–F’06) received the B.S.E.E. degree from the Bangladesh University of Engineering and Technology, Dhaka, Bangladesh, in 1988; the M.S. degree from Clarkson University, Potsdam, NY, USA, in 1991; and the Ph.D. degree from Purdue University, Lafayette, IN, USA, in 1994, all in electrical engineering. He is currently a Professor of electrical and computer engineering with the School of Electrical Engineering and Computer Science, Purdue University, where his research and teaching focus on physics, simulation, characterization, and technology of classical and novel semiconductor devices. From 1995 to 2001, he was with Bell Laboratories, Lucent Technologies, Murray Hill, NJ, USA, as a member of Technical Staff with the Silicon ULSI Research Department. From 2001 to 2003, he was a Distinguished Member of Technical Staff and the Technical Manager of the IC Reliability Group, Agere Systems, Murray Hill. In 2004, he joined Purdue University. He has contributed to more than 150 papers in international journals and has presented many invited and contributed talks at international conferences. His current research interests include stochastic transport theory of oxide reliability, transport in nanonet thin-film transistors, nanobio sensors, and solar cells. Dr. Alam received the IEEE Kiyo Tomiyasu Award for his contributions to device technology for communication systems. He is a Fellow of the American Physical Society and the American Association for the Advancement of Science. IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 4, NO. 4, JULY 2014 Mark S. Lundstrom (S’72–M’74–SM’80–F’94) received the B.E.E. and M.S.E.E. degrees from the University of Minnesota, Minneapolis, MN, USA, in 1973 and 1974 and the Ph.D. degree from Purdue University, West Lafayette, IN, USA, in 1980. He is currently the Don and a Carol Scifres Distinguished Professor of Electrical and Computer Engineering, Purdue University, West Lafayette, IN, USA. His current research interests include the physics of small electronic devices, particularly nanoscale transistors, on carrier transport in semiconductor devices and devices for energy conversion, storage, and conservation. Dr. Lundstrom is a Fellow of the American Physical Society and the American Association for the Advancement of Science, as well as a Member of the U.S. National Academy of Engineering.