Thin Solid Films 480–481 (2005) 307 – 311 www.elsevier.com/locate/tsf Temperature dependence of the diode ideality factor in CuInS2-on-Cu-tape solar cells Johan Verschraegena,*, Marc Burgelmana, Jqrgen Penndorf b a Gent University, Electronics and Information Systems (ELIS), Pietersnieuwstraat 41, B-9000 Gent (B), Belgium b Institut für Solartechnologien (IST), Im Technologiepark 7, D-15236 Frankfurt/Oder (D), Belgium Available online 15 December 2004 Abstract Current versus voltage measurements were done on copper indium disulphide cells fabricated on a continuous copper tape fabricated at IST (Frankfurt/Oder, D). Temperatures were in the range of 90–370 K. We compared the measurements with an ideal diode model and extracted parameters accordingly. The temperature dependence of the diode ideality factor can give valuable information about the main recombination mechanism in the cell. We find a remarkable agreement of the temperature dependence with a theory derived by Padovani et al. We conclude that tunneling currents play an important role inside the cell. D 2004 Elsevier B.V. All rights reserved. Keywords: CuInS2; CISCuT; Capacitance; Modelling 1. Introduction 2. Experimental First reports on the CISCuT (CuInS2 on Cu-tape, fabricated at IST, Frankfurt/Oder, D) cells in literature date from 1998 [1]. Starting from a near endless copper tape, an absorbing layer is made by deposition of In followed by a short sulphurisation step. In this phase, the CuInS2 is formed. The solar cell is completed by adding a CuI buffer layer and a transparent front contact [2,3]. Since the introduction of this cell, the efficiency has improved a lot and has now reached over 9%. Although a very good result already, further improvements should be possible. Interpretation of I–V measurements can give a lot of information about the diode parameters. In particular, the temperature dependence of the diode ideality factor and the dark saturation current gives indications on what the main recombination paths are. In this paper, we will discuss the I–V measurements that were done and try to find the mechanisms which are limiting the opencircuit voltage. The samples used were prepared at IST (D) by the standard CISCuT roll-to-roll procedure until the absorber layer is formed. The cells were completed by spraying a thin CuI buffer layer and by sputtering a ZnO layer as the front contact. The area of the buffer and contact is confined to single dots with area between 9 and 10 mm2. We performed dark and light I–V measurements using a 236 Keithley source measure unit with bias voltage varying between 1 and 1 V. As a light source, an Oriel Xenon arc lamp was used with an infrared filter (but no solar spectrum correction filters). Through the use of neutral density filters, a total of 16 different illumination intensities were achieved, ranging from about 1 to 1500 W/m2. The samples were mounted in a cryostat setup and their temperature was varied between 90 and 370 K. At the same temperatures, we also did I sc versus Voc measurements. 3. The I sc–Voc theory * Corresponding author. Tel.: +32 9 264 8953; fax: +32 9 264 8961. E-mail address: Johan.Verschraegen@elis.ugent.be (J. Verschraegen). 0040-6090/$ - see front matter D 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.tsf.2004.11.006 First, we will discuss the I sc–Voc measurement. While this kind of measurement is well known, there are some 308 J. Verschraegen et al. / Thin Solid Films 480–481 (2005) 307–311 implicit assumptions connected with its interpretation and those will be shown here. We will start from the diode equation, corrected for series resistance R s and parallel conductance G sh: qð V R s I Þ 1 þ Gsh ðV Rs I Þ IL I ¼ I0 exp nkT ð1Þ I L is the light current, I 0 is the dark saturation current and n is the diode ideality factor, which can all be dependent on bias voltage and temperature. For an ideal diode, n is however constant and I 0 is only dependent on temperature. In the case where there is no recombination in the space charge region, n has a value of 1. If on the other hand, the current is dominated by recombination in the space charge region, n will be equal to 2. Usually, n has a value between 1 and 2, but values greater than 2 are also possible. It is not straightforward to extract the diode parameters from a measurement that is influenced by series resistance and shunt conductance. Therefore, we measure the short circuit current I sc=I|V=0 and the open-circuit voltage Voc=V|I=0 for different illumination intensities. Inserting, respectively, V=0 and I=0 in Eq. (1) gives: qRs Isc Isc ¼ I0 exp ð2Þ 1 Gsh Rs Isc IL nkT qVoc 1 þ Gsh Voc : IL ¼ I0 exp nkT ð3Þ If R s is small compared to Voc/I sc (less than 1%) and G shR s is much smaller than 1, then we see by Eq. (2) that I sc=I L and by Eq. (3) that I sc and Voc satisfy the diode law without the influence of series resistance. The influence of the parallel conductance remains, but for high-enough light intensities, it does not affect the open-circuit voltage. It is now easier to extract the diode parameters. A semilogarithmic plot of I sc versus Voc will yield a straight line for high-enough Voc. From the slope, n can be calculated if one knows the temperature. I 0 can be found by taking the intercept of the straight part of the plot with the I sc-axis. A small error in the estimation for the diode ideality factor results in a large error in I 0 due to the use of the logarithmic scale. Therefore, the value of I 0 will generally be less accurate than that of n. One could also try to fit a measurement to Eq. (1) using R s as an extra parameter, instead of making an I sc–Voc measurement, but then there is less confidence in the obtained values, the more because the series resistance might be bias voltage dependent. During the derivation of Eqs. (3) and (4), a few implicit assumptions have been made. First, we assumed the diode parameters to be independent of illumination intensity. This condition has to be fulfilled to get meaningful results from the I sc–Voc measurements; it can, however, be formulated a little less stringent as will be shown further on. Second, we assume I L to be independent of bias voltage. This condition is also crucial, if it is not met, we get: qVoc ð4Þ Isc ¼ I0 exp 1 þ Gsh Voc þ DIL ðVoc Þ; nkT with DI L(Voc) being the difference in light current for zero voltage and open-circuit voltage. We can expand this in terms of Voc. If only the first-order term is not negligible, then we get an effective G sh [4], and we can still use the I sc– Voc curve to extract the diode parameters. 4. Results The cells that were measured have an efficiency of 5– 6%. Best efficiencies for CISCuT cells so far are around 9%. The rather large difference is mainly caused by a much lower short circuit current. For our cells, this is around 14 mA/cm2 (not corrected for contact pin shadowing), while for the best cells, it is around 21 mA/cm2. Therefore, we will focus on the mechanisms that limit the open-circuit voltage. It will be shown further on that there is crossover between the light and dark curves and thus the diode parameters differ for light and dark conditions. According to the theory given in the previous paragraph, this would mean that the I sc–Voc technique is not applicable. However, we will still use it, but the diode parameters thus obtained will not give information about the cell under dark conditions. Still to get meaningful results, the diode parameters have to be illumination intensity independent. This supposes that there are two current mechanisms, one for dark and one for light conditions. In Fig. 1, a number of I–V curves for different light intensities are plotted together with the dark Fig. 1. I–V measurements for different illumination intensities (from 0.1 to 1.5 suns) together with the dark I–V curve. Note the crossover of the dark curve with the others. J. Verschraegen et al. / Thin Solid Films 480–481 (2005) 307–311 curve. The fact that even for low intensities there is already crossover with the dark I–V curve supports the idea of two distinct current mechanisms. This can be verified by starting from the I sc–Voc curve, and pretending that it represents the dark I–V characteristic of the junction, only influenced by parallel resistance. We can then calculate the light characteristic by adding series resistance and a light current. If the diode parameters are independent of illumination intensity and the light current is independent of bias voltage, then this should give a perfect fit with the measured light I–V curve. The result can be seen in Fig. 2. For three different illumination intensities, the I–V curve is fitted using the explained technique. The fit is reasonable but not perfect. Especially for the curve with the largest illumination intensity, there is a notable difference at low bias voltages. Looking at the slope of the I–V curve around 0 V, it seems as if the parallel conductance were much higher. Although this means that the conditions that allow the use of I sc–Voc curves to extract the diode parameters are not fully met, we will still use them and the confidence in the results will be based on the good agreement with the recombination model that will be proposed further on. In Fig. 3, a number of I sc–Voc curves are presented on a semilog scale plot for temperatures going from 90 to 370 K. According to the ideal diode model, one would expect a curve which only deviates from a straight line at low bias voltages due to parallel conductance in the cell. The CISCuT cells do not follow the ideal diode model. In fact, in Fig. 3, there seems to be a transition from one flat region at low bias voltages to another at high bias voltages. This can be interpreted as two current mechanisms being present in the cell. The first one has the weakest voltage dependence and governs the current at low bias voltages. The second one has a stronger bias voltage dependence and will therefore be the main current mechanism at high-enough voltages. It is however also possible that there are not just two distinct mechanisms but that there is a continuous 309 Fig. 3. I sc–Voc curves for temperatures going from 90 to 370 K, in steps of 20 K. Each measurement was done for 16 illumination intensities ranging from about 1 to 1500 W/m2. transition. In what follows, we will determine the diode parameters for the high bias voltage region. After all, this is the region closest to the working condition of the cell and if we can determine, and eventually diminish, the main recombination paths at these voltages, then that would result in the greatest gain in efficiency. From the curves from Fig. 3, we can calculate the ideality factor n and these are plotted in Fig. 4 as a function of temperature. The first thing that strikes is that the values are much greater than 2. This means that the current is not limited by drift and diffusion or by recombination in the space charge region. A number of models have been proposed to explain ideality factors greater than 2. Padovani and Stratton [5] calculated the current for a Schottky diode in the case of tunneling through the contact barrier (thermionic field emission). They found for the ideality factor: E00 E00 n¼ coth ð5Þ kT kT E 00 is a characteristic tunneling energy depending on doping concentration. In Fig. 4, a calculation of n for E 00=50 meV (or E 00/k=580 K) is shown. The fit with the measurement data is remarkable, the more if one realizes that this is a oneparameter fit. From this, we can conclude that tunneling plays an important role in the cell. Konovalov and Tober [6] came to the same conclusion on the basis of open-circuit voltage measurements as a function of temperature and illumination intensity. Their estimation for E 00/k was around 200 K, which corresponds with E 00 equal to 17 meV, a much lower value than what we find. It is interesting to note that they use an approximation of Eq. (5) for the ideality factor [7]: Fig. 2. Comparison of the measured light I–V characteristics (symbols) for different light intensities with the measured I sc–Voc curve of Fig. 3, shifted downward over the light current I L and corrected for a series resistance of 4 V cm2 (line). 1 2 E00 n¼ 1 2 2 : 3k T ð6Þ 310 J. Verschraegen et al. / Thin Solid Films 480–481 (2005) 307–311 temperature is there a reasonable fit. The fact that the accuracy on I 0 is lower than that for the ideality factor n might be a reason for this, but a satisfying explanation is not found yet. Another tunneling model which can explain ideality factors larger than 2 was presented by Riben and Feucht [9]. They considered tunneling enhanced bulk recombination in which electrons/holes first tunnel to a trap and then fall in the valence/conductance band, or first fall into a trap and then tunnel to the appropriate band. The expression for the ideality factor reads n¼ Fig. 4. Ideality factor n versus temperature (symbols). The full line is a fit with Eq. (5), with E 00/k=580 K. The dotted line is a fit with Eq. (7), with T*=580 K. TT ; T ð7Þ The approximation (Eq. (6)) has a negligible error (b2.5%) only for E 00/kTb2/3, corresponding to nb1.15. In our case, n is much higher and the exact expression (Eq. (5)) should be used. From the I sc–Voc measurements, not only the ideality factor can be calculated but also the dark saturation current. The result as a function of temperature is shown in Fig. 5. The two curves are calculations from one measurement cycle with dropping temperature (290–90 K) and another one with rising temperature (90–370 K). A kind of hysteresis is apparent for temperatures below room temperature. This behavior is also apparent in capacitance versus temperature measurements [8], but has not been explained so far. It is important to note that the diode ideality factor is not dependent on temperature history (not shown in Fig. 4). The model from Padovani and Stratton [5] which gives us an expression for n, also gives an expression for I 0. The calculated dark saturation currents from Fig. 5, however, cannot be fitted to that expression, at least not for the whole temperature range. Only for temperatures above room with T* a constant which plays the same role as E 00/k. If we try to fit our calculated ideality factors to this expression, we also get a reasonable fit (see Fig. 4). This is not surprising at all, since Eq. (5) can be simplified to Eq. (7) for T much lower then E 00/k. Fitting the dark saturation current to the expression derived by Riben and Feucht is again not possible. Nonetheless, this gives again a strong indication that tunneling currents are dominating the recombination. In Fig. 6, a band diagram is drawn for the CISCuT cell junction. The band offsets are calculated with the Anderson model by the use of SCAPS [10]. The large spike in the valence band is most likely less pronounced. The current mechanism is possibly tunneling from electrons from the CuInS2 side of the junction to the CuI/CuInS2 interface. As it is drawn in Fig. 6, this seems very unlikely due to the large tunneling distance to the interface. In Ref. [6], it is proposed that, near the interface, there is a larger space charge density due to compensating defects. This would result in a smaller barrier near the interface. Another possibility is tunneling into defect states in the forbidden zone of the CuInS2, followed by hopping from defect to defect in the direction of the interface. Further investigations will be necessary to determine the precise origin of the tunneling currents and to find ways to reduce them. Fig. 5. Saturation current I 0 versus temperature. One branch is for a measurement cycle with dropping temperature, the other is for rising temperatures. Fig. 6. Schematic band diagram for the CuI/CuInS2 junction. The possible tunneling path is indicated. J. Verschraegen et al. / Thin Solid Films 480–481 (2005) 307–311 311 5. Conclusion References From I sc–Voc measurements, we have calculated the diode parameters as a function of temperature. The diode ideality factor n is much greater than 2 and has a strong temperature dependence. It can be fitted remarkably well to an interface tunneling model, from which we conclude that tunneling plays an important role in the cell. This is confirmed by work from other authors. However, fitting the dark saturation current to the same model is problematic. More accurate information about the band diagram is needed to fully characterise and locate these tunneling recombination currents. [1] J. Penndorf, M. Winkler, O. Tober, D. Roser, K. Jakobs, Sol. Energy Mater. Sol. Cells 53 (1998) 285. [2] I. Konovalov, J. Penndorf, M. Winkler, O. Tober, Mater. Res. Soc. Symp. Proc. 668 (2001) H5.15.1. [3] M. Winkler, O. Tober, J. Penndorf, K. Szulzewsky, D. Rfser, G. Lippold, K. Otte, Thin Solid Films 361–362 (2000) 273. [4] W.A. Miller, L.C. Olsen, IEEE T. Electron. Dev. 31 (1984) 654. [5] F.A. Padovani, R. Stratton, Solid State Electron. 9 (1966) 695. [6] I. Konovalov, O. Tober, Proceedings of Pv in Europe: from PV Technology to Energy Solutions Conference, Rome, Italy, 2002 (October), p. 401. [7] U. Rau, A. Jasenek, H.W. Schock, F. Engelhardt, T. Meyer, Thin Solid Films 361–362 (2000) 298. [8] J. Verschraegen, M. Burgelman, J. Penndorf, Thin Solid Films 451–452 (2004) 179. [9] A.R. Riben, D.L. Feucht, Solid State Electron. 9 (1966) 1055. [10] M. Burgelman, P. Nollet, S. Degrave, Thin Solid Films 361–362 (2000) 527, (Also, see http://www.elis.ugent.be/ELISgroups/solar/ projects/scaps.html). Acknowledgements The European CISLINE project (ENK6-CT-2001-00519) (J.V.). The Research Fund of the University of Gent (BOFGOA) (M.B.).