Evidence for grain-boundary-assisted diffusion of sulfur in polycrystalline CdS/CdTe heterojunctions

M. K. Herndon,a A. Gupta, V. Kaydanov and R. T. Collins,b

Physics Department, Colorado School of Mines, Golden, CO 80401


We present a Near-Field Scanning Optical Microscopy (NSOM) study of S interdiffusion in polycrystalline CdS/CdTe heterojunctions. S diffusion from CdS into CdTe leads to the formation of a CdTe1-xSx ternary phase. Because the band gap of CdTe1-xSx varies with S composition, we were able to combine NSOM with a tunable laser source to microscopically identify S-rich regions in the CdTe layer. S composition was found to be very nonuniform and frequently to be greater along grain boundaries than in the grain centers, identifying grain boundaries as locations of enhanced interdiffusion.

a) Electronic mail: mkherndon@mediaone.net

b) Electronic mail: rtcollin@mines.edu

Polycrystalline CdS/CdTe heterojunction diodes are being actively explored for use in low cost, high efficiency, thin-film photovoltaic applications (See, for example, Ref. 1). In these devices, the CdS, which has a band gap ~2.5eV at room temperature, acts as both a window layer and n-type material. Solar absorption occurs primarily in the p-type CdTe. The bandgap of CdTe, 1.5eV, is well matched to the terrestrial solar spectrum.

Post growth annealing of the films has been shown to enhance the performance of these devices. Substantial evidence exists that interdiffusion between CdTe and CdS layers occurs during growth and post growth annealing processes.2 In particular, S diffusion into the CdTe layer leads to the formation of a ternary CdTe1-xSx phase within the CdTe layer. For S concentrations less than ~25% , bowing in the bandgap of the ternary as a function of x reduces the band gap as S content increases. This leads to a decrease in the low energy cutoff of the device relative to pure CdTe.3 Various models for the mechanism of S diffusion have been developed, and it is generally believed that grain boundaries must play a role in the diffusion process. This has been difficult, however, to verify experimentally. In this paper we present the results of using a Near Field Scanning Optical Microscope (NSOM) and the wavelength dependence of the ternary band gap to study the interdiffusion of S into CdTe.

The samples used in these measurements were grown on soda-lime glass coated with a 400nm layer of tin oxide (SnO2) to act as the front contact. Chemical bath deposition was used to grow a ~200nm thick CdS layer on the SnO2. CdTe was electrochemically deposited on the CdS layer to a thickness of 3-4mm. The sample was treated with CdCl2 and annealed at 365oC. The CdCl2 treatment and anneal is known to improve subsequent device performance.2,4 It also enhances interdiffusion.4,5 Previous work has shown interdiffusion can be observed after the 365°C anneal, and increases significantly with temperature. By 450°C, a nearly uniform S concentration is observed throughout the film.5 Gold was evaporated through a shadow mask to form a back contact and define individual devices. Capacitance voltage measurements gave a zero bias depletion width of ~1.5mm for the diodes and a p-type carrier concentration of ~5 x1014/cm3 in the CdTe. More detail on device fabrication and device properties can be found in Ref. 6.

The glass side of the sample was scribed in the region under a device and broken to expose a cross section of the device. It was mounted so that spatially resolved photocurrent and topographic images of the cross section could be measured with the NSOM. Electrical leads were connected to the front and back contacts of the device, and fed into a current amplifier. The NSOM is a home built instrument that operates in air. The experiments were done at room temperature. The NSOM probe was made from a single mode optical fiber which was tapered using a micropipette puller and metal-coated, leaving a ~100nm diameter aperture at the tip.7 The probe was raster scanned across the sample cross section using optically detected shear force feedback to maintain a constant separation of approximately 10nm between the probe tip and sample.8 Monitoring the feedback signal while raster scanning the sample allowed the topography of the cleaved surface to be measured.

A tunable Ti:Sapphire laser and a 632.8nm HeNe laser were simultaneously coupled into the fiber probe allowing it to be used as an excitation source. Each laser was chopped at a different frequency before being coupled into the fiber. The photocurrent generated in the CdS/CdTe diode was detected for each wavelength using two lock-in amplifiers. The outputs of the lock-in amplifiers were recorded at each point as the sample was scanned. This allowed two spatially-resolved measurements of the generated photocurrent, one for each wavelength of light, to be recorded simultaneously with the surface topography.

Scans of the cell cross section began on the glass side of the junction, and progressed toward the back of the CdTe layer. The same area was scanned several times, each with a different wavelength from the Ti:Sapphire laser, in order to see variations in collection as a function of excitation energy. The HeNe images, taken simultaneously with each Ti:Sapphire measurement, were compared from scan to scan. This reference enabled us to ensure that variations in the Ti:Sapphire measurements were due to changes in excitation wavelength and not due to changes in the probe or scanning conditions.

Previous work has shown that the band gap of the CdTe1-xSx ternary phase decreases as a function of x (for x £ 0.25) from ~1.51eV at x=0 to 1.41eV at x@0.25.9,10,11 This dependence of band gap on x allows changes in excitation energy during NSOM measurements to be used to observe spatial variations in the S composition of the

CdTe1-xSx ternary across the heterojunction, and therefore to identify regions of high S content. In the discussion below we have used the bandgap dependence of Ref. 10 to associate optical excitation energies with specific S concentrations. In general, however, the band edge of polycrystalline films shows inhomogeneous broadening due to effects such as strain, defects, and alloying, which can lead to an apparent bandgap which is lower than that for crystalline material. Because of this broadening and the uncertainties in the reported composition dependence of the CdTe1-xSx band gap, relative changes and spatial variations in S composition found from optical measurements are likely to be much more reliable than the numerical S compositions inferred from the measurements. The latter should probably be viewed as correct to within a few percent.

Figure 1 shows a topographic image and three photocurrent images, each produced using a different excitation energy. All images are 4.8 x 4.8mm. The topography measurements showed an average grain size of 1-2mm, in agreement with previous electron microscope measurements.6 A small amount of drift occurred between scans. Comparison of the locations of individual grains seen in the topographic images allows us to track this drift. The layers present in the film have been identified adjacent to Figs. 1(a) and (b). The interface between the glass and film is evident as the horizontal feature near the bottom of the topography image. One obvious grain, which will be discussed in more detail below, has been outlined on the topography plot (Fig. 1(a)). The topography measurements obtained in registry with each photocurrent measurement were used to locate this feature in subsequent scans, allowing this region to be identified in the photocurrent images in Fig. 1(b)-(d).

When the probe was above the glass substrate, photocurrent was generated by light which traveled through the glass and entered the device through the front face (SnO2 layer), as it would under normal operation. This is apparent in Figure 1(b), where photocurrent collection from the glass region is seen at the bottom of the image. The dependence of this signal on excitation energy is in good agreement with the known spectral response for these devices.6

For 1.615eV excitation (Fig. 1(b)), which is a little more than 100meV above the CdTe bandgap, we see the photocurrent is strongly peaked near the junction, and decreases toward the back of the device. This is typical of measurements at energies above bandgap. In general, the photocurrent becomes even more strongly peaked near the CdTe/CdS junction as energy increases. As energy decreases below the CdTe bandgap to 1.463eV (Fig. 1(c)), corresponding to a S composition of 5.7%, the collection efficiency at the back of the device improves relative to that at the interface. We have observed this effect (a widening of the collection region at longer wavelength) in most edge cleaved semiconductor diodes we have studied. We attribute it to a competition between surface recombination and collection by the junction. At longer wavelengths, the light is absorbed further from the surface, allowing more of the carriers to be collected by the junction before recombining at the surface. Similar observations and conclusions were presented in Ref. 12. We note that the carrier collection region at this energy extends from the junction to the back of the device and that the average collection efficiency at the junction and near the back are roughly the same. Simplistically, the width of the collection region is determined by the depletion width and the minority carrier diffusion length. Since our depletion width is ~1.5mm, this indicates a minority carrier diffusion length of more than 2mm.

At 1.422eV (Fig. 1(d)), the excitation energy is 80meV below the CdTe band gap and corresponds to a S content of 16.4%. Here the width of the collection region has once again decreased. This is the result of carriers being generated mainly in the higher S content, lower bandgap, CdTe1-xSx ternary phase, which is most likely to exist close to the CdS/CdTe junction. The S concentrations we would infer near the junction in Fig. 1(d) is larger than the reported immiscibility gap composition of ~6% at 415°C.3 This may in part be due to the uncertainties in quantifying S concentration mentioned above.

For excitation energies below the CdTe band gap, a large amount of contrast was visible across the photocurrent images. For example, localized regions of higher photocurrent are visible near the interface in both Fig. 1(c) and 1(d). Presumably this contrast arises from variation in S content and hence bandgap across these regions. A particularly interesting example of this contrast is seen in Fig. 1(c) where a loop of higher photocurrent is clearly visible within the outlined region. This correlates exactly with the boundary of the grain identified in Fig. 1(a). To clarify the energy dependence of this feature, a series of measurements across this grain was recorded while varying the excitation energy from 30meV above to 60meV below the CdTe band gap. The results are shown in Fig. 2 where each line on the graph represents a single horizontal scan across the grain. Each scan was made at the same location using a different excitation energy.

In agreement with the data in Fig. 1, as energy decreases, the traces in Fig. 2 initially show an overall increase in the average value of the photocurrent which is maximum for energies just below the CdTe bandgap. As excitation energy decreases further, the average value of the photocurrent drops. Nearly all the signal has disappeared in the 1.44eV scan.

Comparison with topographic data shows that the two peaks in the line traces occur at the grain boundaries, indicated by arrows in Fig. 2. The relative peak height, or contrast, is maximum at 1.492eV. The region around the boundaries collects less relative to the boundaries as the excitation energy falls below the CdTe band gap. This is presumably due to the boundary material having a lower band gap, and therefore higher S concentration, than the grain center. The contrast first becomes visible at 1.509eV and finally disappears at 1.451eV, allowing us to estimate that the region away from the grain boundary is nearly pure CdTe while at the boundary the S concentration is ~7.6%.

In the discussion above, we have attributed features in the photocurrent images to spatial variations in absorption which arise from changes in bandgap across polycrystalline grains due to the presence of CdTexS1-x. Contrast can also originate from other sources including coupling between topography and photocurrent, and spatial variations in the collection efficiency of carriers created by the NSOM. We were able to discount the possibility that the enhanced collection was somehow due to topography affecting the coupling of light into the sample, since many such grain boundaries did not show a significant variation in photocurrent. This can be seen in Fig. 1 where several grain boundaries visible in the topography do not show large photocurrent contrast. Our tests also showed that the enhanced contrast at the boundaries was independent of such measurement parameters as probe-sample separation, illumination intensity, and scan speed.

The presence of spatial variations in the efficiency with which carriers are collected is a more interesting possibility to consider. For the photocurrent measurements shown in Figs. 1 and 2, small changes in excitation energy dramatically affect the spatial nonuniformities observed. This leads us to conclude that they are not due to variations in collection efficiency, as we would expect such variations to be observable over a broader range of wavelengths. In some cross sectional photocurrent images of these devices we have, however, observed enhancements in photocurrent near grain boundaries which are present from energies well above the CdTe bandgap to below the gap and which we believe are due to enhanced collection efficiency near the grain boundary. In future work we hope to explore transport effects near grain boundaries in more detail and correlate them with S concentration.

In conclusion, we find that the S concentration across annealed CdTe/CdS heterojunctions is quite nonuniform. We were able to directly observe regions at grain boundaries which have a lower band gap than that at grain centers, indicating a higher S content at the boundary. This provides very strong evidence that S diffuses preferentially along the grain boundaries of polycrystalline CdTe and suggests that the overall lack of uniformity in S composition is a result of this grain boundary assisted diffusion mechanism.

The authors gratefully acknowledge valuable discussions with M. H. Aslan, D. S. Albin, T. Ohno, and P. V. Meyers, and the technical support of J. Kintner and O. Wolf. This material is based upon work supported by the National Science Foundation under Grant No. DMR-9704780. We also would like to acknowledge support for device fabrication from the National Renewable Energy Laboratory's subcontract #XAF-8-17619-28.


1. Harin S. Ullal, Kenneth Zweibel, and Bolko von Roedern, 26th IEEE PVSC, p.301 (1997).

2. D. Mao, L. H. Feng, Y. Zhu, J. Tang, W. Song, R. Collins, D. L. Williamson, and J. U. Trefny in 13th NREL Photovoltaics Program Review Proc., AIP Conference Proc. 353, p.352 (1996).

3. D. G. Jensen, B. E. McCandless and R. W. Birkmire, Proc. 25th IEEE PVSC p.773 (1996).

4. B. E. McCandless, I Youm and R. W. Birkmire, Prog. Photovolt. Res. Appl. 7, 21 (1999).

5. M. H. Aslan, W. Song, J. Tang, D. Mao, R. T. Collins, D. H. Levi, R. K. Ahrenkiel, S. C. Lindstrom and M. B. Johnson, Mat. Res. Soc. Symp. Proc., 485 p.203 (1998).

6. W. Song, D. Mao, Y. Zhu, J. Tang and J. U. Trefny, 25th IEEE PVSC, p.873 (1996).

7. G. A. Valaskovic, M. Holton and G. H. Morrison, Applied Optics 34, 1215 (1995).

8. E. Betzig, P. L. Finn and J. S. Weiner, Appl. Phys. Lett. 60, 2484 (1992).

9. Keiichi Ohata, Junji Saraie and Tetsuro Tanaka, Jap. J. Appl. Phys. 12, 1641 (1973).

10. D. S. Albin, private communication.

11. O. deMelo, M. Melendez-Lira, I. Hernandez-Calderon, L. Banos, and Arturo Morales-Acevedo, Proc. 1st IEEE WCPEC p.369 (1994).

12. M. S. Unlu, B. B. Goldberg, W. D. Herzog, D. Sun and E. Towe, Appl. Phys. Lett. 67, 1862 (1995).