Effects of oxygen partial pressure, deposition temperature, and annealing on the optical response of CdS:O thin films as studied by spectroscopic ellipsometry Free

Cadmium sulfide (CdS) thin films are commonly used as n-type window layers in various thin film solar cell configurations, most notably cadmium telluride (CdTe)^1,2 and copper indium gallium selenide^1,3 photovoltaic (PV) devices. As has been detailed elsewhere⁴ for the case of CdTe cells, overall device performance can be restricted by CdS layers because of a tradeoff between greater short-wavelength absorption in thicker CdS layers and poorer open circuit voltage and fill factor in devices fabricated with thinner CdS layers. One strategy for overcoming such limitations has been to increase the band gap of CdS by incorporating oxygen, thereby allowing optimal thicknesses with a greater amount of the incident solar spectrum reaching the underlying PV absorber. CdTe devices made with oxygenated CdS (CdS:O) window layers have shown promising results,^5,6 with a maximum conversion efficiency of 15.5% reported.⁴ While these studies have confirmed that the use of CdS:O window layers can potentially improve the performance of CdTe solar cells, further investigation on the effects of close-space sublimation (CSS) of CdTe on the properties of underlying CdS:O as is found in the CdTe solar cell configuration is needed. This CSS of CdTe is typically done at temperatures above 500 °C, which is more than sufficient to alter the initial structure and optical properties of CdS:O deposited at lower temperatures. It has been reported by Paudel et al.⁶ that by exposing as-deposited CdS:O layers to the temperature, pressure, and ambient gas composition encountered during the CSS processing, crystallinity in the films increases and the band gap energies revert to values similar to that of CdS. In that study, however, band gap determinations were performed using unpolarized transmittance spectroscopy of relatively thick films (3000–4000 Å) with incident wavelengths limited to the 300–1000 nm range (4.13–1.24 eV). Those films are not completely representative of the <1000 Å thick CdS or CdS:O layers typically used in high performance CdTe solar cells. Additionally, the material-specific optical properties obtained using those measurements are mainly limited to the absorption coefficient (α), and from this the band gap energy (E_g), which provides only a cursory understanding of the optical response of these films both before and after high temperature processing.

Optimization of CdS:O films for opto-electronic devices like thin film solar cells requires a detailed understanding of their optical response. Previous studies of CdS:O films have demonstrated the effectiveness of applying spectroscopic ellipsometry in determining the influence of some fabrication and processing conditions on resultant optical properties.^7–9 These studies, however, have not examined important intermediate oxygen gas flow ratios and high temperature processing consistent with full CdTe device fabrication, and they do not exhibit sufficient sensitivity to accurately model higher energy, above band gap critical point transitions in CdS and CdS:O. This work represents a comprehensive study of the effects of deposition temperature and subsequent processing conditions similar to CSS of CdTe on CdS:O films for use in CdTe solar cells, as evaluated using a physically realistic optical property model developed to retain sensitivity to critical point transitions in disordered, but polycrystalline, CdS:O films.

Two series of CdS:O films have been deposited onto soda lime glass substrates. The glass has been prepared using a detergent-based (Micro-90, International Products Corp.) cleaning procedure assisted by ultrasonication to ensure chemically clean surfaces. Radio frequency (RF) magnetron reactive sputtering (frequency = 13.56 MHz) has been used to deposit the CdS:O in 10 mTorr ambient with the proportion of O₂ gas adjusted by controlling the flow rates of separate streams of pure Ar and 5% volume fraction O₂ in Ar, with the total combined flow rate maintained at 40 sccm. Each series consists of films grown with O₂ gas flow ratios 100% × [O₂]/{[Ar] + [O₂]} = 0% (CdS), 1%, 2%, and 3% volume fractions, with one series deposited with no intentional substrate heating (i.e., at room temperature denoted RT) and the other with the substrate heated to 270 °C. Each sample is annealed at a temperature of 607 °C in 99.5% He + 0.5% O₂ volume fractions at 50 Torr to simulate the CSS CdTe processing conditions that would follow CdS:O deposition in the PV device fabrication sequence. Further details about the fabrication and processing of these samples are described elsewhere.⁶ 

Ellipsometric spectra (in N = cos 2Ψ, C = sin 2Ψ cos Δ, and S = sin 2Ψ sin Δ) for all the samples have been measured at RT over a spectral range of 0.74 to 5.89 eV using a single rotating compensator multichannel ellipsometer^10,11 (Model M-2000FI, J. A. Woollam Co., Inc.) both before and after annealing at 607 °C. A stratified structural model has been developed to extract layer thicknesses and optical properties in the form of complex dielectric function (ε = ε₁ + iε₂) spectra. The parameters describing the model are iteratively fit using a least-squares regression that minimized the unweighted error function¹² between the model simulation and measured ellipsometric spectra as shown in Figure 1. The structural model consists of a semi-infinite glass substrate, bulk CdS:O layer of thickness d_b, and surface roughness of thickness d_s as shown schematically in Figure 1. The optical properties of the roughness layer are described by a Bruggeman effective medium approximation^13,14 consisting of f_v volume fraction void and 1−f_v fraction material identical to the underlying CdS:O. Spectra in ε describing the substrate are described by a Sellmeier oscillator and a constant additive term to ε₁, ε_∞, when fitting to ellipsometric measurements collected for a bare piece of identical soda lime glass.^15,16 The optical properties of the CdS:O are represented by a hybrid parameterization of ε. Spectra in ε₂ are described by the sum of oscillators assuming critical point parabolic band (CPPB) transitions¹⁷ above the lowest energy critical point (E₀) (i.e., photon energies where the films are strongly absorbing) and an Urbach tail¹⁵ below E₀. Then, spectra in ε₁ are calculated as the sum of a high-energy Sellmeier oscillator, a constant ε_∞ term, and Kramers–Kronig integration¹⁵ of corresponding spectra in ε₂ to ensure that ε remains physically realistic. Thus, the full parameterization of ε used for the CdS:O materials can be written

where A_j, θ_j, Γ_j, and E_j are the CPPB amplitude, phase projection factor, broadening, and resonance energy of the jth interband transition, respectively; μ_j is fixed at 0.5 for one dimensional critical points; E_u is the Urbach energy; A_s and E_s are the Sellmeier amplitudes and resonance energies, respectively; and P is the principal value of the Kramers–Kronig integral. It should be noted that x-ray diffraction measurements of similar but thicker CdS:O films⁶ and atomic force microscopy measurements of both these films and the corresponding thicker films confirmed the predominantly polycrystalline CdS:O structure, thereby validating the use of CPPB oscillators to describe ε for photon energies where the films are absorbing. As will be discussed in greater detail later, the sharpness and distinguishability of individual features in ε vary significantly with sample processing conditions, in particular, ambient O₂ gas flow ratio during sputtering. Comparison of spectra in ε for each sample indicates that the film prepared at 0% oxygen flow ratio, 270 °C deposition temperature, and (post-deposition) annealed (0%/270 °C/annealed) most closely resembles ε of highly crystalline CdS films as has been reported elsewhere.¹⁸ Therefore, the phase projection factors θ_j of the three critical points within the measurement range (hereafter referred to as E₀, E₁–A, and E₁–B) are fixed for all samples to the values determined for the 0%/270 °C/annealed sample. This decrease in the number of fitting parameters is essential for reliably extracting ε of the other samples, especially those with broader, less distinct features in ε.

Spectra in ε for each sample are plotted in Figure 2, and the corresponding model parameter values and overall spectrally averaged mean square error (MSE) are reported in Table I. In addition to the E₀ transition, we have independently resolved the E₁–A and E₁–B critical points and modeled their dependence on CdS:O film processing. These closely spaced features are easily distinguished for samples prepared at low oxygen gas flow ratios, but films prepared with increasing oxygen flow ratios generally have increasingly broad absorption features, sometimes making the E₁–A and E₁–B critical points difficult to individually resolve. In the most extreme cases, the indistinguishability of these critical points led to unacceptably high error associated with many of the E₁–A and E₁–B parameters, necessitating the use of a single CPPB oscillator to model the higher energy features in ε, as is seen for the 3%/270 °C/annealed, 3%/270 °C/as-deposited, 2%/RT/as-deposited, and 3%/RT/as-deposited samples. Additionally, d_s could not be determined with acceptable error for three of those samples and is therefore fixed to 0 Å. The location and values of the E₁–A and E₁–B peaks in ε₂ for the 0%/270 °C/annealed sample are in good agreement with the published literature.^18–20 Such agreement with other studies further justifies the identification of that sample as the “best crystalline quality” CdS sample and its suitability for determining the θ_j phase projection factors.

Although the error associated with each fitted parameter is the primary indicator used to assess the models' sensitivity, parameter cross correlation coefficients are also examined. Some correlation coefficients are found to be somewhat high (>0.95). However, a vast majority of these cases are between two parameters that affected the model simulation in very similar ways such as ε_∞ and A_s or f_v and d_s. A few instances of high CPPB parameter cross correlation are found and are almost exclusively limited to A₀ and E₀ correlation, likely a result of the large flexibility of lineshape in the vicinity of E₀ due to the transition from Urbach to CPPB behavior. In general, the occasional correlation between parameters is a side-effect of having an optical-structural model that could be applied to the analysis of all sixteen samples while still retaining sufficient sensitivity to the fine-structure of the films' optical properties and surface structure. Models with fewer fitting parameters have been tested, but result in increased MSE and overall poorer fit quality in all cases. The combination of low MSE and reasonable error bars on all model fits reported in Table I provides justification that the presence of some cross correlation has not rendered the results and subsequent analysis unreliable. Although some parameters may be correlated, the overall resulting line shape of ε represents the characteristics of the particular CdS:O material.

With the parameterization of ε described in Equation (1), the fundamental gap energy E₀ is the first allowed electronic transition and is therefore representative of the material band gap. Figure 3 shows the dependence of each sample's E₀ critical point energy on oxygen gas flow ratio during sputtering. To demonstrate the reliability of using E₀ as the material band gap, the band gap energies have also been determined using the absorption coefficient (α). Using ε, α for each sample is calculated using

where λ is the wavelength of incident light. Finding the zero-intercept energy of a linear fit to α² in the region directly above the absorption onset serves as an extrapolation of each sample's direct band gap energy.²¹ Figure 4 shows the extrapolation over the range of α from 2.5 × 10⁴ to 6.6 × 10⁴cm⁻¹ for the 0%/270 °C/annealed sample, producing E_g = 2.43 ± 0.06 eV, which is in reasonable agreement with other published values for CdS.^22–25 Similar ranges of α (∼10⁴–10⁵cm⁻¹) and, by extension, α² are used to extrapolate E_g for all the other samples. Finally, Figure 5 shows the correspondence between E₀ critical point energies and E_g values linearly extrapolated from α². This correspondence is highlighted by the fact that a linear fit to all sixteen data points (dashed line) is extremely close to the E_g = E₀ line (solid line).

Figures 3 and 5 highlight the fact that the band gap energies for the as-deposited films have relatively large variations, whereas annealing those same films results in all band gaps converging near the CdS band gap of ∼2.4 eV, regardless of oxygen gas flow ratio during deposition. A likely source of the band gap variation for the as-deposited samples is the formation of cadmium oxide (CdO) and cadmium peroxide (CdO₂) compounds within the film. Numerous studies have confirmed that CdO films are readily produced using high temperatures and can have band gaps as much as a few tenths of an eV lower^26–30 than pure CdS with epitaxial CdO films exhibiting a band gap as low as 2.16 eV.³¹ By contrast, CdO₂ films are most stable near RT and feature band gaps greater than 3.3 eV and as high as 4.64 eV.^30,32,33 In this context, lower E₀ seen for both 1%/as-deposited samples is a likely incorporation of oxygen as in CdO. At 3% oxygen flow ratio, however, the band gaps for both of the as-deposited samples are greater than that for CdS suggesting incorporation of oxygen as in CdO₂. The shift from E₀ decreasing to E₀ increasing occurs at lower oxygen flow ratio (2%) for RT deposited as compared to the 270 °C deposited samples (3%), which is consistent with the observation that CdO₂ forms more readily at lower temperatures. Additionally, the scale of E₀ deviations from the CdS value of ∼2.4 eV suggests greater CdO₂-like oxygen incorporation in the RT samples, which is again consistent with the increased stability of CdO₂ seen at lower temperatures. Other studies of CdS thin films have shown that increases in the microstructural tensile strain lead to increases in the band gap energy.^34–38 This effect is another likely contributor to the changes in E₀ observed for the as-deposited samples in Figures 3 and 5; however, studies of strain in CdS³⁴ and in CdTe³⁹ have demonstrated strain-induced critical point energy shifts of only up to ∼0.024 and ∼0.2 eV, respectively, meaning that this mechanism is not likely to be the dominant factor for these CdS:O films. Annealing, however, can be understood to relax the majority of microstructural strain induced by oxygen inclusion as a result of E₀ being close to uniform for all annealed samples. Finally, it should be noted that the uniformity of E₀ for annealed samples also suggests that performance improvements seen in CSS-deposited CdTe devices with CdS:O window layers may not be a result of increased CdS:O band gap. A more likely source of such performance improvement is the general inverse relationship between oxygen gas flow ratio and ε₂ amplitude, which would allow a larger proportion of the incident light to reach the absorber layer due to greater ultraviolet transparency.

The trend of increasingly broad features in ε with increasing oxygen gas flow ratio seen in Figure 2 is in qualitative agreement with the previous reports.^7–9 Figure 6 provides a more quantitative representation of how the fundamental gap transition broadening Γ₀ evolves with oxygen gas flow ratio. It is noteworthy that Γ₀ generally increases with increasing oxygen gas flow ratio and that for all samples Γ₀ decreases after annealing. As has been previously demonstrated for CdS thin films,²⁰ sharper critical point features (i.e., decreased broadening) are correlated with increased crystallographic order. Therefore, it can be reasonably concluded that increased oxygen gas flow ratio during sputtering generally results in increased disorder stemming from smaller grain sizes, a large defect density, or a combination of both. Additionally, as could be expected, the data suggest that both the grain size and crystallinity increase upon annealing for all samples, regardless of oxygen gas flow ratio, which is consistent with previous observations.⁴⁰ 

As described earlier, ε below E₀ is described by an Urbach tail where the Urbach energy E_u is generally indicative of the defect density in the material. Larger E_u results in band tails with greater absorption and further extension into the band gap. Figure 7 shows E_u as a function of E₀ for each film. Despite both annealed series of films (squares and triangles) having nearly uniform E₀ energies, they exhibit a spread in E_u suggesting that significant variations in defect density can persist even after annealing has modified the optical band gaps to be nearly the same for all samples. The Urbach energies of the 270 °C/annealed samples are observed to monotonically increase with oxygen gas flow ratio as opposed to the RT/annealed samples for which a clear trend is not evident. For the pure CdS samples (0% oxygen gas flow ratio), annealing can be understood to decrease the overall defect density, as is demonstrated by the decrease in E_u for both the RT and 270 °C samples upon annealing. This result is consistent with the decrease in Γ₀, observed in Figure 6, and the understanding that annealing improves crystallographic order and therefore decreases defect density. In the case of the 1%–3% oxygen gas flow ratio samples, the effects of oxygen complicate the result of the annealing process since exactly half of the samples show increase in E_u while the other half shows decrease following annealing.

Finally, an analysis of the effective film thicknesses in conjunction with corresponding values of ε₁ and relative material density deduced from ε₁ reveals that the result of the annealing process is more complex than simple densification of the existing material in the films. Figure 8 shows the effective film thicknesses calculated as d_b + (1 – f_v) d_s and plotted as a function of oxygen gas flow ratio. Annealing is seen to result in thinner effective thicknesses for all films except for one pair where the thicknesses are within the associated error. Figure 9(a) plots the value of ε₁ at 0.74 eV for all films as a function of oxygen gas flow ratio. If densification of the films is the sole cause of the thickness changes shown in Figure 8, higher values of ε₁ would be expected where the films are transparent, but this behavior is observed in Figure 9 for only a few sample pairs. The film with the highest value of ε₁ at 0.74 eV is identified as 0%/270 °C/as-deposited CdS. A Bruggeman effective medium approximation consisting of fractions of this relatively “dense” material and void is fit to ε₁ at 0.74 eV for the other samples, with results shown in Figure 9(b). The results show a direct mirror correspondence to ε₁. Consequently, the densification of existing material cannot alone explain the effect of annealing on the films. As a result of the Kramers–Kronig relationship between the real and imaginary parts of ε, the results seen in Figure 9 can be understood in the context of the other variations in ε described in this work, even variations seen in the absorbing part of the measured spectrum. The contributions from specific parameters are often difficult to deconvolute. For instance, an increase in Γ₀ would result in an increase in ε₁ where the film is transparent due to extension of absorption to lower photon energies. Considering the annealed and as-deposited 1%/RT pair of samples, offsetting effects on ε₁ between the densification of the film as shown in Figure 8 and the decrease in Γ₀ from Figure 6 could explain the exact correspondence in ε₁ for that sample pair. However, similar densification and change in Γ₀ are seen for the 2%/RT samples, yet they have significantly different ε₁ values at 0.74 eV, indicating that a more complicated interplay between various physical mechanisms affects the overall film optical properties. Similarly, increases in E₀ shifting peak absorption to higher photon energies, such as those seen in Figure 3 for the 1%/270 °C and 2%/270 °C samples as a result of annealing, can explain the corresponding decrease in ε₁ at 0.74 eV. The relatively close E₀ energies for the as-deposited and annealed 3%/270 °C samples could be a strong contributing factor to convergence of ε₁ for those samples. However, for the 3% RT deposition samples, E₀ is vastly higher for the as-deposited film when compared to its annealed counterpart while ε₁ is nearly the same for both at 0.74 eV. Ultimately, multiple structural and optical characteristics of these films have been shown to be sensitive to the oxygen inclusion but no single parameter dominates the optical response for all films. Further study will be necessary to more comprehensively understand the exact role oxygen plays in the composition of CdS:O films.

Spectroscopic ellipsometry has been used to study the effects on spectra in ε obtained for CdS:O thin films for multiple combinations of oxygen volumetric gas flow ratios from 0% to 3% during sputtering, RT and 270 °C deposition temperatures, and 607 °C high-temperature annealing similar to temperatures found in CSS of CdTe. As expected, ε of films deposited at 0% oxygen gas flow ratio most closely resemble ε of CdS published elsewhere while the critical point features of films with increasing oxygen flow ratio are increasingly broadened. The band gap energies of as-deposited films are sensitive to relative oxygen gas flow ratio during deposition, whereas all annealed films exhibit band gaps resembling that of CdS. In the broader context of full solar cell devices, our results indicate that the optimum fabrication conditions for CdS:O layers will likely differ for different types of PV devices. For instance, only the annealed samples in this study are relevant to CdTe superstrate cells since the high temperature CdTe deposition will follow the CdS:O deposition. For such devices, the most promising candidate may be the 3%/RT/annealed CdS:O since it has the best combination of increased crystallinity as assessed by narrow broadening of the lowest energy critical point, low sub-gap absorption represented by the Urbach energy, and low above-gap absorption aiding in ultraviolet transparency. It is likely that performance gains observed in CdTe solar cells through the use of CdS:O window layers are the result of the desirable characteristics previously mentioned and not the increased bandgap in the CdS:O layers as initially believed.

This work was supported by University of Toledo start-up funds, the Ohio Department of Development (ODOD) Ohio Research Scholar Program (Northwest Ohio Innovators in Thin Film Photovoltaics, Grant No. TECH 09-025), National Science Foundation (CBET-1230246), and Office of Naval Research (11847944).