The impact of Cu on recombination in high voltage CdTe solar cells Free

II–VI semiconductors are used in lasers, light-emitting diodes, detectors, and solar cells.¹ The most common approach to dope II–VI semiconductors is to substitute type IB or type V elements on the cation or anion sites, respectively. Because of the relatively large effective hole mass in CdTe, these states are approximately 50–200 meV above the valence band.² Type IB dopants such as Ag, Au, and Cu are more mobile and easier to insert at lower temperatures than type V dopants.^3–7 However, we find that these relatively shallow acceptors can influence recombination.

This has significant implications particularly for photovoltaics. CdTe solar cells are challenging the dominance of polycrystalline silicon. Record CdTe solar cell efficiency is higher than that for multicrystalline Si (21.5% and 20.8%, respectively),⁸ and costs at current manufacturing efficiencies are already similar or lower.⁹ Yet, CdTe technology has room to improve. The open-circuit voltage (V_oc), fill factor, and efficiency depend on hole concentration, which for CdTe is currently on the order of 10¹⁴cm⁻³. If the hole concentration can be increased by two orders of magnitude without decreasing lifetime, V_OC can be significantly increased and efficiencies of 25% are possible.

Throughout the history of CdTe solar technology, Cu has been critical to attain hole concentration and to form the back contact.^10–13 However, the role of type IB dopants is complex. First-principles calculations indicate that increasing Cu levels can lead to formation of Cu_i donors that compensate Cu_Cd acceptors.¹⁴ Lifetime is affected by Cu levels, and Cu is mobile and can segregate to grain boundaries or adjacent material layers.^{10,13,15–17} Cu concentration is often 10¹⁷–10¹⁸cm⁻³, yet hole concentration is several orders of magnitude lower.^15,17 Here, we construct a recombination model for state-of-the-art CdTe devices and demonstrate that the same Cu_Cd site that increases hole density can also reduce lifetime.

Five polycrystalline films with large grains (6–10 μm) were chosen for photoluminescence (PL) analysis, because for smaller grains inhomogeneous broadening could obscure some features in the PL spectra.^{6,11,12,18,19} The films were grown in a superstrate configuration by vapor-transport deposition, annealed in a CdCl₂ containing atmosphere, and doped with Cu from a wet-applied Cu source followed by a thermal bake. For polycrystalline CdTe, the resulting devices have an extremely high V_oc of 890–900 mV. A reference sample without intentional Cu doping had a much lower V_oc of 777 mV. The high V_oc of these Cu-doped solar cells can be attributed largely to an increase in the hole concentration and long lifetimes in the CdTe absorber layer.^20,21 Capacitance-voltage measurements indicate a net acceptor concentration of 1 × 10¹⁵cm⁻³, which is approximately an order of magnitude greater than typical CdTe solar cells.^10,21 PL and electrical measurements indicate that Cu_Cd sites cause this hole concentration.^4,11,22,23

For PL emission spectroscopy, excitation was at 632.8 nm with a continuous wave (cw) HeNe laser. Time-resolved photoluminescence (TRPL) was measured using time-correlated single-photon counting and laser excitation at 640 nm with 0.3-ps pulses fired with a frequency of 1.1 MHz.²⁴ The photons were detected with a Si avalanche photodiode and appropriate bandpass filters, or with a monochromator and photomultiplier tube.²⁵ The resulting decay curves were fit with a bi-exponential function using deconvolution²⁶ 

The initial component τ₁ is attributed to the combination of interface recombination, drift, and diffusion, whereas τ₂ better reflects the bulk lifetime.^27–30 For the samples studied here, τ₂ ranges from 13 to 24 ns, which is state of the art for polycrystalline CdTe.²¹ 

To understand how hole concentration and lifetime may be further increased, we use energy- and time-resolved spectroscopy to form a detailed recombination model. Electron-hole pairs were generated in a Beer's law distribution with a penetration depth of ∼500 nm (2.3/α_640 nm, where α is absorption coefficient³¹). By measuring time-resolved emission spectra starting from such well-defined electron-hole pair distribution, we can analyze emission signatures and carrier dynamics even when spectral lines are heterogeneously broadened due to the S interdiffusion at the CdS/CdTe interface.¹⁹ Fig. 1 illustrates the time-resolved emission spectra at 5 K for a sample doped with Cu. The PL decay is much faster for the high-energy (exciton) region of the spectrum than for the lower-energy (defect) region. This leads to a time-dependent PL spectral shift attributed to faster decay of the excitonic PL and donor-acceptor pair (DAP) recombination. The red shift at >17 ns is largely attributed to DAP recombination. DAP recombination can be analyzed from³² 

where E_ZPL is zero phonon energy, E_g is bandgap, E_a/E_d are acceptor/donor energies, e is electron charge, ε is dielectric constant, and ε₀ is permittivity of vacuum. As carriers recombine, the average distance between donors and acceptors, r, increases so that the spectral shift toward the red is observed. To determine all terms in Eq. (2), we varied temperature and excitation power.

Figure 2 illustrates the temperature-dependent shift of the exciton band. Data show that the device has different E_g values at the junction and back-contact regions. The difference ΔE_g ≈ 40 meV is attributed to formation of a CdSTe alloy at the junction.¹⁹ Ellipsometry data indicate this energy shift corresponds to a CdS_xTe_1−x composition with x ≈ 0.05.³³ The measurements described below were repeated for both sides of the photovoltaic device: the CdS/CdTe junction where CdSTe alloy properties were evaluated, and the CdTe film side (see inset in Fig. 2). The absolute E_g values were estimated from the data in Fig. 2 by adding the free exciton binding energy (≈10 meV)³⁴ and exciton-defect binding energy (10–20 meV).^35,36 From 50 to 250 K, the E_g shift is measured to be approximately 0.3 meV/K. A similar result was reported for CdTe single crystals.³² 

Next, we determined E_ZPL from the analysis of well-resolved phonon replicas in Fig. 3. Peaks from 1.35 to 1.45 eV can be fit with five Gaussians spaced by 21.3 meV, which is the LO phonon energy in CdTe.^35,36 Therefore, the sub-bands are phonon replicas for a defect with E_ZPL = 1.450 eV. The intensity distribution shown in the inset is well described by the Poisson distribution with the Huang-Rhys factor S = 1.58 ± 0.02. E_ZPL and S values are in agreement with Stadler et al., who derived E_ZPL = 1.45 eV and S = 1.60 for single-crystal CdTe with Cu.³² The data also agree with other reports for the Cu on Cd site.^4,22,23,37 The clear PL signal from Cu_Cd indicates that this dopant is important for the recombination kinetics.

Having determined the E_ZPL, we examined the DAP binding energy by varying the electron-hole injection level. Fig. 4 shows PL emission spectra from the junction side when excitation power was varied from 3 nW to 11 μW. Heterogeneous broadening does not allow observation of the phonon sub-bands,^{6,11,12,18,19} but the DAP band is similar to that in Fig. 3. As shown in the inset, the Coulombic interaction for the DAP leads to the blue shift of the PL emission. The dependence of the broad DAP peak on the excitation power, I_EXC, can be analytically described as³⁸ 

where D is a coefficient, hν_max is the experimental DAP energy, hν_∞ = E_g − E_a − E_d is the limiting photon energy of infinitely distant pairs, and hν_B = E_B + hν_∞ with E_B = e²/4πεε₀R_B (R_B is Bohr radius of the shallow impurity). Data in the inset indicate hν_∞ = 1.336 eV. Eq. (3) predicts that the blue shift due to Coulombic interaction is at most the binding energy of the shallow impurity. For CdTe, shallow donors form with E_d ≈ 10 meV below the conduction band.³² Consistent with this estimate, when excitation power is increased by almost four orders of magnitude, the DAP band shifts by ≈10 meV. Fitting with Eq. (3) gives hν_B = 1.36 eV and R_B = 66 Å. The large Bohr radius is consistent with the small donor ionization energy. Taken together, the data indicate E_a + E_d = 0.16 eV, and most of this energy is attributed to E_a because E_d is small.

Next, we examine the concentration of DAPs from the TRPL data measured at 5 K (Fig. 5). DAP recombination can be described as consisting of two steps—electron (minority carrier) capture by donors, followed by tunneling recombination.³⁹ These processes occur on very different time scales. By examining the DAP PL rise, we find that the donor sites fill at a rate of (400 ± 30 ps)⁻¹ independent of injection (Fig. 5, inset). Consequently, it appears that the donor density is high and the donor states are not saturated for the injection levels used in the experiments. DAP recombination is much slower (>700 ns time scale). We determine the acceptor concentration, N_t, by fitting the data to the DAP recombination model^39,40

where ⟨A(t)⟩ is the ensemble average for the probability that the electron is on the donor, and the recombination rate W(r) exponentially depends on the DAP separation, r, where the maximal rate is W_max^39,40

When fitting data to the model of Eq. (4), the best fit occurs for N_t ≈ 10¹⁶cm⁻³. Experimental TRPL decays for the DAP band were essentially unchanged from 5 to 150 K, and the PL emission intensity was also approximately constant in this temperature range.⁴¹ These data are consistent with DAP tunneling recombination at low temperatures because tunneling rates are expected to have weak temperature dependence.

Shockley-Read-Hall (SRH) recombination at higher temperatures was analyzed by comparing TRPL lifetimes for undoped single-crystalline CdTe, a polycrystalline CdTe device without intentional Cu, and a high-voltage device with Cu-doping (Fig. 6). For the single-crystal reference sample, the lifetime has relatively weak temperature dependence and changes from 360 to 383 ns at 275–200 K.⁴² 

For the Cu-doped device, the lifetime at 170 K is 134 ns, which is on the order of single crystal lifetimes.^20,24,42 An apparent Arrhenius trend with the activation energy E_Arrhenius = 86 ± 8 meV is observed in τ(T) data for the Cu-doped sample in Fig. 6. Arrhenius-like temperature dependence for capture cross-sections has been identified for non-radiative multiphonon transitions due to deep levels in some semiconductors, such as GaAs and GaP.⁴³ Because the lifetime temperature dependence is very different for the single-crystal and Cu doped/undoped polycrystalline CdTe, it does not appear that a common deep defect can explain our experimental data, unless deep levels are created by Cu.

It is plausible that the lifetime decrease from 134 ns (170 K) to 11 ns (295 K) occurs due to relatively shallow recombination centers. The PL emission data indicate the presence of acceptors ≈150 meV above the valence band and shallow donors. Radiative lifetime at this doping is ≈10 μs,⁴⁴ while experimental lifetimes range from 500 to 11 ns. Therefore, SRH recombination is dominant. For a single recombination center in p-type material, the Shockley-Read-Hall equation can be written as

where $σn$ and $vth,n$ are the electron capture cross-section and thermal velocity, respectively, k_b is Boltzmann's constant, and T is temperature. The shift of the Fermi level E_f(T) with temperature will depend on the distribution of defects, and the PL spectra indicate the presence of both donors and acceptors. Time-resolved photoluminescence simulations were performed with code written in Sentaurus Device to analyze the role of compensation on lifetime temperature dependence.^28,30 Without a large concentration of compensating donors, the Fermi level can move both above and below the E_a = 150 meV defect and produce trends that are not observed here. In the case of strong compensation, the Fermi energy temperature variation is reduced, and the Cu_Cd recombination site can generate results consistent with the experimental data (see Fig. 6). Both donors and acceptors have been observed in the PL spectra here, and compensation has been widely reported in CdTe. So the PL and TRPL data can be consistently explained with the recombination model in Fig. 6. At low temperatures, recombination happens by DAP tunneling as described by Eq. (4). At higher temperatures, nonradiative SRH recombination from Cu_Cd states describes the data.

The estimated capture cross-sections and acceptor densities needed to fit these data are on the order of 10⁻¹⁶cm⁻² and 10¹⁵−10¹⁷cm⁻³, respectively, which is typical for neutral traps.⁴⁵ Consequently, it is reasonable that the relatively shallow states can influence recombination. PL spectra for II–VI semiconductors are often dominated by emission from acceptors such as Cu_Cd in CdTe, and our data indicate that these defects can limit lifetime.

This research was funded by NREL-First Solar Cooperative Research and Development Agreement (CRADA). The work at NREL was supported by the U.S. Department of Energy, Energy Efficiency and Renewable Energy, under Contract No. DE-AC36-08GO28308.