Transient Metastable Behavior Caused by Magnesium-Doped Zinc Oxide Emitters in CdSeTe/CdTe Solar Cells

Metastable behavior in highly efficient MZO/CdSeTe/ CdTe solar cells has been reported previously. Different preconditioning procedures have been studied that are used to recover the performance of the devices. 11 wt% of MgO content in the MZO layer has shown to give optimized photovoltaic parameters in the device compared to other MZO compositions. J–V characteristics before preconditioning of the devices with higher MgO content show an “S” shaped behavior, which is removed during preconditioning. However, this recovery remained only for 3 days while the devices were stored under vacuum in the dark. Temperature-dependent J–V and capacitance measurements before and after preconditioning revealed the presence of recombination centers and defect levels at the MZO/absorber interface. Previous studies have shown degradation of MZO occurring if the layer is exposed to ambient atmosphere. Hall effect measurements on the MZO films showed no significant changes after any preconditioning or CdCl2 treatment. Secondary-ion mass spectrometry images show diffusion of oxygen from the MZO layer into the CdSeTe region after CdCl2 treatments. This likely enables the MZO to function as a buffer layer since it will increase the carrier concentration due to the formation of oxygen vacancies. As-deposited MZO thin films are insulating. However, the oxygen vacancies in the MZO layer also increase its reactivity and instability.

technology for low-cost and high-efficiency thin film solar cells. CdTe has a near optimum band gap and a high absorption coefficient [1]. Device efficiencies have improved significantly and the champion cell efficiency is 22.1%, achieved by First Solar, Inc. [2].
Recent research work has focused on introducing new materials for the buffer and absorber layer, which has resulted in significant improvements in the device J sc and power conversion efficiency (PCE). The most significant improvements have occurred by introducing magnesium-doped zinc oxide (MZO) as a highly transparent buffer layer and the introduction of Se into the absorber to form a CdSeTe layer with a reduced band gap. These modifications have improved the PCE above 19% [3]. However, modifying the cell architecture introduces more complexity in understanding the factors that limit performance and has also introduced detrimental characteristics in the PV device, including stability issues as well as transient metastable behavior due to prior exposure history (such as temperature, climate, and irradiance). These characteristics have been observed after introducing the MZO buffer layer [4]. Furthermore, for consistent comparisons between different laboratories when using different PV materials and architectures, it is advisable to adopt a common preconditioning procedure in order to measure and interpret reliable performance data.
We have previously reported on the behavior of MZO/ CdSeTe/CdTe devices with 11 wt% of MgO content in the MZO layer [4]. Here, we report on the metastable behavior of PV parameters observed in devices with MgO content of 5 wt%, 8 wt%, 11 wt%, 14 wt%, and 21 wt% as measured in different laboratories under different ambient conditions. We also report on different preconditioning approaches that have been used to recover the device performance, and to eliminate the transient behaviors observed in high-efficiency devices. Approaches include light soaking at atmosphere, various cooling procedures under vacuum, and annealing studies under different vacuum and atmospheric conditions. One of the possible reasons for observed differences in PV parameters may be due to degradation of the MZO layer, which is sensitive to the presence of atmospheric humidity [5]. Recombination centers and defect levels are found located at the interface using admittance spectroscopy and temperature-dependent J-V measurements. These defects will contribute to metastable behavior and the associated degradation This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ observed in the devices. Hall effect measurements were carried out on the MZO thin films (11 wt% MgO content) to investigate the effect of preconditioning on the film properties and attempt to measure the carrier concentration. This work also reports oxygen diffusion observed in NanoSIMS images from the MZO layer into the CdSeTe region after the CdCl 2 treatment.

A. Sample Preparation
The devices used for this study were fabricated at Colorado State University (CSU) using an in-line vacuum deposition system [6]. The device architecture was as follows: Glass/TCO /MZO/CdSeTe/CdTe/(CdCl 2 &CuCl treatment)/Te/C/Ni. The thicknesses of the as-deposited layers in the p-n junction region were: 100 nm of MZO, ∼ 0.5 µm of CdSeTe, and ∼ 3 µm of CdTe. Details of the fabrication process of CdSeTe/CdTe devices are provided elsewhere [3].
The MZO films for Hall effect measurements (MgO 11%, ZnO 89% by weight) were deposited using an RF magnetron sputtering system [7] on NSG-Pilkington TEC SB glass, which is used as a substrate for offline coatings and does not have a conductive layer. The MZO film thickness was 500 nm. Two sets of samples, as deposited, and annealed were provided. The MZO films were annealed at 620°C under vacuum for 140 s. Compared to a full device stack (which uses 100 nm of MZO as a buffer layer), thicker MZO films were used to ensure a lower resistance film to assist the Hall effect measurements. 0.5 cm × 0.5 cm samples were cut from the substrates, and four contacts were attached at the corners of each sample using an ultrasonic soldering iron. Subsequently, samples were mounted on a holder with thin wires for each contact using a soldering iron.

B. Characterization
The room temperature current density-voltage (J-V) measurements were performed using an in-house built solar simulator setup with a xenon arc lamp at 100 mW·cm −2 with a AM1.5G filter. An unfiltered silicon diode was used as a reference to calibrate the solar simulator. The J-V measurement was taken using a Keithley source measurement unit. The preconditioning was performed using an ABET solar simulator, which has a temperature-controlled stage with a PID controller to maintain the sample temperatures in the operating range of −25°C to 85°C. Preconditioning was carried out at 100 mW·cm −2 using a xenon arc lamp with a AM1.5G filter. The temperaturedependent electrical measurements were carried out in an evacuated closed-cycle helium cryostat with a measurement range of 315 to 45 K. The device temperature was adjusted using a LakeShore 335 temperature controller.
Capacitance measurements were performed with a Keysight E4990 impedance analyzer. Temperature-dependent J-V (JVT) curves were acquired using a Keysight B2902A source measurement and samples were illuminated under 50 mW·cm −2 with a white LED (6500 K) light source.
The distribution of oxygen in MZO/CdSeTe/CdTe devices was mapped in 3 days by high-resolution secondary-ion mass spectrometry (SIMS) measurements with a Cameca 50/50L NanoSIMS. During the measurements, a 0.5-1 pA Cs+ ion beam with a nominal diameter of 60 nm was rastered over the surface and sputtered secondary ions were analyzed with a double-focused magnetic sector mass spectrometer. Masses analyzed were 35 Cl − , 16 O − , 24 Mg 16 O − , and 64 Zn 16 O −. A highresolution map of each of these species was formed from each complete scan.

C. Preconditioning Procedures
Following a similar procedure to NREL's standard protocol for preconditioning and stabilization of polycrystalline thin film PV modules [10], two separate preconditioning procedures have been applied to the MZO/CdSeTe/CdTe devices. These procedures were categorized as vacuum and atmospheric preconditioning and the sequence involved for each procedure is described as follows.
Vacuum preconditioning involves annealing at 65°C (using a substrate heater) under vacuum (in the dark) at a pressure of ∼ 7.8 × 10 −6 torr for an hour. Then, the devices are removed from vacuum and light soaked for 15, 30, and 60 min under 100 mW·cm −2 . Light soaking is performed at open-circuit conditions with no temperature control of the device. Dark and illuminated J-V curves are recorded before and after the annealing and between each time interval.
Atmospheric preconditioning involves annealing and cooling down the device at different temperatures, under ambient atmospheric conditions, while maintaining the device temperature. Preconditioning starts with 60 min of light soaking under 100 mW·cm −2 at 25°C. Then, 30 min of light soaking at 85°C followed by 30 min of light soaking at 10°C and finally, 30 min of light soaking at 25°C. Dark and illuminated J-V curves are recorded before and after the preconditioning and between each temperature change.
MZO films preconditioned using light soaking under 100 mW·cm −2 at 25°C for 60 and 120 min. The as-deposited MZO films were also CdCl 2 treated to identify the effect this device level treatment has on the individual film properties. CdCl 2 was deposited by thermal evaporation at ∼ 1×10 −6 torr for 20 min using 0.5 g of CdCl 2 pellets inside a quartz crucible. A set of samples was then annealed on a hot plate in air at a dwell temperature of 420°C, while another set was annealed in nitrogen under vacuum at ∼ 400 torr. Both samples were heated for 1, 5, 10, and 20 min. After the annealing steps, the MZO films were rinsed with DI water to clean the excess CdCl 2 from the surface. Treated MZO films were also light soaked by the same procedure as annealed and asdeposited films. Additionally, atmospheric preconditioning was applied to the MZO films prior to conductivity and Hall effect measurements.

II. RESULTS AND DISCUSSION
The devices were fabricated at CSU, where a J-V measurement was obtained after fabrication. The devices were then packaged under vacuum and shipped to Loughborough, U.K., by courier. Fig. 1 shows J-V measurements directly after receiving the samples at Loughborough of devices with varying MgO content in the MZO layer. Upon receipt in Loughborough and before any preconditioning, the J-V curves for the devices with MgO content ≥ 11 wt% show an "S" shaped behavior, which is commonly referred to as a "kink" type anomaly in the illuminated J-V curve and occurs in the third or fourth quadrant of the J-V curve [11]. This may be due to a conduction band offset or the presence of a significant current barrier. Since the MZO layer is sensitive to the presence of atmospheric humidity, more MgO content in the film could cause increased degradation of PV parameters and less diode like characteristics in the J-V curves before any preconditioning. The ZnO only devices showed a typical J-V response with better PV parameters, and the MgO only devices and the devices with 11-21 wt% of MgO suffer from high series resistance (R s ) and low shunt resistance (R sh ), resulting in a low fill factor (FF), which could be due to a presence of a significant current barrier or a conduction band offset. The rest of the analysis will focus on the devices with 11 wt% MgO content in the MZO layer, which are reported with the best J-V characteristics and high PCE from CSU measurements directly after fabrication in Colorado, USA. Fig. 2 shows J-V measurements for the device with 11 wt% MgO directly after fabrication (measured in Colorado, labeled as CSU), and before and after 15 min of light soaking after receipt of the devices in Loughborough. The J-V curve from the asdeposited device measured at CSU is well behaved showing the typical J-V response expected for a high efficiency device, with high FF. However, upon receipt in Loughborough and before any preconditioning, the J-V curves show an "S" shaped behavior. After 15 min of light soaking under 100 mW·cm −2 at opencircuit conditions, this "S" shaped behavior reduces significantly and shows better J-V characteristics with improved FF and R s . However, 15 min of light soaking did not recover the device performance fully. Note there is also a slight mismatch of shortcircuit current density (J sc ). This is due to the different reference diodes used at the two laboratories and is not a metastable effect.  Vacuum preconditioning was applied to the same device after the device was maintained under vacuum in the dark for approximately 3 days. J-V curves before and after the vacuum preconditioning are shown in Fig. 3. Before and after the annealing, the J-V curves showed similar characteristics and the "S" shaped behavior was restored. After, with light soaking at atmosphere, the "S" shaped behavior is removed but the performance of the device was not totally recovered. The same approach was applied for the atmospheric preconditioning study, where the same device was kept under vacuum in the dark for approximately 3 days. Prior to preconditioning, J-V curves were obtained, which again showed the "S" shaped behavior. Like vacuum preconditioning, atmospheric preconditioning showed similar J-V characteristics (see Fig. 4.) with improved R s with light soaking.
V oc (T) analysis from the JVT measurements (for temperatures between 315 and 115 K) indicates no significant changes in the activation energy (E a ) of the recombination mechanism before  (E a = 1.43 eV) and after (E a = 1.44 eV) preconditioning shown in Fig. 5. Since the activation energy before and after preconditioning is still E a < E g [12], the main recombination mechanism for the device dominates at the buffer/absorber interface [13]. The buffer/absorber band alignment or a small barrier can play a significant role in the interface recombination. Light soaking can establish a stronger built-in field, which is believed to improve the band alignment at the interface, leading to an increase of the barrier height. Thus, reducing the interface recombination [14], [15]. However, no significant change in E a has been observed after atmospheric preconditioning.
Admittance spectroscopy was performed under equilibrium conditions in the dark from 315 to 45 K. The measurements revealed admittance steps at temperatures between 225 and 135 K. The activation energies corresponding to these admittance steps are found to be E A = 31 meV before and E A = 32 meV after preconditioning from the Arrhenius plots, as shown in Fig. 6. There has been some controversy for many years concerning the origin of these admittance steps. Based on the measured activation energies, the energetic position of these  steps link to the N1 defect, which may possibly originate from the buffer/absorber interface or the Schottky back contact [16], [17]. There are also other reports correlating the N1 step to the potential barriers at grain boundaries [18], [19]. However, the origin of N1 step in our case is not well understood and requires more investigation. The depth profiles of the net carrier concentration before and after preconditioning were investigated using capacitancevoltage (C-V) and drive-level capacitance profiling (DLCP) techniques. For this study, vacuum preconditioning was also applied to investigate the effects of annealing on the depth profile characteristics. All measurements were carried out in the dark at room temperature, with dc bias voltage varied at a fixed frequency. Fig. 7 shows the depth profiles for both C-V and DLCP measurements before and after atmospheric preconditioning. The depth profiles exhibit very similar shape and profile distance for both techniques. Before any preconditioning, the depth profiles showed an unusual double minima, often also seen in some solution-processed CIGS devices, which have a bilayer   [20], [21], [22]. The double minima in the C-V depth profile of solution-processed CIGS devices is usually caused by the bilayer structure in the absorber layer when the depletion width crosses between the two layers during the voltage sweeps. The double minima in this case is likely due to the formation of a back barrier with MZO degradation caused by moisture [5]. Fig. 8 shows the depth profiles for both C-V and DLCP measurements before and after vacuum preconditioning, where a different cell from the same device was used. Vacuum preconditioning resulted in similar effects on the depth profiles. However, annealing did not show any significant changes in the shape of the depth profiles. After both atmospheric and vacuum preconditioning, the second minima toward the back of the device becomes less pronounced. Hence, the depth profiles become more U-shaped commonly reported for CdTe devices. Table I summarises the effects of preconditioning (on different cells but from the same device) on the net carrier concentration for both DLCP and C-V methods (N DLCP and N CV ), deep level defect densities (N defect ), built-in voltage (V bi ), and depletion width (w d ) at 0 V. V bi before and after preconditioning was extracted from the linear portion of the Mott-Schottky plots. An increase in V bi is observed in devices after preconditioning with both atmospheric and vacuum procedures (note that different cells are used for vacuum preconditioning). A similar increase is also observed in indium−tin oxide (ITO)/CdTe devices with light soaking and is believed to improve the band alignment between ITO and CdTe [15]. However, it remained only for a short time period, similar to that observed in MZO/CdSeTe/CdTe devices in this study. The net carrier concentration was extracted at 0 V from the depth profiles and an increase was observed after preconditioning. This increase was expected as devices were light soaked and values after preconditioning are now much more closer to previously reported values for CdTe devices [23]. The depth profiles became narrower after preconditioning and these changes correlate well with the increase observed in net carrier concentration. The defect density was measured by DLCP in the dark at room temperature for varied frequency and oscillating voltage. The net carrier concentration has increased significantly after light soaking, resulting in a small increase in defect density in the absorber. Due to the high net carrier concentration, the small increase in defect density has no significant effect on the device performance. The increased net carrier concentration after light soaking shows a similar effect to a doped absorber layer with an increased V bi and a narrower w d .
Despite all the preconditioning attempts, the recovery of the PV parameters and the J-V characteristics remained only for 3 days while the devices were kept under vacuum in the dark. The capacitance-and temperature-dependent JVT measurements before and after preconditioning showed the presence of recombination centers, and defect levels at/close to the buffer/absorber interface. Therefore, the next appropriate step is to investigate the buffer layer. The characteristics of the buffer layer before and after preconditioning will provide a better understanding in observed metastable behaviors, and any relation between the measured devices.
The MZO films, which were either light soaked, annealed under vacuum, atmospheric preconditioned, or annealed at atmosphere after CdCl 2 treatment, did not show any significant changes in conductivity and were too resistive. Thus, no useful signal was detected for viable Hall measurements. However, the MZO films, which were CdCl 2 treated and annealed in nitrogen under vacuum, showed significant improvements in conductivity, and improved response of the signal from the measurements. Since the MZO films are sensitive to the presence of atmospheric humidity and oxygen, annealing the films at atmosphere after CdCl 2 treatment may introduce more oxygen into the film, causing the conductivity to decrease. However, annealing in nitrogen under vacuum would prevent the introduction of oxygen into the film thereby benefiting the conductivity. Annealing for a longer time in nitrogen under vacuum has shown significant improvements in the linearity of the I-V response compared to those annealed for a shorter time.
Despite all attempts to recover the I-V response of the measurements, the films were still resistive, and at the limit of the system to provide any reliable results. Therefore, no linear sheet resistance response and useful Hall signal were detected. Since the as-deposited MZO films are insulating, MZO as a window\buffer\emitter layer should not perform in the device. It is therefore clear that another mechanism is present that enables the MZO to function. For example, oxygen vacancies could be created within the MZO film during full device processing, probably during the CdCl 2 treatment. Oxygen vacancies create electron donors, which would result in improved carrier concentration in the film [24]. Also, the incorporation of excess chlorine atoms from the CdCl 2 treatment into the ZnO might be another mechanism that enables the MZO to function, as chlorine may substitute the oxygen vacancies in the lattice to improve conductivity [25]. Fig. 9 shows cross-sectional SIMS images of the Cl, O, MgO, and ZnO signal intensity at the front MZO/absorber interface from a working device that has been activated with the CdCl 2 treatment. The chlorine is segregated in the adjacent grain boundaries and is also present across the junction. It can be seen that some oxygen has diffused from the MZO layer into the CdSeTe region, and the oxygen vacancies are likely to have been created in the MZO by diffusion occurring during the CdCl 2 activation treatment. However, oxygen vacancies make the oxide vulnerable to hydrolysis and the formation of hydroxides in a damp atmosphere [26]. Therefore, the formation of oxygen vacancies will make the device vulnerable to the degradation, which is observed in the J-V analysis and the PV parameters.

III. CONCLUSION
The PV parameters of MZO/CdSeTe/CdTe devices have been measured following transport between two laboratories in North America and Europe. Different preconditioning procedures were applied to recover the device performance. Annealing the device did not show a significant improvement. However, the longer light soaking resulted in a significant recovery of the device performance. Despite all the preconditioning attempts, storing the device in the dark for approximately 3 days appears to subsequently degrade the performance again. A reason for these differences is likely due to degradation of the MZO layer since temperature-dependent measurements show recombination centers and defect levels dominate at the MZO/absorber interface. Although using MZO as a buffer layer in CdSeTe/CdTe devices results in high PCE initially, the metastability and degradation are responsible for the variation of PV parameters measured in different laboratories, under different ambient conditions. Diffusion of oxygen from the MZO layer after the CdCl 2 treatment will increase conductivity and enable it to function as a buffer layer due to the formation of oxygen vacancies. The diffusion of oxygen creates electron donors in the MZO resulting in improved carrier concentration in the film. However, these same oxygen vacancies make the oxide vulnerable to hydrolysis resulting in degradation in the film composition and the PV parameters. It is a paradox that the mechanism that means that the MZO works well as a buffer layer also makes it vulnerable to degradation.
As deposited and annealed MZO films before and after any preconditioning and CdCl 2 treatment show nonlinear I-V response, thus no useful Hall signal was detected to extract the carrier concentration and mobility. One potential way of improving the film preparation to extract the Hall signal is to perform a standard fabrication process on a glass/MZO/CdSeTe/CdTe stack and detach the CdSeTe/CdTe layer after CdCl 2 treatment. In this way, any potential effect on the material properties and conductivity of the MZO layer can be analyzed on a device level treatment and provide a more accurate representation of the effect on the MZO layer. However, on exposure to the atmosphere, the MZO is likely to quickly reoxidize and again become electrically insulating.