Spectroscopic Analysis for the Identification of Loss Mechanisms in Back‐Contact Perovskite Solar Cells

Back‐contact perovskite solar cells offer a significant potential to reach high efficiency due to reduced parasitic absorption from the top surface. However, the currently reported efficiencies are considerably lower (<10%) than planar perovskite solar cells (>20%). Herein, back‐contact perovskite solar cells are fabricated to study loss mechanisms that cause low device efficiency. This work spatially resolves the short‐circuit current, open‐circuit voltage, photoluminescence quantum yield, carrier lifetime, and external quantum efficiency of the devices. The results indicate that the front surface recombination, increased nonradiative recombination at hole contact layer/perovskite interface, and the extraction barriers are three main mechanisms limiting devices from achieving high efficiencies.


Introduction
The world record cell efficiency in crystalline silicon photovoltaics (PV) currently stands at 26.7%, just ≈2.1% lower than the maximum achievable power conversion efficiency (PCE) when realistic losses such as Auger and light trapping are considered. [1][2][3] The record performance in silicon PV is mainly attributed to the complete interdigitated back-contact (IBC) heterojunction design. [1,4] In the back-contact (BC) configuration, both the anode DOI: 10.1002/admt.202300241 and the cathode are located at the rear of the photoabsorber layer, rendering the front-facing surface available for antireflection coatings and transparent passivating layers to be deposited. [1] Furthermore, these layers can be deposited without the constraint of requiring electrical conductivity. The strongest advantage of the BC cell is that parasitic absorption and reflection associated with the top electrode in the planar counterpart are nearly eliminated, while the front surface can be completely passivated. These improvements have been shown to increase the absolute short-circuit current density (J SC ) by more than 3.6% in silicon PV. [5][6][7] While IBC Si cells are commercially available, back-contact configurations in emerging, efficient PV such as III-V, CdTe, CIGS, and perovskite devices have been relatively less explored. [8][9][10][11][12][13][14][15][16][17][18][19] The key difference between silicon and such emerging technologies is the nature of the absorber layer. For silicon, which has an indirect bandgap, the minority carrier-diffusion length (L d ) is on the order of 100-1000 μm. [20,21] In comparison, organic-inorganic perovskite materials, which are the primary focus of this paper, demonstrate L d on the order of 1-10 μm. [22][23][24][25] Consequently, advanced processing such as photolithography is necessary to achieve the required scale and positioning of the back contacts for efficient carrier extraction. There are few reports implementing BC configurations in perovskite solar cells, exploiting various BC architectures such as quasi-interdigitated back-contact, honeycomb like structures, and micron-sized V-shaped grooves. [14][15][16][17][18][19] Such works have reported device efficiencies between 3.2% and 10.5%.
An advantage in studying BC structures in novel PV technologies is that the exposed front surface allows us to optically and electrically probe interfaces between the contacts and absorber material. In turn, these spatially resolved scans yield insights into the carrier generation, recombination, and transport mechanisms, thereby enabling us to identify and mitigate remaining loss pathways-for both the BC cell and the corresponding planar configuration. [39] Here, we develop a BC solar cell based on our previously proposed theoretical design. [26] In this architecture, the Ni anode, NiO x hole transport layer (HTL) and Al 2 O 3 dielectric spacer are fabricated in a square grid array with a pitch of 12 μm and linewidth of 4 μm. The grid is fabricated on top of a planar SnO 2 electron transport layer (ETL), with the perovskite layer deposited on top of both contacts. We report a PCE of 4.83% for a 4 mm 2 cell, using a triple cation perovskite active layer with a nominal composition of Cs 0.1 (FA 0.8 MA 0.2 ) 0.9 Pb(I 0.8 Br 0.2 ) 3 . Moreover, we investigate the device performance limitations by spatially resolving the optical and electrical characteristics using a home-built integrating sphere microscopy setup combined with a piezoelectric stage, as described in our previous works. [27,28] We map the local photogenerated current and open-circuit voltage (V OC ), absorption, external quantum yield (EQE), photoluminescence quantum yield (PLQY), and lifetime, across the device surface and use this data to identify loss mechanisms and propose potential points for improvement.

Device Structure
The fabricated BC perovskite device considered in this study is schematically depicted in Figure 1a. Generally, for IBC cells, photolithographic alignment is required to position the different carrier selective contacts across the device. [29,30] As only one contact is patterned in our device, photolithographic alignment is not necessary which significantly simplifies the fabrication procedure compared to the conventional IBC cell.
The full fabrication procedure of the device is outlined in detail in Figure S1 in the Supporting Information. Briefly, 60 nm of SnO 2 is grown on a cleaned FTO glass substrate by chemical bath deposition. The chemical bath growth method has been shown to form conformal, pinhole-free coverage of SnO 2 , resulting in higher performing devices compared to spin-coated SnO 2 layers. [31,32] Potassium iodide is spin-coated on the SnO 2 to passivate the surface, followed by an anneal at 100°C for 10 min in air. [40] To fabricate the grid contact, a photoresist adhesion layer (hexamethyldisilazane) and a negative photoresist (Ma-N 1410) are sequentially deposited on the substrate. The grid pattern is developed through a mask by UV lithography. The grid consists of a 12 μm pitch with a linewidth of 4 μm (Figure 1b). Following the development of the pattern in the resist, 500 nm thick grids consisting of 300 nm of an Al 2 O 3 spacer and 200 nm of Ni are sequentially deposited on the sample by e-beam evaporation. The thickness of grids is optimized to 500nm by comparing efficiencies using a sample set with 420, 460, and 500 nm thick grids. The result is summarized in Figure S2 (Support-ing Information). The sample is kept overnight in N-methyl-2pyrrolidone to lift off the remaining photoresist and the undesired metal. The samples are then UV ozone treated for 45 min to form an oxide seed layer and annealed at 300°C for 1 h under ambient air. The anneal yields the desired core-shell Ni-NiO x structure, where the Ni and NiO x function as the anode and the hole transport layer, respectively. [19] Finally, a solution of Cs 0.1 (FA 0.8 MA 0.2 ) 0.9 Pb(I 0.8 Br 0.2 ) 3 is spin-coated on top of the sample in a N 2 atmosphere and annealed at 100°C for 1 h. [33,34] The details of the perovskite film preparation are described in the Supporting Information. The thickness of the perovskite on the SnO 2 surface is measured to be 418 nm by a profilometer ( Figure S3, Supporting Information). The cross-sectional SEM in Figure 1c indicates that the perovskite film is not planarized across the surface; rather, it follows the topography of the underlying grid. The top surface of the device is intentionally not capped so that device properties can be analyzed without considering parasitic absorption. However, high efficiency BC solar cells generally employ a passivating layer and antireflective coating on top of the photoabsorber to minimize front surface recombination velocity, increase photoabsorption and enhance the overall efficiency of the device. [41] 3. Results

Photovoltaic Cell Performance
The photocurrent density-voltage (J-V) scan was measured at 1 Sun under standard AM1.5G, 1000 W m −2 , 25°C test conditions. The forward (0 V → V OC ) and reverse (V OC → 0 V) J-V sweeps are shown in Figure 2a. The V OC = 0.77 V, J SC = 14.4 mA cm −2 , and fill factor (FF) = 0.38 for the forward sweep, and V OC = 0.79 V, J SC = 12.5 mA cm −2 , and FF = 0.49 for the reverse sweep. These parameters correspond to respective device efficiencies of 4.26% and 4.83%. To identify loss pathways resulting in the low PCE in this device, we measured the spatially resolved electrical characteristics using focused laser scanning microscopy, which enables photocurrent and photovoltage generated from the laser scanning to be mapped. More information is provided in the Supporting Information of this paper. Figure 2b depicts the local current generated from a focused 720 nm laser when scanned across a 30 × 30 μm 2 area of the cell. The local current generated on the perovskite located above the Ni@NiO x grid is ≈17.6% (300 nA) lower than the local current generated above the planar SnO 2 /FTO substrate. The V OC generated upon laser excitation was also mapped. Similar patterns are observed, as shown in Figure S4 (Supporting Information). Further resolving the photocurrent variation between the contacts, Figure 2c plots the average line scan of the photocurrent highlighted in Figure 2b.

Optical Scanning Microscopy
To further investigate the mechanisms behind the observed spatial variations in photocurrent and photovoltage over the device, we measured key optical properties of the device, namely the ab-sorptance, PLQY and PL lifetime. We measure the steady-state optical properties with an integrating sphere microscopy set up, coupled to a supercontinuum laser and a piezoelectric stage. [27,28] Figure 3a shows the measured absorptance of the device, scanned with a 650 nm laser over a 24 × 24 μm 2 area. On average, 70% of the incident light is absorbed by the perovskite layer located directly above the SnO 2 /FTO/glass substrate. The fraction of absorbed light increases to 79% in the region above the Ni@NiO x grid. As the perovskite layer was measured to be 418 nm thick above the SnO 2 and 411 nm thick above the grid ( Figure S2, Supporting Information), the enhanced absorption is not a result of layer thickness. Rather, we attribute the increase to the metallic Ni anode in the grid acting as a back reflector of light. However, despite the higher absorption, we observe the opposite behavior in photocurrent generation (Figure 2b). Given the relatively flat c) The average PL lifetime mapped across the area using FLIM with TCSPC. d) The fitting of differential lifetime on the two contact regions (Ni@NiO x grid and SnO 2 ) over a larger (80 μm × 80 μm) area that includes the map in panel c (for complete area lifetime map see Figure S5 in the Supporting Information). photocurrent profile on the NiO x and that the highest photocurrent is observed in the center of the SnO 2 regions (Figure 2b,c), we discount carrier diffusion as the dominant factor in this spatial modulation (photocurrent would then be expected to peak at the electrode grid edge). Instead, our results point to a higher carrier recombination rate above the Ni@NiO x grid compared to the SnO 2 /FTO contact.
Further examining the optical properties, the PLQY measured across the device under 650nm laser excitation is shown in Figure 3b. The PLQY deviates significantly between the SnO 2 /FTO opening and directly above the Ni@NiO x grid which show the PLQY of 13% and 5% on average, respectively. This is consistent with the higher nonradiative recombination rate above the HTL regions compared to the ETL regions.
From the PLQY-V OC relation: V OC,imp = V OC,rad + V therm ln(PLQY), where V OC,imp is the implied V OC , V therm is the thermal voltage = 25.7 mV at 25°C and V OC,rad is the V OC at the radiative limit. [36][37][38] The actual V OC from the device assuming perfect carrier extraction should be 1.12V based on the relation shown above. However, considering the measured V OC of 0.79V, a big loss exists in our device which indicates a V OC loss from an extraction barrier at the contact.
To test this hypothesis, we conduct spatially resolved PL lifetime measurements under a 485 nm pulsed laser excitation using time correlated single photon counting microscopy (TCSPC). The average PL lifetime-calculated at each pixel using fluorescence lifetime imaging microscopy (FLIM)-is mapped in Figure  3c [35] . The complete 80 μm × 80 μm scan and the separated complete average lifetime histograms are included in Figure S5 (Supporting Information). The calculations deriving the average PL lifetime are included in the Supporting Information Section 6.The transient PL decay curves for each of the separated NiO x and SnO 2 contact regions are plotted in Figure S5b (Supporting Information). A high order fitting procedure was applied to these curves and the differential lifetime is calculated at every time point for the separated contact regions and is shown in Figure 3d [42] . From the definition of the differential lifetime, the plateaus of each of the two curves at low carrier density (that is, at longer times, i.e., after 10 −7 s) are the effective nonradiative recombination lifetimes. This was found to be T NiOx = 58 ns for T SnO2 = 71 ns. Extended details on the differential lifetime definition and decay curve fitting procedure, in addition to the differences in the differential lifetime and the average PL lifetime are provided in the Supporting Information. The 22% longer lifetime on SnO 2 suggests a higher rate of nonradiative recombination on the HTL as compared to the ETL. Assuming that the bulk nonradiative lifetime of the perovskite is the same across both contact regions, it is therefore likely that this difference arises from higher interfacial carrier recombination at the NiO x interface, compared to the SnO 2 interface. Figure 4 shows EQE measurement results for SnO 2 and Ni@NiO x device regions. The EQE was obtained by measuring the photocurrent of the device and varying the incident laser wavelength across the visible range ( = 480-780 nm). As part of the measurement, white bias light is mixed with the chopped laser to help the device reach the injection level expected under solar illumination. The details of the measurement are described in Section 4 of the Supporting Information. Considering the energy barriers of the contacts, there is a marked decrease in absolute EQE over all wavelengths when the laser is scanned over the HTL grid (Figure 4b). At ex = 700 nm, there is a 17% drop in EQE over the HTL grid compared to the ETL substrate (Figure 4c). Notably, we observed a trend of decreasing EQE from longer wavelength 700 nm to shorter wavelength 480 nm. Additionally, the difference between EQE on two contacts becomes smaller toward shorter wavelengths. A current scanning map at 410 nm is performed as a supplement to this observation ( Figure S4, Supporting Information). The map shows a lower current value than at 710 nm and spatially uniform photocurrent and photovoltage maps.

EQE Line Scan
The measured EQE can be integrated over wavelength to calculate the expected short-circuit current density by the standard relation (Equation (1)), where Φ is the spectral photon flux. [36] By integrating the EQE, we obtain the J SC value of 10.17 mA cm −2 which is about 20% lower than the J SC extracted from the J-V curve (Figure 2). This difference is partially explained by the smaller range over which we can measure the EQE (only to 480 nm due to laser bandwidth restrictions) and partially from the difficulty in satisfying the condition of the chopped laser being a small perturbation on the light bias while also having sufficient signal. Nevertheless, the general trend of lower EQE at shorter wavelengths was present for different light bias values (see Figure S6 in the Supporting Information).

Discussion
Combining the results shown above, the loss mechanisms of back-contact perovskite solar cells can be described as the follows. First, we see that the front surface recombination plays a significant role. Due to the lack of a top contact, carriers generated in the perovskite film have to diffuse through the entire layer to reach the back contacts. For shorter wavelengths, the carriers are generated close to the top surface where surface recombination dominates the loss because of the unpassivated top surface. For longer wavelengths, the carriers are generated closer to the back contacts on average, which minimizes the need for vertical diffusion. This hypothesis is supported by the EQE, which shows an increasing EQE value from shorter wavelengths to longer wavelengths everywhere on the device (Figure 4a,b).
Increased nonradiative recombination rates at the HTL/perovskite interface compared to the ETL/perovskite provides a second major loss mechanism. This is supported by the lower PLQY and PL lifetime as well as lower EQE over the HTL regions (Figures 3b-d and 4a,b). The ETL regions have lower nonradiative recombination rates than both the air/perovskite and the HTL/perovskite interfaces. This can be seen from Figure 4b: when carriers are generated closer to the ETL interface (longer wavelengths), EQE almost doubles, while carriers generated closer to the HTL interface only lead to slight EQE enhancements. This effect can be seen even more clearly by comparing the photocurrent map at 720 nm ( Figure 2b)-which shows dramatic spatial variations-to that at 410 nm ( Figure S4, Supporting Information), which is homogeneous. At short wavelengths where absorption happens near the surface, surface recombination dominates, while at longer wavelengths the difference in recombination rates at the ETL and HTL interfaces become relevant.
Carrier extraction is the third major loss mechanism. This is clear from the very high device PLQY-approaching that of record GaAs solar cells-combined with the low V OC , J SC , and FF values. If nonradiative recombination at the contacts, surface or in the perovskite were the main loss mechanism limiting efficiency, it would not be possible to measure such high PLQY values. The absorption measurements also confirm that enough light is collected to reach much higher efficiencies. Measurements of leakage current through the dielectric showed nothing above the noise floor (approximately 5 pA, as shown in Figure S7 in the Supporting Information). Altogether this suggests that the primary problem is a very large extraction barrier at either the ETL, HTL, or both. Given that the SnO 2 has been used extensively for high efficiency perovskite solar cells, we postulate that this barrier is primarily caused by the HTL. The substantial light bias dependence of the EQE ( Figure S6, Supporting Information) is consistent with an extraction barrier caused by poor Fermi level alignment or insufficient HTL doping. In addition, an s-shaped JV curve is observed when scanning the sample over a larger voltage range ( Figure S8, Supporting Information). This further confirms the carrier extraction barrier.
Given the three loss mechanisms described above we can suggest three concrete steps to improve performance in back-contact perovskite solar cells. First, a top surface passivation layer is re-quired to improve the blue response of the cells. The reduced surface recombination will also boost the V OC . [41] If chosen correctly, this layer can serve a dual function as an antireflection (AR) coating and encapsulation layer. Silicon oxide and aluminum oxide have been shown to provide excellent passivation and improved stability, but SiO 2 is likely preferable due to its nearly ideal refractive index for an AR coating. Second, the NiO x HTL needs to be replaced or modified to reduce the interfacial recombination rate and especially lower the extraction barrier. Currently thermal oxidation is used to form the NiO x HTL, but electrodeposition might be a more suitable technique since it can provide much more flexibility in the HTL material choice, including the use of dopants to tune the Fermi level and carrier density.

Conclusion
In this paper, we have fabricated grid-shaped back-contact perovskite solar cells with SnO 2 as ETL and NiO x as HTL and measured their electrical and optical properties with micron-scale spatial resolution. The combination of absorption, photocurrent, photovoltage, EQE, PLQY, and PL lifetime mapping provides a powerful suite of characterization tools to gain insights into the loss mechanisms and develop a path forward for improved device efficiency. Our results suggest the ultimate efficiency of back-contact perovskite devices can approach record efficiency GaAs solar cells since the PLQY of the devices, which is an indication of the maximum achievable V OC , is already approaching the highest value ever recorded for operating solar cells. Moving forward the biggest improvement will come from replacing or modifying the NiO x HTL, which currently shows a large extraction barrier and substantial interface recombination. Higher efficiency, particularly from improved blue response, can also be reached by passivating the surface with a transparent dielectric that can serve a triple function as AR coating and encapsulation layer for improved stability. These results suggest the back-contact perovskite solar cell geometry is promising for reaching the ultimate device efficiency and deserves further study.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.