Irreversible phase back conversion of α-FAPbI3 driven by lithium-ion migration in perovskite solar cells

Abstract

Typically n-i-p structured perovskite solar cells (PSCs) incorporate 2,2′,7,7′-tetrakis (N,N-di-p-methoxyphenyl amine)-9,9′-spirobifluorene (spiro-OMeTAD) as the hole-transporting material. Chemical doping of spiro-OMeTAD involves a lithium bis(trifluoromethyl sulfonyl)imide dopant, causing complex side-reactions that affect the device performance, which are not fully understood. Here, we investigate the aging-dependent device performance of widely used formamidinium lead triiodide (FAPbI₃)-based PSCs correlated with lithium-ion (Li⁺) migration. Comprehensive analyses reveal that Li⁺ ions migrate from spiro-OMeTAD to perovskite, SnO₂, and their interfaces to induce the phase-back conversion of α-FAPbI₃ to δ-FAPbI₃, generation and migration of iodine defects, and de-doping of spiro-OMeTAD. The rapid performance drop of FAPbI₃-based PSCs, even aging under dark conditions, is attributed to a series of these processes. This study identifies the hidden side effects of Li⁺ ion migration in FAPbI₃-based PSCs that can guide further work to maximize the operational stability of PSCs.

INTRODUCTION

The power conversion efficiency (PCE) of perovskite solar cells (PSCs) has dramatically improved from 3.8% to 25.8% during the last decade, owing to continuous advancements in compositional design, crystal growth, and defect engineering.^1–3 Despite continuous improvements in the material and composition of perovskite and electron transporting layer (ETL), 2,2′,7,7′-tetrakis(N,N-di-p-methoxyphenyl amine)-9,9′-spirobifluorene (Spiro-OMeTAD) is still adopted as the hole-transporting material from the first solid-state PSCs to state-of-the-art devices.^1,3–5

The undoped Spiro-OMeTAD has poor carrier mobility (~4 × 10⁻⁴ cm² V⁻¹ s⁻¹) and low doping efficiency (~10 mol% of Spiro⁺ radicals are formed). The doping process to form the Spiro⁺ radicals take advantage of the oxidation mechanism through bis(trifluoromethane)sulfonimide lithium salt (LiTFSI) and 4-tert-butylpyridine (tBP). Thus, a large amount of dopant (~55 mol% LiTFSI, ~330 mol% tBP) is needed to facilitate the fabrication of well-functioning devices with enhanced hole mobility and proper band alignment.^6,7 As shown in Figure 1, the Li reacts with oxygen and water molecules in ambient air to form LiO_x or LiOH, while the TFSI⁻ undergoes a redox reaction to form spiro⁺ radicals.⁸ Prior to generating these byproducts, the tBP interacts with Li⁺ to assist in the doping process. However, these dopants were found to be highly hygroscopic and can induce damage to the underlying perovskite layer.⁹ Thus, replacing these dopants with other alternatives has been widely investigated to achieve more stable PSCs.^10,11 Nevertheless, most state-of-the-art PSCs, including record-setting devices, still incorporate these conventional dopants.³

FIGURE 1. Schematic of the aging process of typical perovskite solar cells(PSCs) with an electrochemical oxidation process of Spiro-OMeTAD.14

Besides participating in undesirable chemical reactions, Li⁺ ions tend to penetrate the underlying perovskite and ETL depending on the aging time.^12,13 As the alkali metal is incorporated as an interstitial dopant within the perovskite, it can passivate and immobilize the charged defects, thus enhancing the performance and stability of the PSCs.^5,14 The ETL can also benefit from the migrated Li⁺ ions, as carrier extraction is enhanced by improved interfacial band alignment and conductivity of the ETL.^15,16 Despite the reported beneficial effects of interstitial doping, our recent study unraveled possible detrimental effects that are caused by induced microstrain, particularly for widely used formamidinium lead tri-iodides (FAPbI₃) based perovskites.⁵ Thus, we were motivated to study side effects caused by the Li⁺ ion migration in FAPbI₃ PSCs.

In this study, we analyzed the aging-time-dependent performance and energetics of FAPbI₃-based PSCs to determine any correlation between Li⁺ ion migration and device performance. Cross-sectional Kelvin probe force microscopy (KPFM) was used to investigate any changes in the interface energetics of the device with different aging times. The impedance and time-resolved photoluminescence (PL), current, and voltage dynamics were correlated with the investigated energetics. Time-of-flight secondary ion mass spectroscopy (ToF-SIMS) confirmed the redistribution of Li⁺ ions in the device, even during dark storage in a dry atmosphere. Finally, a possible mechanism for the detrimental effects of Li⁺ migration is suggested.

RESULTS AND DISCUSSION Aging-dependent PSC performance

We adopted PSCs with the structure FTO/SnO₂/FAPbI₃/Spiro-OMeTAD/Ag to investigate the photovoltaic parameters during dark storage (relative humidity <5%). Typical current density-voltage (J–V) curves, stabilized PCE, and external quantum efficiency measurement data are shown in Figures S1–S3. In the repeated experiments, we observed a reproducible aging trend in the photovoltaic parameters (Figure 2A–E and Table 1) during storage in the dark. The device showed a relatively low open-circuit voltage (V_OC, 1.050 ± 0.020) and fill factor (FF, 0.73 ± 0.01) with insufficient aging (IA, until ~144 h), and V_OC and FF continuously increased and reached maximum values (V_OC = 1.123 ± 0.009, FF = 0.75 ± 0.01) to maximize the PCE with proper aging (PA, ~550 h). With further prolonged storage, the FF rapidly dropped to 0.66 ± 0.01 to degrade PCE while the V_OC decreased slightly to 1.115 ± 0.009; hereafter, we denote this stage as over aging (OA, > 650 h). The short-circuit current (J_SC) remained almost constant for all aging times.

FIGURE 2. Photovoltaic parameters of (A) short-circuit current density (JSC), (B) fill factor (FF), (C) open circuit voltage (VOC), and (D) power conversion efficiency (PCE) for FAPbI3-based PSCs with different aging times under dark conditions (~5 RH% and ~20°C). Red and black dots represent the data obtained from reverse and forward scans, respectively. (E) Representative J–V curves of the corresponding devices with different aging times. Cross section scanning electron microscopy (SEM) images of the device under (F) PA and (G) OA conditions. IA, insufficient aging (~144 h); OA, overaged (> 650 h); PA, proper aging (~550 h).

TABLE 1 Photovoltaic parameters and calculated shunt, and series resistance of devices with different aging times.

Condition	JSC (mA/cm2)	VOC (V)	FF	PCE (%)	Series resistance (Ω cm2)	Shunt resistance (Ω cm2)
IA	24.38 ± 0.20	1.050 ± 0.020	0.73 ± 0.01	18.78 ± 0.57	5.54	7929
PA	24.71 ± 0.24	1.123 ± 0.009	0.75 ± 0.01	20.84 ± 0.35	4.49	9435
OA	24.73 ± 0.25	1.115 ± 0.009	0.66 ± 0.03	18.10 ± 0.89	10.47	10 195

The initial performance enhancement is a typical aging effect owing to the oxidation-assisted doping of spiro-OMeTAD and improved work function alignment at the spiro-OMeTAD/Ag interface.^17,18 However, the origin of the observed rapid fill factor degradation from PA to OA was not elucidated but was reproducibly observed in our study. Upon closer analysis of the J–V curves in Figure 2E, we identified that the degraded FF is dominantly affected by the increased series resistance of 10.47 Ω for OA from 4.49 Ω of PA (Table 1). The enhanced shunt resistance correlated with an almost constant V_OC with the OA.¹⁹ To examine the morphological changes in each layer and their interfaces, cross-sectional scanning electron microscopy (SEM) images of the PA and OA devices are compared in Figure 2F,G. As shown in Figure 2F,G, no clear differences were observed between the two devices.

Aging-dependent energetics

We employed high-resolution cross-sectional KPFM to investigate any changes in the interface energetics in devices with different aging times. It should be noted that the device was mechanically cleaved without the aid of ion milling to minimize any possible damage during sample preparation. Measurements were conducted under constant light exposure in both short-circuit (SC) and open-circuit (OC) conditions. As shown in Figure 3A–C, each layer can be distinguished by phase images obtained in the noncontact mode (NCM) used to measure the contact potential difference (CPD) map (Figure 3D,E). The corresponding topological images are shown in Figure S4. As shown in Figure 3D–F, a clear change in CPD contrast with different aging times was observed, indicating that the energetics at each layer and their interfaces were affected by the aging time. In Figure 3G–I, the E-field line profiles are shown under OC, depending on the aging time, as obtained by differentiating the CPD profile in the vertical direction (white arrows in Figure 3D–F). Initially, in the OA device, negligible potential difference between the spiro-OMeTAD, perovskite, and SnO₂ layers was observed, indicating no induced E-field from the stagnant charges at either interface under IA. With prolonged aging, the potential difference and consequent E-field start to build up under PA (see also Figure S5) and are largely enhanced under OA. Considerable changes occurred at both interfaces.

FIGURE 3. Noncontact mode (NCM) phase images (A–C) and corresponding Kelvin probe force microscopy (KPFM) potential maps (D–F) of the cross-sectional perovskite solar cell (PSC) device depending on the storage time in the dark: (A,D) insufficient aging (IA, (B,E) proper aging (PA), and (C,F) over aging (OA), respectively. The measurement was conducted under mild illumination at OC. All the scale bars represent 500 nm. Electric field line profile of the devices under (G) IA, (H) PA, and (I) OA conditions obtained from corresponding potential maps (white arrows in D, E, and F).

According to the Poisson equation given in Equation (1), the E-field distribution is directly related to the accumulated charge carriers.²⁰ [Image Omitted. See PDF]where the $\phi$ is electrical scalar potential, $\rho$ is charge density, ${\epsilon }_r$ is relative permittivity ${\epsilon }_0$ is permittivity in free space. The local charge carrier density is directly calculated by integrating both sides. Moreover, the estimated carrier density difference between the SC and OC in the same region (Equation 2) represents the accumulated carrier density, acting as a stagnant charge in the device.²¹ [Image Omitted. See PDF]where the ${\phi }_{\mathrm{OC}}$ and ${\phi }_{\mathrm{SC}}$ are potential under OC and SC, ${\rho }_{\mathrm{OC}}$ and ${\rho }_{\mathrm{SC}}$ are charge density under OC and SC, respectively. The calculated charge densities of the PA and OA devices are demonstrated in Figure 4A,B. The PA device shows detectable accumulated charges at both interfaces. The accumulated charges at the interface are known to be detrimental to both the performance and stability of the PSCs.²² Despite the accumulated charge, the PA device shows improved performance relative to IA, probably owing to more dominant improvement due to doping of spiro-OMeTAD and improved work function alignment at the spiro-OMeTAD/Ag interface. However, with OA, the more enhanced charge accumulation would limit device performance, particularly the FF, as observed in the previous study.²² Based on the measured work function profile depending on the aging time (Figures 4C and S6), Figure 4D,E describe the band structure of PA and OA devices. The straight work function line of the device under IA shows an approximately flat junction region where no band bending occurs (Figure 4C). As the device ages from PA to OA, a deep potential well develops at the hetero-interfaces and traps charge carriers, which limits carrier collection at the interface. As shown in Figure 4C, the initial work function of the perovskite layer has a reasonable intrinsic value (~4.85 eV) and is maintained until PA.²³ Under OA conditions, the work function of perovskite decreases, indicating n-type doping of the perovskite layer. At the same time, the work function of spiro-OMeTAD becomes even more upshifted, resulting in ~180 mV vacuum level shift (∆E_vac) at the interface between spiro-OMeTAD and perovskite. The work function of SnO₂ was also slightly shifted to a more n-type value compared to that in the IA condition as previously reported (please also refer to Figure S7 where NCM phase is plotted with the work function profile to distinguish the thin SnO₂ layer).¹² Owing to more pronounced n-type doping of the perovskite layer, ~30 mV of ∆E_vac was developed at the perovskite and SnO₂ interface. The induced ∆E_vac results in potential wells and barriers to cause charge accumulation at the interfaces.

FIGURE 4. Accumulated charge density in (A) proper aging (PA) and (B) over aging (OA) devices measured under mild illumination. (C) The work function profile of the corresponding devices. Schematics showing band diagrams of (D) PA and (E) OA devices under light. The diagrams are not drawn to scale.

Carrier dynamics

To investigate the impact of the changes in band alignment, electrochemical impedance spectroscopy (EIS) measurements were conducted for the PA and OA devices, as shown in Figure 5A. We adopted the equivalent circuit shown in Figure S7, where the high-frequency signature (first semicircle) and low-frequency signature (second semicircle) originate from charge transport and carrier recombination resistance, respectively.²⁴ As the device was aged from PA to OA, the radius of the first semicircle increased, indicating that charge carrier extraction had become inefficient. Additionally, the geometric capacitance (C_g) and surface capacitance (C_s) were calculated from the measured EIS under constant white-light illumination generated by a light-emitting diode (Figure 5B,C). The C_g and C_s values were consistently higher for the OA device, implying the formation of a charge accumulation zone at the interface.²⁵

FIGURE 5. (A) Nyquist plots measured from proper aging (PA) and over aging (OA) devices in the depletion mode. (B) Geometrical capacitance (Cg) and (C) interfacial accumulation capacitance (CS) measured from the corresponding devices. (D) Time-resolved photoluminescence (PL) measured from the active area of the devices under the short circuit condition. (E) transient photovoltage (TPV) and (F) transient photocurrent (TPC) measurements data of the devices.

In Figure 5D, steady-state PL measurements under SC indicate relatively inefficient photogenerated carrier extraction in the OA device relative to the PA device.²⁶ The prolonged average PL lifetime from 8.30 ns for the PA device to 95.58 ns for the OA device correlated with the steady-state PL measurements (inset and Table S1). The carrier dynamics were further examined using transient photocurrent (TPC) and photovoltage (TPV) analyses, as shown in Figure 5E,F, respectively. The photocurrent decay in the OA device under SC was much slower than that in the PA device, in agreement with the PL measurement results. Photovoltage decay was faster in the OA device than in the PA device, suggesting that charge carrier recombination was enhanced in the OA device. In Figure S8, the space charge limited current (SCLC) measurement with an increased trap-filled limit voltage (V_TFL) from 1.148 V for the PA device to 1.239 V for the OA device implies that the charge trap densities increased in the OA device.^27–29

Ion redistribution and phase conversion

To investigate the origin of the change in energetics, ToF-SIMS measurements were performed (Figure 6A–F). As shown in Figure 6A, under IA conditions, most Li⁺ ions were located in the spiro-OMeTAD layer, whereas a small amount of Li⁺ was observed in the perovskite and SnO₂ layers and their interface. The migration of Li⁺ ions from spiro-OMeTAD to the perovskite and SnO₂ layers increased with continuous aging, as shown in Figure 6B,C. In the OA device, the number of migrating Li⁺ ions exceeded that observed in spiro-OMeTAD. Although most of the Li⁺ ions are located at the interfaces between the perovskite, spiro-OMeTAD, and SnO₂, some of the Li⁺ should remain at the interstitial site of the perovskite lattice. The interstitial doping of Li⁺ in the perovskite layer is probably the origin of the induced n-type doping of the perovskite layer.³⁰ Simultaneously, migration of I⁻ from perovskite to spiro-OMeTAD was observed. The intensity of I⁻ increased in spiro-OMeTAD and its interface with the perovskite and Ag layers (Figure 6D–F). The migration of I⁻ into spiro-OMeTAD might induce a redox reaction, thus de-doping the spiro-OMeTAD layer, as previously reported.^31,32

FIGURE 6. Time-of-flight secondary ion mass spectroscopy (ToF-SIMS) depth profile for positive (A–C) and negative (D–F) ions measured from the device with different aging times: (A, D) insufficient aging (IA), (B,E) proper aging (PA), and (C,F) over aging (OA).

In Figure 7A,B, the X-ray diffraction (XRD) patterns of the perovskite layer in the corresponding devices are presented. Compared with that of a fresh perovskite film, the (100) peak of FAPbI₃ gradually shifted toward a lower 2θ angle as the aging time increases, probably due to interstitial doping by Li⁺.⁵ Such a peak shift is particularly pronounced as the aging time increases from PA to OA, which is coincident with overlap of peaks for Li⁺ with Pb⁺ in ToF-SIMS data in Figure 6C, indicating that the interstitial incorporation of Li⁺ mainly occur at this stage. In addition, the δ-FAPbI₃ peak that first appeared in the PA device became more pronounced in the OA device (Figure 7B). It can be inferred that heavy interstitial Li⁺ doping in the perovskite layer induced lattice microstrain, destabilizing the α-FAPbI₃ and facilitating its conversion to δ-FAPbI₃.^5,33 Based on the experimental observations, we suggest the following mechanism to explain correlations between ion migration and device performance (Figure 7C): (I) The Li⁺ cations generated from LiTFSI migrate from the spiro-OMeTAD to the perovskite, SnO₂ layers, and their interfaces. Below a certain threshold, this migration probably helps to passivate defects in the perovskite and enhance the conductivity of the SnO₂ layer; (II) as the aging time increases, heavily concentrated Li⁺ cations in the perovskite layer induce n-type doping and lattice microstrain, which destabilize the α-FAPbI₃ which gets converted to δ-FAPbI₃. Such phase-back conversion is likely to occur at the grain boundaries and the top and bottom surfaces of the perovskite layer, forming heterointerfaces with spiro-OMeTAD and SnO₂ where the migrated Li⁺ cations are concentrated; (III) at the interface between α-FAPbI₃ and the generated δ-FAPbI₃, charged iodine vacancies can be easily generated owing to lower defect formation energy, and they can move through the face-sharing δ-phase^34,35; and (IV) the migrated iodine can undergo redox reaction with oxidized spiro-OMeTAD species to de-dope spiro-OMeTAD layer as observed by the upward shift of the spiro-OMeTAD work function in the OA device.³⁶

FIGURE 7. (A,B) X-ray diffraction (XRD) spectra of perovskite solar cells (PSCs) with different aging times. (C) Schematic showing migration of Li+ and consequent phase conversion and iodine defect generation and migration.

CONCLUSION

In this study, we analyzed the origin of the aging-time-dependent performance of FAPbI₃-based PSCs. Cross-sectional KPFM revealed a change in the work function of perovskite and spiro-OMeTAD toward more n-type one when the device was over-aged. The upshift of the work function induced interfacial potential wells and charge extraction energy barriers, which degraded the photogenerated charge extraction and thus, the PCE of the devices. ToF-SIMS measurements confirmed the migration of Li⁺ ions from spiro-OMeTAD to the perovskite and SnO₂ layers and their interfaces. Heavy interstitial Li⁺ doping induced lattice microstrain that destabilized the α-FAPbI₃ and induced phase conversion to δ-FAPbI₃. The iodine defects generated at the interface between α-FAPbI₃ and the generated δ-FAPbI₃ migrated toward spiro-OMeTAD, where they reacted with the oxidized spiro-OMeTAD species and de-doped the spiro-OMeTAD layer. A series of these processes was the origin of the rapid performance drop of the FAPbI₃-based PSCs, even during storage under dark conditions. This work identified the hidden side effects of Li⁺ ion migration in FAPbI₃-based PSCs that can guide further work to maximize the operational stability of PSCs.

Experimental section Precursor preparation

Unless otherwise indicated, all chemicals were purchased from Sigma–Aldrich. The SnO₂ colloidal solution (Alfa Aesar) was diluted with deionized water at a volume ratio of 1:5. The perovskite precursor was prepared by dissolving 172 mg of formamidinium iodide (Greatcell solar) and 461 mg of lead (II) iodide (Tokyo Chemical Industry) in 500 μL of N,N′-dimethylformamide and 96.4 μL of N-methyl-2-pyrrolidone containing 22.8 mg of MACl (Greatcell solar). The spiro-OMeTAD solution was prepared by mixing 1 mL of 86.8 mg ml⁻¹ spiro-OMeTAD (Luminescence technology corp.) in Chlorobenzene with 34 μL of t-BP and 20 μL of LiTFSI solution (540 mg ml⁻¹ in acetonitrile).

Device fabrication

Fluorine-doped tin oxide glass (FTO) was sonicated successively in acetone and ethanol for 30 min and then dried under N₂ flow. The cleaned substrates were treated with ultraviolet ozone for 30 min. A total of 0.2 mL of diluted SnO₂ colloidal solution was spread onto the cleaned FTO glass and then spin-coated at 4000 rpm, followed by annealing at 150°C for 30 min. The annealed SnO₂ film was subjected to UVO treatment before being coated with the perovskite films. The perovskite precursor solution was spin-coated at 4000 rpm for 20 s, where 300 μL of diethyl ether was dropped after 15 s, followed by annealing at 150°C for 10 min. A spiro-OMeTAD solution was dropped on top of the spinning substrate with the perovskite layer after 5 s and spun for 20 s. Finally, an 80 nm thick silver was thermally evaporated at a deposition rate of 0.5 Å/s. The active area was defined as 0.125 cm² using a metal mask.

Characterization

Current density-voltage (J–V) curves were obtained using a Keithley 2400 source meter under AM 1.5G one-sun (100 mW cm⁻²) illumination generated by a solar simulator (Oriel sol3A, class AAA) equipped with a 450 W Xenon lamp (Newport 6280NS). The light intensity was adjusted using an NREL-calibrated Si solar cell with a KG-5 filter. The J–V curves were measured at a scan rate of 130 mV s⁻¹. The SCLC measurements were performed under the same conditions in the dark. External quantum efficiency (EQE) data were acquired using a QEX-7 series system (PV Measurements Inc.) with a monochromatic beam generated from a 75 W xenon source lamp (USHIO, Japan). X-ray diffraction (XRD) patterns were measured by a Rigaku smart Lab diffractometer with Cu K_α radiation (λ = 1.5406 Å). The scan rate was set to 4°/min, with a step size of 0.02°. To scan the positive ion profile, ToF-SIMS measurements were performed in the spectral mode using a Bi³⁺ ion beam with an ion current of 1.0 pA and a 1 keV O⁻ sputter beam with a current of 30.0 nA (etching area:150 × 150 μm²). In the case of negative ions, a Cs⁻ sputtering beam with a current of 20.0 nA was used. The steady-state PL was measured using a fluorescence spectrometer (Quantaurus Tau C11367-12, Hamamatsu). Perovskite films were photoexcited with a 464 nm laser pulsed at a frequency of 10 MHz, and the emitted PL was collected by a photomultiplier tube (PMT) detector (PMA 182, Pico QuantGmbH) at 816 nm. For the time-resolved PL measurements, a repetition frequency of 2 MHz was used. The EIS measurement data was obtained using an electrochemical working station (Autolab) at a voltage range from 0.0 to 1.0 V, either under illumination, or 0.9 V bias voltage under dark. The frequency ranged from 1 MHz to 0.1 Hz.

Kelvin probe force microscopy

Kelvin probe force microscopy (KPFM) measurements were conducted using an AFM NX-10 (Park Systems) The devices were mechanically cleaved to expose their cross-sectional surfaces without ion beam milling to prevent physical damage during sample preparation. An area with a root mean square (RMS) under 200 nm was selected to minimize the roughness-induced measurement error. For accurate work function measurement, the biased tip was calibrated (scan rate = 0.20 Hz) before and after every measurement using highly ordered pyrolytic graphite, which has a well-known work function of 4.6 eV. The humidity was maintained below 10% relative humidity during the measurements. The NSC36/chromium gold tip was adopted to maximize the KPFM image resolution by considering mechanical coherence with high electrical sensitivity. A handmade device holder was used to mount the prepared samples where sample stage was used as a ground. Electric field, E(z), distribution across the complete device cross section was calculated by differentiating the contact potential difference, V(z), following electrostatic field equation.³⁷ $-\nabla \mathrm{CPD}\left(z\right)=-\nabla \left[V\left(z\right)-\frac{1}{e}{\phi }_{\mathrm{tip}}\right]=E\left(z\right)$ where ${\phi }_{\mathrm{tip}}$ is the work function of tip and $e$ is elemental charge.

AUTHOR CONTRIBUTION

J.-W.L. conceived the idea. S.-G.C. designed and conducted the experiments supervised by J.-W.L. J.-W.L. and S.-G.C. analyzed the data and prepared the manuscript.

ACKNOWLEDGMENTS

This study was supported by a National Research Foundation of Korea (NRF) grant funded by the Korean government (Ministry of Science and ICT) under contract numbers 2022R1C1C1011975 and 2022M3J1A1064315. This research was also supported by the Challengeable Future Defense Technology Research and Development Program through Agency for Defense Development (ADD) funded by the Defense Acquisition Program Administration (DAPA) in 2022 (No. UI220006TD).

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.