Verification and mitigation of ion migration in perovskite solar cells

Multi-functional metal halide perovskite materials have shown outstanding potential for next generation optoelectronic applications including solar cells,¹ light emitting diodes,² resistive memory devices,³ photodetectors,⁴ and lasing devices.⁵ The performance of the state-of-the-art devices is already rivaling or out-performing conventional devices. For example, the record power conversion efficiency (PCE) of perovskite solar cells has reached 23%,⁶ while the external quantum efficiency of perovskite light emitting diodes has exceeded 20%.⁷ Their high performance in combination with low production cost puts the perovskite optoelectronics under serious consideration for possible commercialization by start-up companies.

One of the major concerns is whether or not these materials can sustain their crystal structure, stoichiometry, and thus optoelectronic properties during harsh and prolonged operational conditions. Their high chemical reactivity with oxygen and moisture as well as photo- and heat-induced decomposition has been primary concerns for the stability of the perovskite materials and devices. This degradation caused by extrinsic (or environmental) factors has been significantly relieved by engineering of perovskite materials and application of encapsulation techniques.^8–10 Still, there is a remaining concern that stems from an unprecedented and unique feature of perovskite materials, which is migration of ionic species or charged defects.^11,12 The perovskite materials fabricated by low temperature solution processes inevitably contain high density of defects (10¹⁶–10¹⁸ cm⁻³).¹³ Furthermore, relatively fragile chemical bonds in the lattice lead to low formation energies of the defects, resulting in an active defect formation process with mild external stimuli.¹⁴ Due to high iconicity of precursors and perovskite materials, these defects are ionic in nature. Although characteristic high defect tolerance of perovskite materials allow these defects to be benign or relatively less detrimental to optoelectronic properties and thus performance of fresh devices,^15,16 recent studies have unraveled that migration of the ionic species has critical impacts on long-term operational stability of the materials and devices.^17–20 The migration and accumulation of this ionic and thus reactive species have been found to rupture the crystal structure of perovskite materials and adjacent functional layers, resulting in serious degradation of device performance.

Therefore, in-depth understanding and proper manipulation of this unique physical process is of paramount importance to secure the long-term stability of perovskite optoelectronics. Motivated by this, herein we have reviewed important studies on discovery, quantification, and mitigation of the ion migration process in metal halide perovskite materials. A possible emerging application using the ion migration is also briefly introduced.

In an earlier stage of development, huge research efforts have been devoted to elucidate peculiar properties of perovskite materials and devices. Characteristic current-voltage (I-V) hysteresis was one of the anomalous phenomena observed in perovskite solar cells. Change in a photocurrent profile depending on the scan rate and the direction of applied bias voltage has been observed by several groups.^21–23 The I-V hysteresis has also been observed from conventional solar cells. For example, electrolyte based dye-sensitized solar cells show slow response time (∼200 ms) due to electron trapping in nanocrystalline TiO₂ and/or limited ion transport in the electrolyte.²⁴ Typically, the I-V hysteresis can be relieved by proper pre-conditioning such as light soaking or slower delay time for the measurement. However, in case of perovskite solar cells, a much longer time scale of the transient state was observed (from few hundreds of millisecond to tenths of seconds), which has not been precedent for all-solid-state devices.²² Therefore, it was recognized to be owing to a new intrinsic property of perovskite materials. The existence of the I-V hysteresis has posed difficulty in accurate characterization of device performance because the I-V curves determined from the transient state of the devices are subject to under- or over-estimate the device performance. As a result, measurement of steady-state PCEs was proposed, which is now considered as a standard protocol to evaluate the reliable performance of perovskite solar cells.²¹ 

The characteristic slow dynamic process was found to be highly dependent on the fabrication process and thus the resulting composition and morphology of the perovskite layer, which implies a close correlation between the slow dynamic process and physical processes in the perovskite layer. For example, the I-V hysteresis was found to be significantly varying depending on the grain size of the perovskite films (Fig. 1). The hysteresis index (an index devised to quantify the I-V hysteresis) was increased from 0.059 to 0.362 with smaller grain size of the perovskite film, which was correlated with enhanced capacitance at low frequency (0.1–1 Hz).²³ It was also observed that the hysteresis index is changed with use of a different fabrication process (single step or sequential step) and composition (with different “A” cations such as MA⁺ or FA⁺).²⁵ Because of the high dielectric constant and ferroelectric properties of the perovskite materials, a dielectric relaxation process owing to dynamic “A” cation was suggested to be the origin of the observed I-V hysteresis.^25,26 However, the time scale of observed polarization processes was found to be significantly longer than the time scale of the dielectric polarization induced by molecular orientation; the residence time of ∼14 ps was measured for MA⁺ in MAPbI₃ by using neutron scattering and time scale for a domain wall was suggested to be ∼0.1–1 ms,²⁷ and thus cannot fully explain the I-V hysteresis. The molecular orientation dynamics with ultrafast time scale was later found to have a close relationship with charge carrier dynamics at much shorter time scale.¹⁶ 

With consideration of ionic nature together with low temperature solution based fabrication processes, researchers presume migration of ionic species in the perovskite materials as a feasible origin of the observed I-V hysteresis. The solution processed polycrystalline perovskite films were expected to have high equilibrium defect density. For example, the Schottky defect formation energy for MAPbI₃ was calculated to be as low as 0.14 eV, 0.08 eV, and 0.22 eV per a defect for MAPbI₃, MAI, and PbI₂, respectively, resulting in estimated room temperature vacancy concentration of 0.4% (equivalent to ∼1.6 × 10¹⁹ cm⁻³).¹⁴ Many possible point defects with low formation energy (from about 0.2 to 1 eV) were also identified by Yin et al.²⁸ From photo-physical studies, the polycrystalline perovskite film was indeed characterized to possess the defects with density as high as 10¹⁶–10¹⁸ cm⁻³.¹³ Considering high iconicity of halide perovskite materials, this implies the presence of charged ionic species with high density originated from the point defects. This was correlated with measured low free carrier density under dark (10⁹–10¹⁴ cm⁻³) in the materials regardless of the low defect formation energy and thus high defect density, which can be attributed to self-regulation process accompanied by charge compensation.¹⁴ In fact, the defect density is similar with that of NaCl and KCl, which are considered as a type I ionic conductor.²⁹ 

On the basis of high defect density, Eames et al. calculated the activation energy for vacancy mediated ion migration in MAPbI₃ by using first principle calculation.¹¹ The activation energy is the lowest free energy barrier that an ion should overcome to hop from its original position to neighboring vacancy sites [Figs. 2(a) and 2(b)]. The activation energy for MA⁺, Pb²⁺, and I⁻ were calculated to be 0.58 eV, 2.31 eV and 0.84 eV, respectively. A diffusion coefficient for I⁻ was estimated to be around 10⁻¹² cm² s⁻¹, while that of MA⁺ was calculated to be 10⁻¹⁶ cm² s⁻¹. The lowest activation energy for I⁻ indicates the ion migration at room temperature with short time scale is predominantly originated from the migration of I⁻ vacancies. To support the argument, they correlated the calculated activation energy with the activation energy of photocurrent relaxation, which was measured to be around 0.60 eV. They proposed that the predominant I⁻ vacancies with a positive charge migrate to an interface with a contact under built-in electric field, which in turn screens the built-in electric field to localize the band bending [Figs. 2(c) and 2(d)]. As a result, the charge collection property of the device is affected and thus the photocurrent density changed with respect to the redistribution process.

It has been attempted to identify the migration of ionic species and study its effects on electronic properties of perovskite materials. Xiao et al. observed a switchable photovoltaic effect from many kinds of perovskite materials including MAPbI_3-xCl_x, MAPbBr₃, and FAPbI₃ with symmetric device structure (metal/perovskite layer/metal) comprising different electrode materials. The polarity of the device was tunable regardless of the perovskite composition and metal electrode by applying an electric field of about 1 V/μm, which indicates an intrinsic poling behavior of the perovskite materials under the applied electric field [Fig. 3(a)]. They speculated that the ionic species possibly migrate under the applied electric field to induce the self-doping effect at the interfaces with the metal electrodes, creating a p-i-n homo-junction [Fig. 3(b)]. Their assumption was correlated with the predicted shallow energy states of defects by first-principle calculations, implying the defects can have the self-doping effect.²⁸ The change in a work function upon poling was confirmed using Kelvin probe force microscopy (KPFM). They also presented change in morphology and optical response at the interfaces with the metal electrodes with prolonged poling time, which was attributed to change in stoichiometry of the perovskite layer.

More direct evidences of mixed ionic and electronic conduction in perovskite materials were presented by Yang et al.³⁰ Using Galvanostatic direct current (DC) measurements, they distinguished the ionic conductivity from the electronic conductivity of the perovskite materials where the ionic conductivity of 7.7 × 10⁻⁹ S cm⁻¹ and electronic conductivity of 1.9 × 10⁻⁹ S cm⁻¹ was separately measured under the dark condition. Remarkably, the ionic conductivity is higher than the electronic conductivity under the dark condition whereas the electronic conductivity becomes higher under illumination owing to photo-generated charge carriers. They also performed a series of experiments to identify the moving ions in MAPbI₃ perovskite materials. They confirmed that the iodide spices in MAPbI₃ perovskite materials migrate to react with an adjacent Pb electrode to form a secondary PbI₂ phase at the interface. Yuan et al. performed similar poling experiment using MAPbI₃ in an attempt to find out the origin of the switchable photovoltaic effect in MAPbI₃ films. They measured a spatial resolved photothermal induced resonance (PTIR) microscopy in combination with KPFM measurement. They found a signature of MA⁺ migration in the MAPbI₃ films using the PTIR measurement, which was correlated with work function change measured from the KPFM.³¹ 

The signature of halide migration was also observed during optical response measurements. Varying halide composition has been adopted to control the bandgap and thus optical response of perovskite films [absorption and photoluminescence (PL)].³² It was found that steady-state photoluminescence (PL) spectra of the perovskite films containing the mixed halides changed upon exposure to light illumination.^33,34 For example, a characteristic PL peak of MAPb(Br_xI_1-x)₃ film at around 1.85 eV decreases upon exposure to light with simultaneous appearance of a peak at around 1.65 eV, which was correlated with enhanced sub-bandgap absorption at around 1.7 eV.³⁴ This is an indicative of halide redistribution in the film to induce the segregation of iodide-rich low bandgap phase, which was also confirmed from an X-ray diffraction pattern measurement. Notably, a reversible change in optical response was observed once the light is eliminated.

With identification of the ion migration, there have been noticeable efforts to study the energetics and kinetics of the ion migration. In the subsequent paragraphs, we briefly remind the origin of the ion migration in terms of thermodynamics and review the studies on the energetics and kinetics of the ion migration in the perovskite materials.

An ideal single crystal with no extrinsic defects consists of static atoms at absolute zero temperature of which free energy is primarily dominated by potential energy of the atoms. As the temperature increases, contribution of kinetic energy originating from the atomic vibration becomes higher to increase the free energy of the system whereas the system has tendency to minimize the free energy in turn by increasing the entropy of the system. The crystal defects originated from this tendency as temperature increases because the entropy is indicative of the degree of disorder. Migration of ion in the ionic materials is generally mediated through open space, allowing an ion to jump from one site to another site. This open space can arise from either the defects or vibrational displacement of the ions. Therefore, there is a general tendency of increasing ionic conductivity with higher temperature for ionic materials. Typically, two types of point defects are considered to be important for the migration of ions, which are the Schottky and Frenkel defects. The Schottky defect generates two vacancy sites with opposite charges, whereas the Frenkel defect consists of a pair of a vacancy site and interstitial ion originated from the vacancy. The neighboring ions can transport through the nearest vacancy (vacancy migration mechanism), or the interstitial ion can directly hop into another interstitial site (interstitial mechanism). There is also an “interstitialcy mechanism,” which involves more than two ions.²⁹ 

In most of the experimental situation, the synthesized materials are not an ideal single crystal and contain defects arising from fabrication processes or stoichiometric error, which are considered as “extrinsic defects” whereas the thermally activated defects are termed “intrinsic defects” as it is dependent on materials’ intrinsic properties. From the above discussion, it is intuitive that the ionic conductivity is dependent on the concentration of the defects, temperature of the system, and an energy barrier for the ions to migrate from one site to another site. This free energy barrier that we mentioned above is defined as an activation energy for the migration of ions. The relationship between the activation energy and conductivity can be described by a following Arrhenius type equation given by

where σ is the conductivity at given absolute temperature T, k_B is the Boltzmann’s constant, E_a is the activation energy, and σ₀ is the pre-exponential factor. Due to the presence of extrinsic defects in the materials, there exist two conduction mechanisms depending on the temperature of the system. At lower temperature, not many intrinsic defects are activated, and thus the conduction is dominated by the extrinsic defects. Since the conduction occurs through the extrinsic defects already exist in the materials, the activation energy for the migration is relatively lower. As the temperature of the system increases, the intrinsic defects start to be activated, resulting in transition to the “intrinsic region.” The activation energy for conduction through intrinsic defects is relatively higher because it involves the energies required for activating the defects as well as for migration of ions.

According to the above Eq. (1), the activation energy for ion migration can be derived from a temperature-dependent conductivity measurement of the materials. The activation energy can be calculated from the slope of ln(σT) versus $1T$ plot.^12,29,35 Xing et al. constructed the lateral devices (Au/perovskite/Au) to measure the temperature dependent conductivity of the MAPbI₃ films and crystals [Fig. 4(a)].³⁵ Conductivity of the lateral devices was measured under a constant electric field of 0.2 V/μm. They found that the high temperature activation energy of the perovskite materials is highly dependent on the grain size of the films. The activation energy of the films with small grain (∼300 nm) was measured to be 0.27 eV, whereas that of large grain film (∼1 µm) was measured to be 0.50 eV [Figs. 4(b) and 4(c)]. They concluded that the lower activation energy for the small grain film is probably due to higher density of grain boundaries providing better chances to form ion migration channels. Their argument was supported by much higher activation energy of 1.05 eV measured from the single crystal [Fig. 4(d)]. In fact, the more pronounced local I-V hysteresis at grain boundaries measured by Shao et al. confirms facilitated ion migration through the grain boundaries.³⁶ It was also presented that the activation energies for all the films and crystals significantly lowered upon illumination of light (approximately 2–3 times smaller under 0.25 sun). Similar observation was reported by Zhao et al., where the activation energy for ion migration was decreased by a factor of five (from 0.82 to 0.15 eV) under 0.2 sun illumination.³⁵ Based on the observation of the channel formation [Fig. 4(e)], ion mobility was estimated to be around 10⁻⁹ cm² V⁻¹ s⁻¹ under 0.25 sun. The estimated ion mobility indicates that the part of mobile ion drift with velocity higher than 1 µm s⁻¹ in consideration of typical built-in electric field of ∼2 V μm⁻¹ under one sun illumination. Therefore, the redistribution of ions (with a lower activation energy) in the devices probably occurs within a second upon exposure to light.^12,35

Moisture and oxygen induced degradation has been a primary concern for shelf lifetime, but these factors are probably not very critical once the device is encapsulated under nitrogen atmosphere. On the other hand, the ion migration becomes more severe under the operational condition where light, heat, and built-in electric field provide sufficient energies for activation of the mobile ions and thus will be a primary concern for the operational stability. As discussed before, the lower activation energies of ion migration in perovskite films under illumination imply accelerated defect formation and/or migration under operational condition of devices. Kim et al. recently unraveled that photo-generated charge carrier can also provide excessive chemical potential to generate high density of defects, contributing the ionic conductivity and subsequent photodecomposition.³⁷ The migrated ions will be accumulated at the interfaces with functional contact layers and/or penetrates into their bulk, which will be detrimental to long term operational stability of the devices. In fact, Monojit et al. showed that the degradation of the MAPbI₃ devices in nitrogen atmosphere and under illumination is mainly caused by thermally activated fast ion transport.¹⁸ Migrated ions and their compound have been detected from the degraded devices.^19,20,38,39

The reported J-V hysteresis during the current voltage sweep is originated from the ion migration occurs at a relatively short time scale (0.1 s–100 s). Prolonged operation of the devices under 1 sun illumination at elevated temperature (∼60 °C) has been found to induce long-term migration of ionic species with higher activation energies. Domanski et al. found the presence of fast exponential decay of device performance during the initial operation for about 10 h, which was followed by much slower decay with an almost linear profile [Fig. 5(a)].¹⁷ They found that this initial exponential decay is almost fully reversible when the device is stored under dark conditions.¹⁷ This relatively long-term transient behavior is ascribed to migration of cation vacancies having relatively higher activation energies than anion vacancies [Fig. 5(b)]. While the halide vacancies with lower activation energies migrates and accumulates at the interface with contacts within several minutes, cation vacancies with much higher activation energies migrates much slower and reaches equilibrium states after several hours. The accumulated charged defects forms narrow Debye layers which in turn screening the applied voltage and inhibit the charge extraction to contacts.⁴⁰ This reversible process has been assigned to be an origin of the initial fast exponential decay. After the fast initial decay, the devices shows relatively slower decay of the performance with an almost linear profile, and the slower degradation was attributed to degradation caused by multicomponent. Since the degradation is found to be not reversible, it probably involves irreversible chemical reaction of the accumulated ions with a perovskite active layer and/or adjacent contact materials. Carrillo et al. identified the redox reaction between the spiro-MeOTAD⁺ and I⁻, which was found to be irreversible, and consequently degrading the electrical property of the spiro-MeOTAD layer.^20,41 Also, the migrated I⁻ can penetrate into the contact layers to react with metal electrodes such as silver and aluminum.^19,39

As discussed above, mitigation of ion migration is essential for securing long-term stability of perovskite solar cells. Significant research efforts have been devoted to minimize the ion migration. Here, we review several approaches to reduce the ion migration and thus enhance the operational stability of perovskite solar cells.

A primary attempt for reducing ion migration was to grow perovskite films with high crystallinity and thus low defect density. As discussed previously, the grain boundaries provide pathways for ion migration with the lower activation energy and thus accelerate the degradation of the device performance. In fact, the grain boundary defects have been considered to be relatively less detrimental or even benign for the performance of perovskite solar cells.⁴² However, the accelerated ion migration through the grain boundaries would be critical to operational stability. Therefore, minimizing grain boundaries by controlling crystal growth is of importance to mitigate the ion migration.

There have been tons of studies for development of methods to control the crystal growth of perovskite films. For example, control of supersaturation level of the perovskite precursor solution facilitates the modulation of nucleation rate, where the slower nucleation facilitates the growth of large perovskite grains with higher crystallinity.⁴³ Utilization of solvent annealing facilitates the diffusion of precursors and coarsening of small grains to form larger grains.⁴⁴ Using MAPbI₃ films with different grain size grown by the solvent annealing, Shao et al. studied effects of the grain size on ion migration properties and device performance. The perovskite solar cells made with large grains showed steady-state PCE of 19.0% with no memristive behavior, while the devices made with small grains showed lower steady-state PCE of 16.7% and an obvious memristive behavior (dark current density continuously increased with a scan rate of 0.1 V/s from 0 to 2 V sweeps); the memristic behavior will be discussed in subsequently.³⁶ Furthermore, a lot of additives have been developed to facilitate the growth of larger grains.⁴⁵ 

Passivation of grain boundary defects and/or open space can be another route to mitigate the ion migration.¹² Wei et al. introduced a rubrene as an additive to induce strong supramolecular cation–π interaction with organic cations in perovskite films. They speculate that the added rubrene is formed at the grain boundaries and/or surface accompanied by π interaction with organic cations in perovskite grains of which interaction energy is as high as 1.5 eV. They used a time-of-flight secondary ion mass spectroscopy (TOF-SIMS) measurement to check the effect of rubrene on cation migration in the perovskite layer. After 72 h of 1 V DC biasing under the continuous illumination (90 °C, ambient air), much less inter-diffusion between the PbI₂ and perovskite layer was observed with addition of the rubrene.⁴⁶ Zhao et al. introduced [6,6]- phenyl-C61-butyric acid methyl ester (PCBM) into the perovskite film in n−i−p structured devices by immersing the as-prepared perovskite films in a PCBM chlorobenzene solution.⁴⁷ They suggested that PCBM can also strongly bond with perovskites by forming a donor−acceptor complex, leading to inhibited ionic motion as confirmed by liquid-state H NMR (nuclear magnetic resonance) measurement. The effect of PCBM at the grain boundaries on the ion migration was characterize by using a setup of in situ PL under an applied electric field and standard Galvanostatic measurement. They confirmed that the ion migration is suppressed by the added PCBM. Very recently, Li et al. demonstrated an in situ cross-linking strategy for operationally stable inverted MAPbI₃ perovskite solar cells.⁴⁸ They incorporated a cross-linkable organic small molecular trimethylolpropane triacrylate (TMTA) into perovskite films. The added TMTA was formed at the grain boundaries where carbonyl groups in the TMTA form weak interaction with PbI₂. The in situ cross-linking of the grain boundary TMTA was induced by annealing at elevated temperature (140 °C). The cross-linking significantly increased the activation energy for ion migration from 0.21 eV to 0.48 eV, and consequently led to significantly enhanced thermal and operational stability of the device. While bare MAPbI₃ films exhibited 6-fold higher ionic conductivity of 0.909 × 10⁻⁹ S cm⁻¹ than the electronic conductivity (0.159 × 10⁻⁹ S cm⁻¹), the cross-linked MAPbI₃-TMTA films showed a much lower ionic conductivity of 0.608 × 10⁻⁹ S cm⁻¹ after the cross-linking. As analogous to the small molecular additives, the polymeric additives were also developed. The polymers having coordinating functional groups at repeating units were found to be effective to form long-range ordered interactions with perovskite crystals. For example, Zuo et al. demonstrated PCE exceeding 20% with elongated shelf and operational lifetime by incorporating poly(4-vinylpyridine) (PVP) into the perovskite film.⁴⁹ They confirmed the molecular interaction between the polymer additive and perovskite crystals using density functional theory (DFT) calculation and Fourier transform infrared (FT-IR) spectrum measurement, where the absorption energy was calculated to be as high as 37.70 kJ/mol.

The activation energy for ion migration is highly dependent on chemical bonding and crystal structure of the materials. Therefore, structural engineering of perovskite crystals can be used to modulate the strength and geometry of the chemical bonds and thus enhance the activation energies for ion migration. Zhou et al. demonstrated that the light-induced enhancement of ion migration in MAPbI₃ can be suppressed by substitution of MA cation with Cs cation.⁵⁰ They showed that the activation energy of MAPbI₃ is reduced from 0.45 eV to 0.07 eV upon illumination, whereas that of CsPbI₂Br are almost maintained to be 0.45 eV regardless of the illumination. Senocrate et al. observed that a diminished ionic conductivity and enhanced photo-stability of FA mixed MAPbI₃ comparing with pure MAPbI₃ by measuring ionic conductivity using DC-polarization of lateral devices. The better photo-stability of FA-based films and devices over MA-based devices have been reported by other groups.^9,51 The measured ionic conductivity of MA_0.8FA_0.2PbI₃ was 9 × 10⁻⁹ S/cm, which is smaller than that of MAPbI₃ (3 × 10⁻⁸ S/cm).⁵² The suppression of ion migration with incorporation of FA cation was attributed to fewer tendencies for deprotonation of FA than MA.^9,53 Zhang et al. observed significantly reduced hysteresis with increasing Br concentration in MAPbI_3-xBr_x films owing to local structural distortion. From density functional (DFT) calculation, they found that the local structural distortion enhances the activation energies for migration of halides and MA⁺, resulting in reduced hysteresis and higher performance.⁵⁴ 

Small amounts of alkali metal ions (Na⁺, K⁺, Rb⁺) were found to be able to act as a suppressor of ion migration in perovskite materials. Son et al. developed a universal approach to eliminate the hysteresis from the perovskite solar cells by introducing tiny amounts of alkali metal ions (∼10 µmol) into the perovskite layers.⁵⁵ They found that the sodium, potassium, and rubidium cations similarly act as the suppressor to reduce I-V hysteresis [Figs. 6(a)–6(c)]. Especially, incorporation of potassium ions effectively suppressed the I-V hysteresis irrespective of the composition of perovskite layers. By investigating the defect density and lattice parameter, they argued that the added potassium ions occupy the O_h interstitial site to suppress the formation of halide Frenkel defects and consequently reduce the defect density and the migration of the halide ions [Fig. 6(d)]. Abdi-Jalebi et al. also utilized the potassium doping to suppress the halide migration.⁵⁶ They observed the high and stable photoluminescence quantum efficiency (PLQE) values under continuous illumination from the potassium-added perovskite films, whereas that of bare films showed a slow increase with time owing to photo-induced halide migration. They insisted that the added potassium ion immobilize the surplus halides by forming compounds at grain boundaries while excess iodide originated from the potassium iodide compensate for any halide vacancies.

Utilization of 2D perovskite has been recently emerged as a promising approach to suppress the ion migration. The 2D perovskite was found to have relatively high moisture and ion resistance compared to 3D counterpart owing to bulky and hydrophobic organic cations.^10,57 Chen et al. and Lee et al. introduced 2D phenylethylammonium lead iodide (PEA₂PbI₄) into the precursor solution for MAPbI₃ and FAPbI₃ perovskite films, respectively (Fig. 6).⁵⁸ It was found that the 2D perovskite is spontaneously formed at the grain boundaries once the amount of added 2D perovskite is below certain threshold (∼5 mol. %). The grain boundary 2D perovskite suppresses the ion migration in the perovskite films, which was correlated with reduced I-V hysteresis and enhanced activation energy for the ion migration. The energy barrier for the migration of ions was increased from 0.30 eV to 0.67 eV according to the DFT calculation [Fig. 7(b)]. As a result, the perovskite solar cells with 2D perovskite showed significantly improved chemical stability against a Ag electrode as well as operational stability. The FAPbI₃ solar cells with 1.67 mol. % PEA₂PbI₄ showed T80 lifetime (time at which PCE is degraded to 80% of initial PCE) of 1362 h, which is more than two times longer than that of the device based on bare FAPbI₃ (592 h).

The reversible ion migration can be used to construct a non-volatile resistive switching memory device.^3,59 The resistive memory devices employ altering resistance of active layers to store the information of which on and off (1 or 0) states are referred to high resistance states (HRS) and low resistance states (LRS), respectively. The on and off states can be altered by an applied electric field, which is termed SET (writing of data) and RESET (erasing of data) process, repectively.^60,61 Characteristic of data storage by altering resistance connects resistive switching memory devices to concept of memristor (memory + resistor) suggested by Leon Chua in 1971.^62,63 The multilevel storage characteristic of the memristor was found to be useful for neuromorphic computing by mimicking a synapse in neuromorphic systems.^63,64 Resistive switching devices are divided into several sorts depending on characteristics of a conduction mode in LRS and polarity of electric field for the switching.^60,61 According to the conduction mode in LRS, resistive switching is classified into interface and filamentary types. For the interface type, the switching occurs at the interface between metal and active layers, where ion vacancies or charge carriers are captured at the interface.^60,61,65 On the other hand, in the filamentary type, confined local paths called conducting filaments are formed and ruptured by electric field.^60,61,65 In the filament type, the current in LRS flows only through the localized conducting filaments contrary to the interface type. Conducting filament can be formed by metallic bridge or ion vacancies.⁶⁰ When conducting filament is composed of metallic bridge, the resistive switching memory belongs to an electrochemical metallization (ECM) memory or thermochemical metallization (TCM) memory.^60,66 If conducting filament constitute of ion vacancies, the resistive switching memory is classified as a valence change memory (VCM). In a VCM, applied electric field induces redistribution of anion vacancies (anion) or valence change of metal cation so that the conducting filaments are formed in the active layer.

The resistive switching memory devices require large on-off ratio, large endurance, fast switching speed, and low operating power.^67,68 Various materials have been utilized as the active materials in the devices such as metal oxides, organic materials, oxide perovskite, and chalcogenide for insulator layer.⁶⁹ As the ion migration properties of the metal halide perovskite materials have unraveled, the perovskite materials was also considered for active layer materials for resistive memory devices. The halide perovskite materials are suggested to have potential to overcome limitation of conventional materials such as low on/off ratio and high operating voltage.^3,70,71 Wang et al. first reported the fabrication of a resistive memory device using MAPbI_3-xCl_x, where the device structure of fluorine doped tin oxide (FTO)/MAPbI_3-xCl_x/Au was utilized.⁷⁰ With gradual increase in research efforts, the devices based on the halide perovskite materials have exhibited higher on/off ratio (>10³) and lower operating voltage (<1 V) than those of conventional materials.⁶⁸ The resistive switching effect of halide perovskite materials are generally attributed to redistribution of predominant halide vacancies (or halide) with relatively lower activation energies for migration, where the applied electric field induces redistribution of the vacancies (halide) for formation and rupture of conducting filaments.^70–72 This makes resistive switching memory based on halide perovskite materials being classified as VCM. Typically, the resistive switching mechanism of the filament type devices is dependent on a top electrode used.^60,61,66 For example, the use of active electrode materials such as Ag can induces the electrochemical metallization.⁶⁶ In case of halide perovskite materials, the devices usually showed characteristic VCM irrespective of active top electrodes.^72,73 It was found that the Ag electrode can react with halide perovskite materials to form silver halide compounds on top of the halide perovskite layer such that additional halide vacancies are formed in the perovskite active layer.⁷³ However, the top electrode induced ECM cannot fully excluded since the ECM mechanism was also reported to be involved in some of the devices with active top electrode [Fig. 2(c)].⁷³ 

As we discussed in above, the activation energies and resulting ion (vacancy) migration process is highly dependent on crystal structure of the materials, which implies the formation of conducting filament and thus resistive switching behaviors might be altered depending the crystal structure of the materials. Yang et al. reported dependence of crystal structure on resistive switching behavior of FAPbI₃.⁷⁴ The FAPbI₃ has two polymorphs with different crystal structures [Figs. 8(a) and 8(b)].⁵¹ The yellow 1D hexagonal structure (δ-phase) was found to be stable at room temperature, while it converted to black trigonal structure (α-phase) upon annealing at elevated temperature (∼150 °C). They found that reliable resistive switching occurs only in δ-FAPbI₃ whereas the α-FAPbI₃ do not show a RESET process [Figs. 8(c) and 8(d)]. The DFT calculation revealed that iodine clusters in iotropic α-FAPbI₃ are so stable after forming filaments that the filaments are hard to be ruptured at off state. The resistive memory device based on δ-FAPbI₃ demonstrates endurance of 1200 cycles with on-off ratio of 10⁵. Recently, the resistive switching effect was also reported in all inorganic Pb free bismuth halide perovskite materials, where resistive switching of Rb₃Bi₂I₉ and Rb₃Bi_1.8Na_0.2I_8.6 was compared.⁷⁵ Large on/off ratio (>10⁷) with low operating voltage (∼0.1 V) were reported. The low operating voltage was attributed to grain boundary-mediated ion drift in the materials. Despite the large on/off ratio and low operating voltage; however, the limited endurance of the devices based on perovskite materials should be improved for practical usage.

In this report, we reviewed important studies on understanding of ion migration in metal halide perovskite materials. Various approaches from theoretical to experimental characterization has enabled identification and quantification of the ion migration in the perovskite materials. The activation energies required for formation and/or transport of the ionic defects have been found to be as low as ∼0.3 eV. Consequently, polycrystalline perovskite films with high density of ionic defects are subject to have highly mobile ions contributing to high ionic conductivity of the films. The migration and accumulation of these ionic species have been found to be responsible for characteristic transient and slow dynamic processes in perovskite materials and devices such as a PL aging effect and I-V hysteresis. More importantly, the ion migration has critical impacts on long-term operational stability of the perovskite optoelectronic devices.

We believe that suppression of the ionic defect formation and passivation of possible migration pathways are keys to secure the long-term operational stability of the devices. Recent progress implies that manipulation of the grain boundaries and crystal structure can effectively modulate the defect formation and migration in the perovskite layer. More studies for in-depth understanding of defect chemistry will be helpful to devise possible solution to the ion migration. Finally, precise control over these defects will possibly open-up a new opportunity of utilizing perovskite materials in resistive memory devices toward a next-generation neuromorphic computing application.

S.-G.K., J.-M.Y., and N.-G.P. are grateful for the National Research Foundation of Korea (NRF) grants funded by the Ministry of Science, ICT Future Planning (MSIP) of Korea under Contract Nos. NRF-2012M3A6A7054861 and NRF-2014M3A6A7060583 (Global Frontier R&D Program on Center for Multiscale Energy System), Nos. NRF-2016M3D1A1027663 and NRF-2016M3D1A1027664 (Future Materials Discovery Program) and No. NRF-2015M1A2A2053004 (Climate Change Management Program). J.-W.L. and Y.Y. are grateful for the finical support from the Air Force Office of Scientific Research (AFOSR, Grant No. FA9550-15-1-0333), the Office of Naval Research (ONR, Grant No. N00014-17-1-2484), the National Science Foundation (NSF, Grant No. ECCS-EPMD-1509955), and Horizon PV.