Ion Migration and the Role of Preconditioning Cycles in the Stabilization of the J–V Characteristics of Inverted Hybrid Perovskite Solar Cells

M. De Bastiani, G. Dell’Erba, M. Gandini, V. D’Innocenzo, S. Neutzner, Dr. A. R. S. Kandada, Dr. G. Grancini, Dr. M. Binda, Dr. J. M. Ball, Dr. M. Caironi, Dr. A. Petrozza Center for Nano Science and Technology @PoliMi Istituto Italiano di Tecnologia Milano 20133 , Italy E-mail: mario.caironi@iit.it; annamaria.petrozza@iit.it M. De Bastiani Università degli Studi di Padova Padova 35131 , Italy G. Dell’Erba Dipartimento di Elettronica Informazione e Bioingegneria Politecnico di Milano Milan 20133 , Italy M. Gandini, V. D’Innocenzo, S. Neutzner Dipartimento di Fisica Politecnico di Milano Milan 20133 , Italy Dr. M. Prato Department of Nanochemistry Istituto Italiano di Tecnologia via Morego, 30 , 16163 Genova , Italy


DOI: 10.1002/aenm.201501453
groups have recently and independently demonstrated that, by applying an electric fi eld across a pristine fi lm of 3D hybrid perovskites of different chemical composition, a self-sustained fi eld is induced in the semiconductor as a consequence of ion migration toward the electrode regions. [10][11][12][13] The formation of a self-sustained internal fi eld upon device polarization is also in good agreement with the observations reported by Tan et al. when testing perovskite-based light emitting diodes. [ 14 ] This concept has also been the base of the explanation proposed by Tress et al. for the rate-dependent hysteresis seen in current-voltage scans of solar cells. [ 15 ] So far reports suggest that transient electrical characteristics are due to a polarization response of the perovskite active layer that results in changes in the photocurrent extraction effi ciency of the device. [ 11,15 ] However, it must be noted that a variety of dynamics have been reported, which differ in magnitude and time scale, depending both on the specifi c device architecture and, in particular, on the adopted charge extraction layer. [ 7,9 ] This indicates that contact interfaces have a considerable effect on transients in perovskite based devices.
In this Communication we investigate the role played by charge extracting layers on the slow transient behavior of CH 3 NH 3 PbI 3 perovskite based solar cells. Such transients, which typically affect both short-circuit currents and open circuit voltage of hysteretic devices, are found to notably modify the open circuit voltage also in the very fi rst J -V scans of so-called "hysteresis-free" devices integrating a phenyl-C61-butyric acid methyl ester (PCBM) charge extraction layer. Here a preconditioning of the device, i.e., a repetition of J -V scans, is needed to achieve completely stable J -V characteristics under illumination. In particular, we fi nd that under device operation, iodide ions migrate to the electron extracting layer. We fi rst show that the use of an organic extraction layer such as PCBM, albeit not hampering ions motion, evidently improves charge extraction with respect to interfaces involving compact TiO 2 , in agreement with what is suggested in other seminal investigations, [ 16,17 ] and makes the short-circuit current density virtually insensitive to the transient phenomena related to ions migration. Moreover, while self-doping of the perovskite fi lm close to the contact has been generally put forward in the study of transient phenomena, [ 12,13 ] here we show that ions can specifi cally interact with the organic electron extracting layer, inducing electronic doping and that such I − /PCBM interaction is at the origin of the preconditioning requirement for stabilizing the device and for improving its open circuit voltage with respect to the fi rst scan.
Solution-processable hybrid perovskite semiconductors have risen to the forefront of photovoltaics research, offering the potential to combine low-cost fabrication with high-powerconversion effi ciency. Originally used in dye-sensitized solar cell technology, the fi rst architectures saw the use of TiO 2 , both in the form of mesoporous and compact fi lms, as an electron extracting layer. [ 1 ] Further development has been driven by empirical optimization strategies and testing of a variety of different architectures. [2][3][4][5] One of these promising architectures uses fully organic buffer layers, which, fi rst of all, brings the advantage of low-temperature processing. [ 6 ] The intention of understanding whether some architectures of perovskite-based devices may outperform others in terms of power conversion effi ciency has been the initial driving force for most research groups. However, the observation of slow transient and hysteretic effects observed in perovskite-absorber devices, which severely affects current density-voltage ( J -V ) measurements and efficiency determination, has further stimulated deeper exploration of the working mechanisms of the devices. [7][8][9] Several wileyonlinelibrary.com In Figure 1 we report the J -V characteristics of three different device architectures as a function of the scanning rate and direction. Figure 1 a shows the characteristics of a nanostructured architecture where a compact layer of TiO 2 and 2,2′,7,7′-tetrakis-(N,N-di-p-methoxy phenylamine) 9,9′-spirobifl uorene (spiro-OMeTAD) are employed for the extraction of the electrons and holes respectively, and a mesoporous structure of Al 2 O 3 nanoparticles is employed as a scaffold for the perovskite. Figure 1 b shows a nanostructured architecture, again, where the meso-alumina is substituted with a scaffold of TiO 2 nanoparticles. Then, Figure 1 c shows the so-called inverted architecture where layers of PEDOT:PSS and PCBM for the collection of holes and electrons are adopted, respectively. The fi gures of merit of these devices and details about their testing procedures are reported in Table 1 . The architecture encompassing  the mesoporous alumina layer and a fl at TiO 2 as electron  extracting layer is affected by severe hysteresis features, with a strong reduction of the photocurrent when the scanning rate is slowed down. The phenomenon is attenuated by replacing the mesoporous alumina with a mesoporous TiO 2 extraction layer. This effect fi nally becomes negligible when electron extraction occurs at the interface with PCBM, in an inverted structure. Such a trend already highlights a role of the charge extraction interface in the response of the device to slow transient phenomena. [ 16,18 ] Adv. Energy Mater. 2016, 6, 1501453 www.MaterialsViews.com www.advenergymat.de We want now to emphasize that the curves of Figure 1 have been obtained after a specifi c conditioning. In particular, the curves of Figure 1 a,b are observed only by scanning the device starting from a prebiasing, i.e., a bias voltage applied prior to J -V scans, under light conditions at 1.4 V for 10 s, without which photocurrent is highly suppressed. On the other hand, even the hysteresis-free behavior of the inverted device of Figure 1 c is obtained only after scanning the J -V characteristic of the same device several times, as shown in Figure 2 , while a prebias at 1.4 V under light is highly detrimental. Though it has been recently proposed that PCBM deposited on top of the perovskite reduces the density of trap states, which are indicated as the origin of hysteresis, [ 16,19 ] during the fi rst J -V scans performed on a freshly made device, the latter exhibits clear hysteresis when scanning the voltage from forward-bias to short-circuit and back. By repeating the measurement this difference is reduced, ultimately becoming negligible and producing the hysteresis free scans reported in Figure 1 c. In particular, while the short-circuit current is almost unaffected upon several scans, the open-circuit voltage needs a few cycles to be stabilized: evidently, what is at the base of the transient phenomena does not necessarily affects both parameters at the same time. A similar phenomenon is also present in the dark J -V curves (see Figure S1, Supporting Information). This shows the clear need of some "preconditioning J-V cycles" to reach a stable performance, even in the presence of PCBM.
With the intent of clarifying the role of PCBM in such peculiar behavior, fi rst we investigate the reason behind the short-circuit current stability compared to the cells integrating a TiO 2 layer. Figure 3 a compares the photoluminescence (PL) quenching of CH 3 NH 3 PbI 3 forming a fl at junction with TiO 2 and PCBM. Comparing these dynamics with the PL decay of a CH 3 NH 3 PbI 3 polycrystalline fi lm deposited on a glass substrate, one can immediately notice that the PCBM induces a considerable quenching, while in the presence of TiO 2 quenching is nearly negligible (please note that crystal size effects cannot explain such large change in PL dynamics [ 20 ] ). As we can safely assume that that the perovskite surface will be improved upon PCBM deposition [ 19 ] rather than inducing interface trap states, we can conclude that PCBM provides superior charge extraction properties with respect to TiO 2 . To better contextualize these results with what is generally observed in working devices, in Figure 3 b we show the PL dynamics of CH 3 NH 3 PbI 3 embodied in a full device architecture (i.e., PEDOT:PSS/CH 3 NH 3 PbI 3 /PCBM) at short-circuit condition. In particular we compare the PL decay, at early times after photoexcitation (ps-ns), before and after a prebiasing the device at 1 V. The dynamics do not change dramatically. Therefore we conclude that electron transfer is not affected by polarizing the perovskite fi lm, consistently with the observed short-circuit current stability.
For comparison, in Figure S3 (Supporting Information) we also report the photobleaching (PB) dynamics, in the ps-ns time window, of the same hybrid perovskite, probed in devices  where spiro-OMeTAD was used as hole extracting material, while PCBM or TiO 2 are used as electron extracting layers [ 17 ] and we compare them with the PL dynamics taken from the PEDOT:PSS/CH 3 NH 3 PbI 3 /PCBM device. The photobleach originates from the transparency induced at the onset of the optical absorption of a semiconductor upon photoexcitation after population of the bottom of conduction band and top of the valence band, respectively, thus it will follow the electron and holes population dynamics in the semiconductor. [ 21,22 ] When the PCBM is used, the PB dynamics well follow the PL dynamics of the PEDOT:PSS/Perovskite/PCBM devices and they are not affected by prebiasing of the device. On the other hand, in the presence of TiO 2 , the charge extraction becomes faster, and comparable to those in the presence of PCBM, only upon prebiasing. This is indicating, fi rst of all, that the electron extracting interface is the most sensitive and critical one in state-of-the art perovskite solar cell architectures as electron extraction strongly depends on the nature of the interface and on the thin fi lm polarization condition, while no signifi cant change is observed by changing the hole extracting layer. Then, importantly, we have also the evidence that, differently from fl at TiO 2 , PCBM provides such good electronic contacts at the interface with perovskites that the extraction is not infl uenced by prebiasing of the device. This well explains the steady state photocurrent stability found in the presence of PCBM.
We now move our investigation to the role of preconditioning on solar cell open-circuit voltage. To gain insight into the stabilization process highlighted in Figure 2 , we have performed a systematic investigation on the poling (i.e., polarization of the semiconductor) effects on pristine perovskite ( Figure 4 a)  . a) PL dynamics from CH 3 NH 3 PbI 3 /TiO 2 and CH 3 NH 3 PbI 3 /PCBM bilayer and from CH 3 NH 3 PbI 3 deposited on glass (probed wavelength 780 nm, excitation wavelength 700 nm); b) PL dynamics from CH 3 NH 3 PbI 3 embodied in a solar cell architecture with PEDOT:PSS and PCBM as hole and electron extracting layers, respectively. The device was under short-circuit condition before (black line) and after (red line) a prepolarization (1 V) treatment (probed wavelength 780 nm, excitation wavelength 700 nm).
(5 of 9) 1501453 wileyonlinelibrary.com to be able to disentangle the effects strictly related to the device polarization from eventual light induced effects. All the experiments were performed in inert atmosphere. Without any prebiasing, the current-voltage ( I -V ) characteristics recorded from 0 V to +28 V and from 0 V to −28 V on pristine perovskite and bi-layer samples are symmetric, with similar current values (Figure 4 a,b, black lines). When the two samples are prebiased by applying a fi eld of 1.4 V µm −1 , the symmetry in the curves is lost and the current exhibits a rectifying behavior. This effect is the consequence of the partial positive and negative charges induced at the contacts that hamper the injection of charge at an applied voltage which is opposite with respect to the prebiasing, as suggested by Huang and co-workers. [ 10 ] Furthermore it must be noted that by inverting the polarization between the two electrodes (indicated henceforth as positive and negative poling), the switching of the I -V characteristic is not instantaneous, as it requires minutes-to-hours in order to recover the initial condition. Huang and co-workers reported this effect for the fi rst time for pristine perovskite under light illumination. [ 10 ] They concluded that it is the result of ions drifting through the perovskite, forming two doped regions, n and p , next to the electrodes, according to the voltage applied. Interestingly, Figure 4 b shows that the presence of PCBM deposited on top of the perovskite layer does not apparently hamper the ions drift, as the loss of symmetry in the curves demonstrates. The only difference induced by the presence of PCBM is an increased value of the forward current under poling conditions.
From these results we can summarize that: (i) the ion migration, proposed to be responsible for the formation of an internal fi eld within the planar device with symmetric contacts, is observable not only under light illumination (see Figure S4, Supporting Information) but also in the dark; (ii) under polarization, the presence of PCBM does not prevent the ions drifting. Thus, ions diffusion/drifting is still present in solar cells with a PCBM electron extracting layer, even if the short-circuit current (i.e., charge extraction) is stable upon scanning.
The presence of PCBM in the planar sample increases the absolute values of the forward current. Though one can assume that PCBM diffusion in the perovskite thin fi lm may induce a better morphology (compactness) of the thin fi lm, [ 19 ] we can exclude it as the origin of the enhanced current because, without prebiasing, the current values are similar with or without PCBM.
To pin down the origin of the enhanced forward current when the perovskite/PCBM bilayer is prebiased, in Figure 4 c we investigate the time evolution of the current upon polarization, with and without PCBM, on a time scale of hundreds of seconds for both samples (please note that these are twin fresh samples of the ones studied in Figure 4 a,b). Under poling, the pristine fi lm (full dots, black line) shows an enhancement in current in the fi rst 60 s until a plateau value is reached. We attribute the initial rising to the creation of the doped junction, via ions drifting, which is completed when the current stabilizes. In the presence of PCBM (full dots, red line) the transient dynamic is slowed down; however, the absolute value of the current increases without reaching a stabilized point in our time scale. This behavior is not symmetric when the poling is instantaneously inverted between the two electrodes (hollow circles). In this condition the absolute values of the currents are strongly reduced and slowly increase for both fi lms, again with a slower dynamic for the perovskite/PCBM fi lm. Such slow dynamics clarify that the current enhancement in the presence of PCBM is also related to the migration of ions. Upon fabrication of the perovskite/PCBM bilayer we can assume that the small organic molecules intercalate the perovskite grain of the polycrystalline fi lm, thus forming also lateral PCBM/perovskite interfaces. These observations lead us to hypothesize that upon polarization, ions do interact with the organic compound, increasing the overall conductivity of the fi lm. In the solar cell with PCBM as electron extracting layer, since ions, along with the photogenerated carriers, drift toward the electrodes due to the built-in voltage, an interaction of such species with the extracting layers is possible. Recent experimental [ 13,23,24 ] and theoretical [ 25,26 ] investigations have indicated iodide ions, I − , and methylammonium ions, MA + , as the most probable anionic and cationic species which can diffuse and/or drift within the perovskite absorber here studied.
As a consequence of the built-in fi eld deriving by the different contacts work-function, during cell operation iodide ions, I − , are expected to accumulate close to PCBM, while MA + close to the hole extracting contact. [ 27 ] The actual possibility for MA + to drift under the infl uence of a fi eld is currently being debated. Recent theoretical studies predict a high activation barrier (0.84 eV), [ 25 ] while others derive a barrier consistent with MA + motion (0.46 eV). [ 26 ] Some experimental works [ 13,23 ] agree with the latter, probing an accumulation of MA + upon polarization of MAPbI 3 thin fi lm at the negatively polarized electrode, however these experiments were performed in air or in a polar eviroment. There is instead theoretical agreement on the migration of I − , [ 25,26 ] with however limited experimental evidence in inert atmosphere due to the poor sensitivity of the experimental techniques used. In the following we fi rst address the eventual drifting of iodide as an effect of a fi eld developing across the semiconductor, and we then investigate its interaction with PCBM. To prove the nature of the negatively charged mobile ions migrating within the MAPbI 3 , we biased, in dark and nitrogen atmosphere, a sample composed of a pair of planar and symmetric silver electrodes, on top of which a pristine perovskite layer was deposited. Silver has been chosen due to its selective interaction with iodide, forming metal halides compounds. Figure 5 shows the Scanning Electron Microscopy (SEM) images of the thin fi lm deposited on the metal contacts in pristine conditions (top panel) and upon biasing (bottom panel). After biasing, differently from what observed previously with the samples based on gold electrodes, we are not able to register any I -V curve and we observe that the area corresponding to the positively biased electrodes results damaged, while the negatively biased electrodes appear unmodifi ed. This provides evidence of the electro-migration of the I − ions toward the positively biased electrode and their reaction with Ag.
We move now to the investigation of the interaction of the migrating anions with PCBM. It is well reported in the literature that PCBM can interact with alkyl ammonium salts resulting in a doping of the material itself. [ 28 ] In particular these salts act as n-dopants of the fullerene derivatives, increasing their conductivity by several orders of magnitude.
We verifi ed this effect fi rst by measuring the conductivity of a pristine PCBM fi lm and a PCBM fi lm containing a controlled amount of I − . The latter was realized by adding a Adv. Energy Mater. 2016, 6, 1501453 www.MaterialsViews.com www.advenergymat.de wileyonlinelibrary.com methylammonium iodide (MAI) solution (10 mg mL −1 in isopropanol) to the pristine PCBM solution (30 mg mL −1 in CB) with a ratio of 1:30. In Figure 6 a, we compare the J -V plot of the two fi lms. For this test we chose symmetric gold contacts with a channel length of 200 µm in order to avoid contact resistance effects. The addition of the MAI salt to the PCBM results in a markedly increased conductivity, which changes from 6.43 × 10 −7 S m −1 for the pristine fi lm to 4.35 × 10 −6 S m −1 for the PCBM/MAI fi lm. This effect may be ascribed either to a chemical doping of the PCBM, i.e., an excess of free carriers, or to an improved mobility, or both. In order to get deeper insights we tested the electronic properties of a PCBM fi lm and a PCBM/MAI fi lm in top gate/bottom contact fi eld-effect transistors (FETs). By comparing the measurements (Figure 6 b) of the pristine PCBM and the MAI doped devices, we are able to extract useful information regarding the PCBM interaction with ions that validate the aforementioned PCBM/perovskite interaction. The extracted saturation mobility (at V G = V D = 60 V) results in being 5.9 × 10 −2 cm 2 V −1 s −1 for the pristine PCBM transistor-in agreement with literature values [ 27,28 ] -and 0.15 cm 2 V −1 s −1 for the doped PCBM/MAI device. The latter value is more than two times higher than the pristine PCBM one. The pristine PCBM device shows poorer subthreshold slope values and a higher threshold voltage (36.3 V, which reduces to 8.1 V upon doping), likely owing to deep trap states. The holes current tail exhibited at low gate voltages in the saturation regime by the pristine PCBM device ( V D = 60 V curve in Figure 4 c) disappears in the MAI doped one, resulting in a suppressed hole conduction. At the same time, a marked increase of the nongateable OFF current in saturation in the MAI doped devices denotes a conductivity increase of the semiconducting fi lm, confi rming the results reported above for the two terminal devices (in the linear regime the OFF currents are at the level of leakage and cannot be compared). The generally increased performances ( Table 2 ) of the doped device, with respect to the pristine one, can be attributed according to the literature [ 29 ] to an increased concentration of excess carriers, as a consequence of chemical doping which shifts the Fermi level closer to the LUMO level of PCBM. To confi rm this we used ultraviolet photoelectron spectroscopy (UPS) to investigate the energy levels of the same fi lms used in the transistors (Figure 6 c). The limited penetration depth characteristic of UPS makes this investigation more suitable to explain the FET data, where a nanometer thick channel accumulates at the Adv. Energy Mater. 2016, 6, 1501453 www.MaterialsViews.com www.advenergymat.de  semiconductor-dielectric interface, rather than the two terminal samples data. By linear extrapolation from the high binding energy region of the spectrum we derive the position of the Fermi level for pristine PCBM and for the PCBM:MAI fi lm to be, respectively, at −4.27 and −4.13 eV. This is a relatively small but consistent shift of the Fermi level toward the vacuum level when the methylammonium salt is added to the fi lm, indicating a clear n-doping effect. From the Fermi level position, as a fi rst approximation, we have estimated the charge carrier density at thermal equilibrium in both fi lms on the basis of a single-crystal lattice model. [ 30 ] In the case of the pristine PCBM we determined an electron density of 2.5 × 10 11 cm −3 ; while for the PCBM:MAI fi lm a concentration of 5.7 × 10 13 cm −3 , two orders of magnitude higher respect to the pristine sample (see the Supporting Information for the calculations). Thus, in the transistor deep trap states are fi lled by chemically introduced excess carriers, and as a consequence, for the same applied gate voltage, electrostatically accumulated carriers in the channel occupy more mobile states, thus producing a net shift in the threshold voltage and an increased fi eld-effect mobility. These results confi rm the interaction between ions and the PCBM that occurs at the interface between the perovskite and the PCBM, with the increased conductivity of the PCBM phase owing to an increased number of free carriers induced by I − doping.
We conclude that, while the main cause of hysteretic behavior in perovskite solar cells can be found in the formation of an internal fi eld upon ion migration, the role of the charge extracting layer is of paramount importance in determining the device response to such a transient phenomenon. Effi cient charge extraction is critical to make the devices less sensitive to their J-V scanning history. This is especially true when monitoring the short-circuit photocurrent density: we have shown that by introducing a PCBM electron extracting layer, though a formation of a p-i-n-like junction within the perovskite layer is still expected upon cell operation, the electron transfer rate at the perovskite-PCBM layer results almost unaffected, stabilizing the short-circuit current from the very fi rst J -V scan. On the other hand, we provide evidence for a further phenomenon, not considered so far, which needs to be taken into account both when designing effi cient perovskite solar cells and when attempting their modeling. We show that, upon ion migration, a modifi cation of the electronic properties of PCBM occurs at the interface with the perovskite. In particular, as an effect of the built-in fi eld, iodide migrates toward the PCBM layer and it chemically dopes the organic semiconductor, increasing its conductivity. Upon "preconditioning" J -V cycles, such doping contributes to improve and stabilize the open-circuit voltage, as previously observed also for other photovoltaic technologies, [ 31 ] thus producing an overall stabilization of the device. The latter is a direct consequence of the electron transfer process between the ionic species and the organic molecule, establishing an "electrostatic trap" and reducing iodide back diffusion/drift.
We therefore believe that our work provide relevant indications for the complete rationalization of "hysteresis" by showing that parameters such as the short-circuit current and open circuit voltage are not necessarily affected by the same transient behaviors.
Since hybrid perovskites are ionic solids where ions migration is allowed by either vacancies or interstitials, or the combination of both, charge extracting layers and their interaction with such ionic species have therefore to be carefully considered when designing highly effi cient and stable devices. Furthermore this study wants to highlight that preconditioning cycles are typically required to observe stable J -V curves. Therefore care should be taken when reporting "hysteresis free" perovskite solar cells, which is not a fully appropriate description of such preformed devices.

Experimental Section
Fabrication and Test of Perovskite Solar Cells : MAI was synthesized according to an established procedure [ 2,4 ] and dissolved with PbCl 2 (99.999%, Sigma Aldrich) in Dimethylformamide (DMF) with a ratio of 3:1. Except for metallic electrodes, which were thermally evaporated, all other layers were deposited by spin coating. The TiO 2 compact layer was deposited, according to the literature, [ 32 ] directly on top of Fluorinedoped Tin Oxide (FTO)-covered glasses after a previous cleaning cycle with distilled water, acetone, isopropanol (IPA), and oxygen plasma. The fi nal thickness of the TiO 2 compact layer was 80 nm. The mesoporous scaffolds of Al 2 O 3 (Sigma Aldrich) and TiO 2 (Dyesol) nanoparticles were deposited directly on top of the TiO 2 compact layer with a fi nal thickness of 300 nm. PEDOT:PSS (Clevios 4083) was deposited on clean Indium Tin Oxide (ITO) substrates with a fi nal thickness in of 50 nm. The mixedhalide perovskite (CH 3 NH 3 PbI 3− x Cl x ) was deposited at 2000 rpm for 1 min and annealed at 90 °C for 90 min for the fl at substrates and annealed at 100 °C for 60 min for mesoporous scaffolds. The hole extracting layer spiro-OMeTAD was processed according to the literature [ 2 ] with the addition of tert-butyl pyridine and bis-(trifl uoromethane) sulfonimide lithium salt (Li-TFSI, Sigma Aldrich). PCBM (NanoC) was dissolved in anhydrous chlorobenzene (CB, 30 mg mL −1 ) and for the inverted devices it was covered by a layer of Al-doped ZnO nanoparticles in isopropanol (Nanograde). Gold and aluminum were evaporated in vacuum at 1 × 10 −6 mbar. The J/V curves of the solar cells were recorded with a computer-controlled Keithley 2400 in air under a simulated AM1.5 spectrum provided by a class AAA Newport solar simulator. The intensity of the light was calibrated with a reference diode NREL certifi ed with a spectral mismatch factor 0.99.
Fabrication and Test of Lateral Devices : Gold and silver planar electrodes used for poling the perovskite and perovskite/PCBM bi-layers were realized on precleaned glass substrate by a standard lithography lift-off process. The thickness of the electrodes was 50 nm and the length of the channel was 20 µm. The mixed-halide perovskite (CH 3 NH 3 PbI 3− x Cl x ) was deposited directly on top of the electrodes at 2000 rpm for 1 min and annealed at 90 °C for 90 min. The gold planar electrodes used for measuring the conductivity of PCBM and PCBM:MAI blend were thermally evaporated. The length of the channel between the electrodes was 200 µm. The solutions of the pristine PCBM (30 mg mL −1 in CB) and PCBM:MAI 30:1 blend were spin-coated in a nitrogen atmosphere. Electrical characterization of lateral devices and transistors was Adv. Energy Mater. 2016, 6, 1501453 www.MaterialsViews.com www.advenergymat.de Fabrication of Transistor : 35 nm thick Au source and drain contacts, on top of a 1.7 nm thick Cr adhesion layer, were patterned on a thoroughly cleaned low alkali 1737F Corning glass substrate through a standard lithographic lift-off process. Contacts for single transistors had a channel width to length ratio (W/L) of 10 000 µm/20 µm. Before the semiconductor deposition, substrates were rinsed with acetone and isopropyl alcohol. PCBM and the PCBM:MAI 30:1 solutions were both deposited in a nitrogen atmosphere from a 30 mg mL −1 (CB and CB/IPA) solution by spin-coating at 1000 rpm for 60 s to obtain an ≈40 nm thick fi lm. The semiconductor fi lm was then annealed at 90 °C for 20 min to guarantee solvent drying. Cytop CTL-809M 9 % w/v solution (AGC Chemicals Europe) was adopted as the dielectric layer. A thickness of 600 nm was optimized by spinning at 4000 rpm for 90 s. After the deposition of the dielectric, samples were annealed at 80 °C for 90 min. An 80 nm thick aluminum layer was thermally evaporated through a metal shadow mask to pattern the gate electrodes.
Time-Resolved PL : The experiments were performed using a femtosecond laser source and streak camera detection system (Hamamatsu C5680) operated using a synchroscan voltage sweep module, yielding a maximum temporal resolution of ≈3 ps.
An unamplifi ed Ti:Sapphire laser (Coherent Chameleon Ultra II) operating at 80 MHz was tuned to provide pulses with central wavelengths of 700 nm, energies of ≈50 nJ, temporal and spectral bandwidths of ≈140 fs and ≈5 nm, respectively. The laser light was focused with a 75 mm focal length lens onto the sample, which was kept in vacuum atmosphere during the whole measurement. The emitted PL was collected with a doublet of lenses and focused onto the entrance slit of a spectrograph (Princeton Instrument-Acton SP2300) coupled to the streak camera. The trPL measurements were performed keeping the sample at 0 V polarization, polarizing it at 1.5 V and again at 0 V after the polarization procedure.
Transient Absorption Spectroscopy : The output of a Coherent Micra Ti:Sapphire oscillator in conjunction with a Coherent RegA 9040 amplifi er (800 nm, 40 fs pulse duration and 250 kHz repetition rate) was split into two parts. One part of the laser output pumped a Coherent collinear optical parametric amplifi er to generate pump pulses at 650 nm. The other part was used as probe after white light generation in a sapphire plate. The delay between the pump and the probe pulses was controlled by a motorized delay-stage and the signal was detected using a customized CCD-camera with electronic shutter. For the measurements, the pump beam of 1.6 nJ excitation power was focused on ≈1 mm 2 device area. All measurements were carried out in vacuum, using a continuous fl ow static exchange gas cryostat (Oxford Instruments Optistat CF). For monitoring the current in short-circuit condition and applying a bias voltage, the devices were connected to a Keithley B2912A source/measure unit.
Ultraviolet Photoelectron Spectroscopy : UPS measurements were carried out with a Kratos Axis Ultra DLD spectrometer using a He I (21.22 eV) discharge lamp. A −9.0 V bias was applied to the sample in order to precisely determine the low kinetic energy cutoff. The analyses were carried out with an analysis area of 55 µm in diameter, at a pass energy of 10 eV, and with a dwell time of 100 ms.

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