A mechanistic study of the dopant-induced breakdown in halide perovskites using solid state energy storage devices

Doping halide perovskites (HPs) with extrinsic species, such as alkali metal ions, plays a critical, albeit often elusive role in optimising optoelectronic devices. Here, we use solid state lithium ion battery inspired devices with a polyethylene oxide-based polymer electrolyte to dope HPs controllably with lithium ions. We perform a suite of operando material analysis techniques while dynamically varying Li doping concentrations. We determine and quantify three doping regimes; a safe regime, with doping concentrations of <1020 cm−3 (2% Li : Pb mol%) in which the HP may be modified without detrimental effect to its structure; a minor decomposition regime, in which the HP is partially transformed but remains the dominant species; and a major decomposition regime in which the perovskite is superseded by new phases. We provide a mechanistic description of the processes mediating between each stage and find evidence for metallic Pb(0), LiBr and LiPbBr2 as final decomposition products. Combining results from synchrotron X-ray diffraction measurements with in situ photoluminescence and optical reflection microscopy studies, we distinguish the influences of free charge carriers and intercalated lithium independently. We find that the charge density is equally as important as the geometric considerations of the dopant species and thereby provide a quantitative framework upon which the future design of doped-perovskite energy devices should be based.

significance of the dissolution of the perovskite active material on the measurable electrochemical gravimetric capacity of a conventional lithium ion battery (LIB) using a traditional liquid electrolyte.
(1M LiPF 6 in EC/DEC, 1:1 vol ratio). Fig. S1(a) shows the galvanostatic charge-discharge cycles of the cells, inset are the zoomed axes showing the second cycles onwards. The time indicated corresponds to the length of time left between assembling the cell and then beginning the initial discharge cycle.
All cells were cycled at the current rate of 30 mAg −1 . It is immediately apparent that both the initial discharge capacity and the stabilized capacity are reduced, the longer the cell is left before beginning the cycle test. The best performing cell in both the first and all subsequent cycles was the cell that was cycled immediately upon assembly (0 hours). The cells left for four and eight hours follow a similar drop in capacity. The most prominent change occurs between eight hours and twenty four hours.
The cell that was left for twenty four hours before cycling observed a drop of ≈ 400 mAhg −1 (~80%) compared to the other cells for the initial discharge capacity. Since it is proposed that the dominant processes responsible for large initial discharge capacity are the lithium insertion into and the subsequent conversion reaction of the perovskite structure, it may be hypothesized that the drop in capacity after twenty four hours is due to the absence of these two mechanisms -indeed, no plateaus are observed at 2.1V or 1.5V to contradict this. It could be concluded therefore that after twenty four hours of contact with the electrolyte, the perovskite structure has been dissolved -or at least compromised beyond the point of being able to facilitate lithium insertion or conversion. Traditionally (in non-perovskite cells) it is common to leave cells for a length of time, up to around twenty four hours before cycling in order to allow the electrolyte time to soak the electrode completely, increasing the coverage and thus active surface area.
The final plateaus, below 1.0 V are all approximately equal in capacity irrespective of the time taken between cell assembly and measurement. Since the reaction below 1.0 V is a multistage alloying reaction between lead and lithium -it is unsurprising that the magnitude of capacity contained within this reaction is equal for the samples which, being the same perovskite formulation, inherently contain the same quantity of lead -irrespective of the state of dissolution. For these experiments, a conventional coin cell was used.
A simple, ex situ stability test was carried out by dropping a sample of liquid 1M electrolyte onto a typical film of MAPbI 3 -the iodide analogue of the perovskite used in this work. Upon leaving the sample for 24 hours, the metallic-black characteristic colour and texture of the MAPbI 3 is totally lost and in its place is a yellow solution, as shown in Fig. S1 (b). The remaining yellow solution is indicative of dissolved PbI 2 salts -one of the precursors to making MAPbI 3 . Therefore, it can be concluded that in any battery cell, using liquid electrolytes containing polar solvents, that any OIHP material contained within the electrode will dissolve into its respective PBX 2 lead salt (where X denotes the relevant halide anion).
Finally, Fig. S1(c) shows the corresponding galvanostatic charge-discharge curve of a MAPbBr 3 electrode, in a conventional LIB coin cell, using however, the polymer-based electrolyte described herein. A small capacity decrease of ~30 mAhg -1 (< 10%) after 48 hours is observed, smaller already than the loss observed in the liquid cell after only 4 hours. This indicates an immense improvement to the overall stability of the perovskite/PEO interface relative to the perovskite/liquid electrolyte.

Section II: Calculation of doping concentration and unit conversion charts
Where t (seconds) is the time elapsed during a given current condition.
Since one Coulomb of charge corresponds to 6.24 10 18 charges, the number of charges and × thus Li ions transferred through the cell (N) is given by, For ease of reference, at the used current rate of ± 0.015 mA where a negative corresponds to discharge and therefore Li + /einsertion and a positive corresponds to charge and therefore Li + /eremoval, inserting the correct values yields a charge doping insertion or removal rate of 9.36 x 10 13 Li/eper second.
To calculate the molar ration (Li:Pb %) the number of Pb species per unit volume (cm -3 ) is calculated by assuming 1 Pb per perovskite unit cell, giving a value of 4.81 x 10 20 Pb per cm -3 .

Section III: Additional peaks and GIFs showing the species behaviour in real time
Section IV: Rietveld refinement Rietveld refinement was used to fit the XRD spectra at four different doping stages. Spectra 25 -at the beginning of the first insertion process, spectra 85 -at a low doping concentration of 7.47 x 10 19 cm -3 during the first insertion process, spectra 295 at a high doping concentration of 2.85 x 10 20 cm -3 at the end of the first insertion process and spectra 1225 at a net doping concentration of 3.5 x 10 20 cm -3 at the end of the third insertion process and therefore including two removal processes prior.
A preliminary refinement was used to approximate the spectra before then using a Le Bail extraction [1] to calculate better peak positions for the various components included in the fit. These parameters were then fed back into the Rietveld refinement to achieve the final fit. Due to artefacts introduced during the operando XRD experiment, a perfect refinement was not achievable -this is reflected in the final χ 2 values in Table (S1). The authors would therefore like to exercise caution with overreaching the parameters able to be extracted from the refinement reliably.
However, by limiting the refinement to a closed system comprising only the Pb (0) and MAPbBr 3 species, the relative phase fractions of these two -which are necessarily dependent upon the relative peak area intensities -the shift in ratio between these two present species was extracted with reasonable accuracy. Indeed, the results corroborate the single peak fitting described in the main text. The resulting relative phase fractions are shown in Fig. S6 (c)).
The main parameters used to refine the model include the unit cell length for each species, its µ-strain and domain size. For completeness these parameters and how they vary with each spectra are shown in Fig. S6 and are tabulated in Table S1.

Section V: Photoluminescence measurements and Optical Coin Cell fabrication
A schematic of the setup used to measure the in situ PL spectra of the perovskite material, under various stages of Li doping is shown in Fig. S7 (a). Steady-state PL spectra were recorded by a gated intensified CCD camera (iCCD, Andor Star DH740 CCI-010) connected to a grating spectrometer (Andor SR303i). The pulsed output from a mode-locked Ti:Sapphire optical amplifier (Spectra-Physics Solstice, 1.55 eV photon energy, 80 fs pulse width, 1 kHz repetition rate) was used to produce 400 nm excitation via second harmonic generation in a β-barium borate crystal. The iCCD gate (width 2 ns) was electronically stepped in 2 ns increments, relative to the pump pulse, to enable ns-temporal resolution of the PL decay.
A digital photograph of a fresh MAPbBr 3 optical coin cell, prior to any electrochemical processing is shown in Fig. S7 (b). The orange colour is characteristic of the starting perovskite material. The same cell, after multiple deep insertion and removal processes is shown in Fig S7 (c). The black colouration is characteristic of the final degredation products after multiple cycles, predominantly metallic lead.
The internal stacking of the optical coin cell is similar to that of the pouch cell described in The cell is sealed using a crimping machine at 1000 PSI.

Cell architecture
The optical cell (EL-CELL, ECC-Opto-Std test cell) was assembled as per Fig. S9. First, the perovskite electrode slurry as described in the main text, was drop cast onto a glass microscope cover slip. The cell stack consisted of the coated cover slip, an aluminium mesh current collector, a glass fibre separator, and a lithium metal chip. Once sealed, the cell was filled with 5M LiTFSI in ethylene carbonate/propylene carbonate (1:1 vol ratio) to act as the electrolyte since the polymer PEO system was not suitable for the EL cell configuration. The window of stability for this electrolyte has been studied in [2] and demonstrates how over the time periods required for this measurement do not show degradation (dissolution) effects beyond those due to the cycling itself. The aluminium mesh current collector was electrically contacted by a stainless steel pin.

Data acquisition
Images were acquired with a camera exposure time of 2.7 ms, at a frame rate of 10 Hz. Each recorded image was spatially binned (2×2 pixels, giving an effective pixel size of 69.4 nm px -1 ) and sets of 40 recorded images were temporally binned together to yield an effective frame rate of 0.25 Hz.
Electrochemical control was achieved using a Gamry potentiostat (Interface 1010). Image acquisition and synchronisation of instruments was performed using in-house developed LabVIEW routines.

Differential image analysis
The recorded image stacks were first corrected for stage drift in the xy-plane by selecting a stationary sub-diffraction limited spot and fitting its position over time using a two-dimensional Gaussian function. The extracted centre positions in x and y for each image were subsequently used to correct for stage drift.
Sequential differential images (at each time t) were obtained by dividing the pixel values of pairs of frames separated by a time interval (Δt = 40 s) according to: .
The resulting image contrast displays the fractional intensity change between images, with the contrast centred around 0. This removes constant or slowly varying background contributions and inhomogeneities in the sample illumination, isolating more rapid changes during the interval between pairs of images (40 s). Negative pixel values indicate a reduction in intensity, while positive values indicate an increase, between the two frames. and (c) after the maximal amount of Li ions have been inserted into the electrode -the initial perovskite structure is no longer observed and bright spots of metallic lead appear. The outwards-in loss of perovskite structure is emphasised by considering the differential changes between frames, shown in (d -f). Blue colouration refers to a loss in reflectivity and red to a gain. Therefore, the inward motion of the blue signal refers to the loss of the perovskite, starting from the edges (regions exposed principally to Li ions).

Section VIII: Custom temperature-controlled X-ray Pouch Cell
Fig. S11 Shows a schematic representation of the LIB-inspired material stack used in the X-ray pouch cell used for the operando X-ray experiments of the main text [3]. First, the perovskite electrode slurry (containing the target perovskite phase, super P and PVDF in an 85:10:5 mass ratio) is drop cast onto an indium tin oxide (ITO) coated polyethylene terephthalate (PET) substrate. A glass fibre separator is cut such that it forms a hollow square shape (providing separation between the electrodes via the walls, but allowing full contact with the PEO in the gap) and is placed on top of the electrode.
The polymer electrolyte is then cut to size and placed on top of the electrode/separator stack. Finally, the anode (Li metal chip) is attached to a stainless steel disc which is held on top of the rest of the stack.
A copper contact is placed on the ITO surface near to the perovskite electrode and a nickel contact is placed on the stainless steel disc. These contacts are allowed to protrude from the pouch cell, once sealed, in order to be connected to the potentiostat. The aluminium pouch cell is then heat sealed and is Before being placed into the beamline, the pouch cell is mounted in a custom 3D printed casing which uses a nitrogen "bubble" formed by a Kapton window connected to a gas feed to apply a controlled stack pressure of (~2 bar). The casing also utilizes a copper thermal strip attached to a voltage supply and wrapped around the pouch to heat the pouch to the required temperature. The polymer electrolyte must be heated to above its melting temperature (~70 °C) in order to facilitate significant ionic conductivity [4]. The cell is then placed in the beamline, once connected to the potentiostat for electrochemical control, for the operando X-ray experiments.
Section IX: X-ray single peak analysis New peaks, that are not present inherently in uncycled cells are identified and tracked -for example, those belonging to lead and lead bromide. By tracing the position and relative intensity of all peaks, a picture can be established with regards to the structure and species present and how they are dynamically affected by the insertion and removal of lithium. The following processes were automated using home-built scripts based in Python. The first step is the reduction of the 3D raw spectra to a 1D linear spectra this was done using pyFAI following the python module developed by Giannis Ashiotis et al. [5] .
Second, since the X-ray beam was scanned over 30 positions on the sample to avoid damaging the perovskite from over exposure to the high energy photons, the spectra for each location was siloed into a single set. (Stages shown in Fig. S12 (a -b)). Individual spectra were then examined and each peak indexed to known species by comparison with the relevant cif powder patterns (Fig. S12 (c)).
To extract how the peak intensity varied with the state of Li doping, a background was determined as the linear subtraction between the two points of intersection of the peak. This was then subtracted locally from each peak (example shown in (Fig. S12 (d)). This had to be done since the background variation was too great between each spectra to remove the entire background reliably. The remaining y-values are integrated between the intersection points of the peak by fitting a gaussian profile to the peak. These values were then fed back via the time stamp recorded by the XRD detector to the state of charge determined by the potentiostat.