Multimodal electron microscopy of halide perovskite interfacial dynamics

Nature ( 2026 ) Cite this article
Halide perovskite light-emitting diodes promise high-efficiency 1 , 2 , 3 , low-cost optoelectronics, yet their operational instability remains a critical barrier to practical deployment.

Here we develop a multimodal in situ electron microscopy approach that integrates four-dimensional scanning transmission electron microscopy, energy-dispersive X-ray spectroscopy and atomic-resolution imaging to directly visualize structural and chemical evolution in a working halide perovskite light-emitting diode with nanometre precision.

Our in situ biasing measurements uncover nanoscale structural and chemical transformations initiated at transport layer interfaces, including the formation of metallic lead and lead-rich secondary phases, as well as strain-driven grain fragmentation.

On biasing, we observe the partial transformation of the metallic Al contact to insulating AlCl 3 .

Crucially, whereas the bulk of the perovskite emitter remains relatively intact, our experiment shows that degradation is localized at interfaces.

By comparing in situ and ex situ measurements, these results establish a mechanistic link between interfacial strain, ionic transport and electrochemical reactions in working devices, and provide a broadly applicable framework for nanoscale degradation analysis in complex multilayered optoelectronic systems using multimodal in situ biasing microscopy.

Similar content being viewed by others
Self-supervised deep learning for tracking degradation of perovskite light-emitting diodes with multispectral imaging
Local nanoscale phase impurities are degradation sites in halide perovskites
Interfacial alloying between lead halide perovskite crystals and hybrid glasses
Halide perovskites have emerged as promising semiconductors for optoelectronic applications owing to their high charge-carrier mobility, long diffusion lengths and facile solution processability.

These properties have enabled rapid advancements in perovskite-based photovoltaics 4 , 5 , light-emitting diodes (LEDs) 6 , 7 , photodetectors 8 , 9 and lasers 10 , 11 .

Although substantial progress has been made in photovoltaic stability, hybrid perovskite LEDs remain prone to rapid operational degradation compared to conventional inorganic semiconductors, such as silicon and III–V materials 12 , 13 .

Electric-field-driven ion migration and interfacial electrochemical reactions at interfaces are widely recognized as critical challenges 14 , 15 , 16 , yet the atomic-scale mechanisms driving structural instability and interfacial failure remain poorly understood.

Conventional in situ techniques, such as synchrotron X-ray spectroscopy, offer valuable chemical insight but lack the spatial resolution to capture localized degradation phenomena in nanostructured devices 17 , 18 , 19 .

Consequently, developing comprehensive atomic-scale in situ imaging methodologies is crucial for elucidating degradation processes specific to hybrid perovskite light-emitting diodes (PeLEDs) and designing strategies for enhanced device stability.

The advent of high-speed detectors has enabled time-resolved in situ electron microscopy, allowing precise stimulus control with outstanding spatial, temporal and spectral resolution 20 , 21 .

Early in situ transmission electron microscopy (TEM) studies on perovskite photovoltaics established the importance of field-driven transformations under bias: iodide migration and PbI 2 nucleation with polarity dependence 22 , oxygen exchange at electron transport interface 23 and the bulk amorphization–recrystallization dynamics in perovskite layers 24 .

However, these results were typically obtained under static biasing and focused on bulk perovskite layer, without correlative analysis of the evolution of structure and composition at buried interfaces during device operation (for example, illumination and realistic device drive).

Here we present an in situ multimodal electron microscopy investigation of PeLEDs using aberration-corrected four-dimensional scanning TEM (4D-STEM) combined with low-dose atomic-resolution imaging and energy-dispersive X-ray spectroscopy (EDX) analysis.

We fabricated and analysed a nanoLED device integrated onto a microelectromechanical system (MEMS) chip that enables electrical biasing in situ.

Under galvanostatic conditions, we systematically monitored structural, morphological and compositional evolution at emitters, interfaces and cathode contact.

Our experiments reveal pronounced lattice distortions at emitter–transport layer interfaces, formation of lead-rich phases and the critical role of halide migration, particularly Cl − , in electrode corrosion and material decomposition.

We show that the perovskite device behaves as a nanoscale electrochemical cell under continuous bias, with interface-specific deformation and chemical reactions governing its failure.

These in situ, interface-resolved and correlative insights establish a mechanistic foundation for interface engineering and device design strategies aimed at mitigating degradation and improving the long-term stability of perovskite optoelectronics.

A batch of sky-blue PeLED devices were fabricated for this study 25 ( Methods ).

The photoactive component of the device consists of nanocrystalline islands of Cs 1.2 FA 0.3 Pb(Br 0.65 Cl 0.35 ) 3 (DCDH perovskite) deposited onto a hole transport layer (HTL) composed of poly[bis(4-phenyl)(2,4,6-trimethylphenyl)amine] (PTAA), poly(9-vinylcarbazole) (PVK) and polyvinylpyrrolidone (PVP), and subsequently coated with an evaporated layer of 1,3,5-tris( N -phenylbenzimidazol-2-yl)benzene (TPBi) as the electron transport layer (ETL) (Fig.

1a ).

The current density ( J )–voltage ( V )–luminance characteristics of a representative standard device are shown in Fig.

1b .

The device initially exhibited electroluminescence at 485 nm with a full-width at half-maximum of 15 nm (inset) and a peak luminance exceeding 9,500 cd m −2 at 10.8 V achieved during sweeping J – V measurement at a rate of 0.4 V s −1 .

The operational lifetime ( T 50 ) was approximately 3 min when a constant driving current of 1 mA cm −2 , sufficient to generate an initial luminance of 100 cd m −2 , was applied (Supplementary Fig.

1 ).

One of the as-prepared PeLED devices was cross-sectioned using focused ion beam (FIB) milling, achieving a final lamella thickness of approximately 100 nm.

The cross-sectional device, here referred to as nanoLED, was Pt wire-bonded to the electrodes of a MEMS chip equipped with simultaneous heating and biasing capabilities 26 .

A schematic diagram of the STEM imaging and biasing configuration used to study the nanoLED device is presented in Fig.

1a and Supplementary Fig.

2 .

A high-angle annular dark-field (HAADF)-STEM image of the nanoLED, labelled with the composition of the layers, is shown in Fig.

1c .

Atomic-scale HAADF images of the pristine DCDH emitters show the orthorhombic crystal structure at room temperature (Fig.

1d–f ).

The fast Fourier transform analysis of the lattice reveals an interplanar spacing of 0.58 nm, consistent with the (020) reflection.

STEM-EDX elemental maps (Supplementary Fig.

3 ) verify that the initial composition of the multilayered structure is uniform across the layers and matches the nominal stoichiometry.

Fig.

1: In situ biasing multimodal electron microscopy of a perovskite blue light-emitting diode.

a , Schematic diagram of the STEM detectors used for the galvanostatic study of the nanoLED device, with biasing configuration of the Al/LiF and indium tin oxide (ITO) contacts.

The nanoLED cross-section is 40 μm wide and 100 nm thick.

b , Current–voltage and luminance characteristics of the pristine device, showing an emission peak at 485 nm (inset).

c , STEM-HAADF cross-section of the nanoLED device stack.

d , e , Atomic-resolution STEM-HAADF image ( d ) of the perovskite emitter with the corresponding fast Fourier transform ( e ), showing an orthorhombic structure with a (020) interplanar spacing of 0.58 nm.

f , Ideal orthorhombic lattice structure of the perovskite emissive layer seen along the [101] zone axis.

Scale bars, 200 nm ( c ), 1 nm ( d ), 2 nm −1 ( e ).

Source data Full size image
a , Schematic diagram of the STEM detectors used for the galvanostatic study of the nanoLED device, with biasing configuration of the Al/LiF and indium tin oxide (ITO) contacts.

The nanoLED cross-section is 40 μm wide and 100 nm thick.

b , Current–voltage and luminance characteristics of the pristine device, showing an emission peak at 485 nm (inset).

c , STEM-HAADF cross-section of the nanoLED device stack.

d , e , Atomic-resolution STEM-HAADF image ( d ) of the perovskite emitter with the corresponding fast Fourier transform ( e ), showing an orthorhombic structure with a (020) interplanar spacing of 0.58 nm.

f , Ideal orthorhombic lattice structure of the perovskite emissive layer seen along the [101] zone axis.

Scale bars, 200 nm ( c ), 1 nm ( d ), 2 nm −1 ( e ).

Atomic-scale interfacial strain mapping
Lattice strain is a critical contributor to defect formation in semiconductors, with extensive investigations highlighting its substantial impact on the stability and efficiency of perovskite optoelectronic devices 27 , 28 , 29 .

Geometric phase analysis, a well-established technique for deformation analysis 30 , 31 , 32 , was used following the Hÿtch formalism to evaluate the relationship between lattice displacement and phase differences between emitter and defective Pb-rich phases in high-resolution STEM images through lattice strain maps 33 , 34 , 35 , 36 .

This approach enables us to extract detailed information about deformation fields within the active region of the nanoLED.

Details of mask selection and precision criteria are provided in Supplementary Fig.

4 .

Atomic-resolution images of a DCDH emitter are shown in Fig.

2a–c , for an interface with the ETL, the perovskite centre and the interface with the HTL, respectively.

Whereas the central area has a uniform orientation across the field of view, both ETL and HTL interfaces show misalignments, defects and inhomogeneities.

HAADF-STEM contrast reveals that the high-intensity clusters are lead-rich, potentially including metallic Pb, PbX 2 , and CsPb 2 X 5 phases 37 .

Lattice strain maps computed in Fig.

2d–i reveal continuous deformation spanning out-of-plane ( y direction owing to FIB geometry), in-plane ( x direction) and shear components (Extended Data Fig.

1a–c ), extending from the centre of the emitters to the interfaces.

Out-of-plane strain critically influences charge transport, interface quality and overall device performance by directly affecting vertical electron flow between layers.

Conversely, in-plane strain relates to lateral defects or grain boundaries, influencing structural and functional integrity and film uniformity.

Localized zones of high tensile strain can be identified at the interface with Pb-rich regions, indicating outward pressure due to compositional heterogeneity associated with lattice expansion.

Negative compressive deformation near the ETL/emitter interface (Fig.

2d,g ) corresponds to lattice shrinkage, probably induced by chemical interactions.

The pronounced strain variations (highlighted by red boxes) probably arise from thermal expansion mismatch, or structural distortions induced by phase transitions or ion migration.

A schematic illustration of possible interface degradation mechanisms associated with lattice strain is presented in Extended Data Fig.

1d , highlighting how pre-existing strain and structural inhomogeneities at the interfaces may predispose the device to instability under electrical operation.

Pb-rich phases, grain boundary misalignments and chemical lattice distortions disrupt interfacial crystallographic continuity, intensifying structural instability.

Fig.

2: Interfacial strain and structural distortion revealed by geometric phase analysis.

a – c , Atomic-resolution HAADF-STEM images of the DCDH emitter at the ETL interface ( a ), emitter centre ( b ) and HTL interface ( c ).

Boxes mark regions of elevated HAADF intensity that co-localize with Pb-rich products (for example, metallic Pb, PbX 2 , CsPb 2 X 5 ), indicative of interfacial compositional heterogeneity.

d – i , Corresponding geometric phase analysis strain maps showing the out-of-plane ( ε yy ; y direction set by the FIB lamella geometry) ( d – f ) and in-plane ( ε xx ; x direction) ( g – i ) components, corresponding to the ETL interface ( d , g ), emitter centre ( e , h ) and HTL interface ( f , i ), respectively.

Red (positive) denotes tensile strain and blue (negative) denotes compressive strain; shear components are provided in Extended Data Fig.

1a–c .

Localized tensile zones are observed adjacent to the Pb-rich regions (boxed), consistent with lattice expansion driven by compositional inhomogeneity, whereas compressive deformation near the ETL/emitter interface ( d , g ) suggests chemically induced lattice shrinkage.

The emitter centre ( b , e , h ) remains largely undistorted under pristine conditions, whereas both interfaces exhibit misalignment, defects and inhomogeneity.

An axis compass on panels indicates in-plane ( x ) and out-of-plane ( y ) directions used for strain analysis.

Scale bars, 2 nm.

Full size image
a – c , Atomic-resolution HAADF-STEM images of the DCDH emitter at the ETL interface ( a ), emitter centre ( b ) and HTL interface ( c ).

Boxes mark regions of elevated HAADF intensity that co-localize with Pb-rich products (for example, metallic Pb, PbX 2 , CsPb 2 X 5 ), indicative of interfacial compositional heterogeneity.

d – i , Corresponding geometric phase analysis strain maps showing the out-of-plane ( ε yy ; y direction set by the FIB lamella geometry) ( d – f ) and in-plane ( ε xx ; x direction) ( g – i ) components, corresponding to the ETL interface ( d , g ), emitter centre ( e , h ) and HTL interface ( f , i ), respectively.

Red (positive) denotes tensile strain and blue (negative) denotes compressive strain; shear components are provided in Extended Data Fig.

1a–c .

Localized tensile zones are observed adjacent to the Pb-rich regions (boxed), consistent with lattice expansion driven by compositional inhomogeneity, whereas compressive deformation near the ETL/emitter interface ( d , g ) suggests chemically induced lattice shrinkage.

The emitter centre ( b , e , h ) remains largely undistorted under pristine conditions, whereas both interfaces exhibit misalignment, defects and inhomogeneity.

An axis compass on panels indicates in-plane ( x ) and out-of-plane ( y ) directions used for strain analysis.

Scale bars, 2 nm.

At the emitter/HTL interface (Fig.

2f,i ), compressive deformation is concentrated near the direct contact boundaries between the transport layer and the emitter (marked in blue), while Pb-rich regions exhibit pronounced tensile strain.

In-plane strain often accumulates near grain boundaries and misoriented regions in the perovskite lattice, resulting from mismatches during crystallization or film formation 38 .

These distortions are evident along grain boundaries, with zones of tension and compression clearly demarcated.

Sharp transitions between blue and red regions (highlighted by a red box) probably represent dislocation cores or strain centres within the crystal structure, where lattice distortions are most pronounced.

Shear strain maps (Extended Data Fig.

1a,c ) for interfacial regions enable visualization of lateral displacements of the (perovskite) lattice, with regions of high shear strain indicating sliding interactions (Supplementary Fig.

4 ).

By contrast, the central region of the main emitter (Fig.

2b ) retains structural integrity, exhibiting minimal detectable strain fields under pristine conditions (Fig.

2e,h and Extended Data Fig.

1b ).

The presence of Pb-rich phases in the as-prepared device probably arises from compositional inhomogeneity or interfacial reactions during film formation and processing 36 .

These secondary phases have been reported to introduce local strain, act as non-radiative recombination centres and destabilize adjacent grains, thereby impairing charge injection and electroluminescence stability under operational conditions 35 , 36 .

Given the low electron dose used during imaging (15.6 e − /Å −2 ), which is well below the reported damage thresholds (from 66 to 200 e − /Å −2 ) for halide perovskites, we consider beam-induced effects negligible 39 .

Therefore, we hypothesize that these strain features originate from intrinsic interface mismatch and compositional inhomogeneity in the as-fabricated device.

These pre-existing strain concentrations may serve as precursors to degradation pathways during operation, which we investigate in the following sections.

Structural and compositional evolution
To monitor the structural and compositional evolution of the active layers, a constant current of 0.2 nA was applied to the nanoLED, equivalent to a current density of 1 mA cm −2 for the pristine full device driven at 4 V.

Given the beam-sensitive nature of the emitter, low-dose 4D-STEM datasets (lower than 8 e − /Å −2 per frame) were acquired with a lateral spatial resolution of approximately 2 nm to produce nanoscale diffraction maps after 0, 3, 6, 15, 25 and 35 min of biasing.

Each time point was acquired from a distinct yet equivalent region to minimize cumulative beam damage (Supplementary Fig.

5 ).

Low-current 4D-STEM and EDX spectrum images were acquired from each region, after which the voltage and current were temporarily set to zero to systematically investigate structural, morphological and compositional evolution.

The evolution of the average driving voltage to maintain galvanostatic conditions versus time profile is shown in Fig.

3a .

A substantial nonlinear increase was observed throughout the experiment, indicating a progressive increase in device resistivity due to structural and compositional degradation.

Figure 3b presents summed one-dimensional radial electron diffraction from perovskite emitters, highlighting systematic peak broadening and intensity redistribution from 0 to 35 min biasing, indicative of increased lattice disorder due to accumulated structural defects and heterogeneity.

Pristine perovskite emitters ( t = 0 min) confirm retention of orthorhombic symmetry ( Pnma ) at room temperature, consistent with thermodynamically stabilized perovskite structures 40 .

Initial diffraction peaks correspond primarily to (121), (040) and (042) at 0.245, 0.343 and 0.425 Å −1 , aligning closely with [010] and [100] orientations.

Peaks at (040) and (042) shift towards larger d spacings in early biasing regions, reflecting electric-field-induced lattice distortions.

Intensive peaks belonging to Pb-rich phases, confirmed by a non-negative matrix factorization (NMF) decomposed diffraction pattern (Supplementary Fig.

6 ), are detected after 25 min biasing and shown as red spots.

Quantitative lattice parameter analysis together with the traces Δ a / a 0 , Δ b / b 0 , Δ c / c 0 ( t = 0 baseline) (Fig.

3c , Supplementary Fig.

7 and Supplementary Table 1 ) reveal a non-monotonic orthorhombic lattice evolution.

Both a and b axes reveal an initial expansion during early biasing stages, whereas c exhibits a net contraction before 25 min of biasing.

By 25 min, both b and c axes show a partial return towards pre-bias values, reflecting a local strain accommodation, potentially owing to defect migration or phase reorganization 41 , 42 .

At 35 min biasing, the a and c axes collapse, whereas b remains only weakly perturbed (around ±2%), signalling severe distortions in octahedra and loss of structural coherence.

This late-time structural deterioration coincides with the sharp increase in driving voltage, linking field-driven interfacial chemistry and mechanically mediated lattice failure to the observed electrical throttling.

Fig.

3: Multimodal analysis of structural and chemical degradation of a DCDH nanoLED under electrical biasing.

a , Plot of voltage versus time required to drive the nanoLED at a constant current of 0.2 nA, for each biasing condition.

b , Summed one-dimensional radial electron diffraction patterns for different biasing times, showing the main diffraction features under increasing biasing time at constant current.

The red dots highlight diffraction peaks corresponding to Pb-rich phases.

c , Calculated lattice parameters across all biasing conditions, revealing progressive distortion of the perovskite lattice.

The reported uncertainties (shading) correspond to one standard deviation derived from the covariance matrix of the refinement.

d , STEM-HAADF of the emitter cross-sections (pristine) from 0 to 35 min biasing, and corresponding vDF images taken after biasing from the same area.

Panels labelled ‘pristine’ are low-magnification HAADF context crops taken from a single parent overview of the lamella at t = 0; post-bias images at each time point were acquired once in fresh, distinct regions of interest (ROIs) to avoid cumulative beam dose.

e , The intensity–mass loss and cross-sectional area loss quantified by analysing changes in areas of emissive layer and local image intensity.

Cross-sectional area loss refers to the projected emitter area decrease after rigid registration; intensity (‘mass’) loss is HAADF-integrated intensity within the emitter mask, normalized to ETL as an internal reference to mitigate thickness and brightness drift.

f , Mean intragrain size and mean grain orientation spread (GOS), showing structural evolution with biasing.

Error bars represent the standard deviation of the intragrain distribution within each ROI.

g , Virtual bright-field (vBF) image of the nanoLED stack after 35 min biasing.

h – j , NMF components (factors and loadings), to show the spatial distribution of the original DCDH stoichiometry ( h ), and of phases that evolve during biasing, such as Pb-rich degradation products ( i ) and AlCl 3 ( j ).

Scale bars, 100 nm ( d , g ).

Source data Full size image
a , Plot of voltage versus time required to drive the nanoLED at a constant current of 0.2 nA, for each biasing condition.

b , Summed one-dimensional radial electron diffraction patterns for different biasing times, showing the main diffraction features under increasing biasing time at constant current.

The red dots highlight diffraction peaks corresponding to Pb-rich phases.

c , Calculated lattice parameters across all biasing conditions, revealing progressive distortion of the perovskite lattice.

The reported uncertainties (shading) correspond to one standard deviation derived from the covariance matrix of the refinement.

d , STEM-HAADF of the emitter cross-sections (pristine) from 0 to 35 min biasing, and corresponding vDF images taken after biasing from the same area.

Panels labelled ‘pristine’ are low-magnification HAADF context crops taken from a single parent overview of the lamella at t = 0; post-bias images at each time point were acquired once in fresh, distinct regions of interest (ROIs) to avoid cumulative beam dose.

e , The intensity–mass loss and cross-sectional area loss quantified by analysing changes in areas of emissive layer and local image intensity.

Cross-sectional area loss refers to the projected emitter area decrease after rigid registration; intensity (‘mass’) loss is HAADF-integrated intensity within the emitter mask, normalized to ETL as an internal reference to mitigate thickness and brightness drift.

f , Mean intragrain size and mean grain orientation spread (GOS), showing structural evolution with biasing.

Error bars represent the standard deviation of the intragrain distribution within each ROI.

g , Virtual bright-field (vBF) image of the nanoLED stack after 35 min biasing.

h – j , NMF components (factors and loadings), to show the spatial distribution of the original DCDH stoichiometry ( h ), and of phases that evolve during biasing, such as Pb-rich degradation products ( i ) and AlCl 3 ( j ).

Scale bars, 100 nm ( d , g ).

STEM-HAADF images of perovskite emitters in the six regions at pristine and corresponding virtual dark-field (vDF) images after biasing reveal progressive morphological changes (Fig.

3d ).

These contrast variations and clustered features reflect localized decomposition processes that become most pronounced after 25 min biasing, consistent with phase segregation and microstructural breakdown observed in the diffraction patterns.

We hypothesize that this degradation is primarily driven by ion migration and spatially inhomogeneous Joule heating, where local current densities and resistive hotspots may lead to differential thermal stresses at specific sites.

Although we do not directly measure Joule heating in this work, the correlation between increased resistivity, structural disruption and Pb-rich phase formation supports this mechanism as a contributing factor.

Cross-sectional area and intensity (mass) analysis of the emitter (Fig.

3e ) pre-biasing and post-biasing, show about 10.5% and 24.2% loss by 35 min biasing, indicating substantial volatilization and structural collapse.

The high mismatch of mass and area loss after 25 min of biasing suggests local densification, probably from increased Pb-rich phase formation or porous grain collapse.

Intragrain metrics (Fig.

3f ) support these structural observations.

The median intragrain size and grain orientation spread initially increase in early biasing ( t ≤ 6 min), then sharply decreases after 15 min biasing, indicative of grain fragmentation with partial structural accommodation due to increased defective phase formation (Supplementary Figs.

8 and 9 ).

Comparison of radial profiles between 0 min and 35 min biasing (Extended Data Fig.

2a ) reveals substantial metallic Pb formation after biasing (up to 6 V).

Fourier-filtered imaging (Extended Data Fig.

2b ) identifies metallic Pb nanoparticles with distinct (111) and (002) lattice spacings (0.294 nm and 0.250 nm, respectively), highlighting structural integrity loss.

The formation of metallic Pb is probably initiated by halide oxidation and subsequent electrochemical reduction of undercoordinated Pb 2+ .

This transformation marks advanced degradation, as metallic Pb introduces non-radiative recombination pathways that reduce electroluminescence efficiency 36 .

NMF components derived from STEM-EDX (Fig.

3g,i ) confirm Pb-rich phase formation after 35 min of biasing, further exacerbating structural and compositional instability.

Similar phase segregation is detected from 3 to 25 min biasing (Supplementary Fig.

10 ), demonstrating a progressive trend of Pb enrichment during in situ biasing measurements.

The PbBr 2 phase formation near the HTL indicates the retention and partial reorganization of Br − ions in regions, reinforcing halide segregation as a key driver of phase evolution and degradation 43 .

This bias-induced segregation provides direct structural evidence of electrochemical degradation mechanisms that undermine device stability and optoelectronic performance.

Despite the progressive compositional degradation, substantial portions of the orthorhombic perovskite persist in the bulk (Fig.

3h and Supplementary Fig.

10 ).

The electroluminescence diminishes markedly after 35 min, whereas photoluminescence shows less of a decline ratio (Supplementary Fig.

1a,c ), indicating that radiative recombination centres are largely intact.

Thus, degradation at interfaces between emitter and transport layers rather than bulk perovskite decomposition primarily accounts for electroluminescence loss, impeding charge transport and increasing device resistivity.

The aluminium contact layer (cathode) plays a pivotal role in facilitating electron injection into the perovskite emitter.

In the pristine device, minor Al 2 O 3 signals were detected (Extended Data Fig.

2c ), arising from ambient oxidation during fabrication and storage.

On biasing, new AlCl 3 diffraction peaks emerge, evidencing chemical interactions between aluminium and mobile Cl − ions (Supplementary Fig.

11 ).

This aligns with previous studies of halide-driven corrosion reactions in perovskite optoelectronics 43 , 44 , enabled by higher mobility of chloride ions owing to their smaller ionic radius and lower migration barrier (0.22 eV for Cl − , 0.47 eV for Br − ) 45 .

Correlative STEM-EDX NMF component analyses across all biased regions (Fig.

3j and Supplementary Fig.

10 ) confirm the first presence of the AlCl 3 layer after 6 min of biasing, increasing to around 10.8 nm after 35 min biasing.

The newly formed insulating AlCl 3 layer hinders electron injection, exacerbating Joule heating and accelerating both physical and chemical electrode corrosion.

Such substantial electrode deterioration contributes to localized electroluminescence quenching, observed as dark spots, and shifts the recombination zone towards the cathode 46 , 47 .

Additionally, continual halide ion consumption at the cathode interface destabilizes the perovskite emitter, driving further structural decomposition and local defect densities, and consequently reducing overall device efficiency and lifetime 48 .

To verify that lamella-biasing faithfully captures device-relevant processes, we performed ex situ STEM-EDX analysis on biased bulk PeLEDs (1 mA cm −2 , 15–45 min).

The same interfacial degradation products were observed, including insulating AlCl 3 at the cathode and lead-rich phases at the interfaces between the transport layers and the perovskite, whereas the emitter interior remained comparatively intact.

These post-mortem observations corroborate that the dominant degradation mechanisms are intrinsic and largely geometry independent (Supplementary Figs.

12 – 15 ), highlighting chloride migration and electrode corrosion as key failure mechanisms and underscoring the need for robust interfacial stabilization strategies.

Interfacial degradation under bias
To investigate the evolution of the interfaces between the perovskite and transport layers, we summed diffraction patterns over equivalent areas (150 pixels) of ETL/perovskite and HTL/perovskite across all biasing times.

The summed patterns for pristine and 35 min biasing are shown in Fig.

4a–d .

Over time under bias, the interface of the ETL/emitter loses the orthorhombic CsPbX 3 , instead forming more lead-rich phases, primarily composed of CsPb 2 X 5 and PbX 2 , as can be seen after 35 min under bias.

These defective phases can act as non-radiative recombination centres, reducing device performance.

The diminishing intensity of diffraction spots indicates reduced crystallinity and lattice alignment, which exacerbate charge transport issues (Extended Data Fig.

3 ).

A schematic representation of the interfacial degradation mechanisms under biasing is presented in Fig.

4e,f .

Similarly, the emitter/HTL interface, which initially exhibits high crystallinity with orthorhombic CsPbX 3 without biasing, displays peaks associated with CsPb 2 X 5 and PbX 2 after 35 min of biasing, accompanied by a substantial drop in diffraction intensity.

Further STEM-EDX line spectrum analysis of ETL and HTL interfaces confirms increased lead-rich compositions after biasing compared with the pristine region (Extended Data Fig.

3e–h ).

The HTL, composed of PVK, is prone to degradation owing to its low hole mobility and energetic mismatch with the perovskite emitter.

The small amount of PbX 2 observed at the HTL interface in the pristine device (Fig.

4b ) probably originates from crystallization dynamics during spin-coating.

Under bias, inefficient hole injection and charge accumulation at the HTL interface intensify local electric fields and may promote halide ion redistribution, oxidation (particularly of Cl − ) or volatilization 17 , 37 , 49 .

These processes collectively lead to halide depletion and stoichiometric imbalance, facilitating the formation of Pb-rich phases (Fig.

4d ).

Localized Joule heating exacerbates thermal decomposition, accelerating the formation of degradation products.

Similarly, the ETL material (TPBi) exhibits interfacial chemical and thermal instability, assisting ion migration and chemical reactions, thereby accelerating degradation at the interface (Fig.

4c ).

Combining these observations with evidence of interfacial lattice strain in pristine conditions confirms that interfacial instability during biasing is a primary factor affecting the stability of mixed Br/Cl PeLEDs.

Fig.

4: Interfacial degradation at transport layer–perovskite emitter interfaces under bias.

a – d , Summed diffraction patterns (150 pixels) at the ETL/emitter and HTL/emitter interfaces under pristine conditions ( t bias = 0 min) and after 35 min under bias.

Red labels denote diffraction features assigned to PbX 2 ; yellow labels indicate reflections matched to the pristine orthorhombic perovskite; green labels denote CsPb 2 X 5 .

a , b , Show a good match to the orthorhombic perovskite emitter near the ETL ( a ) and evidence of PbX 2 formation near the HTL ( b ).

c , d , Pb-rich phases (CsPb 2 X 5 , PbX 2 ) emerge at the ETL/emitter interface ( c ) and the emitter/HTL interface ( d ) after 35 min of biasing, which remains oriented along the [100] zone axis.

e , f , Schematic illustrations show the pristine state of the interfaces with emitter at 0 min of biasing ( e ) and progressive phase transformations at the interfaces with the emitter after 35 min of biasing ( f ).

The dashed lines represent the interfaces of ETL/emitter and emitter/HTL.

At the pristine state, the interfaces show a small amount of the Pb-rich regions, which are negligible compared with the biased region.

Scale bars, 2 nm −1 .

Full size image
a – d , Summed diffraction patterns (150 pixels) at the ETL/emitter and HTL/emitter interfaces under pristine conditions ( t bias = 0 min) and after 35 min under bias.

Red labels denote diffraction features assigned to PbX 2 ; yellow labels indicate reflections matched to the pristine orthorhombic perovskite; green labels denote CsPb 2 X 5 .

a , b , Show a good match to the orthorhombic perovskite emitter near the ETL ( a ) and evidence of PbX 2 formation near the HTL ( b ).

c , d , Pb-rich phases (CsPb 2 X 5 , PbX 2 ) emerge at the ETL/emitter interface ( c ) and the emitter/HTL interface ( d ) after 35 min of biasing, which remains oriented along the [100] zone axis.

e , f , Schematic illustrations show the pristine state of the interfaces with emitter at 0 min of biasing ( e ) and progressive phase transformations at the interfaces with the emitter after 35 min of biasing ( f ).

The dashed lines represent the interfaces of ETL/emitter and emitter/HTL.

At the pristine state, the interfaces show a small amount of the Pb-rich regions, which are negligible compared with the biased region.

Scale bars, 2 nm −1 .

Our study demonstrates that coupled structural, chemical and electrochemical processes localized at buried interfaces govern the operational instability of perovskite LEDs.

Multimodal in situ electron microscopy enables direct visualization of the evolution of interface chemistry, lattice strain and phase transformations, establishing that device failure is dictated by interfacial reactions rather than changes in emissive states.

The emergence of bias-induced Al corrosion and halide redistribution further underscores the vulnerability of multilayer device architectures to field-driven transformations.

By resolving these pathways with high spatial and temporal fidelity, this work identifies interfacial strain management, suppression of ion-mediated reactions and stabilization of metal contacts as central principles for enhancing device durability.

More broadly, the methodological framework advanced here provides a rigorous operando platform for interrogating transformation processes in functional optoelectronic systems, thereby guiding the rational development of robust perovskite emitters and supporting progress towards technologically viable next-generation optoelectronic devices.

Cesium bromide (CsBr, 99.999%), lead bromide (PbBr 2 , 99.999%), lead chloride (PbCl 2 , 99.999%), PVP (Mw 60,000), 4,7,10-trioxa-1,13-tridecanediamine (TTDDA, 97%), PVK (Mn 25,000–50,000), lithium fluoride (LiF, 99.99%), dimethyl sulfoxide (DMSO, 99.9%), N , N -dimethylformamide (DMF, 99.8%), chlorobenzene (CB, 99.8%), 2-propanol (IPA, 99.5%) were purchased from Sigma Aldrich.

Formamidinium bromide (FABr) was purchased from Greatcell Solar Materials.

2,2′,2′′-(1,3,5-Benzinetriyl)-tris(1-phenyl-1 H -benzimidazole) (TPBi) and PTAA were purchased from Ossila.

Preparation of the perovskite solution
Perovskite precursors (CsBr:FABr:PbBr 2 :PbCl 2 :TTDDA ratio of 1.2:0.3:0.3875:0.6125:0.1) were dissolved in DMSO with a Pb concentration of 0.1 M.

Precursor solutions were stirred at 65 °C before spin-coating.

ITO substrates were cleaned with deionized water, acetone and isopropanol for 15 min, followed by an ultraviolet ozone treatment for 20 min and then were transferred into a nitrogen-filled glovebox.

PTAA and PVK were deposited in sequence on the ITO substrates and annealed at 120 °C and 100 °C for 10 min, respectively.

Then, a thin layer of PVP was deposited and annealed at 80 °C for 10 min.

All the above-mentioned HTLs were spin-coated at a speed of 4,000 rpm for 30 s.

Finally, the perovskite precursor was spin-coated at 4,000 rpm for 40 s.

Immediately after spin-coating, the films were transferred into an enclosed petri dish with 20 μl DMF.

After DMF treatment for a certain time, the perovskite films were annealed at 80 °C for 8 min.

Then, all the substrates were loaded into an evaporator with a pressure of 10 −6 mbar for evaporation.

Then, 50 nm TPBi, 1.5 nm LiF and 100 nm Al were deposited with shadow masks via thermal evaporation.

The active area of the LEDs is 4.5 mm 2 .

PeLED device current density–voltage–luminance measurements were carried out in atmosphere with encapsulation.

The current density–voltage–luminance characteristics were measured using a Keithley 2400 source meter and a calibrated silicon photodetector for luminance calibration.

The LED metrics were calculated considering the responsivity of the silicon photodetector.

The electroluminescence spectra were obtained using an Ocean Optics Flame spectrometer.

Photoluminescence was measured using an Edinburgh FLS1000 spectrometer with an excitation wavelength at 405 nm.

The spectrum was captured with a step of 1 nm and an exposure time of 0.2 s at each wavelength.

The TEM lamella cross-section was lifted out and extracted from a conventional sky-blue PeLED with a device architecture of ITO (150 nm)/PTAA (10 nm)/PVK (10 nm)/PVP (2 nm)/Cs 1.2 FA 0.3 Pb(Br 0.65 Cl 0.35 ) 3 layer (40 nm)/TPBi (50 nm)/LiF (1.5 nm)/Al (100 nm) using an FEI Helios Nanolab Dualbeam FIB/SEM following improved protocols, with step-down milling (16–8 kV) and a final 2–1 kV polish to minimize implantation and amorphization, achieving a thickness around 100 nm (Supplementary Table 2 ).

The lamella was then transferred to a MEMS system in situ biasing HennyZ chip, as shown in Supplementary Fig.

2 .

Only two of the eight contacts were connected to the Al cathode and ITO anode with the Pt column during the FIB used for the in situ biasing experiments.

The HennyZ device was then connected to a Keithley power supply for operation at constant current.

During scanning electron diffraction (SED) microscopy, a two-dimensional (2D) electron diffraction pattern was measured at every probe position of an electron beam in STEM mode.

SED data were acquired on the JEOL ARM300CF E02 instrument at ePSIC (Diamond Light Source).

A monolithic Merlin/Medipix direct electron detector was used to acquire fast, low-dose SED.

These direct electron detectors allow for better signal-to-noise ratio under lower doses owing to the superior quantum efficiency compared to traditional charge-coupled devices.

The detector was set to 6-bit, to maintain the targeted electron fluence and fast acquisition readout rate.

An acceleration voltage of 200 keV, nanobeam alignment (convergence semi-angle of 1 mrad), electron probe of 5 nm, probe current at 1.3 pA, scan dwell time 0.6 ms and camera length of 20 cm were used.

An estimated electron fluence of lower than 8 e − /Å −2 per frame was acquired when approximating the beam cross-section as a disk.

This accumulated dose is lower than the previously reported damage threshold for hybrid perovskite compositions 39 , 50 .

To verify that cumulative beam effects were negligible, we performed a repeat-acquisition (beam-only) control on a fixed field of view, which showed small Bragg-intensity attenuation without measurable Δ d / d shifts (full details are provided in Supplementary Fig.

16 ).

Post-processing of SED diffraction data was done using py4DSTEM v.0.14.14 (an open-source set of Python tools for processing and analysis of 4D-STEM data) 51 .

Virtual bright-field (vBF) images were referred as the images reconstructed from taking the intensity integrated solely from the direct beam as a function of probe position, thus only containing information from the electrons recorded at zero scattering angle.

Virtual dark-field (vDF) images were reconstructed from only taking the intensity from a Bragg-diffracted spot from the 2D diffraction pattern as a function of probe position, thus containing information solely on electrons scattered to specific Bragg angles.

All diffraction patterns were distortion corrected and calibrated with an Au cross-grating.

Dead pixels and detector junctions were masked.

No sample drift correction was necessary.

Finally, the background noise in all diffraction patterns was filtered, setting a threshold of detector pixels with single counts to zero.

Crystal orientation mapping was performed on 4D-STEM datasets using the py4DSTEM software package 51 .

Orientation indexing was achieved by template matching to a library of simulated electron diffraction patterns based on an orthorhombic perovskite structure (space group P nma , point group mmm ), yielding Euler angles ( Φ 1 , Φ , Φ 2 ) at each scan position in the Bunge convention.

Grains were segmented by clustering pixels with misorientation angles below 2°, consistent with perovskite grain boundary criteria.

Grain size was defined as the equivalent circular diameter of each segmented region.

Intragranular disorder was quantified using the grain orientation spread, calculated as the mean misorientation between individual pixels and the average grain orientation, with quaternion-based averaging applied 52 , 53 .

All analyses accounted for orthorhombic symmetry operations.

A parent low-magnification HAADF overview was recorded at t = 0 to register ROI positions (Supplementary Fig.

5 ).

All subsequent time-point datasets were acquired in non-overlapping, adjacent ROIs without revisiting earlier ROIs.

Mass loss in the emissive perovskite layer was evaluated by comparing pre-bias, low-dose HAADF-STEM and post-bias, vDF images from matched regions.

Image alignment was achieved via rigid-body registration, and pixel-wise intensity differences were normalized to a structurally stable reference region (typically the ETL).

Under the assumption of uniform projection, local intensity change was used as a proxy for volumetric and compositional variations.

A Gaussian smoothing filter was applied before segmentation.

Control measurements in non-biased areas confirmed reproducibility within ±2%.

All diffraction features were indexed to an orthorhombic perovskite phase with lattice constants a = 8.45 Å, b = 11.88 Å and c = 8.10 Å.

While the emitter composition was Cs 1.2 FA 0.3 Pb(Br 0.65 Cl 0.35 ) 3 , the complex alloyed structure was approximated using the dominant CsPbBr 3 sublattice with empirically adjusted parameters for diffraction indexing.

STEM-HAADF image and STEM-EDX spectrum
HAADF imaging was acquired before 4D-STEM acquisition using the JEOL ARM300CF E02 in nanobeam alignment using an annular dark-field detector at camera length of 20 cm.

The collection angle for HAADF was between 74.3 ± 0.3 and 233.3 ± 1.5 mrad.

Energy-dispersive X-ray spectroscopy (STEM-EDX) maps were acquired using an Oxford Instruments XMAX 100 detector with 10 eV per channel spectral resolution and around 2.4-nm spatial sampling.

A beam current of 50 pA and 0.5-ms dwell time per pixel were used.

The STEM-EDX data were denoised using principal component analysis and non-matrix factorization with HyperSpy v.2.0 (ref.

54 ).

Atomic-resolution HAADF-STEM images were acquired ex situ before and after biasing using a 14-mrad convergence angle and 1.6-pA current to minimize beam damage.

The data underlying all figures in the main text are publicly available from the University of Cambridge repository at https://doi.org/10.17863/CAM.124015 .

Source data are provided with this paper.

Xiong, W.

et al.

Controllable p- and n-type behaviours in emissive perovskite semiconductors.

Nature 633 , 344–350 (2024).

Article ADS CAS PubMed Google Scholar
Li, M.

et al.

Acceleration of radiative recombination for efficient perovskite LEDs.

Nature 630 , 631–635 (2024).

Article ADS CAS PubMed PubMed Central Google Scholar
Peng, C.

et al.

Weakly space-confined all-inorganic perovskites for light-emitting diodes.

Nature 643 , 96–103 (2025).

Veldhuis, S.

A.

et al.

Perovskite materials for light-emitting diodes and lasers.

Adv.

Mater.

28 , 6804–6834 (2016).

Article CAS PubMed Google Scholar
Burschka, J.

et al.

Sequential deposition as a route to high-performance perovskite-sensitized solar cells.

Nature 499 , 316–319 (2013).

Tan, Z.

K.

et al.

Bright light-emitting diodes based on organometal halide perovskite.

Nat.

Nanotechnol.

9 , 687–692 (2014).

Xu, W.

et al.

Rational molecular passivation for high-performance perovskite light-emitting diodes.

Nat.

Photonics 13 , 418–424 (2019).

Fang, Y., Dong, Q., Shao, Y., Yuan, Y.

& Huang, J.

Highly narrowband perovskite single-crystal photodetectors enabled by surface-charge recombination.

Nat.

Photonics 9 , 679–686 (2015).

Liu, T.

et al.

High responsivity circular polarized light detectors based on quasi two-dimensional chiral perovskite films.

ACS Nano 16 , 2682–2689 (2022).

Article CAS PubMed PubMed Central Google Scholar
Elkhouly, K.

et al.

Electrically assisted amplified spontaneous emission in perovskite light-emitting diodes.

Nat.

Photonics 18 , 132–138 (2024).

Zou, C.

et al.

Electrically driven lasing from a dual-cavity perovskite device.

Nature 645 , 369–374 (2025).

Zhao, X.

et al.

Operationally stable perovskite solar modules enabled by vapor-phase fluoride treatment.

Science 385 , 433–438 (2024).

Bracher, C., Freestone, B.

G., Mohamad, D.

K., Smith, J.

A.

& Lidzey, D.

G.

Degradation of inverted architecture CH 3 NH 3 PbI 3− x Cl x perovskite solar cells due to trapped moisture.

Energy Sci.

Eng.

6 , 35–46 (2018).

Watanabe, T., Yamanari, T.

& Marumoto, K.

Deterioration mechanism of perovskite solar cells by operando observation of spin states.

Commun.

Mater.

1 , 1–9 (2020).

Zohar, A.

et al.

In operando, photovoltaic, and microscopic evaluation of recombination centers in halide perovskite-based solar cells.

ACS Appl.

Mater.

Interfaces 14 , 34171–34179 (2021).

Ahmad, M.

I.

et al.

Rapid thermal processing chamber for in-situ X-ray diffraction.

Rev.

Sci.

Instrum.

86 , 13902 (2015).

Mundt, L.

E.

& Schelhas, L.

T.

Structural evolution during perovskite crystal formation and degradation: in situ and operando X-ray diffraction studies.

Adv.

Energy Mater.

10 , 1903074 (2020).

Johannes, A.

et al.

In operando X-ray imaging of nanoscale devices: composition, valence, and internal electrical fields.

Sci.

Adv.

3 , eaao4044 (2017).

Article PubMed PubMed Central Google Scholar
Kosasih, F.

U.

& Ducati, C.

Characterising degradation of perovskite solar cells through in-situ and operando electron microscopy.

Nano Energy 47 , 243–256 (2018).

Dellby, N.

et al.

Ultra-high resolution EELS analysis and STEM imaging at 20 keV.

Microsc.

Microanal.

29 , 626–627 (2023).

Ishikawa, R.

et al.

High spatiotemporal resolution STEM imaging at high temperature.

Microsc.

Microanal.

30 , 1397–1398 (2024).

Jeangros, Q.

et al.

In situ TEM analysis of organic–inorganic metal-halide perovskite solar cells under electrical bias.

Nano Lett.

16 , 7013–7018 (2016).

Joon Jung, H.

et al.

Stability of halide perovskite solar cell devices: in situ observation of oxygen diffusion under biasing.

Adv.

Mater.

30 , 1802769 (2018).

Kim, M.

C.

et al.

Imaging real-time amorphization of hybrid perovskite solar cells under electrical biasing.

ACS Energy Lett.

6 , 3530–3537 (2021).

Karlsson, M.

et al.

Mixed halide perovskites for spectrally stable and high-efficiency blue light-emitting diodes.

Nat.

Commun.

12 , 361 (2021).

Tyukalova, E.

et al.

Challenges and applications to operando and in situ TEM imaging and spectroscopic capabilities in a cryogenic temperature range.

Acc.

Chem.

Res.

54 , 3125–3135 (2021).

Tennyson, E.

M., Doherty, T.

A.

S.

& Stranks, S.

D.

Heterogeneity at multiple length scales in halide perovskite semiconductors.

Nat.

Rev.

Mater.

4 , 573–587 (2019).

Dommann, A.

& Neels, A.

The role of strain in new semiconductor devices.

Adv.

Eng.

Mater.

11 , 275–277 (2009).

Liang, Z.

et al.

Homogenizing out-of-plane cation composition in perovskite solar cells.

Nature 624 , 557–563 (2023).

Wu, D., Xie, H., Dai, X.

& Wang, R.

A novel method to fabricate micro-gratings applied for deformation measurement around a crack in a thin film.

Meas.

Sci.

Technol.

25 , 025012 (2014).

Liu, Z.

et al.

Synthesis of multi-walled carbon nanotube–hydroxyapatite composites and its application in the sorption of Co(II) from aqueous solutions.

J.

Mol.

Liq.

179 , 46–53 (2013).

Zhang, H., Liu, Z., Wen, H., Xie, H.

& Liu, C.

Subset geometric phase analysis method for deformation evaluation of HRTEM images.

Ultramicroscopy 171 , 34–42 (2016).

Hÿtch, M.

J., Snoeck, E.

& Kilaas, R.

Quantitative measurement of displacement and strain fields from HREM micrographs.

Ultramicroscopy 74 , 131–146 (1998).

Hÿtch, M.

J.

& Minor, A.

M.

Observing and measuring strain in nanostructures and devices with transmission electron microscopy.

MRS Bull.

39 , 138–146 (2014).

Shen, X.

et al.

Passivation strategies for mitigating defect challenges in halide perovskite light-emitting diodes.

Joule 7 , 272–308 (2023).

Liang, J.

et al.

Origins and influences of metallic lead in perovskite solar cells.

Joule 6 , 816–833 (2022).

Baumann, F., Raga, S.

R.

& Lira-Cantú, M.

Monitoring the stability and degradation mechanisms of perovskite solar cells by in situ and operando characterization.

APL Energy 1 , 011501 (2023).

Jariwala, S.

et al.

Local crystal misorientation influences non-radiative recombination in halide perovskites.

Joule 3 , 3048–3060 (2019).

Ferrer Orri, J.

et al.

Unveiling the interaction mechanisms of electron and X-ray radiation with halide perovskite semiconductors using scanning nanoprobe diffraction.

Adv.

Mater.

34 , 2200383 (2022).

Akkerman, Q.

A., Rainò, G., Kovalenko, M.

V.

& Manna, L.

Genesis, challenges and opportunities for colloidal lead halide perovskite nanocrystals.

Nat.

Mater.

17 , 394–405 (2018).

Wang, H.

et al.

Impacts of the lattice strain on perovskite light-emitting diodes.

Adv.

Energy Mater.

13 , 2202185 (2023).

Orr, K.

W.

P.

et al.

Imaging light-induced migration of dislocations in halide perovskites with 3D nanoscale strain mapping.

Adv.

Mater.

35 , 2305549 (2023).

Ji, K.

et al.

Self-supervised deep learning for tracking degradation of perovskite light-emitting diodes with multispectral imaging.

Nat.

Mach.

Intell.

5 , 1225–1235 (2023).

Li, X.

et al.

Chemical anti-corrosion strategy for stable inverted perovskite solar cells.

Sci.

Adv.

6 , eabd1580 (2020).

Karlsson, M.

et al.

Role of chloride on the instability of blue emitting mixed-halide perovskites.

Front.

Optoelectron.

16 , 37 (2023).

Vashishtha, P.

& Halpert, J.

E.

Field-driven ion migration and color instability in red-emitting mixed halide perovskite nanocrystal light-emitting diodes.

Chem.

Mater.

29 , 5965–5973 (2017).

Rong, Y.

et al.

Challenges for commercializing perovskite solar cells.

Science 361 , eaat8235 (2018).

Back, H.

et al.

Achieving long-term stable perovskite solar cells via ion neutralization.

Energy Environ.

Sci.

9 , 1258–1263 (2016).

Yang, S.

J., Lee, Y.

H., Tateno, K.

& Dou, L.

Elemental segregation and dimensional separation in halide perovskite light-emitting diodes.

Prog.

Quantum Electron.

98 , 100537 (2024).

Chen, S.

et al.

Atomic scale insights into structure instability and decomposition pathway of methylammonium lead iodide perovskite.

Nat.

Commun.

9 , 4807 (2018).

Article ADS PubMed PubMed Central Google Scholar
Savitzky, B.

H.

et al.

py4DSTEM: a software package for four-dimensional scanning transmission electron microscopy data analysis.

Microsc.

Microanal.

27 , 712–743 (2021).

Bridger, A., David, W.

I.

F., Wood, T.

J., Danaie, M.

& Butler, K.

T.

Versatile domain mapping of scanning electron nanobeam diffraction datasets utilising variational autoencoders.

NPJ Comput.

Mater.

9 , 14 (2023).

Thronsen, E.

et al.

Scanning precession electron diffraction data analysis approaches for phase mapping of precipitates in aluminium alloys.

Ultramicroscopy 255 , 113861 (2024).

de la Peña, F.

et al.

hyperspy/hyperspy: v2.4.0.

Zenodo https://doi.org/10.5281/zenodo.18379337 (2026).

We acknowledge the support of the Wolfson Electron Microscopy Suite and the use of Thermo Fisher Spectra 300 TEM, funded by EPSRC grant EP/R008779/1.

We thank the Diamond Light Source for access and support in using the Electron Physical Science Imaging Centre (Instrument E02 and proposal number MG37292), which contributed to the SED results presented here.

We acknowledge M.

Danaie and C.

Allen for their assistance in the in situ biasing experiment.

We acknowledge F.

Su for his assistance in decomposition of the SED dataset.

This work was supported by the Henry Royce Institute for Advanced Materials through the Equipment Access Scheme, which enabled access to the FIB and TEM at Cambridge; Cambridge Royce facilities grant CAM-YR8-UI-042-REAS.

Schematic figures were created using Figdraw.

This work was supported by EPSRC EP/V06164X/1 (R.H.F., S.D.S., T.L.) and the Royal Society and Tata Group UF150033 and URF\R\221026 (S.D.S.).

T.L.

acknowledges the funding from Leverhulme Trust Early Career fellowship (ECF-2024-217).

Department of Materials Science and Metallurgy, University of Cambridge, Cambridge, UK
Xinjuan Li, Wei Huang, Simon M.

Fairclough & Caterina Ducati
Department of Chemical Engineering and Biotechnology, University of Cambridge, Cambridge, UK
Cavendish Laboratory, University of Cambridge, Cambridge, UK
Richard H.

Friend, Samuel D.

Stranks & Tianjun Liu
Search author on: PubMed Google Scholar
X.L.

conceptualized the study and designed the in situ biasing methodology under the supervision of C.D.

Q.G.

fabricated the perovskite thin films and LEDs, characterized device performance and performed electroluminescence and photoluminescence stability measurements under the supervision of S.D.S.

X.L.

and S.M.F.

prepared the TEM lamellae from a bulk PeLED and incorporated in the MEMS biasing chip.

X.L.

and W.H.

performed in situ and ex situ biasing data acquisitions in the Electron Physical Science Imaging Centre, Diamond Light Source and in Cambridge.

X.L.

analysed and interpreted all in situ biasing datasets under the supervision of C.D.

T.L.

and R.H.F.

provided feedback on data interpretation.

T.L.

performed current-density-dependent device stability and electroluminescence measurements under the supervision of R.H.F.

X.L.

wrote the manuscript with C.D.

All authors discussed the results and contributed to the substantive revision of the manuscript.

Correspondence to Tianjun Liu or Caterina Ducati .

The authors declare no competing interests.

Nature thanks Antonio Guerrero, Sonia Ruiz Raga, Ignacio Sanjuan and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data figures and tables
Extended Data Fig.

1 Complementary shear strain components and intrinsic interfacial degradation pathways.

a–c , The Geometric Phase Analysis (GPA) in shear direction reveals pristine deformed strains at interfaces and main emitter, indicating mechanisms induced interfacial degradation.

The region detected is consistent with areas present in Fig.

2 .

d , Schematic illustration of intrinsic degradation pathways in ETL/perovskite emitter/HTL systems, including formation of Pb-rich phases, point defects, interfacial chemical attacks, and lattice strain.

Extended Data Fig.

2 Device operation induces phase changes in the perovskite emitter and Al cathode.

a , Summed 1D radial diffraction profiles from the perovskite emitter before and after 35 min of operation, showing the emergence of metallic Pb peaks at ( \(\bar{11}1\) ), (002), (311), (400), and (133).

b , Atomic-resolution STEM-HAADF image of Pb 0 nanoparticles located at grain boundaries after biasing, with corresponding FFT confirming the metallic phase.

Scale bar, 1 nm.

c , Summed radial diffraction profiles of pristine and biased Al cathodes, indicating the formation of AlCl 3 via ion migration, with characteristic peaks at ( \(\bar{2}20\) ), (003), and ( \(\bar{33}3\) ).

Extended Data Fig.

3 Interfacial structural and compositional evolution under galvanostatic bias.

a–d , Summed diffraction patterns (integration over 150 pixels) from the ETL (a,c) and HTL (b,d) interfaces at successive biasing times (t = 3, 6, 15, 25 min), progressive structural changes with increasing bias duration.

Colour legend, red labels denote features assigned to PbX 2 ; yellow tick marks indicate matched planes of the pristine orthorhombic perovskite; green labels denote CsPb 2 X 5 .

Patterns at each time point were acquired in fresh, non-overlapping ROIs to avoid cumulative dose history; intensities are normalised to the elastic background for comparison.

e , g , STEM-EDS line profiles across the ETL (e) and HTL (g) interfaces at t = 0 min, establishing baseline elemental distributions in the pristine state.

f , h , Corresponding profiles at t = 35 min, showing enhanced Pb-rich signal near both interfaces, consistent with the emergence/growth of Pb-rich secondary phases under extended operation.

Supplementary Information (download DOCX )
Supplementary Figs.

1–16, Tables 1 and 2, and Notes 1–3.

Source Data Fig.

1 (download XLSX )
Source Data Fig.

3 (download XLSX )
Version of record : 11 March 2026
DOI : https://doi.org/10.1038/s41586-026-10238-8