In situ TEM unveils the role of residual local strain on light-induced phase segregation in halide perovskites

Abstract

Deciphering the mechanisms governing photoinduced phase segregation in mixed halide perovskites is essential to unlock their full potential in stable, high-performance optoelectronic applications. We uncover the mechanism by which residual local strain acts as a key driving force of light-induced phase segregation. By combining in situ transmission electron microscopy with photoluminescence spectroscopy, we observe structural evolution and photocarrier behavior during phase segregation and after re-mixing. Although halide segregation is compositionally reversible, the perovskite lattice retains residual local strain, a “memory” of previously segregated halide domains, which evolves spatiotemporally with each phase segregation cycle. Residual local strain subsequently serves as a driver for successive phase segregation by trapping photocarriers and acts as the nucleation sites for iodide-rich domains. Our findings identify local strain as an intrinsic, evolving driving force of phase segregation, which offers a paradigm for improving the long-term stability of halide perovskites through strain management and compositional engineering.

The evolving local strain in halide perovskite lattices serves as the driving force of light-induced phase segregation.

INTRODUCTION

Mixed halide perovskites have emerged as promising candidates for next-generation optoelectronic applications due to their tunable bandgaps, high absorption coefficients, and long carrier diffusion lengths. However, their commercial viability remains hindered by the intrinsic instability caused by photoinduced phase segregation, wherein the homogeneous mixed-halide phase spontaneously separates into iodide-rich and bromide-rich domains under illumination or electrical bias (1, 2). This phenomenon leads to a redshift in the absorption edge and degrades device performance, particularly in tandem solar cells and light-emitting diodes (3).

Phase segregation in mixed halide perovskites stems from a complex structural and chemical landscape involving the interplay of compositions, photocarriers, lattice dynamics, halide ion migration, and defects. Thermodynamically, compositions (4), photocarriers (5), and lattice dynamics (6) influence the free energy of the system, thereby affecting the driving forces for phase segregation. Kinetically, the type and concentration of defects (7, 8), as well as the ion migration barrier (9), influence the rate of phase segregation. To explain this phenomenon, multiple mechanisms have been proposed, each capturing different aspects. Thermodynamic models attribute phase segregation to photocarrier-induced free energy minimization, whereby charge carriers preferentially accumulate in lower-bandgap iodide-rich regions (5, 10–12). In the absence of illumination, the mixed halide phase is thermodynamically stable due to favorable entropic mixing. The polaron model introduces strain gradients as the driving force for carrier funneling and halide migration, with strain relaxation facilitating halide mixing (6, 13, 14). More recently, the halide oxidation model has implicated photooxidative processes in generating mobile, neutral halogen species that disrupt local halide stoichiometry and initiate compositional phase segregation (15–17). Although these models offer critical insights, they are primarily focused on the forward phase segregation process and assume that the process is fully reversible. Many of the morphological and spectroscopic phenomena suggest possible nonreversible structural behavior that cannot be fully attributed to material degradation (18).

A wide range of in situ characterization techniques, including photoluminescence (PL) spectroscopy (1, 5, 19–21) and imaging (6, 17, 22–26), ultraviolet-visible absorption (27–30), ellipsometry (31), and cathodoluminescence (13, 31), have been used to study light-induced halide phase segregation and recovery. These methods have revealed rich spatial and temporal halide migration behaviors and their dependence on temperature, illumination intensity, and environment. In situ PL measurements (32, 33) also provide insight into the evolution of surface defects and their influence on phase segregation. However, these techniques fundamentally lack access to structural information. Such structural insight is crucial because many open questions, such as cycle-dependent behavior and the appearance of cracks, stem from the absence of a microscopic structural picture of phase segregation models. X-ray–based probes provide structural information but lack the spatial resolution needed to capture microscale compositional redistribution during phase segregation. Transmission electron microscopy (TEM) offers access to local structural information via selected-area electron diffraction (SAED), but its application to in situ phase segregation studies has been limited by the challenges of performing controlled illumination inside the microscope and by the beam sensitivity of halide perovskites. Therefore, a physical picture of the phase segregation and its reversal process at the microscopic structural level has yet to be established.

Here, we discover that although phase segregation is compositionally reversible, the perovskite lattice undergoes some irreversible changes. Specifically, the phase segregation process leaves behind residual local strain, which persists after halide remixing. We further uncover the mechanism by which the perovskite lattices influence phase segregation, where residual local strain facilitates subsequent phase segregation by serving as preferential sites for iodide-rich domain formation. Interface regions of local strain trap photocarriers, thereby enhancing local carrier densities and accelerating subsequent phase segregation.

Our observation was enabled by a multimodal characterization technique combining optical in situ TEM and PL spectroscopy. Through careful calibration and control of the total electron beam dose, the phase segregation environment, sample thickness, and photocarrier density, we were able to investigate the structural changes and behavior of photocarriers associated with phase segregation with both temporal and spatial resolution. Our findings establish local strain as both an irreversible structural remnant of phase segregation and an intrinsic driving force of phase segregation. Our study provides a framework to understand the interplay between the perovskite lattices, the halide ions, and photocarriers that influences phase segregation during and after its occurrence, as well as over multiple phase segregation cycles. These insights suggest that managing residual local strain through tailored compositional design and intermittent strain relaxation techniques to fully reverse phase segregation may offer effective pathways to enhance the stability of mixed halide perovskites.

RESULTS

Reversible structural evolution during phase segregation

In situ TEM combined with PL spectroscopy was used to track structural changes during light-induced halide perovskite phase segregation and subsequent remixing. The [1$\overline{1}$0]-projected SAED of the as-prepared film (Fig. 1A) confirmed an orthorhombic structure (Pbnm) (details are described in Supplementary Text and fig. S1). As halide perovskites are sensitive to electron beam irradiation, all TEM experiments were conducted within a total electron dose threshold of ~403.2 e⁻ Å⁻² for a total time of 112 min to avoid beam-induced degradation. Structural evolution during three phase segregation cycles was captured by SAED patterns (probe diameter of 2 μm), as shown in Fig. 1 (B to D) and fig. S3C. Compared with the SAED of the sample in mixed phase (Fig. 1A), we observed diffraction spot splitting (Fig. 1, B and C, and fig. S3), elongation (Fig. 1C), and the appearance of new diffraction spots (Fig. 1D). Reversible structural evolution of phase segregation is evidenced by the emergence and disappearance of these diffraction features under repeated illumination and dark recovery cycles (figs. S3C and S4).

Fig. 1. Light-induced phase segregation behavior of CsPbBr_2.1I_0.9 films.

(A) The [1$\overline{1}$0]-projected SAED pattern before illumination. (B to D) SAED patterns during illumination. Scale bar, 1 nm⁻¹. The angle θ between the solid lines indicates the angle between the (110) and (002) crystal planes. The white lines in (B) and (C) mark the direction of diffraction spot splitting. Insets in (B) and (C) show enlarged images of the splitted ($\overline{1}\overline{1}$1) and (111) spot. The red circles in (D) highlight newly appeared diffraction spots. (E) Evolution of interplanar spacings d₀₀₂, d₁₁₀, $d_{\overline{11}1}$, and d₁₁₁ during three cycles of phase segregation. The inset in the middle illustrates inter-octahedral distortion of the lead sublattice observed in cycles I and II. The right inset illustrates lead sublattices with 90° rotational twinning along [110] observed in cycle III, where the green line indicates a twin boundary on the (001)/(1$\overline{1}$0) plane. (F) Schematic diagrams showing the distribution of the mixed phase, iodide-rich phase, and bromide-rich phase before and during phase segregation in the three-cycle experiment.

Structure factor analysis revealed that the (110) and (002) diffraction spots primarily originate from the lead sublattice, whereas the (111) and ($\overline{1}\overline{1}$1) diffraction spots are mainly due to the halide sublattice (details are described in Supplementary Text, fig. S6, and table S1). Calculated interplanar spacings under phase segregation in Fig. 1E showed that the spacings of neighboring diffraction spots (002), (110), and ($\overline{1}\overline{1}$1) were larger than those in the mixed phase, indicating the presence of an iodide-rich phase. For the split diffraction spots, the ones with larger spacings (e.g., d_002-1 = 6.010 ± 0.010 Å, d_110-1 = 5.830 ± 0.010 Å, and $d_{\overline{11}\mathrm{1-1}}$ = 5.260 ± 0.004 Å) correspond to the iodide-rich phase, while those with smaller spacings (d_002-2 = 5.980 ± 0.010 Å, d_110-2 = 5.800 ± 0.010 Å, and $d_{\overline{11}\mathrm{1-2}}$ = 5.230 ± 0.004 Å) correspond to the parent phase.

As shown in Fig. 1B, the angle between the (002)-1 and (110)-1 spots (iodide-rich phase) was 91.64°, and between (002)-2 and (110)-2 (parent phase) was 88.12°, deviating by +1.60° and −1.92°, respectively, from the original (002)-(110) angle of 90.04°. As the angular measurement uncertainty is <1°, these pronounced deviations indicate octahedral distortions in the PbX₆ framework (illustrated in Fig. 1E, center inset). In Fig. 1C, elongation of the (110) diffraction spot and splitting of the ($\overline{1}\overline{1}$1) and (111) diffraction spots were observed. Increased spacing in the ($\overline{1}\overline{1}$1)-3 spot and decreased spacing in the (111)-1 spot suggest the simultaneous presence of both iodide-rich and bromide-rich phases. As the (110) diffraction spot originates from the lead sublattice and the halide-related diffraction spots also show splitting, the changes in diffraction observed here is consistent with the octahedral distortions seen in Fig. 1B. In Fig. 1D, interplanar spacings of the neighboring diffraction spots were smaller than the initial values, indicating the detection of a bromide-rich phase. In addition, new diffraction spots were observed as indicated by the red circle. The new diffraction spots are superpositions of diffraction patterns from the [1$\overline{1}$0] and [00$\overline{1}$] zone axes (details are described in Supplementary Text and fig. S7). These patterns closely resemble those observed in CsPbBr₃ and CsPbBr₂I twinned crystals (34, 35), suggesting the formation of 90° rotational twinning along [110] during phase segregation (Fig. 1E, right inset).

As the three phase segregation cycles progressed, diffraction contributions from the 2 μm probe region transitioned from primarily iodide-rich and parent phases in the first cycle, to a mixture of iodide-rich, bromide-rich, and parent phases in the second cycle, and predominantly bromide-rich and parent phases in the third cycle. PL data confirmed that reversible phase segregation occurred in all three cycles (fig. S3B). On the basis of the correlation between in situ TEM and PL data, we infer that the positions of the iodide- and bromide-rich domains shifted during each cycle, such that different compositions dominated the probed region across cycles, as illustrated in Fig. 1F.

Irreversible local strain gradient facilitates phase segregation

After each phase segregation cycle, after keeping the sample in the dark for 24 hours, the splitting, elongation, and emergence of new diffraction spots disappeared, indicating that the octahedral distortions and twinning induced by phase segregation had recovered (fig. S3 and S4). PL spectra further confirmed halide remixing. However, a comparison of the SAED patterns before illumination and after recovery revealed that the positions of neighboring diffraction spots (002), (110), ($\overline{1}\overline{1}$1), and (111) did not fully return to their original positions. The shifts of these diffraction spot positions, as indicated by dashed yellow lines, relative to the central (000) diffraction spot are presented in Fig. 2A, along with corresponding diffraction patterns and cross-sectional views. The evolution of interplanar spacings after recovery is shown in Fig. 2B. The (002) diffraction spot, primarily associated with the lead sublattice, exhibited an increase in interplanar spacing after cycle I, a decrease after cycle II, and a further decrease after cycle III. Similar trends were observed for the halide-sublattice-dominated ($\overline{1}\overline{1}$1) and (111) diffraction spots.

Fig. 2. Irreversible local strain after phase segregation recovery.

(A) Enlarged SAED patterns after phase segregation recovery from cycles I to III, along with intensity profiles of the (002), (110), ($\overline{1}\overline{1}$1), and (111) diffraction spots relative to the central (000) diffraction spot. (B) Postrecovery evolution curves of interplanar spacings d₀₀₂, d₁₁₀, $d_{\overline{11}1}$, and d₁₁₁ during the three-cycle experiment. (C) Schematic diagram of lead sublattices between the strain-free zone, tensile-strain zone, and compressive-strain zone after phase segregation recovery. (D) Local strain distribution map of each cycle after remixing, corresponding to the zones in (C). (E) Strain and strain gradients after remixing of cycles I, II, and III.

These variations in interplanar spacing suggest the presence of persistent local lattice expansion or contraction after recovery, likely arising from incomplete structural relaxation following the formation of iodide-rich or bromide-rich domains during phase segregation. Given that phase segregation occurs across the entire sample (illumination beam diameter 2 mm) and the SAED probe diameter is 2 μm, we infer that the spatial location of iodide-rich and bromide-rich domains varies across cycles. During all cycles, we probed the same region of the sample, and the halide phase evolution proceeded as illustrated in Fig. 1F: In cycle I, the region was iodide-rich, causing lattice expansion (Fig. 1B); in cycle II, both iodide-rich and bromide-rich phases coexisted, resulting in zones with lattice expansion and contraction (Fig. 1C); and in cycle III, the region transitioned to bromide-rich, resulting in lattice contraction (Fig. 1D).

These phases leave behind residual lattice expansions or contractions, even after remixing, as consistently indicated by the trends of interplanar spacings in Fig. 2B. Therefore, the region of local strain, which is zone with residual lattice expansions or lattice contractions, is present. The residual local strain zone retains a “memory” of the iodide-rich and bromide-rich domains formed during previous phase segregation. On the basis of previous experimental characterizations (17), each zone with local strain spans a few micrometers, which is comparable to the 2-μm SAED probe area. As illustrated by the local strain distribution map in Fig. 2D, the probed region (dashed circle) was primarily located within a zone with tensile strain of 0.40% along the [002] direction after cycle I, contained zones with both tensile and compressive strain after cycle II, which result in strain of −0.77%, and was dominated by a zone with compressive strain of −1.52% after cycle III. The residual strain after each cycle is shown in Fig. 2E (top) and fig. S8 (top).

Local strain gradient exists due to lattice mismatch at the interface between the region with local strain and the region that is strain-free. Local strain gradient of the lead sublattices between the strain-free zone, the compressive strain zone, and the tensile strain zone are illustrated in Fig. 2C. Local strain gradients exist at these interfaces and can be estimated from the interplanar spacings in Fig. 2B. The strain gradients between zones are shown in Fig. 2E (bottom) and fig. S8 (bottom). After cycle I, the strain gradient between the residual tensile strain zone and the strain-free zone is 2.00 × 10⁶ m⁻¹; after cycle II, the strain gradient is 3.85 × 10⁶ m⁻¹ when both the tensile strain zone and the compressive strain zone exist; after cycle III, the strain gradient between residual compressive strain zone and the strain-free zone is 7.60 × 10⁶ m⁻¹. On the basis of our observations shown in Fig. 2, it is evident that phase segregation results in irreversible local strain and strain gradients, manifested as localized zones of tensile strain and compressive strain within the sample. Three additional samples (samples 2 to 4) were investigated under the same phase segregation, recovery, and electron beam characterization conditions. As shown in figs. S12 to S15, the phase segregation behavior and residual strain after phase segregation recovery were similar to those shown in Figs. 1 and 2. We also performed control experiments on CsPbBr₃ thin films under the same conditions. As shown in fig. S16, under illumination, no changes in diffraction patterns and interplanar spacing were observed. As shown in fig. S17, under the dark, no changes in diffraction patterns and interplanar spacing were observed compared to pristine samples. These control experiments verify that phase segregation and residual strain were exclusive to mixed-halide perovskites.

The presence of local strain and strain gradient influences the evolution of phase segregation, as evidenced by PL characterizations for cycles I to III. As shown in Fig. 3 (A to C), as the number of phase segregation cycle increases, the saturated PL peak, λ_PL at 40 min, redshifts and the phase segregation rate accelerates (also verified by sample 2 in fig. S12). These observations suggest an increase in the proportion of iodine and a reduction in iodide-rich domain size as the number of phase segregation cycles increases. The above evidence indicates that residual local strain and strain gradients promote and dominate the subsequent phase segregation.

Fig. 3. Mechanistic illustration of strain gradient–facilitated phase segregation.

(A to C) Evolution of PL spectra and the extracted PL peak wavelength (λ_PL) as a function of illumination time during multiple cycles of phase segregation. (D) Schematic illustration (left) of the residual strain gradient within the halide sublattice after halide remixing of cycle I, and a model (right) showing the preferential distribution of iodide-rich domains along the strain gradient under illumination in cycle II. This model involves three coupled driving mechanisms: (i) The initial phase segregation cycle induces the formation of interfacial strain gradients (black dashed circle). (ii) The strain gradient drives the accumulation of photogenerated charge carriers. (iii) The carrier accumulation promotes ion migration and accumulation of iodine ions. Interfaces of iodide-rich and bromide-rich domain indicated by red dashed circle and green dashed circles, respectively. h, hours.

We illustrate the mechanism of the local strain/strain gradient facilitated phase segregation in Fig. 3D. The remixed halide ions after phase segregation cycle I are illustrated in the left panel of Fig. 3D, where a zone with tensile strain persists due to the prior presence of iodide-rich domain in the probe region. At the interface of the local strain, local strain gradient exists, as indicated by the dashed circle. Upon re-illumination in the following phase segregation cycle, photocarriers are generated and tend to accumulate in these strain-gradient zones, likely due to the flexoelectric effect (36–38). To maintain charge neutrality, iodide ions diffuse toward the regions of carrier accumulation, facilitating the preferential formation of iodide-rich domains at the interfaces with strain gradient (black dashed circles), as illustrated on the right of Fig. 3D. The newly formed interfaces of local strains (red and green dashed circles) then act as nucleation seed for iodine accumulation for the subsequent phase segregation cycle. This aligns with the experimentally observed dynamic repositioning of the iodide-rich regions during repeated cycles of phase segregation.

The photocarrier accumulation in strain gradient regions increases the local carrier concentration, which thermodynamically promotes phase segregation. Specifically, higher carrier concentrations shift the binodal points of the free energy diagram outward, favoring the formation of phases with higher iodine (or bromine) content (10). This results in a redshift of the phase segregated PL peak position, as observed over successive phase segregation cycles. Furthermore, increased carrier concentrations accelerate the phase segregation process. The rapid accumulation of iodine leads to smaller iodide-rich domains as indicated in Fig. 3D, consistent with the reduced domain size observed experimentally during cycle II and cycle III. With each successive cycle, the residual strain gradients become more pronounced, increasing both the phase segregation rate and the iodine content in iodide-rich domains. This leads to increased photocarrier accumulation, thereby reinforcing the phase segregation process and further redshifting the emission.

Growth-induced local strain promotes phase segregation

To further elucidate the role of local strain in phase segregation, we validate the proposed mechanism by using samples with preexisting strain that were accumulated during sample growth. The thin-film sample exhibits visible contrast fringes indicative of strain gradients, as shown in Fig. 4A. Atomic-scale high-angle annular dark-field scanning TEM (HAADF-STEM) imaging shown in Fig. 4B and corresponding fast Fourier transform analysis (fig. S9) reveal a slight deviation in the crystallographic zone axis within these fringed regions. As shown in Fig. 4C, the Pb─Pb bond lengths near the zone-axis deviation region are shorter than those farther away. Strain and strain gradient calculations based on these bond length variations confirm that the fringed regions are strain and strain gradient–concentrated zones, with strain gradient at similar orders of magnitude compared to residual strain gradient from phase segregation.

Fig. 4. Phase segregation facilitated by growth-induced local strain.

(A) A bright-field image of a CsPbBr_2.1I_0.9 thin film grown on a Si₃N₄ substrate. Scale bar, 1 μm. Yellow arrows indicate contrast fringes observed in the pristine film. The region enclosed by the white circle corresponds to the HAADF-STEM image shown in (B). (B) HAADF-STEM image of the fringe region. The yellow dashed line marks the boundary between regions with aligned zone axes and those with axis deviation. Scale bar, 2 nm. (C) Variation in Pb-Pb bond lengths along the direction parallel to the yellow dashed line, and the corresponding strain and strain gradient profile calculated from bond length changes. Bond length values are averaged over eight data points along the orange arrow direction in (B). (D) Bright-field image of the film before illumination. The circles denote areas I and II, where PL spectra were collected. Scale bar, 2 μm. (E) During the first illumination cycle, evolution of PL spectra from areas I and II in (D) and the extracted peak wavelength λ_PL as a function of illumination time. (F) Bright-field image of the film after the first phase segregation cycle. The circles again denote areas I and II used for PL measurements. (G) During the second illumination cycle, evolution of PL spectra from areas I and II and the corresponding λ_PL as a function of illumination time.

We then investigated the phase segregation behavior in two regions: area I, located within the strain-concentrated fringes, and area II, situated farther away from them, as indicated by the dashed circles in the bright field image in Fig. 4D. Time-dependent PL spectra of both areas were recorded and shown in Fig. 4E. After phase segregation, area I exhibited two PL peaks at 631.7 and 676.8 nm, while area II showed peaks at 621.2 and 673.5 nm. The more redshifted PL wavelength, together with significantly higher PL intensity in area I suggests a greater degree of phase segregation and a higher proportion of iodide-rich domains (5, 6), indicating that phase segregation was more pronounced in the local strain-concentrated region.

After phase segregation, TEM imaging in Fig. 4F and optical microscopy (fig. S10) revealed the formation of cracks in the film. These cracks relaxed the strain gradients. For samples with cracks, we again monitored the time-dependent PL spectra in areas I and II. The results are shown in Fig. 4G. After 40 min of illumination, the PL peak of area I stabilized at 644.6 nm and at 687.1 nm in area II. Compared to precrack behavior, area I exhibited a blueshift in the maximum PL wavelength, while area II showed a slower phase segregation rate. These results indicate that relaxation of the strain gradient suppresses phase segregation. A comparison of phase segregation behavior before and after strain relaxation, as well as between strain-concentrated and strain-free regions, confirms that local strain facilitates phase segregation.

DISCUSSION

In summary, we discover that while phase segregation may appear to be compositionally reversible, changes in the perovskite lattice is not fully reversible and influences phase segregations over multiple cycles. We reveal that local strain plays a critical role for phase segregation, acting both as a structural remanent of prior phase segregation and as a driving force for subsequent phase segregation. Through optical in situ TEM, we observe that the structural evolution such as octahedral distortion and twinning accompanying phase segregation is reversible. However, phase segregation leaves behind residual local strain gradients, arising from lattice mismatches at the interfaces of strain regions that retain the memory of previously segregated halide domains. Correlating structural data with PL spectroscopy suggests that strain gradients at the interfaces of these residual local strain regions act as traps for photocarriers, thereby serving as preferential nucleation sites for iodide-rich domains in subsequent phase segregation cycles. The role of local strain is further confirmed in samples exhibiting growth-induced local strain. Regions with concentrated local strain and strain gradients show accelerated phase segregation, whereas strain relief via crack formation effectively suppresses halide phase segregation. This provides strong evidence that local strain not only accompanies but also governs the spatial and temporal evolution of phase segregation in mixed halide systems.

Our findings uncover the mechanism by which the lattice, the halide ions, and photocarriers influence phase segregation over multiples cycles, offering insights into engineering the stability of mixed-halide materials. Phase segregation exhibits spatiotemporal evolution across repeated illumination cycles, sometimes resulting in extreme structural changes such as crack formation (18, 39). Therefore, strategies to engineer the lattices to minimize strain accumulation during phase segregation, as well as to actively relieve residual strain intermittently between operation cycles, may be critical for maintaining device performance and longevity. We envision that our characterization method, which combines optical in situ TEM and PL spectroscopy, can be broadly applied to investigate mechanisms in materials whose structure and function are actively modulated by light. Moreover, the observation that residual local strain exhibits a temporal evolution, where one phase segregation cycle influences the next, introduces the concept of structural memory. This concept is an essential principle for artificial synaptic devices. Thus, temporal evolution of local strain after phase segregation recovery in mixed-halide perovskites may offer opportunities for neuromorphic applications.

MATERIALS AND METHODS

Materials synthesis

Single-crystalline films were synthesized via chemical vapor deposition (CVD) on 20-nm-thick free-standing silicon nitride membranes. The precursor was prepared by mixing the appropriate atomic ratio of CsBr (Sigma-Aldrich, 99.999%), CsI (TCI, >99.0%), and PbBr₂ (TCI, 98.0%) at 620°C, atmospheric (atm) pressure in a home-built CVD setup. The CVD setup has a “tube-in-tube” structure as a reactor fixed in a furnace (Thermo Fisher Scientific TF55035C-1) connecting to a nitrogen flowmeter (Beijing Sevenstar Electronics Co. Ltd. D07-19B). The inner diameter of the outer and inner quartz tubes are 2.1 and 1.7 cm. After mixing the precursor, a 20-nm-thick amorphous Si₃N₄ grid (SiMPore, SN100-A20Q33) substrate and the previously mixed precursor were placed in the CVD at a distance of 14.2 cm (precursor was placed at the center of the heating zone of the furnace while the substrate was placed on the downstream of the nitrogen carrier gas). The sample deposition was done at 590°C for 7 min with a flow of 10 sccm, atm. A TEM image of the sample together with thickness measurement was shown in fig. S11.

Optical measurements

The morphology of the samples was imaged by a fast-mapping micro-Raman spectrometer (WITec RAS300). The PL spectrum was measured on a fast-mapping micro-Raman spectrometer (WITec RAS300). We used a 450-nm unfocused continuous-wave laser beam to constantly illuminate the sample for several minutes and then collected the pre-illumination, post-illumination, and post-dark relaxation PL spectrum of samples. Illumination power density (unit: W/cm²) was estimated by I = 4P/(πd²), where I is the power density, P is the illumination power, and d is the laser beam diameter.

In situ TEM characterizations combined with PL spectroscopy

The bright-field TEM (BF-TEM) images and SAED patterns were acquired on a 200-kV transmission electron microscope (Thermo Fisher Scientific Talos F200X G2), with a probe diameter of 2 μm. After initial structural characterization, the sample was placed in a vacuum environment for subsequent measurements: pre-illumination PL characterization, phase segregation characterization, and post-illumination PL characterization. Phase segregation was induced by continuous 40-min illumination with a 450-nm laser (1.06 W/cm², beam diameter of 3 mm), during which time-resolved PL spectra were acquired every 10 min. Then, after placing samples in the glove box in the dark for 24 hours, samples were transferred into the TEM. In situ TEM experiments were carried out on a single-tilt optical holder whose optical conditions were controlled by an external laser connected via internal optical fiber (Fischione Mode 2550SPR). The sample was irradiated under the same optical conditions for 1 hour (beam diameter of 2 mm) before BF-TEM image and SAED pattern characterizations with a dose rate of 0.06 eÅ⁻² s⁻¹. Post-dark relaxation PL and TEM characterization were performed after keeping the phase segregated sample on the TEM sample holder in the vacuum environment in the dark for 24 hours. BF-TEM image and SAED pattern characterizations were performed on the re-mixed sample subsequently. The sample was then taken out for PL characterizations. Each experimental cycle consisted of one PL characterization and one in situ TEM characterization, as illustrated in fig. S3A. Three complete cycles were performed on the same sample.

Ex situ TEM characterizations combined with PL spectroscopy

The sample presented in Fig. 4 was measured with ex situ TEM characterization combined with PL spectroscopy. Ex situ TEM characterizations were carried out on a double-tilt holder to record BF-TEM images before illumination and after dark relaxation. PL spectroscopy was performed under the same condition as the previous section described.

STEM imaging and EELS acquisition

The HAADF-STEM images and low-loss electron energy-loss spectroscopy (EELS) spectrum were acquired using a spherical aberration-corrected transmission electron microscope (Thermo Fisher Scientific Spectra Ultra) operating at 300 kV. The probe convergence semi-angle was set to 25.2 mrad and the annular detector collection semi-angle range was 43 to 200 mrad. The beam current was around 45 pA and the dwell time is 20 μs.

Supplementary Materials

This PDF file includes:

Supplementary Text

Figs. S1 to S19

Tables S1 to S4

References