Direct observation of nanoscale dynamics of ferroelectric degradation

Ferroelectric materials are widely used in electronic nanodevices due to their polarization reversal under applied electric fields, but repeated cycling leads to polarization reversal failure (ferroelectric degradation), compromising reliability and endurance. Degradation alters key properties such as coercive field, remnant polarization, and resistivity under cyclic loading. Multiple mechanisms have been proposed, including domain wall pinning by accumulated charges/defect dipoles with possible competitive unpinning, suppression of domain nucleation/growth near surfaces due to injected charges from electrodes, and microcrack formation initiated at grain boundaries or domain walls. Although excess space or injected charges are often implicated and thought to accumulate at domain walls under cycling, the pathway by which these charges cause degradation is unclear. Direct, real-time nanoscale observation of charge accumulation and domain evolution under cyclic fields is therefore essential. Advances in in-situ TEM and MEMS-based nano-chip platforms now enable stable, repeatable biasing experiments. In this study, the authors investigate Pb(Mg1/3Nb2/3)O3–0.38PbTiO3 (PMN-0.38PT) single crystals using in-situ biasing TEM and STEM-DPC to track charge distribution and domain switching during cyclic in-plane electric fields, aiming to clarify the role of injected/space charges in ferroelectric degradation.

The paper reviews longstanding hypotheses for ferroelectric degradation: (1) domain wall pinning, wherein charge carriers and/or defect dipoles accumulate at walls and lock domains, sometimes countered by detrapping (unpinning); (2) domain nucleation suppression, where injected carriers from electrodes inhibit near-surface nucleation and growth; and (3) microcrack formation due to long-term cycling. Prior work generally attributes degradation to excess charges (space or injected) that accumulate at domain walls and reduce switchable polarization. However, direct evidence linking local charge accumulation and the evolution of domain structures under cycling has been lacking. Recent progress in in-situ TEM and MEMS-based biasing has enabled real-time nanoscale observation of these phenomena, motivating the present study.

Materials and sample preparation: Single-crystal PMN-0.38PT grown by a modified Bridgman method was used due to its simple and stable tetragonal structure at room temperature. Lamellae (≈8 μm × 4 μm × 1 μm; length × width × thickness) were prepared by mechanical grinding and focused ion beam (FIB), then thinned to ≈100 nm. The lamellae were mounted across a 4 μm gap between inner contacts of MEMS-based biasing nano-chips (DENSsolutions) via Pt deposition, aligning the in-plane bias along ±[100]. In-situ biasing TEM experiments: Conducted in a JEM-2100 TEM at 200 kV using a DENSsolutions Lightning holder. A bipolar triangular waveform of ±4.7 V at 1/30 Hz was applied across the 4 μm electrode gap, producing a peak field of ±1.175 MV/m assuming homogeneous field distribution. Domain configurations were characterized by TEM imaging and selected-area electron diffraction to identify polarization orientations and domain types (a1, a2, c) and their evolution with cycling. STEM-DPC electric field mapping: Electric field imaging was performed on an aberration-corrected FEI Themis Z STEM at 300 kV. STEM-DPC relates the differential phase shift of the transmitted probe (derived from CBED center-of-mass shifts) to the projected electrostatic field. Detector segmentation followed a quadrant configuration (per Supplementary Fig. 8). Imaging parameters: semi-convergence angle 17.9 mrad, collection angle 9–51 mrad, beam current 50 pA, dwell time 10 μs/pixel, image size 1024×1024. To avoid domain switching during DPC acquisition, a separate PMN-0.38PT sample was cycled with a sub-switching bipolar triangular bias (peak 3.0 V, 1/8 Hz) along ±[100]; images were recorded before and after 100 cycles. Focus and tilt stability, and defocus insensitivity within a range, were verified (Supplementary Figs. 1–2). EELS analysis: Electron energy loss spectroscopy measured plasmon peak shifts at domain walls before and after cyclic loading to quantify changes in local charge density. An observed plasmon energy shift of 1.2 eV corresponded to an increase of approximately 3.76 electrons/nm³ at domain walls (calculation in Supplementary Note 1). Data analysis: Time-lapse TEM imaging under repeated cycling quantified nucleation and growth of c domains; projected c-domain area versus cycle count was extracted to assess kinetics. Ferroelectric hysteresis loops of switchable domain area versus bias were obtained from thin TEM samples at different cycle counts to evaluate degradation of switchable fraction.

- Under cyclic in-plane electric fields (±1.175 MV/m), new domains nucleate from original a1/a2 90° ferroelastic domain walls in PMN-0.38PT; diffraction analysis identifies these as c domains with out-of-plane polarization. - c domains propagate with increasing cycles, growing along [010] within a1 domains and along [100] within a2 domains; domain walls between a1/a2 and c remain straight 90° ferroelastic walls with head-to-tail configurations. - The projected area of c domains increases with cycle count, exhibiting an accelerating (approximately exponential) growth rate over 0–280 cycles. - STEM-DPC electric field maps show increased local field amplitude at domain walls after cycling, while field directions remain unchanged, indicating increased local charge density concentrated at walls. Subtractive imaging highlights intensity increases primarily at walls. - EELS plasmon peak analysis reveals a 1.2 eV energy shift at domain walls after cycling, corresponding to an increase of ~3.76 electrons/nm³ in charge density, corroborating charge accumulation. - Mechanism: Accumulated mobile charges at polarization transition zones (ferroelastic walls) force metastable polarization in the wall region to rotate out-of-plane (±[001]), nucleating c domains. Surface compensation charges (rearranged accumulated charges) reduce depolarization fields and stabilize these c domains. - Frozen behavior: c domains are largely unresponsive to subsequent in-plane electric fields during a full bias cycle; while a1/a2 domains undergo 180° switching at threshold biases, c-domain areas remain unchanged, evidencing a frozen state that persists even 18 months post-cycling. - Hysteresis analysis shows decreasing switchable domain area with increasing cycles, and the fraction of unswitchable (frozen c) domains increases, indicating ferroelectric degradation. - Stress-induced nanodomains can form in some a2 domains due to mechanical constraint of the lamella ends, leading to multi-step switching via intermediate ferroelastic states, but c domains remain unaffected by the in-plane field. - Implication: The dynamic sequence from charge accumulation at walls to stabilized, frozen c domains provides a direct route explaining how charges cause ferroelectric degradation in nanoscale ferroelectrics.

The study elucidates a direct, nanoscale mechanism by which space/injected charges drive ferroelectric degradation. Cyclic in-plane fields accumulate charges at ferroelastic domain walls, altering local electric fields and forcing polarization rotation within transition zones to out-of-plane, nucleating c domains. Compensation charges at surfaces weaken depolarization fields and simultaneously screen external in-plane fields, stabilizing c domains and rendering them immobile (frozen) under further cycling. The growth of these frozen, unswitchable regions reduces the switchable domain fraction, accounting for observed degradation (reduced polarization reversal, altered hysteresis). This mechanistic link bridges prior qualitative attributions of degradation to charges with direct observations of charge accumulation and domain freezing. While demonstrated in nanoscale PMN-0.38PT lamellae, the mechanism likely extends to thin films and nanodevices, especially at polarization transition zones such as domain walls, phase boundaries, interfaces (film/substrate, ferroelectric/electrode), surfaces, and grain boundaries where charges preferentially accumulate. Understanding this pathway informs strategies to mitigate degradation by managing charge injection, engineering interfaces, and controlling domain wall densities and configurations.

This work directly visualizes the nanoscale dynamics of ferroelectric degradation in PMN-0.38PT under cyclic in-plane electric fields. STEM-DPC and EELS demonstrate charge accumulation at ferroelastic domain walls, which nucleates and stabilizes out-of-plane c domains via compensation charges. These c domains exhibit a frozen response to in-plane fields, expand with cycling, and reduce the switchable domain fraction, thereby degrading ferroelectric performance. The findings provide a dynamic, mechanistic picture linking charge activity to degradation in nanoscale ferroelectrics and offer guidance for device design. Future research could: (1) investigate strategies to limit charge injection and accumulation (e.g., electrode/interface engineering); (2) explore materials and architectures that minimize polarization transition zones or stabilize in-plane domain configurations; (3) test the generality across different ferroelectric systems, thin films, and device-relevant geometries; and (4) quantify kinetics of c-domain formation under varying field amplitudes, frequencies, temperatures, and mechanical constraints.

The experiments were conducted on thin lamellae (≈100 nm thick) under in-situ TEM conditions with mechanical constraints at the electrodes; thus, the observed behavior may primarily apply to thin-film ferroelectrics and nanodevices rather than bulk materials. Stress-induced nanodomains in constrained regions can complicate local switching behavior. Although electron beam effects were minimized (beam blocked during biasing) and focus/tilt stability was verified, in-situ TEM conditions may still differ from operational device environments. STEM-DPC was performed on a separate sample with sub-switching biases to avoid switching during imaging, necessitating correlation across experiments.