Neural field enhanced phase retrieval of atomic-scale structural dynamics in radiation sensitive materials

Atomic-scale imaging of radiation-sensitive materials has been a challenge for both materials science and life science. While low-dose transmission electron microscopy (TEM) is particularly useful for minimizing the radiation damage, the noisy images with poor resolution make it extremely difficult for the purpose of fine structure analysis. Here, this work presents a phase retrieval method to achieve high-quality atomic-scale imaging of radiation-sensitive materials under low-dose TEM conditions. By integrating neural fields (NF) with traditional exit wave reconstruction (EWR), it is able to reveal atomic details from limited low-dose experimental data. Taking the radiation-sensitive organic-inorganic hybrid halide perovskite CH₃NH₃PbI₃ (MAPbI₃) as an example, the EWR-NF method demonstrates superior performance in reconstructing the pristine atomic structure using as few as just three low-dose images, which is beyond the limits of conventional methods. In this manner, EWR-NF enables higher temporal resolution to reveal intermediate states during irradiation-induced decomposition. An example of stacking of MAPbI₃ with its as-decomposed product is shown. EWR-NF offers a promising tool for atomic-level structure analysis of sensitive halide perovskites and understanding irradiation-induced structure changes, with implications for a wide range of applications in materials science and beyond.