Perovskite opto-ionics for in-sensor computing

Professor Chow, Chi Yung Philip   (Principal Investigator (PI))

Zhou Yuanyuan   (Project coordinator)

Professor Tan Chaoliang   (Co-Investigator)

36

2024-06-30

360000

Perovskite opto-ionics for in-sensor computing

""Metal halide perovskite"", ""Memristor"", ""Microstructure"", ""In-sensor computing"", ""Machine learning""

Materials Sciences

Physical Sciences (P)

C2001-23Y

Collaborative Research Fund (CRF) - Group Research Project 2023/2024

2024

On-going

1. Elucidating the multiscale mechanisms that underpin the memristive behavior of metal halide perovskites under electric biasing or light. Specifically, we aim to visualize the morphological evolution of conducting filaments in perovskites at high spatiotemporal resolutions based on combined uses of advanced electron microcopy, X-ray, and optical characterizations, and reveal the interplay between photon, electrons, and ions in perovskites, and understanding the atomistic origins of (opto-)ionic phenomena in perovskites and its relation to conduction properties via multiscale simulation based on an original dressed dynamics framework. 2. Achieving tailored synthesis of perovskite thin films with targeted 3D and low-D phase distributions to confine the formation and dissolution of conducting filaments in corresponding memristor devices. Specifically, we aim to achieve two types of thin films: one showing 3D perovskite grains of controlled numbers distributed in a polycrystalline low- D perovskite matrix, and the other showing continuous 3D perovskite phases aggregated along the grain boundary network of low-D perovskite polycrystalline matrix. The specific targeted parameters: (i) ~1V SET/RESET voltage with 0.05V device-to-device and cycle-to-cycle variation; (ii) a switching lifecycle larger than 106 for optically-operated memristor and 108 for electrically-operated memristor; (iii) less than 1nJ per switching event and 3. Achieving a prototypical in-sensor reservoir computer using the optoelectronic perovskite memristors, which will be leveraged to solve representative edge learning tasks with pattern classification for a proof- concept demonstration. Specifically, we aim to achieve >90% accuracy in classifying the representative MNIST handwritten digits dataset, while demonstrating an on-chip energy efficiency of >10 TOPS/W.