Monolithic Integration of Sensing, Memory, and Processing in ScAlN Ferroelectric Optoelectronic Transistors for Multimodal Fusion.

Abstract

The biological nervous system achieves highly efficient cognitive functions through the seamless integration of sensing, memory, and computation. However, conventional artificial neuromorphic systems suffer from separated functional modules, resulting in considerable data transmission overhead and limited capability for multimodal information processing. Here, we demonstrate a monolithic integration platform based on ScAlN ferroelectric transistors, which implement sensing, memory, and processing functions, respectively. Leveraging ferroelectric polarization as a programmable internal field, the device can be reconfigured between an optoelectronic logic gate (OELG) mode for multimodal sensing and signal encoding and a ferroelectric field-effect transistor (FeFET) mode for non-volatile synaptic storage and computation. The FeFET exhibits stable analogue conductance modulation with an on/off ratio exceeding 104, endurance over 104 cycles, and retention beyond 104 s, enabling synaptic weight updating for neuromorphic computing. By integrating these functions, we further demonstrate audio-visual fusion for multimodal recognition, achieving enhanced perception performance with an accuracy exceeding 99%. This work establishes a reconfigurable ferroelectric optoelectronic platform and provides a hardware foundation for energy-efficient multimodal neuromorphic systems requiring integrated perception and computation.