Hybrid Optical-Electrical Pulse Programming in FeFETs for High-Precision Neuromorphic Computing

초록

The emergence of artificial intelligence (AI) has led to a dramatic increase in the volume of data being processed, exposing various limitations in traditional von Neumann?based computing architectures. To solve these challenges, there is growing interest in neuromorphic computing, which can process vast amounts of data at high speed and low power consumption. However, conventional neuromorphic systems that rely solely on electrical pulses face difficulties in achieving precise weight tuning, making it challenging to faithfully transfer weights obtained in software to hardware. In this study, we therefore fabricated a light-responsive ferroelectric non-volatile memory device to enable finetuning of synaptic weights. To optimize the photoresponse, the device layers were deposited on a glass substrate, and a ferroelectric P(VDF-TrFE) polymer layer was employed as the gate insulator to realize a ferroelectric FET (FeFET) memory element for neuromorphic computing. By using a hybrid approach of optical and electrical pulses, we rapidly optimized the network to achieve highly precise weight configurations. Moreover, energy?consumption analysis during the fine?tuning phase demonstrated that using optical pulses significantly reduced power consumption compared to electrical pulses. More details and simulation data will be discussed in the meeting.