神经形态工程学
计算机科学
晶体管
电子工程
光子学
材料科学
人工神经网络
光电子学
人工智能
瓶颈
计算机体系结构
响应度
逻辑门
多路复用
突触后电流
作者
Jiajun Sun,Jiawei Yang,Kesheng Wang,Shuo Cheng,Hainan Zhang,Huanhuan Wei,Shanshan Jiang,Gang He
标识
DOI:10.1109/ted.2025.3611371
摘要
In recent years, neuromorphic synaptic transistors have emerged as core components for building efficient neural network computing systems due to their unique ability to emulate biological synaptic plasticity. This article presents an oxide semiconductor-based neuromorphic synaptic transistor. The gate dielectric layer is fabricated using a simple solution spin-coating method, while an indium gallium zinc oxide (IGZO) channel is deposited via sputtering. Experimentally, the device demonstrates excellent optical responsivity and significant accumulation effects in response to optical signals. These photonic properties enable it to achieve single-pulse-induced excitatory postsynaptic currents (EPSCs), paired-pulse facilitation (PPF), learning-forgetting memory behaviors, and to emulate visual neural regulation under bright light conditions. Furthermore, image processing capabilities, including frequency-domain filtering, cutoff frequency testing, and high-pass filtering, have been demonstrated. A neuromorphic computing platform was constructed and trained using a custom-built dataset. Unlike traditional pattern recognition, which relies on manually annotated datasets and passive learning, our approach achieves near-perfect (>99%) testing accuracy after a defined number of training cycles. This work provides new insights for neuromorphic computing and offers a promising pathway for co-optimizing device design and system architecture in developing low-power, high-parallelism neuromorphic hardware.
科研通智能强力驱动
Strongly Powered by AbleSci AI