油藏计算
计算
晶体管
计算机科学
材料科学
非线性系统
活动识别
逻辑门
计算机数据存储
电子工程
光学计算
国家(计算机科学)
序列(生物学)
嵌入式系统
人工智能
模拟计算机
功率(物理)
空格(标点符号)
功率消耗
图层(电子)
人工神经网络
计算机硬件
分布式计算
混合动力系统
面子(社会学概念)
智能决策支持系统
复杂系统
数据采集
数码产品
作者
Cong Peng,Weidong Xie,Enlong Li,Di Liu,Xifeng Li,Huipeng Chen
标识
DOI:10.1002/adfm.202523188
摘要
Abstract With the increasing demand for intelligent systems in transportation and healthcare, conventional vision‐based human activity monitoring and recognition (HAMR) systems face challenges. The physical separation of sensing, storage, and computation within these systems leads to exorbitant power consumption and significant latency. Moreover, they struggle to handle multidimensional data effectively and are severely limited when dealing with visual occlusions. In this research, a multifunctional opto‐electronic synaptic transistor (OEST) is developed. The OEST can seamlessly switch between volatile and non‐volatile functions via optical input and gate modulation. Leveraging the unique characteristics of the multifunctional OEST, an intra–sensor reservoir computation (ISRC) system is constructed. Under optical stimulation, OEST exhibits nonlinear short‐term memory characteristics of 4‐bit reservoir states, which can effectively map multi‐dimensional sequence information to the reservoir state space as a reservoir; under gate programming/erasure, 200 non‐volatile conductance states are realized for constructing the readout layer unit. Based on multidimensional information acquisition from multiple sensors, this ISRC system achieves 96.2% accuracy in human activity recognition, providing a new solution for future high‐performance intra‐sensor computing applications.
科研通智能强力驱动
Strongly Powered by AbleSci AI