发表机构
Saha Institute of Nuclear Physics; Homi Bhabha National Institute; University Science Instrumentation Centre, The University of Burdwan; Variable Energy Cyclotron Centre（萨哈核物理研究所; 霍米·巴巴国立学院; 布爾德萬大學大學科學儀器中心; 變能回旋加速器中心）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究基于Ag/TiO$_x$/NiO$_x$/p$^{++}$-Si忆阻器，利用晶态TiO$_x$/NiO$_x$异质结实现多功能突触学习，其构建的人工神经网络在MNIST识别中准确率近95%，还可完成巴甫洛夫学习等任务。
AI 中文摘要
本文尝试通过研究Ag/TiO$_x$/NiO$_x$/p$^{++}$-Si忆阻器结构在直流和脉冲偏置下的不同输运特性，来模拟生物突触的多种特性。研究发现，具有较小增益的晶态NiO$_x$有助于在其顶部沉积的TiO$_x$获得晶态特性和更大的晶粒。该异质结构在反向偏置条件下呈现稳定的双极型、无需形成过程的非易失性阻变开关特性，具备渐进式的 set 和 reset 特性。这些器件还能成功实现经典巴甫洛夫学习、人工伤害感受器研究以及莫尔斯电码检测。当结合由测得的电导编程脉冲关系导出的突触权重时，人工神经网络在MNIST手写数字识别任务中达到了近95%的准确率。简言之，NiO$_x$/TiO$_x$异质结在获得此类可重复的$I-V$特性以模拟生物突触的不同特性方面发挥了关键作用。
英文摘要
An attempt is made here to mimic different properties of biological synapses using Ag/TiO$_x$/NiO$_x$/p$^{++}$-Si memristor structure by studying its different transport properties under dc and pulsed bias. The presence of crystalline NiO$_x$ with smaller gains is found to be helpful to get TiO$_x$ deposited on its top with crystalline properties and larger grains. The heterostructure offers stable bipolar forming free non-volatile resistive switching characteristics under reverse biased condition with gradual set and reset features. These devices are also able to successfully implement the classical Pavlov's learning, study artificial nociceptor and Morse code detection. While incorporating synaptic weights derived from the measured conductance programming pulse relationship, an artificial neural network achieves nearly 95\% accuracy in MNIST digit recognition. In brief, NiO$_x$/TiO$_x$ heterojunction plays the pivotal role in getting such reproducible $I-V$ characteristics for mimicking different properties of biological synapses.
