Associative learning of a three-terminal memristor network for digits recognition

查看全文 作　　者：Yiming [1]REN;Bobo [1,2]TIAN;Mengge [1]YAN;Guangdi [1]FENG;Bin [3]GAO;Fangyu [1]YUE;Hui [1]PENG;Xiaodong [1]TANG;Qiuxiang [1,2]ZHU;Junhao [1]CHU;Chungang [1,4]DUAN 机构地区：[1]Key Laboratory of Polar Materials and Devices(MOE),Ministry of Education,Department of Electronics,East China Normal University,Shanghai 200241,China;[2]Zhejiang Lab,Hangzhou 310000,China;[3]Institute of Microelectronics,Beijing Innovation Center for Future Chips(ICFC),Tsinghua University,Beijing 100084,China;[4]Collaborative Innovation Center of Extreme Optics,Shanxi University,Shanxi 030006,China 出　　处：《Science China(Information Sciences)》索引2023年第66卷第2期,共10页 基　　金：supported by National Key Research and Development Program of China (Grant No. 2021YFA1200700);National Natural Science Foundation of China (Grant No. 62174053);Open Research Projects of Zhejiang Lab (Grant No. 2021MD0AB03);Shanghai Science and Technology Innovation Action Plan (Grant Nos. 21520714100, 19JC1416700);Shanghai Pujiang Program (Grant No. 19PJ1402900);Fundamental Research Funds for the Central Universities。 摘　　要：Imitating the associative intelligence of the biological brain is attractive but is poorly achieved in hardware because the complex tunable connection in neural networks is difficult to reproduce. We develop a circuit composed of a three-terminal memristor network to reproduce the biological conditioning process artificially. The synaptic weight between co-firing neurons is strengthened simultaneously by generating a feedback signal from the integrate-and-fire neuron to the gate of the synaptic memristor. The network allows the multi-associative capacity of recalling more than one digit in one circuit. Both single and multi-associative learning for recalling digital images are achieved. Furthermore, all 10 digital images from “0” to “9” are successfully recalled in an associative network with such paralleling circuits. Assisted by this associative layer, a typical classification network effectively improves the recognition rate for fragmentary digital images.Our work sheds light on brain-inspired artificial associative memory and provides a strategy for applications,such as object recognition with partial features and similar scenes. 关 键 词：artificial intelligence associative learning MEMRISTOR classification network