摘要
文中提出了一种采用计数器存储权值的人工神经网络的实现方案。数字权值采用计数器存储,突触电路和神经元电路用模拟电路来实现。数字权值经脉冲宽度调制电路转换为脉冲信号作为模拟突触电路的输入信号。因而权值可以长期存储,对权值的修改易于实现,突触神经元电路结构简单,融合了人工神经网络模拟实现和数字实现的优点。对于智能计算机的实现具有重要的意义。
A proposal is brought forward to implement an artificial neural network by using counters in weight storage.Digital counter is utilized to store the weight ,while synapse and neuron are analog circuits.Through Pulse Width Modu-lation circuits,the digital weight is converted into pulse signal as the input of the analog synapse circuits.The weight can be stored for a longer time and can be easily adjusted,the synapse and neuron circuits have smaller size,as means this method combines the advantages of both analog and digital realization of the artificial neural network.
出处
《计算机工程与应用》
CSCD
北大核心
2001年第20期22-24,87,共4页
Computer Engineering and Applications
基金
国家自然科学基金(编号:69783008
国家教委博士点基金(编号:98056117)资助项目
