一种基于多态忆阻器的电压型神经网络电路

A Voltage Feedforward Neural Network Circuit Based on Multiple Resistance States Memristor

提出了一种基于多态忆阻器的神经网络电路硬件实现方法。采用2~8 bit的惠普忆阻模型来构建存储权重的双忆阻稳定结构,结合了低功耗轨到轨运放技术以及寄存器技术,设计了模值与极性分离的绝对值电路,以及以忆阻器为核心、可进行正负浮点数运算的权值网络矩阵电路。通过Verilog-A编写激活单元,实现了多层忆阻神经网络。该电路采用并行输入和模拟信号处理方式,控制简单,无需中间数据缓存。实验结果表明,该方法有效提升了以忆阻器为核心的人工神经网络的稳定性和运行效率。

A hardware implementation method of neural network circuit based on multiple resistance states memristor was proposed.A 2~8 bit HP memristor model was used to construct a dual memristor stable structure with storage weight.Thanks to combining with low power rail-to-rail operational amplifier technology and register technology,on the one hand,an absolute value circuit with modulus and polarity separation was designed.On the other hand,a weighted network matrix circuit using the memristor as its core which could perform positive and negative floating point arithmetic was also developed.The feedforward neural network based on multiple resistance states memristor was realized due to the activation unit was written by Verilog-A.This circuit using parallel input and analog signal processing was simple to control and no intermediate data buffer was needed.The results showed that the proposed method effectively improved the stability and efficiency of the artificial neural network with the memristor as its core.