双极性脉冲忆阻桥电子突触神经网络及图像处理

Neural networks based on doublet generator synapses and its applications in image processing

忆阻器是一种具有天然信息存储能力和纳米级尺寸的类突触元件,由忆阻器构成的桥式突触电路结构简单、控制精确,有望应用于人工神经网络、超大规模集成电路、图像处理、模式识别等领域,然而传统的忆阻桥突触电路在突触模拟的过程中,由于单极性脉冲造成的通量积累会对权值模拟产生一定的误差.本文提出了一种新型的忆阻突触桥电路,该电路有效克服了传统桥电路忆阻值漂移的问题,由该电路构成的新型忆阻神经网络应用于图像处理中有着更为明显的优势,文中通过仿真实验验证了该方案的可行性,说明基于双极性脉冲发生器忆阻桥神经网络具有更加高效的仿生特性,而且集成度更高、模板更易更换,有望更有效解决实时的复杂的智能问题.

Memristor is a nanoscale device with memory and similar synaptic properties. The memristor bridge synaptic circuit, which has the advantages of a simple structure and precise control, can be applied to artificial neural networks, ultra-large-scale integrated circuits, image processing, pattern recognition, etc. However, traditional memristor bridge synaptic circuitry will cause some errors in the synaptic simulation process due to the accumulation of flux caused by unipolar pulses. In this paper, we propose a new synaptic bridge circuit, which effectively overcomes the memristance drifting problem of traditional bridge circuits. Then we apply the new synaptic bridge circuit to a synaptic neural network; the application in image processing has advantages that are more obvious. The feasibility of the structure was verified through simulation experiments, confirming the efficient biomimetic properties of neural networks based on doublet generator memristor bridge synapses. In addition,its higher degree of integration and the ability to replace templates more easily make it more effective in solving real-time complex intelligent problems.