论文部分内容阅读
互连的光子脉冲神经元通过权重器件联系在一起,为了实现神经网络的大规模计算,权重器件的实现至关重要。利用微机电系统可调光衰减器(VOA),研制了一种可以自动调节光子脉冲神经元的权重器件。该权重器件包括VOA、光电探测器、单片机、模数转换器、数模转换器和放大器等模块,可以根据接收的光信号快速计算查表,可对VOA的衰减值进行实时在线调整。该权重器件效率高,且容易实现。该权重器件配合脉冲时间依赖的可塑性(STDP)光路使用,可以实现光子脉冲神经元的STDP学习机制。当STDP曲线窗口高度为0.2时,对权重器件进行了测量,实现了4种STDP学习。实验测量结果与理论计算结果一致。
Interconnected photon pulse neurons are connected by weighting devices, and in order to realize the large-scale calculation of neural networks, the realization of weighting devices is very important. Using MEMS (Micro Electro Mechanical System) dimmable attenuator (VOA), a kind of weighting device that can automatically adjust the photon pulse neurons was developed. The weight device includes modules such as VOA, photodetector, singlechip, analog-to-digital converter, digital-to-analog converter and amplifier, which can quickly calculate the look-up table according to the received optical signal and adjust the VOA attenuation value in real time. The weight device is efficient and easy to implement. The weighting device can be used with STDP (Pulse Time Dependent Plasticity) optical path to achieve the STDP learning mechanism of photon pulsed neurons. When the STDP curve window height is 0.2, the weighting device is measured and four kinds of STDP learning are realized. The experimental results are in good agreement with the theoretical ones.