Realization Of A Weight Device For Optical Spiking Neuron-variable Optical Attenuator

Neural network employs neurons as the basic computational units and aims at emulating the human brain at the level of physiological.It has a wide range of applications in the fields of pattern recognition,image processing,signal processing,machine learning etc.Hardware implementations of neural network have been a hotspot in recent years.Since photonics have the advantages of wide bandwidth,high parallelism,being capable of processing large amounts of information quickly,low energy consumption,etc.,it is an ideal way to realize neural network.Optical spiking neural networks are composed of multiple interconnected Optical spiking neurons.The Optical spiking neurons are connected by weight devices,so the hardware implementation of the weight devices is critical to realize the large-scale photonic neural networks.In this paper,using a MEMS(micro-electromechanical systems)VO A(variable optical attenuator),a weight device for optical spiking neuron,which can be automatically adjusted,is developed.The weight device includes a MEMS VO A,a photo detector,a microcontroller,an analog to digital(AD)module,a digital to analog(DA)module,an amplification module,etc.The weight can be tuned online by looking up in the weight-voltage and has the advantages of high efficiency and easy implementation.Using the optical spike-timing-dependent plasticity(STDP)circuits and the proposed weight device,STDP learning rules for optical spiking neural networks can be achieved.The weight device was detected under the condition of 0.2 and 1 heights of STDP curves,and it can achieve four forms of STDP learning curves under the condition of 0.2 height of STDP curves.The experimental results are in good agreement with the theoretical result.