Quantized Weight Transfer Method Using Spike-Timing-Dependent Plasticity for Hardware Spiking Neural Network

A hardware-based spiking neural network (SNN) has attracted many researcher’s attention due to its energy-efficiency. When implementing the SNN, offline training is most commonly used by which trained weights a software-based artificial (ANN) are transferred synaptic devices. However, it time-consuming map all as scale of increases. In this paper, we propose method for quantized weight transfer using spike-timing-dependent plasticity (STDP) SNN. STDP an online learning algorithm but utilize method. Firstly, train SNN Modified National Institute Standards and Technology (MNIST) dataset perform quantization. Next, mapped devices STDP, connected neuron simultaneously, reducing number pulse steps. The performance proposed confirmed, demonstrated that there little reduction in accuracy at more than certain level quantization, steps substantially decreased. addition, effect device variation verified.