Spiking neural networks (SNNs) that enable low- power design on edge devices have recently attracted significant research. However, the temporal characteristic of SNNs causes high latency, high bandwidth and high energy consumption for the hardware. In this work, we propose a binary weight spiking model with IF-based Batch Normalization for small time steps and low hardware cost when direct training with input encoding layer and spatio-temporal back propagation (STBP). In addition, we propose a vectorwise hardware accelerator that is reconfigurable for different models, inference time steps and even supports the encoding layer to receive multi-bit input. The required memory bandwidth is further reduced by two-layer fusion mechanism. The implementation result shows competitive accuracy on the MNIST and CIFAR-10 datasets with only 8 time steps, and achieves power efficiency of 25.9 TOPS/W.

This paper presents a new circuit technique named as “residue oversampling,” which is suitable for high-resolution analog-to-digital converters (ADCs). By adopting this technique and simplifying dynamic element matching (DEM), the impacts of capacitor mismatch and noise upon the successive-approximation register (SAR) ADCs are diminished significantly without calibrations. The proof-of-concept prototype was fabricated in a TSMC 40-nm CMOS technology. At 40-MS/s and 10-MS/s sampling rates, the measured peak signal-to-noise-and-distortion ratios (SNDRs) are 66.84 dB and 69.78 dB, respectively.