A 44.1TOPS/W Precision-Scalable Accelerator for Quantized Neural Networks in 28nm CMOS

Supporting variable precision for computing quantized neural network in a hardware accelerator is an efficient way to reduce overall computation time and energy. However, in the previous precision-scalable hardware, bit-reconfiguration logic increases the chip area significantly. In this paper, we demonstrate a compact precision-scalable accelerator chip using bitwise summation and channel-wise aligning schemes. The measurement results show that the peak performance per compute area is improved by 5.1-7.7x and system-level energy-efficiency is improved by up to 64% compared to previous precision-scalable accelerators.