A Mixed-Signal Time-Domain Generative Adversarial Network Accelerator with Efficient Subthreshold Time Multiplier and Mixed-Signal On-Chip Training for Low Power Edge Devices

Zhengyu Chen, Sihua Fu, Qiankai Cao, Jie Gu

Abstract: This work presents a low-cost mixed-signal time-domain accelerator for generative adversarial network (GAN). A significant reduction in hardware cost was achieved through delicate architecture optimization for 8-bit GAN training on edge devices. An area efficient subthreshold time-domain multiplier was designed to eliminate excessive data conversion for mixed-signal computing enabling high throughput mixed-signal online training demonstrated in a 65nm CMOS test chip.