Details
Abstract
A highly energy-efficient Compute-in-Memory (CIM) processor for Ternary Neural Network (TNN) acceleration is proposed. Most of the previous works suffered from poor linearity of analog computing and energy-consuming ADCs. To resolve the issues, we propose a Ternary-CIM (T-CIM) processor with 16T1C ternary bitcell for good linearity with the compact area, and a Charge-based Psum Adder (CPA) circuit to remove system energy-consuming ADC. Furthermore, configurable data mapping enables execution of the whole convolution layers with 16x smaller bitcell memory capacity. Designed with 65nm CMOS technology, the proposed T-CIM achieves 1,316 GOPS of peak performance and 823 TOPS/W of energy efficiency.