Accelerating Quantized DNN Layers on RISC-V with a STAR MAC Unit

To support quantized neural networks in low-end CPUs, we propose STAR MAC, a reconfigurable multiply-and-accumulate unit based on a modified Baugh-Wooley architecture that operates at a variable reduced precision. We integrated it in a small RISC-V processor called Ibex obtaining an acceleration up to 5.8 \(\times \) in Fully-Connected (FC) layers, 3.7 \(\times \) in 2D-Convolution (2DConv) layers, and 2.8 \(\times \) in Depth-Wise Convolution (DWConv) layers, with respect to the original Ibex core (Orig.), and up to 4.5 \(\times \) in FC layers, 3.0 \(\times \) in 2DConv layers, and 2.3 \(\times \) in DWConv layers, against a modified Ibex core supporting standard 32-bit MAC operations (Orig.+MAC). Area and power in a 28-nm technology with 200 and 600  MHz target clock frequency are 0.015 and 0.017 mm \(^2\) , and 1.5 and 4.3 mW, respectively, with a limited overhead within 10% and 3% with respect to Orig., and within 3% and 3% against Orig.+MAC.