Dianzi Jishu Yingyong (Apr 2023)
An energy efficient radix-4 Booth encoding parallel multiplier design
Abstract
Common-used Convolutional Neural Networks (CNNs) contain billions of multiplications, which is the bottleneck of hardware implementation of CNNs. To reduce energy cost of multiplier, an energy-efficient radix-4 Booth encoder multiplier is proposed. By improving the partial product module, the compensation bits in conventional multipliers are eliminated, which reduces the delay and energy cost of multiplier. Post simulation indicates that the proposed multiplier reduces the area, delay and energy cost by 5.2%, 6.3% and 10.8% respectively. The proposed multiplier can be used in neural network accelerators and breaks the energy efficiency bottleneck.
Keywords