Efficient Approximate Ternary Multipliers for Emerging Nanodevices
Abstract:
In this paper, we present efficient designs of approx-imate ternary multipliers applicable to several emerging nanode-vices. The proposed multipliers are motivated by the multiply-and-accumulate (MAC) operation in convolutional neural networks (CNNs). In particular, CNN applications in imaging are resilient to errors and it is therefore advantageous to examine methods that save energy and reduce the delay. Two approximate single-digit ternary multipliers are proposed. The single-digit approximate multipliers are used to develop an approximate 3 × 3 and 6 × 6 ternary multipliers. The proposed approximate 6 × 6 multiplier saves energy in the range of 22% to 40% over recent approximate designs. Further, there is a reduction of delay of roughly 21% with the proposed multipliers over the best existing design. The multipliers are based on their exact counterparts which are, in turn, developed using an efficient exact ternary carry adder (TCAD) that generates the sum of two carry outputs of a single ternary digit multiplier. The application of the approximate multipliers to CNN-based imaging is then demonstrated. In particular, the proposed approximate multipliers have excellent performance for CNN-based image denoising. Further, the approximate multipliers show good performance on MNIST and CIFAR-10 datasets. Simu-lations for Carbon Nanotube FET (CNTFET) reveal energy savings in excess of 50% over the best existing multipliers.
” Thanks for Visit this project Pages – Register This Project and Buy soon with Novelty “
Efficient Approximate Ternary Multipliers for Emerging Nanodevices
