Rounding technique analysis for power-area & energy efficient approximate multiplier design

Date

2020-08

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Approximate computing drecipitates a paradigm shift in energy-efficient systems design and operation, based on the idea that we are hindering computer systems’ efficiency by demanding too much accuracy from them. Interestingly, perfect answers are unnecessary in a large number of application domains, such as DSP, statistics, and machine learning. Approximate computing is suited for efficient data processing and error resilient applications, such as signal and image processing, computer vision, machine learning, data mining etc. Approximate computing reduces accuracy, which is acceptable as a cost of changing other circuit characteristics, depends on the application. Desirable accuracy is the threshold point for controlling the tradeoff between accuracy and circuit characteristics under the control of the circuit designer. In this work, the rounding technique is introduced as an efficient method for controlling this trade off. In this regard, multiplier circuits as a critical building block for computing in most of the processors have been considered for the evaluation of the rounding technique efficiency. The impact of the rounding method is investigated by comparison of circuit characteristics for three multipliers. These three multipliers are the conventional Wallace tree accurate multiplier, DRUM the recently proposed approximate multiplier, and the round based approximate multiplier proposed in this work. Simulation results for three selected technologies show significant improvement in the circuit characteristics in terms of power, area, speed, and energy for the proposed multiplier in comparison with its counterparts. Input data rounding pattern and the probability of the repetition for rounded values has been introduced as two essential items to control the level of the accuracy for each range of the data with minimum cost on the hardware. The proposed approximate multiplier algorithms are shown with Support Vector Machine classifier, which is therefore an attractive option for low power mobile applications.

Description

Keywords

Approximate multiplier, Rounding technique

Citation