ATM: Approximate Toom-Cook Multiplication for Speech Processing Applications

摘要

Approximate Computing has paved way for elaborate savings in design area and latency of modern system architectures processing images or signals, by a deliberate yet tolerable loss of functional accuracy. This paper thus proposes a design of an approximate multiplier based on the efficient Toom-Cook algorithm, that has a lower complexity of O(N^logd(zd−1)) than O(N²), for order d. Inherent integer divisions in the algorithm has restricted its feasibility in hardware, unless without suitable approximation. On an average, the proposed multiplier achieves 53%, 18% and 57% improvements in area, delay and power only with less than 1% mean error. Owing to these benefits due to lower computational complexity, the multiplier can be configured to achieve significant savings with a high quality output and that suits well to the nature of the speech processing systems, hence the design works well for the epoch extraction system in speech.