Permutation-Based Approximate Multiplier with High Accuracy.

Kunlong Li,Yunfei Dai,Zhen Li,Lingli Wang

DOI: https://doi.org/10.1109/ASICON58565.2023.10396418

2023-01-01

Abstract:In computationally complex applications, the tolerance of errors and the restriction of energy consumption stimulate the research into approximate computing, a promising methodology to trade off the hardware cost and accuracy loss. In this paper, we propose an automated design method to generate approximate substitutes for a widely-used arithmetic operation—multiplication. The partial products of two rows are compressed by logical operations, and the partial products in adjacent columns can also be permutated. Our proposed 8-bit multiplier can achieve 19.6% higher accuracy, 3.8% lower power, and 11.3% less area than the best reproduced approximate multiplier under the uniform distribution. Compared with an exact multiplier, the reduction in power and area is 36.2% and 32.8%, respectively. To evaluate the design, we carry out a simulation in image blurring with a 5*5 Gaussian kernel by replacing all the multiplications. The Peak Signal-to-Noise Ratio (PSNR) between the exact multiplier and ours can attain 50.10 dB, which outperforms all other reproduced approximate multipliers.