ABSTRACT

Approximate computing can be considered a well-promising method for improving the critical performance criteria, such as latency and power consumption at the cost of accuracy penalty. The (4,2) compressor is a vital block at the heart of various arithmetic circuits like multipliers. This paper proposed two approximate (4,2) compressors with power and delay improvements with the error-correcting module. The simulation results based on 7 nm FinFET technology show that the proposed designs compared to the most efficient related approximate (4,2) compressors, reduce delay, power, and energy consumption. Furthermore, according to the comprehensive Monte Carlo simulations, the robustness of the proposed designs in the presence of significant FinFET process variations compared to the previous compressors is proven. In addition, an image multiplication process assesses the effectiveness of the proposed approach in the qualitative assessment of the inexact multiplier. The obtained results indicate acceptable accuracy of inexact multiplier based on the proposed approximate (4,2) compressor.

摘要

近似计算可以被认为是一种很有前景的方法，可以以牺牲精度为代价来改善延迟和功耗等关键性能标准。(4,2) 压缩器是乘法器等各种算术电路核心的重要模块。本文提出了两种近似 (4,2) 压缩器，通过纠错模块提高了功率和延迟。基于 7 nm FinFET 技术的仿真结果表明，与最高效的相关近似 (4,2) 压缩器相比，所提出的设计减少了延迟、功耗和能耗。此外，根据全面的蒙特卡洛模拟，与以前的压缩机相比，在存在显着 FinFET 工艺变化的情况下，所提出的设计的稳健性得到了证明。此外，图像乘法过程评估了所提出的方法在不精确乘数的定性评估中的有效性。获得的结果表明基于所提出的近似（4,2）压缩器的不精确乘法器的精度是可接受的。