Approximate computing provides a promising way to achieve low power design at the cost of acceptable error. As a core component in a processor, the performance of the multiplier is important. This study presents designs of approximate-truncated Booth multipliers (ATBMs) using proposed approximate modified radix-4 Booth encoders (AMBEs), approximate 4-2 compressors (ACs) and gradually truncated partial products. The accuracy of the ATBMs is adjustable with the so-called approximation factors that indicate the number of AMBEs and ACs used. The normalised mean error distance and the product of the power and delay are used to evaluate the error and the hardware performance of the multipliers. The results show that the proposed ATBMs outperform previous approximate Booth multipliers. Their validity is also shown with case studies of image processing, K-means clustering and handwritten digit recognition.

中文翻译：

---

近似截断的Booth乘法器的设计，评估和应用

近似计算提供了一种以可接受的误差为代价实现低功耗设计的有前途的方法。作为处理器的核心组件，乘法器的性能很重要。这项研究提出了使用建议的近似改良的radix-4 Booth编码器（AMBE），近似4-2压缩器（AC）和逐渐删减的部分乘积的近似截断的Booth乘法器（ATBM）。可以通过所谓的近似系数来调整ATBM的精度，该近似系数表示使用的AMBE和AC的数量。归一化的平均误差距离以及功率和延迟的乘积用于评估误差和乘法器的硬件性能。结果表明，提出的ATBM优于以前的近似Booth乘数。影像处理的案例研究也显示了它们的有效性，