A newton raphson method based approximate divider design for color quantization application

Abstract

Approximate computing in general offers high per-formance with a compromise in the exactness of the results achieved, that has grabbed utmost attention in designing artificial intelligence (AI) on chip applications. A novel reciprocal and iterative approximation process using Newton Raphson method on integer data is proposed, where the design not only achieves results close to exact value, but also makes it performance efficient. The novel divider design for both 8-bits and 16-bits were realized in 45 nm CMOS technology. The 8-bit and 16-bit approximate divider using newton raphson (ADNR) method demonstrated an average accuracy of 90.30%, and 100% re-spectively, and achieved 60% improvement in performance over the existing approximate divider implementation. The ADNR method was further logically implemented for K-means color quantization application in Matlab, and the results were found to be comparable to that of exact divider.