Approximate Population Count Using the Harley-Seal Method for Error-Tolerant Applications

Abstract

Population count is a necessary process in several fields, such as cryptography, database search, data mining, and machine learning. Real-time problems have very large datasets, which require enhanced performance. Therefore, the objective of this paper was to propose a modified approach to the hamming weight algorithm to allow the implementation of approximate calculations on error-tolerant applications, such as machine learning and database searches, which do not require precise results. The proposed approach used approximate adders, rather than the exact carry save adder computations used in the Harley-Seal methodology of bit counting, along with modified imprecise error-tolerant adder type II (ETAII), named Approximate Harley-Seal with Modified ETAII (AHS-METAII). The precise versus imprecise designs of Harley-Seal approach were tested, evaluated, and compared to show that implementing partial logic functions instead of fully logic functions resulted in 27% power reductions with a slight decrease (9%) in the accuracy level over traditional adders on a 64-bit stream. The simulation results demonstrated that the proposed approximate approach model using Verilog was faster than the exact methods by 3% and consumed 16% less area.