An Approximate Floating-Point Multiplier Based on Partial Product Probability Analysis

Abstract

The invention discloses an approximate floating-point multiplier based on partial product probability analysis, which includes a sign bit XOR module, a mantissa approximate multiplication module, a normalization module, a rounding module, an exponent addition module, an exponent adjustment module, and special case processing module and result output module; the mantissa approximate multiplication module includes truncation and compensation unit, low OR gate compression unit, approximate 4‑2 compressor and precise compressor; The weight bit is compensated, and every two partial products of the compensation bit and the first intermediate weight bit are compressed into one bit using an OR gate, and each four partial products of the second intermediate weight bit are approximately compressed; Precise compression. The invention can effectively simplify the structure of the compressor and generate as few errors as possible, adjust the input order without generating additional errors, and reduce the complexity of the compression structure while ensuring the accuracy of the multiplier.

Description

technical field [0001] The invention relates to the technical field of approximate circuits, in particular to an approximate floating-point multiplier based on partial product probability analysis. Background technique [0002] With the gradual slowdown of the development of semiconductor process technology and the failure of Dennard's scaling law, the energy consumption and efficiency of integrated circuits are facing severe challenges. As an emerging computing paradigm, approximate computing provides a new idea for solving the high energy consumption problem of integrated circuits, that is, by sacrificing appropriate accuracy in exchange for considerable energy consumption and area gains. For applications with certain fault-tolerant capabilities such as data recognition, image processing, machine learning, and wireless communication, reasonable results can be produced even if the introduction of approximate calculations will cause some accuracy degradation. As an arithmet...

AI Technical Summary

Problems solved by technology

[0003] There is little research on the approximate floating-point multiplier in the existing technology. At present, the approximate floating-point multiplier is mainly designed for the mantissa multiplication part, without considering the specific distribution of the partial product in the mantissa multiplication. Introduce additional error when the probabilities are different

Images