Analysis and design of approximate inner-product architectures based ondistributed arithmetic

Distributed arithmetic (DA) based architectures are popularly used for inner-product computation in various applications.Exi sting literature shows that theuseof approximate DA-architectures in error resilient applications providesas ignificant improvement inthe overall efficiencyofthe system. Based on precise error analysis, wefind that theexi sting methods introduce large truncation error inthe computation ofthefinal inner-product. Therefore, tohavea suitable trade-off between the overall hardware complexity and truncation error, aweight-dependent truncation approach is proposed inthis paper. The overall efficiencyofthe structureis further enhanced by incorporating an input truncation strategyinthe proposed method. It is observed that the area, time and energy efficiencyofthe proposed designs are superiortothee xisting designs with significantly lower truncation error. Evaluation intheca se ofnoi sy image smoothing application isalso shown inthis paper.