High accurate approximate adders using hybrid gates

Approximate adders (AxAs) are at the core of image processing algorithms, of which high-order bits are exactly designed and low-order bits are loosely realized. In this paper, a carry prediction circuit using an AND gate and a majority gate is proposed to generate a carry signal propagated from the approximate part to the exact part. AxAs designed using the carry prediction circuit have been significantly improved in terms of computing accuracy and overall performance. Compared to previous designs, the mean squared error (MSE) and overall performance values for 16-bit AxAs with 4 and 8 approximated bits have been improved by up to 88.48 % and 87.91 %, 63.69 % and 62.01 % on average, respectively. To investigate the efficacy of the AxAs in real applications, they are applied in three image processing algorithms, including image addition, Gaussian smoothing, and edge detection. The results show that the proposed AxAs generate better image quality for three applications than previous AxAs in general.