Article ID: 13.20160814
Approximate computing has drawn recent attention as it can reduce power without significantly sacrificing the output quality for arithmetic operations in error-resilient applications. However, the error characteristics evaluation often depends on time consuming random simulation and estimation. In this paper, a formal framework of evaluating approximate adders is proposed based on a theoretical model. The proposed framework can accurately measure mean squared error (MSE) and mean absolute error (MAE) without using functional simulations. An approach has been adopted to estimate the final PSNR value in terms of MSE and MAE for DCT-IDCT image compression system. Experimental results have validated the accuracy and usefulness of the framework.
