A novel method for reliability-aware
techniques in approximate arithmetics (adders and multipliers) for Internet of
Things
As transistors shrink down to nanometer scales, improving the energy-efficiency of computer systems has become more complex than ever. Approximate Computing (AC) is a promising approach to address energy-efficiency problems of approximation/imprecision-tolerant systems, by performing a trade-off between the computation accuracy and circuit metrics (i.e., power consumption, area occupation, and the delay). Making approximate computing specific to user requirements is crucial to system performance, energy-efficiency, and reliability. However, developing hardware for such optimization becomes a significant challenge due to the high cost of examining all potential choices while exploring a large design space. Therefore, the efficiency of evaluating error metrics, e.g., Mean Error Distance (MED), Mean Squared Error (MSE), Mean Error (ME), and Error Probability (EP), is a determinant aspect of the design space exploration. Since computing these error-metrics is quite time-consuming, efficient calculation approaches are essential. Since computing these error-metrics is quite time-consuming, efficient calculation approaches are essential. Digital adder is one key component and is widely used in approximation-tolerant applications. This thesis proposes a novel formal approach to accurately compute the MED, MSE, ME,and EP of approximate low power adders for any input pattern at a linear time and space complexity. Our experimental results indicate that the proposed approach can accurate compute error-metrics of large approximate adders at a 150 times faster speed compared to the Monte Carlo sampling methods. We then develop AxMAP, a design tool based on the proposed error-metrics computation that generates energy-efficient approximate adders for any given input pattern. AxMAP produces more than 200 different designs for adders that achieve superior performance and energy-efficiency compared to the existing state-of-the-art approximate adders, up to 50%. When applied to image processing applications (i.e., image addition), AxMAP outputs achieve superior performance and energy-efficiency compared to the existing state-of-the-art approximate adders. |