AMCAL: Approximate Multiplier With the Configurable Accuracy Levels for Image Processing and Convolutional Neural Network

In algorithms for machine learning and signal processing, multiplication is an essential arithmetic operation. In order to save energy and space, this research suggests a unique technique called Approximate Multiplier with Configurable Accuracy Levels (AMCAL). The suggested approach manipulates the leftover bits and truncates the least significant bits to carry out the multiplication operation. The remaining input bits have been altered with a smaller bit width in order to minimize or eliminate the error that results from truncating the bits. Additionally, it has demonstrated that, in contrast to other algorithms, we are less concerned with the size of the error introduced into the input. When compared to the Wallace multiplier, the suggested AMCAL multiplier reduces energy consumption .Furthermore, the suggested multiplier performs better than other approximation multipliers in the same class, including DRUM and DSM, in terms of latency, power, and area. In terms of power and area efficiency, the AMCAL multiplier outperforms new approximate multipliers like DSI and TOSAM. Ultimately, it is demonstrated that the image quality produced by four image processing programs—smoothing, sharpening, JPEG encoding, and face alignment—is unaffected by such a small computational error.