Design and Implementation of Area Efficient Approximate MAC Unit for Deep Neural Network based Architectures and Applications

The Multiply Accumulate Unit, which is utilized to boost the processor's overall speed, is the subject of this essay. Applications for digital signal processing that use multiply-accumulate units include convolution, digital filters, image, video, and audio, among others. Accuracy is not given priority when image and video processing applications are taken into account, hence an approximate multiply-accumulate unit is built. This approximate multiply-accumulate unit, compared to a floating point multiply-accumulate unit simulated in Xilinx ISE 14.5, and various parameters like area, delay, and speed are compared between floating and approximate MAC consisting of an approximate multiplier efficient, an approximate adder, and an approximate accumulator. The approximate multiply-accumulate unit uses less space (66% less) and has a 75% shorter delay, all of which contribute to its high speed.