Approximate computing is a recent trend that reduces energy dissipation and design complexity in a variety of applications by sacrificing computational accuracy. But because of NNs' and their applications fault tolerance, approximate computing methods can be used to lower implementation costs. In this research, a novel Truncated Approximate Multiplier based Strip Convolutional neural network (TAM-SCNN) has been proposed to tackle the increased hardware cost. The proposed method utilizes a truncated approximation multipliers instead of exact multipliers, the Strip CNN can perform better in terms of power, area, and speed. In addition, a approximate 4:2 compressor and adder allow for significant gate-level simplicity, critical route latency reduction, and complexity reduction. Area, power, delay accuracy analysis, PSNR, and MSSIM are used to assess the suggested method's performance. Xilinx ISE 13.2 is used to simulate image processing applications and verify the performance of the suggested multipliers. The proposed TAM-SCNN multiplier achieves 35.2% reduction in area, 31.8% reduction in power consumption, and 43.9% reduction in delay compared to the existing methods. In addition, the propose multipliers outperform existing multipliers in terms of MSSIM and PSNR.
Antony et al. (Fri,) studied this question.