Low-Complexity Deep Neural Network Accelerator on VLSI Using Stochastic Computing

Abstract

Convolutional Neural Networks continue to be dominant research inside the areas of GPU acceleration using Programmable Arrays, demonstrating their efficacy in a variety of technical vision tasks such as extraction of features, image analysis, person detection, and rear bridge alert, among many others. Nonetheless, there are other constraints to deploying the Deep network on FPGA, such as limited on-chip recall, DNN dimensionality, and parameters. This study proposes Tv commercial, a powerful DNN prototype based on the baseline Alexnet prototype. The proposed framework makes use of a Commercial engine, which is a developed version of the insight different and independent randomization unit. Designers also propose a GPU integration model that supports the initiation characteristics Mish and also Rectified linear. Regardless of the fact that only a minimal quantity of computer equipment was employed, the practical results were attained. When compared to state-of-the-art techniques, the given scheme offers a comparatively high detection accuracy while requiring fewer computer system resources. Furthermore, when compared to the estimate technique, the proposed framework helps to reduce system resources even more than others.