MUMBAI, India, Aug. 10 -- Intellectual Property India has published a patent application (202641089784 A) filed by Srm University-Ap on July 23, 2026, for Hardware-Accelerated Convolutional Neural Network Inference System For Handwritten Digit Image Classification.
Inventors include Baranala Sai Subrahmanya Tejesh; Ravisankar Dakupati; M. Sivaji; Dr. Pradyut Kumar Sanki; Dr. Saswat Kumar Ram; and Dr. Ramakrishnan Maharajan.
The application for the patent was published on July 31, 2026, under issue no. 31/2026.
Abstract: The present invention relates to a hardware-accelerated convolutional neural network (CNN) inference system (100) and a method comprising: a FPGA-based SoC platform (102) comprising an embedded processing system and programmable logic; an AXI Interconnect (104) configured to function as an internal routing switch fabric; a Memory Sub-system (106) configured to transfer image data between an external memory and programmable logic; a CNN Accelerator (108) implemented as a hardware intellectual property (IP) core within programmable logic of the FPGA-based SoC platform (102); and a System Reset Engine (110) configured to distribute reset signals across the Memory Sub-system (106) and the CNN Accelerator (108) to clean hardware initialization wherein the system and method demonstrate an classification accuracy of 94.54% across most handwritten digit classes and execution latency of 2.990 ms per image while requiring only 10.46% of the Look-Up Tables (LUTs) on FPGA-based SoC platform (102).
Disclaimer: Curated by HT Syndication.
