Machine learning accelerator design in HDL

Summarise with VIRAT
Authors: Yutong Li
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The Convolutional Recurrent Neural Network (CRNN) delivers strong performance in sequence recognition tasks. However, its hybrid architecture and computational complexity hinder its deployment on embedded devices. Therefore, achieving efficient hardware implementation of CRNN is a key challenge. To address this, this project designs an FPGA-based hardware acceleration system for CRNN. Through a modular architecture, dataflow optimization, and fixed-point implementation, the system achieves efficient hardware mapping for CRNN inference. Compared with an unoptimized baseline, the proposed shared RAM with unified scheduling reduces on-chip memory usage by approximately 20%–30%, the line-buffer and sliding-window scheme lowers bandwidth demand by about 60%–70%, a caching mechanism tailored to pooling operations reduces overall memory accesses by roughly 85%, and a shared MAC array improves DSP utilization by approximately 35%. Simulation verification in Vivado demonstrates that the system operates stably and correctly executes CRNN inference. This work provides a feasible solution for implementing CRNN on FPGA and establishes a foundation for future optimization and deployment.

Fields of Study & Subjects

Embedded Systems Design Techniques Advanced Neural Network Applications Embedded Systems and FPGA Applications

Publication details

Journal / source
DR-NTU (Nanyang Technological University)
Publication date
1 January 2026
Language
EN
Document type
Thesis