CGRA4ML: A Hardware/Software Framework to Implement Neural Networks for Scientific Edge Computing

View PDF HTML (experimental)

Abstract:The scientific community increasingly relies on machine learning (ML) for near-sensor processing, leveraging its strengths in tasks such as pattern recognition, anomaly detection, and real-time decision-making. These deployments demand accelerators that combine extremely high performance with programmability, ease of integration, and straightforward verification. We present cgra4ml, an open-source, modular framework that generates parameterizable CGRA accelerators in synthesizable SystemVerilog RTL, tailored to common ML compute patterns found in scientific applications. The framework supports seamless system integration through AXI-compliant interfaces and open-source DMA components, and it includes automatic firmware generation for programming the accelerator. A comprehensive verification suite and a runtime firmware stack further support deployment across diverse SoC platforms. cgra4ml provides a modular, full-stack infrastructure, including a Python API, SystemVerilog hardware, TCL toolflows, and a C runtime, which facilitates easy integration and experimentation, allowing scientists to focus on innovation rather than dealing with the intricacies of hardware design and optimization. We demonstrate the effectiveness of cgra4ml to implement common scientific edge neural networks using ASIC and FPGA design flows.

Comments: Accepted for publication in ACM TRETS 2026

Subjects:

Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI)

MSC classes: 68T07, 68U01, 65Y05

ACM classes: B.7.1; C.1.3

Cite as: arXiv:2408.15561 [cs.AR]

(or arXiv:2408.15561v4 [cs.AR] for this version)

https://doi.org/10.48550/arXiv.2408.15561

arXiv-issued DOI via DataCite

Related DOI:

https://doi.org/10.1145/3801097

DOI(s) linking to related resources

Submission history

From: G Abarajithan [view email] [v1] Wed, 28 Aug 2024 06:24:13 UTC (2,489 KB) [v2] Thu, 29 Aug 2024 01:26:50 UTC (1,519 KB) [v3] Wed, 4 Feb 2026 19:27:47 UTC (2,579 KB) [v4] Thu, 26 Mar 2026 22:59:53 UTC (2,519 KB)