Open-source Stand-Alone Versatile Tensor Accelerator

2025-09-24

AI Summary

Machine learning deployment in safety-critical domains (e.g., avionics) faces dual challenges of high computational demand and stringent certification requirements (e.g., DO-178C). Method: This paper introduces an open-source, standalone VTA tensor accelerator compiler pipeline that eliminates dependence on TVM. It proposes the first modular, Python-based compiler architecture explicitly designed for airworthiness certification compliance—enabling verifiable and extensible hardware–software co-design. The pipeline targets FPGA platforms and supports end-to-end compilation, simulation, and execution of CNN models. Contribution/Results: We fully implement and validate LeNet-5 on the VTA simulator, confirming functional correctness and performance feasibility. The pipeline demonstrates scalability to more complex CNN architectures while maintaining certification-aligned modularity and transparency—thereby advancing deployable, certifiable ML acceleration for safety-critical systems.

Technology Category

• Machine Learning: Hardware-aware ML
• Computer Vision: Large Vision Models
• Natural Language Processing: Safety and Robustness

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Security and Privacy: Security and privacy of machine learning and AI applications
• Search and Retrieval-Augmented AI: Vertical and domain-specific search

Abstract

Machine Learning (ML) applications demand significant computational resources, posing challenges for safety-critical domains like aeronautics. The Versatile Tensor Accelerator (VTA) is a promising FPGA-based solution, but its adoption was hindered by its dependency on the TVM compiler and by other code non-compliant with certification requirements. This paper presents an open-source, standalone Python compiler pipeline for the VTA, developed from scratch and designed with certification requirements, modularity, and extensibility in mind. The compiler's effectiveness is demonstrated by compiling and executing LeNet-5 Convolutional Neural Network (CNN) using the VTA simulators, and preliminary results indicate a strong potential for scaling its capabilities to larger CNN architectures. All contributions are publicly available.

Problem

• Developing standalone compiler for VTA FPGA accelerator without TVM dependency
• Addressing certification compliance challenges for safety-critical ML applications
• Enabling modular extensible compilation pipeline for aerospace ML workloads

Innovation

• Standalone Python compiler pipeline for VTA
• Designed with certification requirements and modularity
• Compiles and executes LeNet-5 CNN on VTA simulators