Speaker
Description
TASTE is an open-source framework for Model-Based development of accelerators used in space applications. It provides a seamless workflow that integrates design, development, and deployment while bridging multiple existing technologies instead of introducing a new modeling language.
This presentation explores the integration of ONNX-based machine learning (ML) models into the TASTE toolchain as part of an ongoing OSIP initiative. The objective is to extend TASTE’s capabilities to support ML-driven space applications, including onboard data processing and autonomous decision-making. The approach enables the seamless incorporation of ML models alongside traditional modeling tools such as OpenGeode and MATLAB Simulink.
A key focus is the integration of the open-source PandA Bambu High-Level Synthesis (HLS) tool with the ONNX-MLIR framework to generate efficient FPGA-accelerated ML code. This combination provides an automated path from high-level ML model descriptions to optimized hardware implementations, addressing the need for high-performance and energy-efficient computing in space systems.
The presentation will cover the methodology, key technical challenges, and implementation details, along with experimental results on selected ML models. Attendees will gain insights into the benefits of this approach and its potential impact on future space missions.
Affiliation of author(s)
Politecnico di Milano (Italy)
| Track | High Level Synthesis and Model Based Design |
|---|
