autoAx: An Open-Source Automated Design Space Exploration Framework for Approximate Accelerators in FPGAs and ASICs

autoAx: An Open-Source Automated Design Space Exploration Framework for Approximate Accelerators in FPGAs and ASICs

Software

The automated generation of approximate circuits and accelerators has been a useful design strategy to achieve energy efficiency and/or performance improvements. In this work, we propose a framework, autoAx, that leverages machine learning models that evaluate the state-of-the-art approximate components to explore the architecture space effectively. These accelerators are modeled at RTL and optimized using an evolutionary algorithm. The AutoAx framework is extensible, open-source, and can assist in exploring new directions in high-level approximation. 

The automated generation of approximate circuits and accelerators has been a useful design strategy to achieve energy efficiency and/or performance improvements. In this work, we propose a framework, autoAx, that leverages machine learning models that evaluate the state-of-the-art approximate components to explore the architecture space effectively. These accelerators are modeled at RTL and optimized using an evolutionary algorithm. The AutoAx framework is extensible, open-source, and can assist in exploring new directions in high-level approximation. 

approximate computing, high level synthesis, machine learning

approximate computing, high level synthesis, machine learning

https://github.com/ehw-fit/autoax

Využití výsledku jiným subjektem je možné bez nabytí licence (výsledek není licencován)

https://github.com/ehw-fit/autoax