SuperCode: Sustainability PER AI-driven Co-design Invited paper

Currently, data-intensive scientific applications require vast amounts of compute resources to deliver world-leading science. The climate emergency has made it clear that unlimited use of resources (e.g., energy) for scientific discovery is no longer acceptable. To address this challenge, the use of future and emerging computing architectures promises to be much more energy efficient. However, without well optimized code these cannot reach their full potential. Effectively using emerging architectures has proven challenging due to excessive cost and time involved in porting and optimising existing code.We propose a generic AI-driven co-design methodology, using specialized Large Language Models (like ChatGPT), to effectively generate efficient code for emerging computing hardware. Instead of conventional KPI s like computational efficiency or runtime, we propose sustainability as KPI, to emphasize our commitment to do more science with fewer resources. We validate our methodology with two challenging radio astronomy use-cases, terrestrial (LOFAR, SKA) and space-based (OLFAR). The primary transverse goal of SuperCode is to reduce the environmental impact of data-intensive applications by unlocking the use of emerging efficient hardware architectures, through a novel approach of AI-driven co-design. In contrast to normal co-design, where computational performance or efficiency is used as Key Performance Indicator (KPI), we introduce a sustainability score instead. 

We present the SuperCode project here in this form to introduce the vision behind the project and to disseminate the work in the spirit of Open Science and transparency. An additional aim is to collect feedback and invite potential collaboration partners and use-cases to join the project.