Imec Cambridge UK gets off to a flying start with a funded project to develop cutting-edge Design Space Exploration (DSE) framework for AI training

LEUVEN (Belgium), 25 October, 2024— This week, imec, a world-leading research and innovation hub in nanoelectronics and digital technologies announces it has secured support from the UK’s Advanced Research and Invention Agency (ARIA) to develop a predictive system-level software framework for training of large AI models, incorporating future emerging compute, memory and connectivity technology models. Accessible to the broader research community, the cutting-edge simulation platform aims to facilitate research into hardware and technology for AI training. It is the first funded project awarded to imec Cambridge UK building upon its expertise in system-technology co-optimization (STCO).

Training (the model learns from data) and inference (the trained model makes predictions on unseen data) are two fundamental stages in the lifecycle of artificial intelligence (AI) and machine learning (ML) models. While current design space exploration (DSE) frameworks have primarily targeted AI inference, and inference hardware is being widely developed in academic research communities, the exponential growth of ML applications calls for more efficient AI training processes. Moreover, training complex models is time-consuming, involves large compute infrastructure, that results in high operating costs and significant carbon emissions, thereby raising profound sustainability concerns. Imec Cambridge UK will leverage its expertise in system-technology co-optimization (STCO), to understand the broader impact of new technologies on the overall efficiency of these large systems. STCO enables to tackle challenges like power/thermal limits, memory constraints and bandwidth bottlenecks, and achieve optimal energy usage.

James Myers, program director STCO at imec Cambridge UK: “We are pleased to be selected for this exciting project and keen to get started. Thanks to the £3M ARIA funding, we will be able to create a robust DSE framework that can be used in early-stage explorations of optimal system and technology configurations for AI training, thereby validating or motivating new technology specifications. The new DSE will leverage established component-level simulators available in the academic community that will be augmented with emerging technology models such as short-range optical interconnects or future nanosheet and CFET CMOS nodes.” Bruno Jansen, regional managing director: “With this project, imec Cambridge UK also intends to grow its local team and expand its involvement with the UK research ecosystem. We believe imec's advanced technologies have a strong role to play and remain open to other collaboration opportunities.”

Imec anticipates leveraging its knowhow in emerging semiconductor technologies to ensure that the new DSE framework remains supported by state-of-the-art technology capabilities. This will guarantee the continued relevance and applicability of its outputs as the field of AI evolves.