Program

From Low Power Tiny Machine Learning to Energy Efficient Generative Edge AI

Abstract: On 2018, the challenge to bring artificial intelligence at the edge was represented by the quest to achieve under one mW machine learning with tight constraints on computational and memory requirements. Initially solutions were made by standard micro controllers, then sensors and then dedicated neural processing unit. That heterogenous hardware required unified software technologies to achieve productive deployment of pretrained neural networks as viable approach for the developers.

While the associated developments are industry endorsed, a new challenge is impacting the edge: generative artificial intelligence. It is dramatically changing the edge land scape by prioritizing energy efficiency rather than low power. Furthermore, new software frameworks are appearing linking small and medium language models with reasoning skills.

All of that represents a huge opportunity for new chip developments.