Can we cut the energy consumption of current AI technologies?

As artificial intelligence, commonly referred to as AI, becomes increasingly integrated into various aspects of our daily lives, the energy demands of AI systems continue to rise. Despite AI being inspired by the human brain, it is significantly less energy efficient because most tasks are implemented on computer hardware that is not specifically tailored for this purpose. The human brain remains an exemplar of energy-efficient computation, a standard that modern technology strives to replicate.

INL researchers are studying advanced materials and devices aimed at revolutionising AI systems. Carlos Rosário, ERA fellow at INL (Marie Skłodowska-Curie actions (MSCA)), focuses his research on reducing the substantial energy consumption required by current AI technologies.

Carlos’ work is grounded in the field of neuromorphic computing, which seeks to mimic the brain’s intricate network of neurons and synapses using artificial equivalents. The goal is to create an electronic brain capable of performing complex computations with minimal energy use.

One of the key areas of Carlos Rosário’s research is the exploration of 3D topological insulators for novel electronic devices. These cutting-edge materials exhibit unique properties: while they are insulators and do not conduct electricity through their bulk, their surfaces behave like metals, allowing for efficient electrical conduction (among other peculiar physical phenomena).

Topological insulators hold significant promise for the future of electronics. Their distinctive characteristics could lead to the development of next-generation neuromorphic computing devices that are not only more efficient but also consume considerably less energy. Carlos Rosário explains that “one of the current challenges being targeted is the integration of the topological insulators with other common (topologically trivial) materials and micro- and nanofabrication technologies needed to fabricate the devices.”

“This research could pave the way for major advancements in how we build and use AI systems,” says Rosário. “By leveraging the special properties of topological insulators, we aim to create devices that bring us closer to the energy efficiency of the human brain.”

The ongoing project at INL exemplifies the potential of material science to address some of the most pressing challenges in AI development. As Carlos Rosário and his team continue their groundbreaking work, the future of AI could see significant improvements in sustainability and efficiency.

Carlos recently participated in ‘90 Segundos de Ciência’, a series of science communication episodes with daily broadcasts on Antena 1 Portuguese public radio, where Carlos explains the motivation behind the project TopRooT in just 90 seconds. You can listen to the episode (in Portuguese) here.

Text and Photography by Catarina Moura, Science Communication Officer