Researchers discover building blocks that could 'revolutionize computing'

The researchers at UL's Bernal Institute have discovered new ways of probing, controlling and tailoring materials at the most fundamental molecular scale.

The results have been used in an international project involving experts worldwide to help create a brand-new type of hardware platform for artificial intelligence that achieves unprecedented improvements in computational speed and energy efficiency.

The research has just been published in the scientific journal Nature.

The UL team, led by Damien Thompson, Professor of Molecular Modelling at UL and director of SSPC, the Research Ireland Centre for Pharmaceuticals, in an international collaboration with scientists at the Indian Institute of Science (IISc) and Texas A&M University, believe that this new discovery will lead to innovative solutions to societal grand challenges in health, energy and the environment.

Professor Thompson explained: "The design draws inspiration from the human brain, using the natural wiggling and jiggling of atoms to process and store information. As the molecules pivot and bounce around their crystal lattice, they create a multitude of individual memory states.

"We can trace out the path of the molecules inside the device and map each snapshot to a unique electrical state. That creates a kind of tour diary of the molecule that can be written and read just like in a conventional silicon-based computer, but here with massively improved energy and space economy because each entry is smaller than an atom.

"This outside the box solution could have huge benefits for all computing applications, from energy hungry data centres to memory intensive digital maps and online gaming."

To-date, neuromorphic platforms -- an approach to computing inspired by the human brain -- have worked only for low-accuracy operations, such as inferencing in artificial neural networks. This is because core computing tasks including signal processing, neural network training, and natural language processing require much higher computing resolution than what existing neuromorphic circuits could offer.

For this reason then, achieving high resolution has been the most daunting challenge in neuromorphic computing.

The team's reconceptualization of the underlying computing architecture achieves the required high resolution, performing resource-intensive workloads with unprecedented energy efficiency of 4.1 tera-operations per second per watt (TOPS/W).

The team's breakthrough extends neuromorphic computing beyond niche applications in a move that can potentially unleash the long-heralded transformative benefits of artificial intelligence and augment the core of digital electronics from the cloud to the edge.

Project lead at IISc Professor Sreetosh Goswami said: "By precisely controlling the vast array of available molecular kinetic states, we created the most accurate, 14-bit, fully functional neuromorphic accelerator integrated into a circuit board that can handle signal processing, AI and machine learning workloads such as artificial neural networks, auto-encoders, and generative adversarial networks.

"Most significantly, leveraging the high precision of the accelerators, we can train neural networks on the edge, addressing one of the most pressing challenges in AI hardware."

Further enhancements are coming, as the team works to expand the range of materials and processes used to create the platforms and increase the processing power even further.

Professor Thompson explained: "The ultimate aim is to replace what we now think of as computers with high-performance 'everyware' based on energy efficient and eco-friendly materials providing distributed ubiquitous information processing throughout the environment integrated in everyday items from clothing to food packaging to building materials."