A research team at the University of Limerick has made a major discovery by designing a molecule that could revolutionise computing.
Researchers at UL’s Bernal Laboratory have discovered new ways to investigate, control and manipulate materials at the most fundamental molecular scale.
These findings have been used in an international project involving experts around the world to help create an entirely new type of hardware platform for artificial intelligence, which will enable unprecedented improvements in computational speed and energy efficiency.
The study was just published in the journal Science. nature.
The UL team, led by Damien Thompson, Professor of Molecular Modelling at UL and Director of the Irish Pharmaceutical Research Centre SSPC, and in collaboration with international scientists from the Indian Institute of Science (IISc) and Texas A&M University, believe that this new discovery will lead to innovative solutions to enormous societal challenges in health, energy and the environment.
Professor Thompson explained: “The design is inspired by the human brain, which processes and stores information by using the natural wobble and jolt of atoms. As the molecules rotate and bounce around the crystal lattice, they create numerous individual memory states.
“We can trace the paths of molecules inside the device, mapping each snapshot to a unique electrical state. This creates a sort of tour diary of the molecules that can be written to and read from just as in a conventional silicon-based computer, but here with dramatically improved energy and space economy because each item is smaller than an atom.
“These out-of-the-box solutions could provide significant benefits to all computing applications, from energy-hungry data centers to memory-intensive digital maps and online games.”
Until now, neuromorphic platforms, a computing approach inspired by the human brain, have only worked on low-fidelity tasks such as inference in artificial neural networks. This is because core computing tasks, including signal processing, neural network training, and natural language processing, require much higher computational resolution than what existing neuromorphic circuits can provide.
For this reason, achieving high resolution has become a major challenge in neuromorphic computing.
The team reimagined the underlying compute architecture to achieve the required high resolution and perform resource-intensive workloads with unprecedented energy efficiency: 4.1 tera operations per second per watt (TOPS/W).
The team’s groundbreaking achievement has the potential to expand neuromorphic computing beyond niche markets, unlocking the long-heralded transformative benefits of artificial intelligence and powering the core of digital electronics from the cloud to the edge.
“By precisely controlling the molecular motions across a wide range of states, we have created the most accurate and fully featured 14-bit neuromorphic accelerator ever integrated onto a circuit board,” said IISc project leader Professor Sritosh Goswami. “It can handle machine learning workloads such as signal processing, artificial neural networks, autoencoders and generative adversarial networks.”
“The most important thing is that we can train neural networks at the edge by leveraging the high precision of accelerators, which solves one of the most pressing challenges in AI hardware.”
Additional improvements will be made as the team works to expand the range of materials and processes used to build the platform, and to further increase processing capabilities.
“The ultimate goal is to replace what we think of as computers today with high-performance ‘everything devices’ built on energy-efficient and environmentally friendly materials, providing ubiquitous information processing distributed throughout the environment, integrated into everyday objects, from clothing to food packaging to building materials,” explained Professor Thompson.
