2023-10-29 22:08:19
Artistic rendering of a photonic chip with light and RF frequency coding data. Credit: B.Dong / University of Oxford
Researchers have developed integrated photonic-electronic hardware capable of processing 3D data. This innovation significantly improves the parallelism of data processing for AI tasks.
A revolutionary development in photonics and electronics hardware might significantly increase processing power for AI and machine learning applications. The approach uses multiple radio frequencies to encode data, allowing multiple calculations to be performed in parallel. The method shows promise for outperforming state-of-the-art electronic processors, with further improvements possible.
New advances in photonic-electronic hardware for AI
In an article published on October 19 in the journal Natural photonicsResearchers from the University of Oxford, along with collaborators from the Universities of Münster, Heidelberg and Exeter, report on their development of integrated photonic-electronic hardware capable of processing three-dimensional (3D) data, significantly improving the parallelism of data processing for AI tasks.
Challenges related to today’s computing power and the role of photonics
The processing efficiency of conventional computer chips doubles every 18 months, but the processing power required by modern AI tasks currently doubles approximately every 3.5 months. This means that new computing paradigms are urgently needed to cope with growing demand.
Artistic rendering of a photonic chip with light and RF frequency coding data. Credit: B.Dong / University of Oxford.
One approach is to use light instead of electronics, which allows multiple calculations to be performed in parallel using different wavelengths to represent different sets of data. Indeed, in groundbreaking work published in the journal Nature in 2021, many of these same authors demonstrated a form of integrated photonic processing chip that might perform matrix vector multiplication (a crucial task for AI and machine learning applications) at speeds far greater than electronic approaches the fastest. This work led to the birth of the photonic AI company Salience Labs, a spin-out from the University of Oxford.
Parallel Processing Innovations and Real-World Applications
Now, the team has gone further by adding an additional parallel dimension to the processing capacity of its matrix-vector multiplier photonic chips. This “higher dimensional” processing is made possible by leveraging several different radio frequencies to encode data, taking parallelism to a level far beyond that previously achieved.
As a test, the team applied their new hardware to the task of assessing the risk of sudden death from electrocardiograms of patients with heart disease. They managed to simultaneously analyze 100 electrocardiogram signals, identifying the risk of sudden death with an accuracy rate of 93.5%.
Future outlook and expert opinions
The researchers further estimated that even with moderate scaling of 6 inputs × 6 outputs, this approach can outperform state-of-the-art electronic processors, potentially providing a 100x improvement in power efficiency and density. calculation. The team plans further improvements to computational parallelism in the future, exploiting more degrees of freedom of light, such as polarization and mode multiplexing.
First author Dr Bowei Dong from the Department of Materials at the University of Oxford said: “We previously thought that using light instead of electronics might only increase parallelism by using different wavelengths – but then we realized that using radio frequencies to represent data opens up yet another dimension, enabling ultra-fast parallel processing for emerging AI hardware.
Professor Harish Bhaskaran, from the Department of Materials at the University of Oxford and co-founder of Salience Labs, who led the work, said: “This is an exciting time to be conducting research into AI hardware on a fundamental scale, and this work is an example of how what we thought was a limit can still be exceeded.
1698619111
#photonicelectronic #hardware #revolutionizes
