From square to cube: AI hardware processes in higher dimensions

 
 
 
Artistic rendering of a photonic chip with both light and RF frequency encoding data

"This is an exciting time to be doing research in AI hardware at the fundamental scale, and this work is one example of how what we assumed was a limit can be further surpassed".

Professor Harish Bhaskaran

In a paper published today (19 October 2023) in Nature Photonics, researchers from the University of Oxford, along with collaborators from the Universities of Muenster, Heidelberg and Exeter, report on their development of integrated photonic-electronic hardware capable of processing three-dimensional (3D) data, substantially boosting data processing parallelism for AI tasks.

Conventional computer chip processing efficiency doubles every 18 months, but the processing power required by modern AI tasks is currently doubling around every 3.5 months.  To cope with this we need new computing paradigms.  One approach is to use light instead of electronics - this allows us to carry out multiple calculations in parallel using different wavelengths to represent different sets of data. 

Indeed, in ground-breaking work published in the journal Nature a couple of years ago, many of the same authors demonstrated a form of integrated photonic processing chip that could carry out the key AI task of matrix vector multiplication at speeds far outpacing the fastest electronic approaches - work that resulted in the birth of the photonic AI company, Salience Labs.
 
Now the team has gone further by adding an extra parallel dimension to the processing capability of their photonic matrix-vector multiplier chips.  This 'higher dimensional' processing is enabled by exploiting multiple different radio frequencies to encode the data, propelling parallelism to a level far beyond that previously achieved.
 
As a test case, the team applied their novel hardware to the task of assessing the risk of sudden death from electrocardiograms of heart disease patients.  They were able to successfully analyse 100 electrocardiogram signals simultaneously, identifying the risk of sudden death with a 93.5% accuracy.
 
The researchers further estimated that even with a moderate scaling of 6 inputs x 6 outputs, this approach can outperform state-of-the-art electronic processors, potentially providing a 100-times enhancement in energy efficiency and compute density.  The team anticipates further enhancement in computing parallelism in the future, by exploiting more degrees of freedom of light, such as polarisation and mode multiplexing.
 
First author, Dr Bowei Dong, here in Materials, said:
 
"We previously assumed that using light instead of electronics could increase parallelism only by the use of different wavelengths - but then we realised that using radio frequencies to represent data opens up yet another dimension, enabling superfast parallel processing for emerging hardware".
The work was supported by the EU Horizon 2020 PHOENICS project (Grant No. 101017237) and the EU Innovation Council Pathfinder project HYBRAIN (Grant No. 101046878).
 
The full paper, 'Higher-dimensional processing using a photonic tensor core with continuous-time data' is published in Nature Photonics.