Turing AI Fellowship: PHOTONics for ultrafast Artificial Intelligence

In today's society, the massive deployment of smart devices, the popularity of online services and social networks, and the increasing global data traffic, makes the ability to process large data volumes absolutely crucial. Demand for Artificial Intelligence (AI) has therefore exploded, fuelled by an increasing number of industries (e.g. energy, finance, healthcare, defence) heavily relying on the efficient processing of large data sets. Nonetheless, the ever-growing data processing demand creates a pressing need to find new paradigms in AI going beyond current systems, capable of operating at very high speeds whilst retaining low energy consumption.

The human brain is exceptional at performing very quickly, and efficiently, highly complex computing tasks such as recognising patterns, faces in images or a specific song from just a few sounds. As a result, computing approaches inspired by the powerful capabilities of networks of neurons in the brain are the subject of increasing research interest world-wide, and are in fact already used by current AI platforms to perform these (and other) complex functions.

Whilst these brain-inspired artificial neural networks (ANNs) are supported to date by traditional micro-electronic technologies, photonic techniques for brain emulation have also recently started to emerge due to their unique and superior properties. These include very high speeds and reduced interference, among others. Remarkably, ubiquitous photonic devices such as vertical-cavity surface emitting lasers (VCSELs), the very same devices used in supermarket barcode scanners, computer mice and in mobile phones for auto-focus functionalities, can exhibit responses analogous to those of neurons but up to 1 billion times faster. VCSELs are also compact, inexpensive and allow practical routes for integration in chip modules with very low footprints (just a few mm2) making them ideal for the development of ultrafast photonic ANNs using ultrafast light signals instead of electric currents to operate. This permits exploring radically new research directions aiming at exploiting the full potential of light-enabled technologies for new paradigms in ultrafast AI. This Fellowship project will focus on this key challenge to develop transformative photonic ANNs using VCSELs as building blocks capable of performing complex computational tasks at ultrafast speeds, using data rates below 1 billionth of a second to operate. These will include the ultrafast prediction of complex data signals, of interest for example in meteorology forecasting, to very high speed data classification of interest in green-energy systems (e.g. analysis of wind patterns in off-shore wind-energy farms).

The research milestones of this programme are: (1) the design and fabrication of photonic ANNs using coupled VCSELs as building blocks, emulating the operation of the human brain at ultrafast speeds; (2) the development of chip-scale modules of VCSEL based photonic ANNs; (3) the demonstration of complex data processing tasks with photonic ANNs at ultrafast speeds (at data rates below 1 billionth of a second); (4) the delivery of photonic systems for AI, tackling key functionalities across strategic UK economic sectors (e.g. energy, defence).

In summary, by bringing together the hitherto disparate fields of brain-inspired computing and photonics, this programme proposes unique pioneering research in photonic ANNs for future ultrafast AI technologies.