Low-Latency Coherent Optical Networks for AI Data Centre Clusters

Project ID: 2531bd1663

(You will need this ID for your application)

Research Theme: Information and Communication Technologies

Research Area(s): Information and communication technologies theme
Optical communications
Digital signal processing

UCL Lead department: Electronic and Electrical Engineering (EEE)

Department Website

Lead Supervisor: Filipe Ferreira

Project Summary:

Why this research is important

AI models are advancing faster than the technology connecting the GPUs that train them. The training of the largest models now demand full high-end datacentres (~100MW), pushing local energy grids to their limits. Reducing latency and energy consumption in the optical interconnects linking regional data centres is therefore vital to sustain AI growth and enable next-generation AI technologies. Current coherent optical transceivers, which can take hundreds of milliseconds to (re-)establish connections, are far too slow for distributed AI workloads. This project tackles that challenge directly, transforming optical transceivers to ‘instantaneously’ adapt to new fibre connections by embedding machine-learning models directly into the signal-equaliser chip.

Who you will be working with

You will join the Optical Networks Group at UCL, a world-leading centre for optical communications research, supervised by Dr Filipe Ferreira, UKRI Future Leaders Fellow. Industrial advisors include BT Labs, Corning, and Oriole Networks, providing exposure to real-world innovation in network systems, fibre technology, and optical switching. You will also collaborate with UCL Advanced Research Computing and the National Dark Fibre Facility (NDFF), where field trials are expected to take place.

What you will be doing

You will design and experimentally demonstrate new low-latency optical transceiver architectures for AI data-centre connectivity. The work will involve developing real-time FPGA&ASIC implementations of machine-learning-driven digital signal processing, capable of nanosecond-scale response and low-power operation. You will validate your designs on UCL’s high-speed optical testbeds under realistic AI traffic and contribute to a prototype demonstration achieving sub-microsecond latency.

Who we are looking for

We welcome motivated applicants with backgrounds in electronic or electrical engineering, physics, or computer engineering, and an interest in photonics, communications, or AI-hardware. Experience with signal processing, FPGAs, or machine learning is helpful but not essential: curiosity, creativity, and enthusiasm for cross-disciplinary research are key.