About the project
This PhD project will research how advanced system-technology co-optimisation (STCO) can be harnessed to enhance the efficiency and sustainability of inference, exploring full-stack co-design including the AI model itself.
As AI becomes increasingly embedded in everyday technology, there is growing demand for systems that are faster, more accurate, and more intelligent. These place significant strain on hardware, and the 50-years of relentless improvements that Complementary Metal-Oxide-Semiconductor (CMOS) scaling has offered are reaching their limits.
Instead, industry is shifting toward heterogeneous integration, including chiplets, advanced 2.5D/3D packaging, application-specific accelerators, and novel memory technologies. This emerging design philosophy—sometimes referred to as “CMOS 2.0”—relies not on a single technology but on mixing-and-matching technology ingredients to unlock system scaling bottlenecks. However, this significantly increases the complexity of the design space, and exacerbates thermal issues resulting from increased power density.
This PhD project will research how advanced system-technology co-optimisation (STCO) can be harnessed to enhance the efficiency and sustainability of inference, exploring full-stack co-design including the AI model itself, e.g.:
- Developing methods to incorporate AI model design into STCO workflows, e.g. using neural architecture search (NAS)
- Demonstrating design points to showcase how such full-stack STCO could be leveraged to optimise the system for inference accuracy/latency, efficiency, thermal dynamics and power density, and/or embodied carbon;
- Exploring how sustainability and efficiency trade-offs (e.g. achieving sufficient rather than maximal AI accuracy) can influence system design choices.
The project is supported by Imec UK, providing industrial co-supervision, expertise on STCO, and access to technology details (e.g. 3D pitches, PDKs, NVM datasheets, embodied carbon estimates, etc).
You will join a dynamic and diverse cohort, benefiting from expert mentorship and interdisciplinary collaboration.
This PhD project is part of the UKRI AI Centre for Doctoral Training in AI for Sustainability (SustAI), a 4-year integrated programme (iPhD).
The programme includes comprehensive training in
- sustainability
- AI and machine learning
- digital design
preparing students for a career at the forefront of research in this area.
Students will have access to state-of-the-art facilities and resources, fostering an environment of innovation and excellence.
