Samuel Stark

PhD thesis title: Investigating heterogeneous capability domains in the context of CPUs and GPUs

“Big data” computation, which relies on cooperation between CPUs (the main “brains” of a computer system) and accelerator cards such as GPUs, has become ubiquitous. These computations handle large amounts of data, some of which may be sensitive personal information, and as such security is extremely important.

The CHERI project at Cambridge describes how one can add extra memory protection to CPUs, but has not yet fully investigated how to best extend this protection to GPUs. On top of that, external GPUs control their own memory, and it’s important to ensure that CPUs can’t access external GPU memory without permission (and vice versa). I’m interested in guaranteeing security under various trust models for systems made up of these independent “memory domains”.

Alongside my research, I am working with other postgrads to develop an app/game designed to help young paediatric cancer patients understand their disease and treatments.

Who or what inspired you to pursue your research interests?
My initial interest in computer architecture was piqued during my gap year, where I worked at a startup building VR games and technology. The senior programmers would design algorithms based on how the underlying hardware functioned, which led me to change my bachelors degree subject from Computer Science to Computer Systems Engineering, so I could better understand architecture in particular.

My bachelor’s project was GPU-focused, and my MPhil project introduced me to hardware security/CHERI. I thoroughly enjoyed both, and am excited to combine these interests for my PhD.