Our research exploits the capabilities of programmable hardware devices such as FPGAs tightly coupling optimised accelerator architectures with high throughput interfaces to host computers, external memory, and the network. This elevates these accelerators to primary components, thereby enhancing their impact on overall system performance. We consider three dimensions of research that are closely related.
Accelerator Virtualisation
Enable rapid integration and sharing, with data from network, memory, storage
Conduit to application experts leveraging FPGA acceleration more scalably and easily
Accelerator Design
Extend overlay concepts with compilation flows and integrate with virtualisation
Focus on domain-specific overlays for emerging workloads
Enable rapid prototyping of accelerator designs
Distributed Acceleration
Embedded edge working with accelerated datacenter core and network
Distributed/federated machine learning support
Collaborative autonomous systems, sensor processing