Key Highlights

Compute-heavy application benefits from specialized hardware to execute application-specific computations. For example, machine learning application needs hardware acceleration mechanism, such as using GPU, to perform simpler but large number computation in parallel. Across the scientific community, the drive is to design hardware and software together (hardware/software co-design) to optimize computation according to a specific application's need. Given the chips that can support such a computation model is expensive, cloud vendors such as amazon web services provide hardware and tool rental where a user rents tools to develop and deploy their designs. While such a model is inexpensive, hardware sharing may allow information to leak to the other user even if the entire hardware is sanitized before renting it to another user. As a result, cloud vendors do not allow users to know the exact identity of a hardware module. However, even though each FPGA chip behaves the same from a logical perspective, the underlying circuit structure varies because of silicon manufacturing variation. This project aims to develop a digital system to extract the hardware [This is an ongoing project.]