Approximate Computing (AC) trades accuracy for performance and energy efficiency. AC is increasingly attractive in various computation-intensive applications such as imaging processing, audio recognition, information search, and artificial intelligence. However, recent work shows that the unique approximate behavior and computational uncertainty in AC systems attract new attack opportunities. Compromised AC modules will sabotage the integrity and security of the entire computing system because AC components closely interact with non-AC modules in the same system. This project will identify the risk of applying approximate mechanisms in computing systems and explore effective methods to protect AC systems.
The objective of this project is to develop holistic, hardware-software integrated methods to secure AC systems. More specifically, the research tasks include: (1) performing a thorough study on the security vulnerabilities of the hardware and software for AC systems, (2) developing methods to locate the vulnerable AC modules, (3) exploring hardware and software obfuscation methods to secure AC systems, and (4) evaluating the attack resilience of a cross-layer proactive defense framework in practical applications.
The success of this project will facilitate the security of large-scale energy-efficient computing systems used in military, government, transportation, and commercial applications. This project is expected to support female Ph.D. students and engage more women students via the Workshop for Women in Hardware and Systems Security (WISE). The project will also promote undergraduate research and train undergraduate students to compete in international security competitions.
The output of the project (source code, documentation, experiment log, and scholarly publications) will be managed by the version control system Git. New developed curriculum materials will be managed by the course system Canvas at the University of New Hampshire (UNH). A local copy of all the data and documents will be stored in the backup servers at UNH. Data will be retained for at least three years beyond the award period. The scholarly publications, presentations, and open-source code will be available on the project homepage at https://mypages.unh.edu/qyu/secure-approximate-computing-systems.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
