CRII: OAC: Building Explainable Computer Architecture with Simulation and Visualization Techniques

College Of William And Mary, Williamsburg VA

Investigators

Abstract

Computing systems are getting more and more complex to achieve high performance. On the one hand, different devices are combined in one heterogeneous platform so that specialized applications (e.g., artificial intelligence) can execute more efficiently. On the other hand, the scale of devices and systems is getting much larger. The heterogeneity and the massive scale of computing systems make it challenging for researchers to understand, interpret, and explain the behavior of each hardware component that participates in application execution. The lack of explainability of existing computer architectures can potentially lead to functional bugs, performance drawbacks, security vulnerabilities, and reliability problems. This project advocates the importance of building Explainable Computer Architecture by enhancing computer architecture simulators and developing visualization tools that help analyze performance issues. The research outcome of this project reduces the barriers for students, especially students from underrepresented groups, to participate in computer architecture design, development, and research. This project initiates the Explainable Computer Architecture research direction, which can accelerate innovation in computer architecture research and design. This project contributes to a universal data format and instrumentation library for computer architecture simulators to collect execution traces for visualization purposes. Additionally, this project develops a novel data visualization tool for network-on-chips system performance analysis, facilitating the identification of performance issues in large-scale and highly-coupled computing systems. Moreover, this project delivers an educational game, based on the simulator and the visualization tool developed in this project, to support K-12 computer science education. Finally, using a research-through-design approach, the experiences learned from this project guide the community to design future ``easier-to-explain'' computer architectures. 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.