Machine Learning for Computer Architecture
[ISCA 2022] SNS’s not a Synthesizer: A Deep-Learning-Based Synthesis Predictor
- Ceyu Xu, Chris Kjellqvist, and Lisa Wu Wills. In The 49th Annual International Symposium on Computer Architecture (ISCA 22).
- https://doi.org/10.1145/3470496.3527444
- Design a deep learning model for making fast and accurate predictions of the power, performance, and area of largescale arbitrary hardware designs.
- Received Honorable Mention of IEEE Micro Top-picks Honorable Mention 2023, recognizing exceptional research in the field of computer architecture.
[Micro 2023] Fast, Robust, and Transferable Prediction for Hardware Logic Synthesis (SNS v2)
- Ceyu Xu, Pragya Sharma, Tianshu Wang, and Lisa Wu Wills. In 2023 56th IEEE/ACM International Symposium on Microarchitecture (MICRO).
- https://doi.org/10.1145/3613424.3623794
- Extended the original SNS work to enhance transferability and robustness. Proposed a unique circuit transformationbased contrastive pre-training method, which enabled a general-purpose circuit encoder.
[CODES/ISSS 2023] Special Session: Machine Learning for Embedded System Design
- Erika S. Alcorta, Andreas Gerstlauer, Chenhui Deng, Qi Sun, Zhiru Zhang, Ceyu Xu, Lisa Wu Wills, Daniela Sanchez Lopera,Wolfgang Ecker, Siddharth Garg, and Jiang Hu. In 2023 International Conference on Hardware/Software Codesign and System Synthesis (CODES/ISSS ’23 Companion).
- https://doi.org/10.1145/3607888.3608962
- Invited Survey paper in which SNS is discussed.
Computer Architecture for Machine Learning
[ASPLOS 2022] ProSE: The Architecture and Design of a Protein Discovery Engine
- Eyes Robson*, Ceyu Xu* (Co-first author), and Lisa Wu Wills. In Proceedings of the 27th ACM International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS 2022).
- https://doi.org/10.1145/3503222.3507722
- Designed and validated a hardware accelerator based on heterogeneous systolic arrays to speed up large, longsequence language models, including protein analysis and ChatGPT-like models.
- The accelerator has 1 to 2 orders of magnitude higher higher energy efficiency than NVIDIA GPUs.
Homomorphic Encryption
[ISPASS 2023] PyTFHE: An End-to-End Compilation and Execution Framework for Fully Homomorphic Encryption Applications
- Jiaao Ma, Ceyu Xu, and Lisa Wu Wills. 2022 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS).
- https://doi.org/10.1109/ISPASS57527.2023.00012
- Developed the PyTFHE fully homomorphic encryption framework, which provides an end-to-end solution including compilers and distributed backends. This framework supports execution on both CPUs and GPUs, and maintains stateof-the-art performance among other frameworks.
- ISPASS 2023 Best Paper Award