Wujie Wen Receives William J. McCalla ICCAD Best Paper Award
The NC State Computer Science Department congratulates Dr. Wujie Wen, Associate Professor of Computer Science, for co-authoring a paper that was recently awarded the William J. McCalla ICCAD Best Paper Award in the Backend Category at the 2023 42nd ACM/IEEE Conference on Computer-Aided Design (ICCAD).
The paper, “Improving Realistic Worst-Case Performance of NVCiM DNN Accelerators Through Training with Right-Censored Gaussian Noise,” was co-authored by Wen’s collaborators from the University of Notre Dame: Zheyu Yan, Yifan Qin, X. Sharon Hu, and Yiyu Shi. The paper was entered in the competition alongside 750 other submissions. Of those submissions, 172 were accepted, and six of those were nominated for the Best Paper Award - three in the frontend category, and three in the backend category. Ultimately, in these two categories only two papers in total are selected as the William J. McCalla ICCAD Best Papers, making it an extremely competitive and highly honorable achievement.
The abstract follows:
Compute-in-Memory (CiM), built upon non-volatile memory (NVM) devices, is promising for accelerating deep neural networks (DNNs) owing to its in-situ data processing capability and superior energy efficiency. Unfortunately, the well-trained model parameters, after being mapped to NVM devices, can often exhibit large deviations from their intended values due to device variations, resulting in notable performance degradation in these CiM-based DNN accelerators. There exists a long list of solutions to address this issue. However, they mainly focus on improving the mean performance of CiM DNN accelerators. How to guarantee the worst-case performance under the impact of device variations, which is crucial for many safety-critical applications such as self-driving cars, has been far less explored. In this work, we propose to use the k-th percentile performance (KPP) to capture the realistic worst-case performance of DNN models executing on CiM accelerators. Through a formal analysis of the properties of KPP and the noise injection-based DNN training, we demonstrate that injecting a novel right-censored Gaussian noise, as opposed to the conventional Gaussian noise, significantly improves the KPP of DNNs. We further propose an automated method to determine the optimal hyperparameters for injecting this right-censored Gaussian noise during the training process. Our method achieves up to a 26% improvement in KPP compared to the state-of-the-art methods employed to enhance DNN robustness under the impact of device variations.
Also during the conference, another paper co-authored by Dr. Yiran Chen from Duke University's Electrical and Computer Engineering Department was awarded a William J. McCalla ICCAD Ten Year Retrospective Most Influential Paper Award. Chen was Wujie Wen's PhD Advisor when they both were at the University of Pittsburgh. The paper, "Reduction and IR-Drop Compensations Techniques for Reliable Neuromorphic Computing Systems", and Wen's award-winning paper focus on the same technology - the next generation processing-in-memory architecture-based AI hardware accelerator. (See related story here.)
The International Conference on Computer-Aided Design (ICCAD) is jointly sponsored by IEEE and ACM, and it continues to be the premier and most selective conference devoted to technical innovations in design automation. It is the premier forum to explore new challenges, present leading-edge innovative solutions, and identify emerging technologies in the electronic design automation research areas. ICCAD covers the full range of CAD topics – from device and circuit level up through system level, as well as post-CMOS design. ICCAD has a long-standing tradition of producing cutting-edge, innovative technical programs for attendees. It is a top CSRankings conference for design automation (https://csrankings.org/#/index?all&us).
For more information on Dr. Wujie Wen, click here.
Return To News Homepage