我与导师 Jason Cong 教授和 Guangyu Sun 教授合作发表在《IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems》的期刊论文《Caffeine: Towards Uniformed Representation and Acceleration for Deep Convolutional Neural Networks》获得Donald O. Pederson 最佳论文奖，成为首个获得该奖项的中国大陆学者。PKU Report(English)，北京大学报道（中文），UCLA大学报道，IEEE报道。

Award Brief

Prof. Jason Cong (CECA director), Dr. Chen Zhang (CECA alumni, now at Microsoft Research Asia), Prof. Guangyu Sun (Associate Professor at CECA), Prof. Zhenman Fang, Dr. Peipei Zhou, and Dr. Peichen Pan have received the 2019 Donald O. Pederson Best Paper Award from the IEEE Council for Design Automation (CEDA). The awarded paper is “Caffeine: Towards Uniformed Representation and Acceleration for Deep Convolutional Neural Networks”, which is published in the IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems (IEEE TCAD) in Oct. 2018.

Donald O. Pederson Best Paper Award is dedicated to award the best paper published in IEEE TCAD in the recent two calendar years. Current Associate Editors of the IEEE TCAD nominates the best paper candidates first. Among the papers published in the past two years, the most referenced or downloaded papers are nominated automatically by the entire editorial board for review and voting.

The editorial board nominated five papers this year, and another nine papers are automatically nominated for receiving highest downloads in the past two years. After the voting, a confidential review committee reviews the top five papers before deciding the final winners. The selection committee unanimously agreed to declare two of the candidates to be co-winners. The award is recognized at the Design Automation Conference (DAC) in Las Vegas on Jun. 4th, 2019.

This is the first time for researchers from mainland of China to win the IEEE TCAD Donald O. Pederson Best Paper Award (prior winners are listed in the chart below). Dr. Zhang, Prof. Sun, and Prof. Cong published a pioneering paper on FPGA-based accelerators for deep learning in FPGA’15. This paper has been referenced for almost 800 times in the past four years, including TPU. They extend the FPGA’15 work, and design Caffeine, a software-hardware co-design tool to map Caffe-described deep neural networks onto FPGA-based accelerators.

Background:

• IEEE TCAD and DAC are top journal and conference of electronic design automation, respectively. Both are A-class papers on Computer Architecture / Parallel and Distributed Computing / Storage Systems recommended by CCF.

• Download the Caffeine paper: https://doi.org/10.1109/TCAD.2017.2785257