First Place in ICCAD Competition
The iEDA Team Won the First Place in the 2022 ICCAD Competition
On November 3, 2022, the final rankings of the CAD Contest algorithm competition were announced at the ICCAD, a grand event in the field of electronic design automation held in San Diego, California, USA. After several months of intense competition, the iEDA 3Dplacer team from the Open Source EDA Research Group of Peng Cheng Laboratory won the first place in one of the three major tracks of this event. A total of 166 teams from renowned universities and research institutions at home and abroad participated in this year's CAD Contest@ICCAD algorithm competition, such as the University of Tokyo, National Taiwan University, The Chinese University of Hong Kong, Peking University, National Tsing Hua University of Taiwan, The University of Texas at Austin, Fudan University, etc.
1. Competition Background
Electronic Design Automation (EDA) tools refer to a set of computer-aided design software for the entire automated chip design process, including functional design, logic synthesis, physical design, simulation verification, etc. As one of the strategic basic pillars of the integrated circuit industry chain, EDA actively promotes technological innovation in the chip industry and greatly improves the efficiency of the semiconductor industry.
The International Conference On Computer Aided Design (ICCAD) began in the 1980s and is one of the top academic conferences in the field of electronic design automation. The CAD Contest@ICCAD algorithm competition is the most influential international academic competition in the EDA field. The competition questions all come from the real business scenarios of globally renowned EDA/IC companies such as Synopsys, Cadence, Siemens EDA, Nvidia, IBM, and Dharma Academy, expecting to develop better solutions for the actual problems that the current industry urgently needs to solve. The results of the competition can be directly transformed into industrial solutions, which greatly promote the development of the field of computer-aided design of integrated circuits and attract high attention from both industry and academia every year.
A total of 166 teams from renowned universities and research institutions at home and abroad participated in this year's CAD Contest@ICCAD algorithm competition, such as the University of Tokyo, National Taiwan University, The Chinese University of Hong Kong, Peking University, National Tsing Hua University of Taiwan, The University of Texas at Austin, Fudan University, etc.
2. Interpretation of the Competition Question
This year's CAD Contest@ICCAD algorithm competition released three questions. Since all the members of the research group have been involved in physical design-related research, we finally chose the 3D Die-to-Die Placement problem, which has a more fitting research background. Die-to-Die technology is one of the most widely used Chiplet technologies at present and is a key technology to break through 3D chip design in the future, which can bring higher yield, better timing, lower cost, etc. And placement is an extremely important part of physical design. The quality of the placement of unit positions directly affects the PPA (Performance, Power, Area) of the chip to a large extent. The competition question requires the participating teams to divide the netlist into two parts and determine the specific position of each unit in its respective layer on the basis of considering many complex constraints such as different processes in the upper and lower layers, layer utilization, the number and spacing constraints of hybrid bonding terminals, so as to minimize the total length of the routing. Solving this problem has a key improvement for enhancing the design quality of 3D chips.
3. Competition Process
3D Die-to-Die placement was brand new for our research group, which made us very interested in exploring the automated placement solution in this new scenario. We started working on the competition question at the end of June and submitted the code by early September, which took more than two months. Due to the late start and the desire to achieve a good result, it meant that we had to race against time, make a good time plan and division of labor and collaboration.
During the competition, we constantly had new ideas and quickly implemented them, always being passionate about solving the competition questions and optimizing the algorithm. On many late nights, we were still communicating the trial and error results and new ideas in the group chat. To alert and motivate each group member, we also made a score record sheet to compare with the then first place and constantly refreshed our best record. A tense插曲 occurred that a few days before the final version submission, the server environment image of ours was mistakenly deleted, which meant that our results could not be uploaded to the server of the organizer and further led to the scores being invalid. After discovering the problem, we immediately投入 into the repair of the server. Fortunately, there was no danger in the end. Two days before the submission, our apartment was controlled due to the epidemic. In this environment, we did not slack off and continued to jointly carry out the final performance improvement and process inspection. Our ability to stand out among many excellent participating teams is inseparable from the support of the research group and the joint efforts of each of our team members. In addition, the team used the super computing power of "Peng Cheng Cloud Brain II" to realize the learning of key parameters in the algorithm and finally won the first place.
4. Introduction to the iEDA Research Group
The research group mainly conducts research around open-source EDA tools, intelligent chip design methods, open-source EDA system platforms, and open-source EDA benchmark test sets, creating a four-in-one AI+EDA technology ecosystem (problems, data, platform, computing power), developing open-source and open chip design solutions, improving chip design efficiency, improving chip design quality, reducing the chip design threshold by orders of magnitude, attracting and promoting innovation and entrepreneurship in the chip field, cultivating chip professionals, and promoting the diversified development of the chip industry.
The research group is committed to developing the EDA tool chain for the physical implementation of digital chips, forming a technology research and development platform for industry-university-research cooperation and communication, connecting the academic and industrial circles, and forming an important part of the industrial closed loop (innovative ideas -> verification -> academic transformation -> application iteration -> algorithm improvement). The goal is to provide a more agile and efficient R & D system for the research community; provide a real talent training platform for the academic community; and output more direct and effective key technologies and talents to the industry. The project is expected to break through the physical implementation of domestic 28nm process Netlist-GDSII chips, build an EDA data set, support the industry to polish EDA tools, and evaluate key technologies and train intelligent models in the research community. Currently, problem decomposition, system platform, breaking through 110nm physical implementation and completing tape-out verification have been completed.
In this CAD Contest@ICCAD competition, the team's instructors were Teacher Xingquan Li and Dr. Zhipeng Huang, and the participating students were from four different units and majors (Shijian Chen: Joint-training doctoral student of Peng Cheng Laboratory and Institute of Computing Technology, Chinese Academy of Sciences, Computer Science and Technology; Xueyan Zhao: Doctor of Institute of Computing Technology, Chinese Academy of Sciences, Computer Architecture; Jiangkao Li: Master of Minnan Normal University, Operations Research; Yihang Qiu: Master of Guangdong University of Technology, Control Science and Engineering). Most of them are students visiting or interning in the research group and are all simultaneously undertaking the research and development of some point tools in physical design. With their love and vision for open-source EDA, they all gathered in the iEDA research group of Peng Cheng Laboratory and gave full play to their respective disciplinary background advantages, learned from each other and made progress together.
Here, the iEDA research group sincerely invites all friends interested in the research and development and use of open-source EDA projects to participate and give us more suggestions and support. Interested friends can join our WeChat communication group.