"Basic manufacturing and high-end manufacturing complement each other, and the development of advanced manufacturing must pay attention to the transformation of traditional manufacturing." Li Guojie, academician of the Chinese Academy of Engineering and chairman of Shuguang, made a report at the 2021 Greater Bay Area Science Forum Intelligent Industrial Software Forum with the title of "Vigorously Developing Open Source Intelligent Industrial Software". Focusing on EDA (Automated Electronic Design) software, one of the "three mountains" in industrial software, he said that EDA software for integrated circuit design is a typical representative of intelligent manufacturing. However, EDA software development, chip design and processing enterprises should work closely together to open up the upstream and downstream of the industry and obtain the necessary data from the feedback of manufacturers to ensure the effectiveness of the design. In this regard, China still needs to open up this industrial chain.
Li Guojie, academician of the Chinese Academy of Engineering and chairman of Sugon
Li Guojie proposed that there are three new trends in the development of EDA software: First, intelligent EDA, which essentially seeks the optimal IC design scheme under given constraints, and the ultimate goal is to achieve "full automation". "It's like autonomous driving, and eventually there will be unmanned automatic design." The second is open source EDA, forming an open tool cluster and industry ecology, and jointly formulating open standards by system manufacturers, chip manufacturers and EDA manufacturers, which greatly reduces the threshold for professionals. The third is the high-computing power of EDA, that is, EDA is combined with cloud computing to provide EDA services "on the cloud".
From the perspective of the development of intelligent EDA, looking at today's chip design, its demand is getting higher and higher, more and more types, and the scale is getting larger and larger, and the corresponding design cost is getting higher and higher, and the design cycle is getting longer and longer. Li Guojie pointed out that the design cost of a 5-nanometer chip alone reaches hundreds of millions of US dollars, and the most delicate chip design costs up to 500 million US dollars.
To this end, Li Guojie said that the first challenge to achieve "full automation" is to solve the efficient problem of the whole process, "to achieve artificial intelligence full process design chip (AIDA), rather than artificial intelligence assisted EDA design." He said that this requires the use of artificial intelligence models to learn expert knowledge, greatly reduce the threshold of chip design, improve the efficiency of chip design, achieve end-to-end rapid unmanned design, and greatly compress the original chip design time calculated by year to weekly calculation.
The second challenge is to cross the process generation gap. Li Guojie mentioned that in chip design, AIDA crosses the constraints of breaking the layering and blocking in the original chip design process, breaking the module boundary in the horizontal direction for horizontal optimization of cross-modules, and breaking the design level in the vertical direction for cross-level vertical optimization.
The third challenge is to cross-process methods. The complexity of design rules increases rapidly with the process, and the design rules introduced by different node processes increase exponentially, which poses new challenges to chip design. Li Guojie said that this requires the design of cross-process methods, both to extract design technology that is independent of the process, so that the basic technology of AIDA can be generalized to different process nodes, rather than redesigning each process, training AIDA models and algorithms, and extracting process-related technologies for different processes to optimize the design to improve the chip design effect.
In the report, Li Guojie pointed out that the technology used in chip design in the past was mainly artificial mathematical modeling, turning scientific and engineering knowledge into programs, writing explicit knowledge into a form of expert systems, or using point technologies such as deep penetration into the network on some sub-problems, and did not fully realize the potential of machine learning.
What is Machine Learning? He explains that machine learning is a way for computers to do all kinds of work using data instead of executing instructions, that is, using existing data (experience) to come up with a certain model (late laws) and using this model to predict the future.
"Chip learning technology can replace chip design to meet the above challenges, that is, to use machine learning methods to complete the whole process of chip from logical design to physical design." Li Guojie said that the goal of chip learning is to make chip design completely unnecessary for professional knowledge and design experience through learning, and can be completed efficiently in a short period of time and without human participation.
Nowadays, open source technology is widely used in a wide range of applications, including EDA software design in addition to operating systems, compilers, artificial intelligence, big data systems, etc. Li Guojie said that various successful experiences show that open source and openness are the necessary foundation for building a prosperous technology ecology and industrial ecology. Open source EDA provides new ideas for gathering superior forces and promoting scientific research and talent training.
"Open source is not only in software research and development, but also in hardware research and development. To establish a new open source ecosystem, and finally be able to use open source software to do open source hardware, the entire design cycle will be greatly shortened, the cost will be greatly reduced. Li Guojie said.
Written by: Nandu reporter Mo Zhihua