Different from the iterative development process of the three giants for decades, today’s local EDA enterprises are born in a good era of high-efficiency computing supported by “cloud +ai”. These mature technologies can provide many emerging EDA enterprises with rich algorithms and computing resources. EDA enterprises can take advantage of AI and cloud computing to greatly save costs and take many “detours”.
For example, AI, Baigeng, chief technology officer of Guowei Group Co., Ltd., told the reporter: “EDA is facing a new era of AI. How can AI technology be better integrated into the traditional EDA platform? This is also a question we often consider. Therefore, in the process of setting up colleges and universities to train graduate students from 2020 to 2021, we have five projects, four of which are related to AI. In this way, we have found a scenario suitable for developing AI algorithms on the EDA platform, and can implement AI algorithms on the EDA platform.”
Therefore, in the process of developing EDA products, Bai Geng said: “Based on the unified data structure and the unified general service engine core, we will establish an AI training application layer on the outer layer. The data required by all AI engines for training can be input from our sign off or physically implemented tool engine. After training, the test results of AI engines can be verified with our sign off tool. After training, a good AI engine can be put into an optimized engine for actual IC Optimize to verify whether the AI algorithm is practical, and then continue to iterate to see if the AI engine can be improved to make the next better optimization. This is an overall good algorithm and scene for in-depth learning. Although this is just an example, it shows that our entire platform is open to advanced technologies. We welcome good domestic algorithms and tools to be integrated into this platform, making it easier to enter the actual IC design process, so that good algorithms and tools can be quickly iterated and industrialized. This is also one of our group’s intentions. “
Indeed, AI for EDA is a very popular concept at present. In fact, the two can complement each other. Some insiders told the reporter that the use of AI training, such as the process of optimizing the circuit, generally requires a lot of expert knowledge to tell the system how to efficiently execute commands. Through this automatic machine learning method, we can quickly calculate and efficiently optimize the data. Of course, the EDA method can also assist AI in data calculus in turn. For example, in the look-up table network, because arbitrary Boolean functions can be realized, the system does not know what functions are realized at the beginning, but it can be inferred through some training sets. Compared with the mainstream network, this method can also quickly approach the optimal solution, so as to improve efficiency.
For example, the hot spot correction in DFM, Bai Geng gives an example: “Often in the chip layout, many graphics are photolithographic hot spots, which will produce defects in the manufacturing process. In order to avoid such defects, the graphics should be corrected and avoided in the early manufacturing stage. If a graphics mode is not defined, this photolithographic hot spot will be missed. Using AI algorithm, the algorithm can be deduced according to the existing graphics in the process, and there may be The new graphics of lithography hot spots can be seamlessly fed back to the layout and routing tools through the unified database through the method of in-depth learning, so as to make very efficient corrections. “
After AI, in order to seek more resources, “going to the cloud” has also become another fast lane for domestic EDA players to catch up with speeding. Linkaipeng, an industry veteran, told reporters: “Whether it is software simulation or hardware simulation, the demand for resources is very strong and huge. Therefore, the most popular way is to put it on the cloud for cross regional sharing. However, for EDA, a special industrial application, especially the verification scenario, it has hardware, which is different from the general private cloud. Therefore, source code security is very important for every team. Therefore, we have conceived a three-tier architecture through this Local cluster, realize CPU or hardware simulation and prototype verification, and build a private EDA dedicated cloud to realize HPC computing power or more EDA software cloud control. In this link, sensitive data or a large amount of computing data can be put on the public cloud. “
“Whether it is software simulation or hardware simulation, the demand for resources is very strong and huge. Therefore, the most popular way is to put it on the cloud for cross regional sharing. However, for EDA, a special industrial application, especially the verification scenario, unlike the general private cloud, it has hardware, so source code security is very important for every team.”
However, this integrated approach also needs to solve many problems, Linkaipeng pointed out: “For example, one of them is how cloud management can reasonably manage and schedule software, hardware, personnel and resources. These different resources may be placed on different devices, some on CPUs. With the scheduling of resources, data should be moved from one place to another synchronously and adjusted accordingly. On the other hand, data security should also be considered. Data should be considered in hierarchical data management or design stage Security issues, and in the process of data encryption, decryption or user data integrity need to be considered. We have a case where all prototype verification systems are stacked in the computer room to build a cloud system, which can realize the operation of many users and projects. At present, it seems very stable. ” It can be seen that “cloud +ai” has become the core “killer mace” of local EDA enterprises.
However, even so, it is still too early to talk about “overtaking on Curves”, and even “unrealistic”. After all, “Rome was not built in a day”. The accumulation of “global monopoly” of international giants today is not to say that it can be simply surpassed. It must be that they have experienced a variety of problems and challenges that even the current local EDA manufacturers have never thought of, as well as more integration and mergers and acquisitions. Therefore, from a more pragmatic point of view, the editor believes that the current focus should not be on how to achieve transcendence or how to compete with major international manufacturers. What local EDA players need now is how to make good use of the advanced technologies available in China, such as AI and cloud computing, make breakthroughs from “point” and “surface” at all levels, step by step from point tools to surface tools, and make products with the thinking of commercialization and profitability. This is because, after all, the “transcendence” of domestic EDA does not depend on the efforts of each player, but more on group warming, integration and M & A. only in this way can we truly create a Chinese version of EDA giant with the strength to fight with major international manufacturers in the future.
Editor in charge: Tzh