Hendrik_2000
Lieutenant General
Via Xinhui EDA should not be a problem with China
Without a good EDA flow, China’s semiconductor investment is at risk. While its commercial EDA ventures have not been successful, it’s highly likely that work continues.
October 29th, 2020 - By: Brian Bailey
popularity
If you saw this headline and thought you missed a press release, don’t panic. China has not, at this point, announced to the world that it has a suite of EDA tools ready to roll. The rest of the world is content to look at the substandard attempts it have made so far and write them off as not being capable of developing competitive EDA software. But in all likelihood, given the current political climate, China is hard at work on it.
The country is committed to building world-class fab capacity using all of the latest technology. The government is putting together the necessary money to ensure that happens, and it has the staying power to make that happen. China already has caught up to 14nm finFETs, and it is working on 7nm. It is even developing its own EUV capability so as not to be reliant on Western nations’ equipment.
But it still has one huge vulnerability. It relies on American EDA tools for back-end design software. Around 14nm is the point at which EDA says that the fabs can’t do it on their own, and that there has to be close cooperation between fab and EDA. When China gets to that point, it will not be able to use pirated tools anymore. Instead, it would have to partner with U.S-based EDA companies, and that could be problematic. But even without that, if the United States was to decide that it would no longer sell EDA software to China, or to any company that wanted to have their chips fabricated in a Chinese foundry, then China’s foundries would sit idle. I don’t think that is going to happen.
I am sure China is working on its own government-sponsored EDA tools at this point. Those tools may not be very good at the moment, or create as optimal designs, but just like manufacturing, you have to start somewhere and learn. China doesn’t need to develop EDA software to sell to anyone else. It just needs to ensure that its requirements can be supported, and that its design companies would have a fallback position if the West were to continue to raise barriers. Thus, I believe they will keep quiet about any such program and continue to buy the best tools on the market until they have become competitive enough. Pre-announcing would buy them nothing.
Now let’s focus on the advantages they may have. First, while EDA companies get to see a lot of designs, they have to be very careful about what they do with that data. If company secrets were to be released, they would get into huge trouble. Not so for the Chinese. Their government can insist that their EDA research teams get to see the entirety of every design being worked on in China. With the emergence of artificial intelligence as a way to drive tools, rather than the traditional heuristics, this is certainly an advantage for China. China also may be ahead in various AI fields, given the number of patents being filed, and it has access to much more data that can be used for training.
The second advantage for China is that it does not have to deal with all of the legacy that has built up in the commercial EDA tools. There are countless tweaks and workarounds inserted in the tools because a company once did something very strange and unique, and that becomes part of the legacy of a tool that cannot be removed – even after that company no longer needs it. When you are starting from a clean sheet of paper, you can create a clean tool. It’s also possible that China may not allow such one-off crazy tricks to be pulled by their design companies. If a company develops something really unique and good, that technology is more likely to be shared with others. Cooperation often leads to faster innovation.
The third advantage is that the EDA landscape is changing rapidly. Twenty years ago we had logic synthesis. Then it became impossible to close timing and synthesis had to be completely re-imagined to take layout and timing into account. Thus, physical synthesis was born. We are on the cusp on another of those changes that may cause the tools to be rewritten again – power. It has risen to become a primary optimization factor, but even more importantly, heat can kill chips or cause them to prematurely age. You cannot design a chip without regard to power and thermal anymore. In addition, place-and-route tools that are knowledgeable about power potentially could produce much better layouts, allowing the chips to run cooler and faster. Power has to be considered at the system level, as well, especially if we get to heterogeneous silicon integration where thermal models and heat/power analysis have to run across chip stacks.
So if EDA companies are going to think about a fundamental retooling, they will have to invest the same amount of money and effort as Chinese companies to make that happen. Compared to the cost of a semiconductor manufacturing program, an EDA program would almost be pocket change for them. Not so for the individual companies in the US that have to fund that out of profits.
EDA development teams are not particularly large, and the time it takes to get a tool up and running is not that long. This is why EDA startups were so common in the ’80s and ’90s. Yes, with the close integration between tools and methodologies of today, it is a lot more difficult than the creation of point tools back then, but when you have the time and the money and little pressure to announce tomorrow so you can start to see revenue, Chinese companies would be crazy if they were not working on it.
Without a good EDA flow, China’s semiconductor investment is at risk. While its commercial EDA ventures have not been successful, it’s highly likely that work continues.
October 29th, 2020 - By: Brian Bailey
popularity
If you saw this headline and thought you missed a press release, don’t panic. China has not, at this point, announced to the world that it has a suite of EDA tools ready to roll. The rest of the world is content to look at the substandard attempts it have made so far and write them off as not being capable of developing competitive EDA software. But in all likelihood, given the current political climate, China is hard at work on it.
The country is committed to building world-class fab capacity using all of the latest technology. The government is putting together the necessary money to ensure that happens, and it has the staying power to make that happen. China already has caught up to 14nm finFETs, and it is working on 7nm. It is even developing its own EUV capability so as not to be reliant on Western nations’ equipment.
But it still has one huge vulnerability. It relies on American EDA tools for back-end design software. Around 14nm is the point at which EDA says that the fabs can’t do it on their own, and that there has to be close cooperation between fab and EDA. When China gets to that point, it will not be able to use pirated tools anymore. Instead, it would have to partner with U.S-based EDA companies, and that could be problematic. But even without that, if the United States was to decide that it would no longer sell EDA software to China, or to any company that wanted to have their chips fabricated in a Chinese foundry, then China’s foundries would sit idle. I don’t think that is going to happen.
I am sure China is working on its own government-sponsored EDA tools at this point. Those tools may not be very good at the moment, or create as optimal designs, but just like manufacturing, you have to start somewhere and learn. China doesn’t need to develop EDA software to sell to anyone else. It just needs to ensure that its requirements can be supported, and that its design companies would have a fallback position if the West were to continue to raise barriers. Thus, I believe they will keep quiet about any such program and continue to buy the best tools on the market until they have become competitive enough. Pre-announcing would buy them nothing.
Now let’s focus on the advantages they may have. First, while EDA companies get to see a lot of designs, they have to be very careful about what they do with that data. If company secrets were to be released, they would get into huge trouble. Not so for the Chinese. Their government can insist that their EDA research teams get to see the entirety of every design being worked on in China. With the emergence of artificial intelligence as a way to drive tools, rather than the traditional heuristics, this is certainly an advantage for China. China also may be ahead in various AI fields, given the number of patents being filed, and it has access to much more data that can be used for training.
The second advantage for China is that it does not have to deal with all of the legacy that has built up in the commercial EDA tools. There are countless tweaks and workarounds inserted in the tools because a company once did something very strange and unique, and that becomes part of the legacy of a tool that cannot be removed – even after that company no longer needs it. When you are starting from a clean sheet of paper, you can create a clean tool. It’s also possible that China may not allow such one-off crazy tricks to be pulled by their design companies. If a company develops something really unique and good, that technology is more likely to be shared with others. Cooperation often leads to faster innovation.
The third advantage is that the EDA landscape is changing rapidly. Twenty years ago we had logic synthesis. Then it became impossible to close timing and synthesis had to be completely re-imagined to take layout and timing into account. Thus, physical synthesis was born. We are on the cusp on another of those changes that may cause the tools to be rewritten again – power. It has risen to become a primary optimization factor, but even more importantly, heat can kill chips or cause them to prematurely age. You cannot design a chip without regard to power and thermal anymore. In addition, place-and-route tools that are knowledgeable about power potentially could produce much better layouts, allowing the chips to run cooler and faster. Power has to be considered at the system level, as well, especially if we get to heterogeneous silicon integration where thermal models and heat/power analysis have to run across chip stacks.
So if EDA companies are going to think about a fundamental retooling, they will have to invest the same amount of money and effort as Chinese companies to make that happen. Compared to the cost of a semiconductor manufacturing program, an EDA program would almost be pocket change for them. Not so for the individual companies in the US that have to fund that out of profits.
EDA development teams are not particularly large, and the time it takes to get a tool up and running is not that long. This is why EDA startups were so common in the ’80s and ’90s. Yes, with the close integration between tools and methodologies of today, it is a lot more difficult than the creation of point tools back then, but when you have the time and the money and little pressure to announce tomorrow so you can start to see revenue, Chinese companies would be crazy if they were not working on it.
