Chinese semiconductor industry

dengyeye said:

So China was only 2 years behind in 2000. China's method of mass importing over indigenous creation must have been one of the greatest mistakes post Deng. Perhaps nobody could have foreseen how important and complex semiconductors would evolve to become, but this seems like a massive oversight.

Click to expand...

Not really. SMIC achieved that initial success by importing machine tools and licensing processes from someone else. Back then the global industry had much more uniform chip manufacturing processes as it was all planar CMOS. So moving up was a matter of fiddling around improving existing processes. FinFET was a huge barrier for the industry and most of the players dropped out. When multiple-patterning became commonplace, tool costs rose through the roof, it is no small feat SMIC could even get past that hurdle.

dengyeye said:

Why did SMIC's market share increase from 2015 to 2020 though? If in 2014 they were near the top, and in 2020 they are a generation or two behind, shouldn't we see the opposite trend?

Click to expand...

There is such a huge gap between supply and demand in China SMIC still have lots of room for expansion. And you cannot assume production ramp up is immediate.

dengyeye said:

Not saying you are wrong, perhaps SMIC was still trailing TSMC in terms of mass production of chips?

Click to expand...

TSMC was having a difficult time back then after Morris Chang stepped down into retirement and his successors made a bunch of stupid decisions which nearly ended the company. They were losing ground to Samsung and UMC so quickly a lot of people thought they would cease to exist and many of its clients were leaving in droves. Morris had to come back and straighten things out.

dengyeye said:

Why did SMIC's market share increase from 2015 to 2020 though? If in 2014 they were near the top, and in 2020 they are a generation or two behind, shouldn't we see the opposite trend?

Not saying you are wrong, perhaps SMIC was still trailing TSMC in terms of mass production of chips?

Click to expand...

Leading edge fabrication is mostly winner take all as you see from Samsung's struggles. Most of SMICs revenue still comes from much older nodes.

Also as @gelgoog said finFET was a huge barrier not only for SMIC but for everyone else as well.

FairAndUnbiased said:

Leading edge fabrication is mostly winner take all as you see from Samsung's struggles. Most of SMICs revenue still comes from much older nodes.

Also as @geloog said finFET was a huge barrier not only for SMIC but for everyone else as well.

Click to expand...

Does this mean market share isn't the best way (or even a good way) to judge China's self sufficiency in semiconductors? After all the purpose of MIC 2025 isn't to increase exports, it's to ensure China can't get sanctioned out of key technologies.

dengyeye said:

Does this mean market share isn't the best way (or even a good way) to judge China's self sufficiency in semiconductors? After all the purpose of MIC 2025 isn't to increase exports, it's to ensure China can't get sanctioned out of key technologies.

Click to expand...

It is a good indirect proxy in that the price of semiconductors raises a lot the closer you are to the leading edge. But you have remember than when companies like Samsung first got into the business a lot of people laughed at them. They started building huge amounts of cheap chips to put in appliances. Those were several generations behind state of the art.
What Samsung got was volume. And the semiconductors business is all about volume. If you can't deliver you are dead and customers will move somewhere else.

Look at Samsung today. At one point they were a junior partner in IBM/AMD/Samsung process design alliance to compete against Intel. IBM and AMD dropped out and Samsung is the one still standing. Samsung has more advanced process than Intel. IBM manufactures their CPUs at Samsung. AMD moved CPU production to TSMC.

Problem with Samsung is that they seem to have soaked up some of IBM's design philosophy i.e. leading edge approach with risky design approaches which has poor manufacturability. Intel used to be more conservative and have better yields and hence more profitability. But Intel seem to have dropped the ball for like a decade. TSMC is now the conservative one. TSMC made 16nm process with planar logic, and made 16nm FinFET when rest of industry wanted to use FinFET with 14nm and skip 16nm intermediate node altogether. TSMC went for this dual approach with intermediate node and kicked everyone's butt.

gelgoog said:

It is a good indirect proxy in that the price of semiconductors raises a lot the closer you are to the leading edge. But you have remember than when companies like Samsung first got into the business a lot of people laughed at them. They started building huge amounts of cheap chips to put in appliances. Those were several generations behind state of the art. Look at them today. At one point they were a junior partner in IBM/AMD/Samsung process design alliance to compete against Intel. IBM and AMD dropped out and Samsung is the one still standing. Samsung has more advanced process than Intel. IBM manufactures their CPUs at Samsung. AMD moved CPU production to TSMC.

Click to expand...

Thoughts on China's progress in SME, particularly SMEE? Lots of people have been saying that they're more than a decade behind, because the 90nm DUV has zero market share.

dengyeye said:

Thoughts on China's progress in SME, particularly SMEE? Lots of people have been saying that they're more than a decade behind, because the 90nm DUV has zero market share.

Click to expand...

90nm DUV has zero market share because adding immersion to the machine tool does not increase the tool cost that much and you get a lot better density. Having the option to buy both which one would you pick? You can also use all the same materials.
EUV tools are hugely expensive. They require new materials. Maintenance is finicky. Energy costs are huge due to poor performing light source. They only became a benefit over DUV in terms of price/performance recently and just by a thin margin. There is a chance the next generation high NA EUV with high intensity light source ASML is producing will basically wipe the floor with DUV in terms of economics. But it is not available yet.

SMEE needs to keep up with Nikon and improve its DUV tool to enable 5nm fabrication. They need to massively ramp up tool production and solve any issues it has in production. At the same time they need a massive EUV program. I am not sure SMEE can tackle EUV with its existing structure. They are perpetually behind the race. They need multiple simultaneous engineering teams working on all the problems at the same time, including how to compete with next generation ASML technology not just catching up with current generation ASML technology.

dengyeye said:

because the 90nm DUV has zero market share.

Click to expand...

Nobody not for sure, SMEE doesn't provide sales data, it could be a hit in non critical and back-end processes. Also is obvious that they use it for military applications. Is more complicated than that.

gelgoog said:

90nm DUV has zero market share because adding immersion to the machine tool does not increase the tool cost that much and you get a lot better density. Having the option to buy both which one would you pick? You can also use all the same materials.
EUV tools are hugely expensive. They require new materials. Maintenance is finicky. Energy costs are huge due to poor performing light source. They only became a benefit over DUV in terms of price/performance recently and just by a thin margin. There is a chance the next generation high NA EUV with high intensity light source ASML is producing will basically wipe the floor with DUV in terms of economics. But it is not available yet.

SMEE needs to keep up with Nikon and improve its DUV tool to enable 5nm fabrication. They need to massively ramp up tool production and solve any issues it has in production. At the same time they need a massive EUV program. I am not sure SMEE can tackle EUV with its existing structure. They are perpetually behind the race. They need multiple simultaneous engineering teams working on all the problems at the same time, including how to compete with next generation ASML technology not just catching up with current generation ASML technology.

Click to expand...

As China also has to produce all the equipment for DUV and EUV, would it be correct to say SMEE is 12-15 years behind ASML?

tokenanalyst said:

Nobody not for sure, SMEE doesn't provide sales data, it could be a hit in non critical and back-end processes. Also is obvious that they use it for military applications. Is more complicated than that.

Click to expand...

Yeah, when you have no other choice you have to stick with it. For example look at the Russians. You go to Mikron's site and it claims they manufacture at 90nm/65nm. But before the West started sanctioning semiconductors in China it said 55nm. Funny eh? Some people said it was a typo. Some even say their 65nm process is highly inefficient and basically is not used. They developed it but it isn't used because it would ruin output volume. However I also heard rumors back then that Mikron had got multiple exposure KrF lithography working in small batches. This is a technology no one in the West uses. They just jump straight into ArF.

There were rumors some time ago China was using dry ArF with multiple exposure lithography. Again a process no one else uses. I assume this is true for military and government chips. And it is like you said I doubt these sales show up on their balance sheet.

dengyeye said:

As China also has to produce all the equipment for DUV and EUV, would it be correct to say SMEE is 12-15 years behind ASML?

Click to expand...

It is hard to say because it is a bit opaque. SMEE is a private company and the news we get are in Chinese. These news are often small snippets in local news reports no one bothers to read in the West and are not widely spread around in China either.
You could say that with regards to what is in their website at least.