Chinese semiconductor industry
- Thread starter Hendrik_2000
- Start date
- Status
14 nm fabs are much more expensive and there may be insufficient customer demand. FinFET requires different design rules than planar transistor and more patterning steps. If the customer volume is insufficient, you don't build more fabs.
As for why demand is insufficient, moving to FinFET is a huge leap that many companies are unwilling to pay for, especially in ASICs and microcontrollers.
LAM/AMAT are actually more replaceable companies. Not to say what they do isn't hard but AMEC and Naura exist and are roughly at parity. ASML is much less replaceable.
28nm and above are planar, 14nm-5nm are all FinFET，below 5nm are GAA
Since SMIC doesn't have anything under 14nm yet，28nm/FinFET is 28nm/14nm
There is always a strong demand for advanced nodes，majority of TSMC's revenue coming from advanced nodes rather than mature nodes.
FinFET has been around over decade now，it isn't something new，many Chinese fabless have been using TSMC's FinFET service for many years already.
LAM/AMAT are replaceable or not，does not change the fact that at this moment Chinese fabs still cannot get rid of them，just look at latest bidding infos released by the fabs you will see it
advanced nodes are winner takes all unfortunately because the volume is taken by a few large customers i.e. Apple and Qualcomm, rather than a large number of small customers. So TSMC has first mover advantage. If you look at sales numbers, Samsung also has a low market share compared to TSMC yet they have EUV and everything.
People posted bidding info of a late build fab by a Chinese IDM here. It had next to no US equipment in the list. It was all Chinese except for ASML lithography machines.
SMIC has the capabilities to produce anything if they had relevant access to necessary machine tools. They have already proven this more than once. SMIC isn't the problem. SMIC has 14nm FinFET in full production with high yields and 10nm is in production as well.
And, of course, much like with Chinese exascale supercomputing, SMIC isn't going to go around saying what are their actual yields. Even in the best of times this is typically considered corporate secret information. When you see such numbers from companies like Intel they are typically only provided a year or two after mass production rate was achieved. This information is typically only presented to limited audiences. Giving out this information basically tells your competitors what is your production capacity and production cost per wafer. They can then use this information to squeeze you out.
14nm and lower require more passes and to be efficient you need more lithography machines. You might require twice the machines or more. Currently ASML has been having delays with machine deliveries and they already burned SMIC more than once. Given the extra expense in building such production facilities it makes sense for SMIC to focus on 28nm.
SMIC has the capabilities to produce anything if they had relevant access to necessary machine tools. They have already proven this more than once. SMIC isn't the problem. SMIC has 14nm FinFET in full production with high yields and 10nm is in production as well.
And, of course, much like with Chinese exascale supercomputing, SMIC isn't going to go around saying what are their actual yields. Even in the best of times this is typically considered corporate secret information. When you see such numbers from companies like Intel they are typically only provided a year or two after mass production rate was achieved. This information is typically only presented to limited audiences. Giving out this information basically tells your competitors what is your production capacity and production cost per wafer. They can then use this information to squeeze you out.
14nm and lower require more passes and to be efficient you need more lithography machines. You might require twice the machines or more. Currently ASML has been having delays with machine deliveries and they already burned SMIC more than once. Given the extra expense in building such production facilities it makes sense for SMIC to focus on 28nm.
Last edited:
what about sales figures for AMEC and Naura? They are going 30%+, and not just through raising prices but through physical shipments.
Neither are they growing simply by taking on debt, since they have asset:debt ratios >>1
So their growth is proven to be sales and technology driven, not debt or price driven.
Furthermore, here one of our resident SME @tinrobert believes that the new "28 nm" fab will not be limited to 28 nm production.
So I have good hopes for AMEC, Naura and SMIC based on current trends.
Seeing the biddings looks like Chinese semi equipment are getting most of the orders along with their Asians counterparts, Singapore Koreans and Japan. American companies are probably getting less orders than used to be but is hard to tell because some fabs sign secret contracts with suppliers.
Some companies like LAM and KLA are starting to manufacture their equipment overseas to evade sanctions and restore their reputation as reliable suppliers.
Its looks like for its size, being the biggest SME company, Applied Materials, their reputation as reliable supplier is suffering a hit.
ASML is getting high score, they are too much a "too big to fail" company, that has been allowed to monopolize the industry.
Like you said，large customers get most of TSMC’s capacity，small fabless would have to wait on the long queue. If SMIC can build new 14nm capacity that will be a relieve for these small fabless，so they don't have to wait for their turn at TSMC.
As for Samsung，Samsung always have yield issue at their advanced nodes，there is tons of report on this. But SMIC has no yield issue on FinFET，right？
So demand is there and yield is not a problem，then why doesn't SMIC build new FinFET fab？
- Status