Ansy, the chart that Alb posted show that N+2 (SMIC's version of 7nm) is superior to any TSMC/Samsung 7nm variant and is comparable to Samsung 5nm technology. It also shows that N+2 Improved is equivalent to Samsung 4nm or TSMC 5nm technology. N+3 would be TSMC 3nm technology. There, all SMIC needs to do is get to N+2 improved. It has a good chance to get there in 2 years.
I just highlight the most important part from this article. Mostly, it was just rehashing of the earnings call and transcript, but with a couple of important part.
Here is the important part. Retaining talent, training talent and keep up with expansion plant during the worst part of the demand cycle. That's what makes great companies. Or at least in this case, one with a lot of resources.
Here is the more important part. SMIC is fully capable of not only mass producing 7nm and 5nm technology.
I think there is a good possibility that some of the Beijing/Shenzhen production lines will be using all domestic equipment, but other ones will still be using foreign equipment like ASML.
I have often wondered how much advanced node production they can actually do, because SN1/SN2 were originally only designed for 70k wpm when fully completed. As in the case with many Chinese fabs, the eventual capacity often grow past the originally designed capacity as more money and space become available. It's also possible that Lingang will do more of the 22/28 production since it's capex for equipment is about 25% higher than Beijing/Tianjin plant. Regardless, I think it's a fair assumption that Lingang is more likely to see foreign equipments.
Which brings us back to the original question of how much they can expand SMSC's SN1/SN2 fabs. They have gone completely radio silent in order to avoid further sanctions. we know that they intentionally kept FinFet away from other fabs, so that possible Western sanctions of ASML/Japanese tool makers do not affect any of their other projects (since US gov't is trying to keep any 14nm tools away). As such, any sanctions that ASML/Japanese may agree to (based on FT article) would only affect the SMSC fab among SMIC fabs.
We know that their increased capex of $1.6 billion are prepayment for ASML machines (or at least most of it). Since ASML is doubling their capacity while other chipmakers are cutting back on their capex, the idea that they need to put in money right now to reserve spots for 18 months from now does not pass the smell test. More importantly, $1.6 billion would give you 25 to 30 Arfi scanners. For prepayments, $1.6 billion is probably enough for like 50 arfi scanners. They are likely to add 150k wpm of 12-inch capacity over the next 3 years between Beijing/Shenzhen/Lingang. You do not need 50 arfi scanners for 150k wpm of mostly 28 nm and more mature process. If 50k wpm of 28nm node requires 10, then 150k wpm of 28 nm to 180 nm likely would only require 15 to 20. And we know some of those front end scanners will be SMEE machines.
Which tells me that the $1.6 billion additional capex is mostly for SMSC fab and possibly Lingang fab (although that one won't start production until 2023 I think?) It makes the most sense for them to buy up as many of the latest ASML Arfi scanners now, because they might not be available a year from now. On top of that, slots just opened up as other fabs reduced their capex. Let's say that $1.4 out of $1.6 are used for fast tracking ASML deliveries in 2023 and 2024 (if they are lucky) and they normally would spend about $1.5 billion a year from ASML, this would double their deliveries and possibly even more if they keep up their increased capex spending. $3.0 billion over the next year from ASML could mean 35 Arfi scanners (at $70 million each since they will buy a mix of 1980i and the more expensive 2100i) and 40 to 50 other front end scanners. That is on top of probably 15 or more arfi scanners they are taking in this year that are likely to mostly into SN2 fab. Assuming that the machines for SN1 production have mostly already been purchased in previous years for finishing to ramp up SN1 production this year. 15 + 35 = 50. Let's say 15 out of 50 will be set aside from SN1, Beijing and Lingang plants in 2023 to 2025.
The remaining 35 ends up with SN2 and other future SMSC expansion. Based on figures provided by tinrobert before of 20 Arfi scanners per 50k wpm at 7nm, 35 of these could allow for more than 75k wpm. What I'm saying is that they are spending enough money to probably have 100k wpm of SMSC capacity eventually (more than the 70k that was announced at the start of the project). The yield on N+2 probably won't be too great without EUVs, but China really doesn't have other alternatives right now. Out of that 100k, you can have
25k of 14 nm
25k of N+1
50k of N+2
It would be interesting to hear the capex number from SMIC next year. If they can continue to spend $6 to 7 billion on capex, then they could go even further than this. Maybe Lingang plant will get 14nm nodes production lines also. What do you think
@tokenanalyst and
@european_guy ? imo, pretty big move by SMIC here.
