(c) N3B gives, among other improvements, a density boost of 1.7x and a power reduction of ~25%
N3E gives, among other improvements, a density boost of 1.6x but a power reduction of ~34%
(d) It APPEARS to be the case that N3B began volume production in late September:
TSMC’s 3nm era set to begin in the coming weeks
www.tomshardware.com
(e) It APPEARS that this ceremony refers to N3E, but it's honestly impossible to tell whether N3B was delayed three months, or N3E is being rushed into production ASAP. My GUESS is that it's the latter, N3B was on schedule in September, but N3B is yesterday's news, an experiment that pushed things too far and that is being abandoned by TSMC as soon as contractual obligations to Apple (and anyone else?) are done.Since N3E will be the only option going forward, that's the one they want to get all the media oxygen.
Doubtful this the N3E. Just several months ago TMSC 'pulled forward' N3E to Q2/Q3 2023 from Q3/Q4 2023 as when HVM would kick off. A quarter ahead of schedule on a variant that is a scaling back on complexity is creditable. To a couple months later to hand wave and say pulling forward yet another quarter would be a stretch.
Decent chance that TSMC is just talking about the same N3 (N3B) as they were in September. September was win a full set of 'blank' wafers went in and this is just when there is a full production pipeline of when finished wafers are coming out the other end. It is "start" of where HVM N3 will start getting paid for. Which is what Wall St. actually really care about ("when is N3 going to start to contribute to revenues").
Also aligns with guidance outlined by TSMC a quarter call (or so) ago were they said there would be N3 wafer revenue at end of Q4 '22 (i.e., now) , but it wouldn't make substantive difference in overall revenue. And that the revenue flow would start in Q1. (customer(s) taking receipt of finished N3 wafers. ) For TSMC to finish substantial numbers of N3 wafers in mid-late Q1 they would have needed to start filling the production pipeline around September/October and all the production pipeline kinks worked out by now.
What is dubious about Thursdays exact date for N3 (or N3E) is that it also was covered in the press release that it was the 'topping off' for Phase 8 at the 'Gigafactory' site were N3 is being made. ( the 8th fab building finished at that site). The likelihood that a physical construction phase would just happen to finished on the exact same date as fab processs went HVM is probably up there with Nottingham Forest finishing on top of the Premiere League this season ( or NY Jets winning Super Bowl 2023). This specific date is more likely about that factory topping off and the fab status stuff is being 'sprinkled' around the ceremony for PR and symbology reasons. It probably isn't about announcing yet another status change for N3E.
(If this theory is correct, it's something of a change for TSMC. In the past supposedly "unsuccessful" experiments, like 10nm, were not just useful learning exercise, but stuck around for a while for others customers to use if they wanted. N3B looks like TSMC have no interest in keeping it around longer than necessary.
Initial N3 being whispered to be "bad" is good reason why they'd need to bump up the PR about it shipping in volume.
Tomshardware had a write up about the N3 ceremony yesterday
TSMC's 3nm production is behind Samsung, but ahead of Intel.
www.tomshardware.com
But they also have an article today saying that Nvidia, AMD, Intel, and others are chopping orders.
Large customers revise orders to TSMC due to the economy slowdown.
www.tomshardware.com
" ... , the utilization rate of TSMC's N7-capable lines (7nm, 6nm-class technologies) will decline to around 50% in early 2023. Furthermore, even TSMC's N5/N4-capable lines will be underutilized, though this may not come as a surprise since these are used to make leading-edge products, ..."
The days of 100% capacity and large backlogs of orders is basically over. So when the N3 wafers are going to start contributing to the revenues is an even more urgent issue for TSMC at this point.
My guess [once again a guess] is that the TSMC engine broke down because of covid. Problems with N3B that, in the past, could have been figured out via engineers shutting between Taiwan, the Netherlands, and Japan, had to be handled more slowly and inefficiently via Zoom calls and eventually TSMC made the call to just abandon the most aggressive aspect of the design [the EUV double-patterning] and switch to a less dense, and less aggressive, set of design parameters.
Except a healthy chunk of N3 was started before Covid-19 even infected anyone.
July 2019.
"... As it appears, the manufacturing technology is out of its pathfinding mode and TSMC has already started engaging with early customers. ..
...
... TSMC said that it had evaluated all possible transistor structure options for 3nm and came out with ‘a very good solution’ for its clients. ..."
Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
www.anandtech.com
August 2020 [ while after Covid-19 the N5 family extension of N4 highly likely was not scheduled in 2019 to arrive at the same time , or before , N3 . ]
" ... We’ll be seeing N4 risk production start in 4Q21 for volume production later in 2022.
...
...N3 is planned to enter risk production in 2021 and enter volume production in 2H22 ... " . " ...
Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
www.anandtech.com
Even early on N3 was never a Q1-Q2 22 target. (and therefore not a good iPhone SoC target). Is TSMC landing at the end of the range estimate they gave ( 2H22)? Yes.
N3's FinFlex was always something that a number of design shops were going to eschew because it was going to be more complexity than they needed .
Many product designs are a series of compromises. Take for example, electric vehicles, wouldn’t it be great to have a 1000-mile or 1600-kilometer range in an EV? But consider what compromises would need to be made to achieve this goal. The battery would be massive and heavy, leading to an...
www.tsmc.com
There probably always was going to be a 'simpler track' N3 that cost less to design with because not everyone needs to pack 'everything and the kitchen sink' onto a single, expensive monolithic die.