There’s a new report out of China claiming that Intel and AMD are now fighting over TSMC’s 7nm capacity that implies a slugging match between the three companies that may actually be more of a happy mutual coincidence.
Note: The Google Translate version of the text refers to “Supermicro” at several points. This may confuse English readers who are familiar with SuperMicro, the motherboard company. According to the native Mandarin speaker we consulted, the naive translation “Supermicro” is actually a reference to AMD. This also makes the most sense contextually within the article.
According to the story, Intel will specifically license the 6nm variant of TSMC’s 7nm technology that the company announced last year. 6nm is expected to deliver a density improvement over 7nm/7nm , though TSMC has not disclosed improvements on power or performance compared with the other node. AMD, meanwhile, is expected to become TSMC’s largest customer next year on the 7nm node, likely in part because Apple is moving to 5nm and below.
The reason this may be less of a problem and more of a happy coincidence is because of the Trump Administration’s ban on TSMC doing business with Huawei. This opened an obvious hole in TSMC’s capacity utilization that Intel, apparently, will be happy to fill. TSMC announced their 6nm process last year, it typically takes about 12 months to ramp up for volume manufacturing, and that means the node could be ready for production soon.
In a recent story, I pointed out that Ponte Vecchio’s position on Intel’s roadmap had shifted. While the card is still coming to market, Intel no longer refers to it as the leading 7nm part. Instead, according to Swan, “We now expect to see initial production shipments of our first Intel-based 7-nanometer product, a client CPU in late 2022 or early 2023.” As for Ponte Vecchio:
Yes. On Ponte Vecchio, originally the architecture of Ponte Vecchio includes an IO based die, connectivity, a GPU and some memory tiles, all kind of package together… From the beginning, we would do some of those tiles inside and some of those tiles outside, and again leverage the packaging technology as a proof point of how do we mix and match different designs into one package. So, that was the design from the beginning.
This is a beautiful example of a statement that is probably literally true, yet simultaneously leaves the reader with the wrong impression. What Swan is trying to imply is that Intel is tapping TSMC for some minor additional work. The fact that PV is no longer positioned as the leading candidate for 7nm, however, means that the GPU itself has left the metaphorical and literal building. It’s true that Intel was always going to fab some of PV at an external foundry, because Ponte Vecchio uses HBM, and Intel doesn’t manufacture it.
TSMC’s 6nm process is reportedly density-optimized and intended for high-performance computing, making it a great fit for Intel’s data center GPU. There’s also the fact that launching a GPU on TSMC rather than a CPU has less impact on Intel’s overall reputation. People expect Intel to lead in CPU development. They don’t necessarily care if Intel leads in GPU, where the company has no established reputation.
How much of an impact this has on Ponte Vecchio’s launch will depend on when Intel initiated this process. It takes about a year, best-case, to port a design from one foundry to another. Ponte Vecchio was supposed to launch in 2021, so if Intel kicked off the transition quickly enough, it might still get the card out in that timeframe. The China Times story says that TSMC will begin to convert 7nm to 6nm in the second half “of the year” and bring the node online for volume manufacturing at the end of the year, but doesn’t specify if this applies to 2020 or 2021. If 2021, it means Ponte Vecchio could slip to 2022 or 2023. If it’s a reference to 2020, it means Intel has a shot at getting the card out in 2021 – 2022.
Reportedly AMD will contract for 200,000 wafers over the whole of next year, making it TSMC’s largest customer on 7nm. The wafer order numbers are speculations by the China Times, not factual reports of order size from TSMC. And while the article says the two companies are fighting, it also emphasizes that the cancellation of Huawei’s orders left a giant hole in TSMC’s capacity the company very much wants to fill.