The more I ponder, it seems like the Huawei / SMIC 7nm breakthrough might have a massive impact on industry.
Other than mobile, the use case for advanced nodes beyond 5nm is really not obvious as others have said - costs are extremely high, value relatively low. And Apple primarily leveraged their dominant market position in high end phones to keep pushing the edge here.
What happens if/when Apple loses 1/3 of their revenues globally due to a Huawei phone that invests in features outside of shrinking gate sizes? How does TSMC get enough orders to fill their Arizona fabs?
We aren’t going around driving cars with ever increasing # of horsepower, and I suspect we don’t need more EV range beyond a certain number of miles. I suspect the same will be true in smartphones.
Yes, business case for 5nm and beyond isn't great given the 2 factors we talked about (SRAM can't shrink any further and dark silicon limitations)
However, we may want to pause a little bit in wondering about Apple or TSMC demise. I would worry less about TSMC in Arizona and more about how Chinese semi industry can manufacture enough chips for Huawei.
As I discussed, if we anticipate a full cutoff, then they would need for 2024
60 million SoCs for phones (their goal for 2024) + 8 million for pads (quarterly sales routinely around 2 million)
) + 5 million for desktops/laptops (Q3 2022 sales was a little over 1 million in China alone
) + 2mill server CPUs + 100k GPUs
Do you want to guess how many 7nm wafers they need for that? My estimate so far is over 20k wpm, which they certainly don't have right now. So over next year at least, they are burning through backlog of chips they have and same with DRAMs and such.
Of course, there are things Huawei is really good at, so it can overcome some of these problems.
But the reality is that it's suppliers need to massively increase investment and capacity/production to satisfy 2024 demand let alone further down the road when Harmony devices get sold outside of China
And this is just 1 company. Do we want to guess how many wafers all of China's OEMs need?
probably 150k wpm. Will it really hurt apple, qualcomm & TSMC if it all goes to SMIC? Of course, but we are a long way from that. Things don't ramp up over night. you need to train people for the extremely complicated multi-patterning process. domestic tools need to continue to improve....
The reality is we hit the first key in a long process.
surely they can easily afford a $2,000 Iphone....lol
