You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
but that's not fair - you can't compare a HD library with a HP library. It'd be interesting to compare N5 with Intel 3 HD library, but I assume those tech details are not know.
According to an Angstronomics article (linked below) Intel 4 is denser than TSMC N5:
So, if 18A is only slightly more dense that N5, then it must be less dense, or equally dense as Intel 4. Can this be true?
https://www.angstronomics.com/p/the-truth-of-tsmc-5nm
That's quite a bombshell, but it would explain a few things. It must have been a lot easier for Intel to get GAA and back side power to yield when the pitches are so relaxed. Also, if 18A is less than N3 then it must be only marginally more dense than Intel 4?
I think next year IFS will lose even more money. By then probably 50% client processors sold will be Meteor lake / Arrow lake and both these processors are mostly manufactured by TSMC. Intel's external customers are all low volume, so Intel's fabs will be mostly idle.
I notice IFS loss was around $5 billion for both 2022 and 2021 - that is despite Intel making $12 billion less revenue in 2022. Really IFS loss should be more for 2022, because they are manufacturing less products.
50% is just a figure I pulled out of the air, but the price difference is due to details in their contract, not just "hot runs". Reserving wafers on the latest node probably effects the price. Also, some customers reportedly pay for working chips, rather than wafers...
But still there are a range of customers for the latest node, Nvidia, Apple, Intel, AMD, Qualcomm, etc. The price range that TSMC charges each one could vary by as much as 50%.. yet these sources are just saying "$30K for a wafer", regardless of the volume or contract conditions of the customer
I would assume for some one like Nvidia who is buying 10k/month wafers the price per wafer is much lower than a small customer, who is only buying 1000 wafers a month. Maybe $30K is the price quoted to a small customer?
Supposedly Lunar Lake is just two tiles and the compute tile includes CPU and GPU, that's why they can't use Intel 20A (20A is missing high density libraries needed for the GPU)
Another thing of interest, Intel was meant to benefit from the chiplet design, by reusing SoC and IO tiles from Meteor lake in Arrow Lake, but it turns out those rumours are not true.
If you compare the Arrow lake die shot here: , with the Meteor lake dies shot...