[The Exiled]

TSMC wants to make Intel dependent on external manufacturing

TSMC likes to say they are everyone’s foundry and that they will supply everyone, but TSMC has their favorites. This doesn’t preclude TSMC from selling to everyone though. To be clear, TSMC is everyone’s foundry, but the terms of their agreements are not the same everywhere. TSMC gives quite different terms to their various clients. The most simple is volume pricing, but others include earlier access to technology or other terms. The best way to explain these differences are by first diving into some of the terms of TSMC’s most important customer, Apple. By all measures, Apple is TSMC’s most important customer. Apple is the largest customer on the most advanced node. Apple utilizes advanced packaging more than any other firm in the form of high-density integrated fan outs (InFO). Apple is 3 times larger than TSMC’s 2nd largest customer. Apple is the rich friend that TSMC has had for a long time. Apple buys bleeding edge wafers and TSMC knows Apple will pay for them. TSMC knows Apple’s plans intimately, so TSMC will build out the capacity Apple asks for without demanding prepayments.

Apple’s integration with TSMC is so incredibly tight that they work together on defining the original process development kits (PDK) for new nodes. TSMCs other major customers get to have some customization, but Apple drives much of the base PDK. Apple gets to buy capacity, but they also have the right to move the capacity they need up and down across the year according to their business needs. The exact details are unknown, but it’s abundantly clear Apple does not do prepayments for capacity like other customers. Likewise, Apple does not seem to pay charges when they reduce demand early in the year or increase demands mid-year. The advantages of being the tier 1 customer are immense.

AMD and MediaTek are the two other preferred TSMC customers. They are mostly exclusive on the leading edge and therefore they do not deal with having to prepay large amounts for capacity. They get most the leading-edge wafer capacity they need, and the issues for their respective supply chain hinges on other aspects. For AMD, these supply issues deal more with substrates and externally at server and notebook ODMs for components such as BMC’s and WiFi. For MediaTek, these supply issues deal more with PMIC and RFFE. As such, both firms’ pre-payment for supply agreements with TSMC are close to non-existent. In Q3 2021, AMD only notched up to $355M of pre-paid long-term supply agreements despite being amid the largest semiconductor supply crunch in decades. Most of this prepayment is dedicated to substrates.

Both MediaTek and AMD also work very closely with TSMC on customizing process nodes. MediaTek has been in the driver seat for the N6 and N4 node evolutions. MediaTek has been the first customer to use these process nodes by about 6 months. AMD also designs their own custom libraries. They worked on this some at N7, but especially so at the N5 process node. AMD is more so a trailblazer on packaging. TSMC’s has a testbed and ramp partner for 3D hybrid bonding packaging. This is AMD. Volumes here very small, but AMD has a more than 6 months head start on shipping the technology.

Now, onto Intel. Let’s start by dispelling this fiction that Intel is buying up capacity to stiff AMD. AMD is getting the capacity they want from TSMC, on the time frame they want. Another fact, Bob Swan has long since agreed to some very large deals. This deal included capacity at N6, N5/4, and N3. This is nothing new. TSMC is smart, calculated, and is playing the long game. Yes, TSMC could be helping Intel get back on the horse as Intel works through internal process node issues. TSMC has a longer-term vision, where only they can supply leading edge capacity. In 2022 and 2023, Qualcomm and nVidia recognize this fully, but are keeping Samsung in the back pocket for the future. They want to do the same to Intel.

As Intel has moved to more industry standard SoC design flows, the ability for them to design to different process nodes increases. Bob Swan kept using the word “optionality” for the change. Intel is keeping their core IP blocks portable and even designing them to multiple processes. Their roadmap includes chiplets on in-house and external processes. There is a battle brewing with Intel on internal process versus external. TSMC wants to convert the Intel internal design and product teams to choosing TSMC nodes despite the lower margins. These teams have mostly used Intel nodes for majority of their history despite having external manufacturing account for 20% of total wafer supply. If TSMC can convince Intel design and product organizations that their nodes are better, they can cut the throat of Intel’s in-house manufacturing.

Pat Gelsinger has obviously pitched a very different story. He has been investing heavily in manufacturing and is even pushing a foundry service story. SemiAnalysis can confirm that Intel has stopped selling old tools into the market as they will be preserving older node capacity that they cannot transition. While the deals Bob Swan signed are very much binding and do include some pretty harsh terms to prevent cancellation, the long-term story is quite different.

Izvor: SemiWiki i SemiAnalysis