June 18, 2019
Joel Hruska

Intel May Tap Samsung for 14nm Manufacturing Support

There are fresh rumors that Intel might tap Samsung for support at 14nm, though there are also reasons to be dubious of them. If true, it suggests that Santa Clara will be stuck on 14nm for a significant period of time for at least some parts, recent discussions of Ice Lake notwithstanding.

According to SE Daily (via Google Translate), Intel and Samsung are in the final stages of negotiating for additional capacity. Intel is said to have chosen to work with Samsung over TSMC due to concerns about Huawei and AMD’s improved competitive performance. TSMC has said that it believes it can continue to manufacture chips for Huawei, and this has supposedly spooked Intel into preferring Samsung as a partner, due to the potential for further retaliatory trade decisions targeting companies that do business with Huawei.

I don’t want to go so far as to imply that this is untrue, but the timeline seems highly compressed. Negotiations for foundry capacity between two massive firms aren’t going to be hammered out over a weekend, and the US government’s total blockade of Huawei is still fairly new. Furthermore, taking action against TSMC for believing it can still fabricate SoCs for Huawei would be, in some regards, overkill. Huawei is facing massive issues bringing products to market for reasons that have nothing to do with its ability to source SoCs. Even with perfect support on foundry, its manufacturing supply chain is under existential threat, to say nothing of its access to software and support tools.

The idea that Intel would opt to use a foundry other than its chief competitor, AMD, is possible. Intel might be sensitive to the idea that it had gone, hat-in-hand, to the same firm that supplies its competition. Partnering with Samsung — whose 14nm node is in generally excellent shape and was used for AMD hardware at GlobalFoundries after GF licensed the tech years ago — is a bit less direct.

euv_adv-1

Samsung has been putting a heavy push on its 7nm EUV technology, but Intel might tap the company for its older 14nm nodes.

The biggest reason to look askance at this rumor is that it suggests Intel would launch a 14nm “Rocket Lake” competitor on Samsung silicon. In the past, Intel has signed agreements with TSMC to produce Atom CPUs or (as was rumored) chipsets. Building “big-core” CPUs at a rival foundry would be a major shift. That’s one reason I don’t want to put a ton of weight on this rumor, but there is a way in which it could all make sense.

One of the difficulties of bringing up a new process node in an existing factory is the disruption to ongoing production. If you want to replace 14nm capacity with 7nm, you’re going to have to turn the lines off at some point to do the upgrades. Intel has always run its fab lines in tight cadences to allow for this, but we know that demand for 14nm has run extremely high. As recently as last year, Intel announced it would allocate additional funds to beef up 14nm production. At the same time, the long delay on 10nm has bottlenecked Intel’s facilities. The company has forecast a relatively quick shift to 7nm (with production scheduled by 2021), which means it needs a fairly quick volume rollout at a time when 14nm demand may already be running hot.

If this rumor is true, it may be true inasmuch as Intel cut a deal with Samsung to run some products out of its own factories while it aggressively upgrades its own plants. The company undoubtedly wants to re-establish the narrative of process supremacy that it enjoyed for nigh on 20 years prior to its 10nm slide and might prefer to sprint for 7nm by taking advantage of a competitor’s production capability rather than going it entirely alone.

The SE Daily hints at another reason why Intel and Samsung might cut this kind of deal: Pricing. From the story:

Recently, Samsung foundry has announced that it has introduced some 60% unexpected production unit price compared to TSMC for some companies. Samsung has offered TSMC a full set of mask sets that are cheaper than the “multi-layer mask” (MLM) set introduced to reduce costs in small-volume production. A mask is a kind of film that is used to draw a circuit on a wafer.

While the dramatic cost reductions we’ve heard about were in relation to 7nm, it’s entirely possible that Samsung and Intel would cut a deal on 14nm as well. Samsung Foundry is likely hungry for customers and building for Intel would be a prestigious win. Intel (again, assuming this rumor is accurate) would obviously want a good deal for products and might find Samsung more palatable than TSMC — or simply be concerned about more prosaic questions of part availability.

Right now, Intel has given limited windows into its roadmaps on 10nm and 7nm. The company has stated that 10nm++ and 7nm will overlap in 2021 and that it will lead with a GPU on 7nm. Ice Lake shipments in notebooks are expected to begin in June, with volume shipments by the end of the year. No timeline has been provided for desktop parts and the roadmaps that have leaked (which may or may not be accurate) show 14nm hanging around on desktop through 2020. With AMD launching 7nm in a matter of weeks, the ride may about to get bumpy for Intel.

Update (6/18/2019): There’s reason to believe that if a deal like this goes through — and nothing has been publicly announced yet — it might be for the same sort of adjacency products that Intel has sometimes built through partners before. This kind of allocation is the type of maneuver we’d expect Intel to make while trying to maximize building its highest-margin parts in-house with limited foundry space.

Now Read: