Late last month, the Wall Street Journal leaked news of a partnership with TSMC, in an agreement that put a formal signature on what sources have previously implied was years of behind-the-scenes collaboration. Apple has explored its options with foundry partners that aren't Samsung for several years as its relationship with that company has grown sour. Last year, Apple reportedly attempted to buy a significant stake in a particular fab for a billion dollars, but CEO Morris Chang rebuffed the idea (or at least, the price tag), saying that TSMC preferred to retain flexibility and control of its own technology roadmap.
Now there are reports that Apple is considering a similar alliance with GlobalFoundries, while SemiAccurate is reporting that Apple just bought a fab. That's a great deal of churn in such a short time -- so what the heck is going on?
The Difference Between Talking and "Talking"
talking to GlobalFoundries? Undoubtedly, yes. GloFo reportedly has unused 28nm capacity and the company would love to add Apple to its list of contracts. Sources claim that the current conversation with GloFo is more serious than previous conversations have been and that Apple may be looking to add second-source capacity. This makes a great deal of sense. TSMC'
s 40nm ramp was disastrous and the 28nm production cycle, while much smoother, still hit roadbumps and snags. At 20nm, the outlook is grimmer -- for the first time in 40 years, 20nm transistor costs are not expected to drop below 28nm levels, period.
A second-source option at GlobalFoundries
would give Apple flexibility if TSMC has production trouble without forcing it back to Samsung on short notice. It might also be easier to ramp 28nm parts at GF as compared to TSMC, given that Samsung and GloFo are part of the Common Platform Alliance. It's possible that Apple might be looking at GloFo as a partner to pick up some of its older designs from Samsung while it moves newer projects to TSMC, or it might simply want additional flexibility at future nodes. GloFo is still banging the fully-depleted SOI (FD-SOI) drum, which reportedly offers FinFET like performance on planar silicon, but there are concerns about total production cost as compared to conventional bulk silicon or FinFET
As for the rumor of Apple buying a fab, this goes under the "entirely possible, but raises more questions than it answers" check box. Price isn't the issue -- Apple, even after its stock declines of the past six months, could buy or build a fab with its spare change, but fabs take years to build and years to ramp. Apple may have bought a fab from someone, but the companies with high-capacity modern fabs probably can't afford to sell them -- GlobalFoundries, TSMC, Samsung, Intel -- none of these manufacturers have an "extra" leading-edge node, save for Intel -- and Intel knows that such a move would play disastrously in the financial press.
Still more questions than answers on this one, but an Apple/GlobalFoundries secondary deal would be a nice boost for the foundry and could provide Apple with much-desired security.