Intel's Tri-Gate gamble: It's now or never
Deep dive into Chipzilla's last chance at the low end
Analysis There are two reasons why Intel is switching to a new process architecture: it can, and it must.
The most striking aspect of Intel's announcement of its new Tri-Gate process isn't the architecture itself, nor is it the eye-popping promises of pumped-up performance and dialed-down power. And it certainly isn't the Chipzillian marketeers' risible emphasis on "3-D".
No, it's that Intel has not only the know-how and ready cash, but also the never-surrender cojones to pull off such a breakthrough. The world's number-one microprocessor designer is not content to merely dominate some sectors of the market for silicon brains, it wants its parts to be in every thinking device, from the lowliest embedded systems to the brainiest HPC clusters.
And Intel is willing to invest – bet? – big to make that happen.
The move to a 22nm Tri-Gate process architecture is an important step for Intel's entire microprocessor line, but it's especially critical for the company's desire to enter the low-power world of tablets and smartphones – and whatever consumer-level world-changers might appear next.
But although Tri-Gate is undeniably a breakthrough – more on the deep-tech details in a moment – it was not an unexpected one. The basic idea behind what Intel calls Tri-Gate is amalgamated into a concept that the rest of the known universe calls FinFET – cute geek-speak for a vertical "fin" of silicon poking up into a field-effect transistor's gate.
A fin in a FET – get it? There are a number of FinFET-ish architectures under study in labs around the world. Intel calls theirs Tri-Gate, due to the fact that the fin has a left, right, and top surface upon which charge can flow through the gate.
FinFET-based process architectures are far from new. Intel has been futzing around with the concept since 2002, and the Taiwanese chip-baking giant TSMC demonstrated a 25nm FinFET design it dubbed "Omega" at about the same time.
What Intel has now done, however, is not merely demo another FinFET concept, but to throw the full weight of its manufacturing prowess and financial clout behind
FinFET Tri-Gate, and move its entire microprocessor line to the new process.
Intel's breakthrough isn't conceptual and lab-based, it's real and market-based.
And it's very, very expensive. Intel is spending $8bn to upgrade four fabs in Oregon, Arizon, and Israel to 22nm Tri-Gate, and to create a fifth one, also in Oregon, from scratch. At the same time, by the way, Intel is also investing a good chunk of its $10.2bn 2011 capex budget on the development fab for its upcoming 14nm process, and bulding that fab bigger than originally planned.
Eight billion dollars for 22nm Tri-Gate fabs is not chump change – especially to a company that recently bought McAfee for $7.7bn and Infineon for $1.4bn, and that spent over $7bn since early 2009 on other fab upgrades.
Oh and let's not forget the little $1bn annoyance that was the Cougar Point chipset flaw this winter.
But Intel has the resources it needs to move to 22nm Tri-Gate. After all, acquisitions, capex, and "oops" are not the only ways that the company is doling out cash these days. The company also spent $4bn to repurchase 189 million shares of common stock in its most recent quarter – a quarter during which it also paid just under $1bn in dividends to stockholders. (Are you listening, Steve Jobs?)
Not that Intel has a cash hoard equivalent to Apple's $66bn: its cash and cash equivalents, short-term investments, and trading assets total just under $12bn. But compare that with its closest microprocesser rival, the fabless AMD, which has just $1.75bn in the bank. Intel's market capitalization is $127bn; AMD's is $6bn.
Intel's other main rival, of course – and one that's growing in importance – is also fabless and also goes by a TLA: ARM. If Intel's engineering chops and deep pockets are the proof of the "because it can" reasoning behind its move to Tri-Gate, ARM – and, to a lesser extent, AMD – is the driving force behind "because it must".
Let's back up a few years to 2007, the year in which Intel introduced its most recent Really Big Thing™ in process technology. That's the year when the company replaced the traditional silicon dioxide gate dielectric in its microprocessors' transistors with a high-k metal gate, which increased gate capacitance, thus improving performance while significantly reducing current leakage.
The high-k metal gate helped make it possible to shrink that generation's process technology to 45nm. And we're all familiar with the process-shrinking mantra: smaller processes mean lower power, faster performance, less heat, Moore's Law lives to fight another day, and blah, blah, blah.
The difference, however, between 2007 and today was that four years ago Intel was riding high with nary a serious threat in sight (sorry, AMD fans). The company's move to the 45nm "Penryn" line was arguably prompted mostly by a desire to induce upgrades from users of the previous generation of 65nm "Conroe" processors, and to sell data center folks on the cost savings of lower-power parts.
That was then. Things have changed.