Intel 32nm arrives right on schedule in the first days of 2010, a clean shrink of the 45nm process into Westmere, launched at CES, and the only remarkable thing about it is how unremarkable it has become for Intel to be a full node ahead of everyone else. Second-generation high-k metal gate, smaller transistors, better power, delivered on the exact cadence the roadmap promised two years ago. The metronome ticks. Nobody else in the industry can set a watch by their own factories the way Intel can.
32nm is a tick, the shrink that follows the 45nm tock, and it refines rather than reinvents: the high-k metal gate transistor Intel pioneered, now in its second generation, tighter and more efficient. It also quietly makes the tool change Intel skipped last time, moving to immersion lithography on the critical layers, the same immersion that gave TSMC such grief at 40nm, except Intel folds it in on schedule and without drama. The launch parts are Clarkdale and Arrandale, the first Core i3 and i5 chips, a 32nm processor sharing a package with a 45nm graphics-and-memory die. Sandy Bridge is coming on this node within the year, and it will be the chip that defines the early decade, the one that makes laptops fast and cool enough that the desktop starts to feel optional. The process is the foundation under all of it, and the process is, as usual, the best in the world.
The distance to everyone else is almost comic. TSMC only just reached 40nm, and it got there through a yield crisis that nearly broke the node, while Intel is shipping 32nm in volume with the calm of a company that has done this many times. The foundries are building other people’s designs a generation or two back, and AMD, now making its processors at the GlobalFoundries it spun out last year, is still finishing its own move to 45nm. Intel is building its own designs on the most advanced line on earth, and the integrated model, design and manufacturing under one roof, looks not just successful but permanent.
The money makes the case better than any engineer could. Intel’s gross margins sit north of sixty percent because it sells the most advanced processors in the world and builds them in factories it has already paid off, on a node nobody can match. Every wafer of Intel silicon earns more than a wafer of anyone else’s because the chips on it command a premium the competition cannot touch. This is why Intel keeps its fabs to itself. Renting them to outside customers would mean filling premium capacity with lower-margin work, and from inside margins like these, that looks like a mistake only a desperate company would make.
There is one thing on the horizon Intel is not winning, and it is easy to wave away in early 2010. Phones. The processors going into the smartphones that are just starting to matter are ARM designs built by foundries, the outsourced model TSMC and its rivals run, with Qualcomm and Texas Instruments turning out the parts that go in them, and Intel has no real answer. It is not for lack of history. Intel owned a perfectly good ARM phone-chip line called XScale and sold it to Marvell in 2006, the year before the iPhone redefined what a phone chip was for. Its answer now is Atom, an x86 part it is trying to squeeze into a phone power budget the architecture was never drawn for. Intel’s position is that phones are a toy market with thin margins and that real computing will always need real processors. It is a reasonable position. It is also the first time in years Intel is not the obvious leader in something, and it is choosing not to notice.
None of this dents the confidence, and at 32nm the confidence is earned. Intel is two years ahead on manufacturing and its tick-tock cadence has turned chip-making from an art into a schedule. Perso, I would not bet against the metronome in 2010; it has been right for a decade and there is no crack in it yet. The phone thing nags, though. Not because Intel is losing money on it, it is not, but because the one market Intel is ignoring is the one growing fastest, and companies that own the present have a way of being slow to notice when the future sets up somewhere else. For now the watch keeps perfect time.