At yesterday's Common Platform technology day, IBM
and the other members of the Common Platform Initiative made a major announcement. While plans to use gate-first
technology at the 32nm/28nm node remain unchanged, the coalition will move to gate-last technology when it makes the jump to 20nm production.
As semiconductor manufacturing has moved to ever-smaller process nodes, the difficulty of managing each transition has increased markedly. As a result, the major foundries have adopted divers methods of improving product yields and minimizing gate leakage. Examples of these methods include AMD's adoption of immersion lithography at 45nm, Intel's simultaneous decision to use double-patterning at 45nm but not introduce immersion lithography until 32nm, and TSMC's
choice to extend traditional Poly/SiON technology to the 28nm node. All of these decisions have carried a degree of risk—Robert Palmer, current member of AMD's Board of Directors and former Digital CEO purportedly remarked that "Designing microprocessors is like playing Russian roulette. You put a gun to your head, pull the trigger, and find out four years later if you blew your brains out."
He That Is First Shall Later Be Last
The major semiconductor manufacturing question of the last year has been whether gate-first or gate-last technology offered the smoothest transition to 28nm and the nodes below it. According to an IMEC paper available here
, "The terminology 'first' and 'last' refers to whether the metal electrode is deposited before or after the high temperature activation anneal(s) of the flow."
Up until now, Intel and TSMC were the major proponents of gate-last technology based on its supposed long-term manufacturing advantage. IBM, GlobalFoundries, and other members of the Common Platform Initiative, had instead opted to pursue gate-first 28nm technology. The advantage of gate-first, at least in theory, is that it doesn't require customers to re-architect their products and doesn't negatively impact chip density.
talked up the advantages of GF over GL on multiple occasions, including during the company's September 2010 technology day
There have been rumors that Samsung, a member of the Common Platform Initiative, was less-than thrilled with the gate-first approach, but yesterday's announcement was the first sign that IBM and GlobalFoundries were switching their positions on such an important issue. We spoke with GlobalFoundries' Jon Carvill, the VP of Global Communications, and were told that the decision to use GL at 20nm has nothing to do with the foundry's current 32nm ramp. Carvill explained that the change "is due to the design rules and desired scaling our customers want at 20nm. Gate Last was the best option to hit it and Gate First was the better option at 32/28nm. We have multiple customers engaged on Gate First 32nm/28nm as does Samsung (who made this disclosure with us yesterday)."
The timing of the announcement and the reason for the change both make sense. It's been less than two years since GlobalFoundries opened for business; the company's executives likely decided it wasn't realistic to ramp 32nm SOI and multiple bulk silicon production nodes while simultaneously adopting the new rules and restrictions that governm 28nm gate-last manufacturing.
Now that GF is shipping 40nm-LP wafers and ramping 32nm SOI, 28nm-HP, 40nm-G and 28nm-SLP smoothly, it presumedly has more than enough time to begin work on the eventual 20nm gate-last transition. Yesterday's announcement won't have any impact on GlobalFoundries' 2011 roadmap and is not expected to disrupt the company's plans to move below the 28nm node.