FABULOUS TIDINGS are at hand for those who like a good old fashioned competitive skirmish, as it would appear that Intel will be going head-to-head with its manufacturing rivals over the coming months.
A recent article in EETimes discussed whether Intel should be referred to as a fab broker or breaker, pointing to the chip giant's recent investments in IC-equipment vendors like ASMI, which provides Chipzilla with both high-k and low-k tool tech. The article clearly stated that Intel could not afford to have a weak semiconductor equipment supply chain, and would invest in and prop up its partners where necessary to stop this from happening.
It certainly isn't a secret that the semiconductor industry has taken a heavy hit during the current economic crisis and that many fab firms have had to drastically cut back on spending and R&D.
Intel, it would appear, has done no such thing, however, announcing just a few months ago that it would be channeling seven billion dollars into building and expanding its US fabs over the next two years and retooling for its ramp at 32nm. The firm is also expected to choose its tools for its upcoming 22nm process transition by the end of July, which is something the tool vendors are apparently clamouring over much like bridesmaids trying to catch a bouquet.
But if the economic downturn has stopped everyone in their tracks, what's the hurry? The answer to this question lies partly with IBM's fab alliance, which includes AMD spinoff GlobalFoundries. Contrary to EETimes' implication that AMD "has got enough on its hands", GlobalFoundries is indeed pressing ahead with equipment purchases and investing in new tools for 32nm and beyond. Just like Intel.
GloFo's Fab 2 is a $4.2 billion fab and the company is also known to be expanding its capabilities on its Dresden campus, where it hopes to bring 32nm bulk silicon technology online early in 2010.
Intel, realizing that it cannot rest on its laurels, is therefore rushing full steam ahead with 450mm wafer technology, automated test equipment (ATE) and Extreme Ultra Violet (EUV) lithography.
"We're pleased to see Intel following our lead and investing in immersion lithography," Tom Sonderman, GlobalFoundries' vice president of manufacturing systems and technology, told the INQ. "AMD was first to introduce high-volume immersion at 45nm and GlobalFoundries sees a bright future for this technology at the 32nm node and beyond," he added.
This may not be entirely accurate according to some other industry sources. ‘Immersion'lithography uses Deep Ultra Violet at 193nm ArF under a water coating between the bottom lens and the wafer being patterned. Intel has been a key sponsor of EUV as the replacement for 193nm DUV. Intel also regarded the first production immersion tools not meet required performance and throughput levels when selection time was required at the 45nm node migration. This would have been a year before production started and over a year later than AMD had to make the same choice.
Still, whereas the two fab firms may agree on the importance of lithography - a technology many in the industry still remain dubious about due to its high R&D costs and uncertain returns - they certainly don't agree on another controversial move, that of migrating to 450mm wafers.
As EETimes points out, Intel is not alone in shooting for 450mm "prototype" fabs by 2012, with TSMC and Samsung both having jumped on the 450mm bandwagon. But EETimes also makes the assertion that GlobalFoundries, or AMD as it is referred to in the article, "cannot afford to be preoccupied with 450mm fabs" at the moment. This, says Sonderman, is the wrong conclusion.
"Those rushing to 450mm are making a silent acknowledgement that they are out of ideas for improving fab productivity using traditional engineering practices," he told us. "At present, 450mm is largely a distraction given the abundance of productivity opportunities that still exist to make 300mm tools and fabs more efficient."
Sonderman said the improvements he referred to would likely "extend the lifecycle of 300mm investments throughout the industry while paving the way for a smooth long-term transition to 450mm over the next 10-15 years."
Intel's argument appears to be that moving to 450mm fabs will advance the march of Moore's Law and give it a stronger lead in the x86-based processor rat race, but this doesn't seem to make sense either economically or practically.
For a start, there appears to be little or no correlation between 450mm wafers and Moore's Law. The move to 450mm wafer diameter just offers more dies per wafer rather than any innovations at the circuit element density, die technology or fab operations levels. Then there is also the important question to be asked about who will have to pay for the R&D involved.
In a climate where vendors are struggling as it is, we find it hard to believe that many will be willing to make R&D commitments to 450mm when the benefits to be gained appear to be so minimal.
Either way, the battle of the chip fabs just got more interesting. It should be a fabulous contest to watch unfold, so to speak. µ
