I was wrong on 450mm.
But first, a caveat. When speaking out on the subject in 2007, I wasn’t necessarily wrong. Nor in 2008, 2009, or even in 2010, though there were some signs my position was weakening. 2011? I was busy marketing a book. (Shameless plug… buy “Dormant Curse”.) By 2012? Definitely wrong.
Specifically, how was I wrong? 450mm will have more productivity, in comparison to 300mm, than I had given it credit. At least it does now. Now in 2012. How is that possible? Times have changed and so have circumstances. Up until last year, in our (MIT/SEMI) economic model, we had been analyzing parallel paths, that is, modeling what the semiconductor universe would look like if it continued on indefinitely with only 300mm wafers and what it would look like if 450mm were injected at various points in time. These were both reasonable paths to consider at the time, and our economic model and its productivity and cost multipliers reflected as much.
However, I believe that’s now changed. We’ve moved from the theoretical, where parallel 450mm and non-450mm universes could exist, to where Schrödinger has now opened his proverbial box and the cat is alive. His cat’s name is 450mm.
The New Question
In a way, it’s somewhat unfair, because the question posed has been changed. The question at one time, at least for many, was:
“Is it more productive for the industry to stay with just 300mm wafers or add in 450mm at some point?”
Now the question most definitely is:
“When the industry moves to 450mm wafers, which factory will be more productive, 300mm or 450mm?”
You’ll see the nuance when you think about it for a bit.
The New 450mm Productivity Multipliers
So, in this new universe, how will the 450mm productivity multipliers actually change? While it was argued during the heat of the 450mm debate, possibly to appease suppliers and 300mm chipmakers who were naysayers, that future technology improvements would be available on 300mm toolsets just the same as they would be on 450mm, most of us believed otherwise. We argued that major technology improvements would stop for 300mm. There was already scarce funding to pay for everything: factory automation improvements, technology R&D, e-diagnostics, APC, PPM, etc., let alone wafer size increase. The semiconductor industry wasn’t growing like it used to, and investment money, internal, not to mention venture capital, was hard to find. There simply was not funding to keep R&D going on two parallel platforms. History bears this out—200mm tools stop at 90nm technology. You won’t find FOUPs or AMHS for 200mm. One can get a 65nm lithography tool for 200mm, but it will effectively be a 300mm retrofit. Many of the large productivity gains attributed to the 300mm transition had little to do with the actual wafer size increase. While I can’t guarantee this will be the case for 450mm, the latest EUVL slip almost ensures something similar.
Consider the following chart publically provided by ASML. If you assume that double and quad patterning are available for 300mm tools and the “real” EUVL is only available for 450mm—that is, true production quality tools, not just experimental—then we already have a big boost which is available for 450mm-only. Here is some background: when we first analyzed “beam-tool” productivity for 300mm/450mm, we examined theoretical 450mm productivity gain assuming the exact same tool, substituting only a larger wafer size (parallel universes). An issue for beam tools is that a tool can only process a given amount of area per second, assuming the same underlying physics are at work. This being the case, productivity improvement from a larger wafer comes solely from moving more die in and out of the tool at one time. So, using 450mm wafers gave us a productivity improvement of only 1.18x for lithography in 2008, based on the particular tool analyzed. This compared to a productivity multiplier of 2.31x calculated for “non-beam” tools which process a whole wafer at once.
A 450mm Litho Productivity Multiplier Greater than 2.5x?
Now, with this in mind, think about the potential impact of comparing 300mm, quad patterning vs. 450mm EUVL, if this truly ends up being the comparison case in 2018 or 2019. This 450mm multiplier is then quite different than the one originally calculated in 2008 because we are now comparing different underlying technologies. In this case, we don’t necessarily compare the tool throughput in terms of raw die per hour processed, as the reduction in process steps more than compensates for throughput variations. The chart shows EUVL has a 5.0x reduction in process steps over quad patterning and 4.0x reduction over double patterning. This is a 5.0x productivity multiplier vs. a 1.18x multiplier previously used for 450mm lithography. We also have material usage multipliers which are impacted significantly. Granted, this is just one of the beam tools in the fab (others include inspection and ion implant and will pull this number down somewhat), but lithography is a significant cost driver.
Put a Fork in It
So, if, due to cost, timing (particularly, if any further slips occur), and demand, EUVL is a 450mm-only tool, then the answer to:
“When the industry moves to 450mm wafers, which factory will be more productive, 300mm or 450mm?”
Is likely:
“450mm!”
About that Train
I still have some numbers to run, and I’m still concerned about wafer cost, but yes, you read right correctly. This may contradict what you may have previously heard from me, and I probably owe someone an apology.
“Frank Robertson, I was wrong. Sort of. Well, I am wrong, now. The train has left the station…“
But in all fairness, Frank, in December 2007, the 450mm train hadn’t left the station. And if we had all jumped on back then, we all would have lost our shirts in the downturn. When did the train leave? It may have been in the Fall of 2011, when G450C was announced. Or, it may have been when EUVL officially slipped to 10nm or 7nm (is it official?)? Or, it may have been when all the chipmakers’ timelines magically aligned to 2018 in September this year. Or, it may have been when Intel and friends invested in ASML this past summer.
In any case, the train has left the station, Schrödinger’s 450mm cat is alive, and the 300mm-only parallel universe is gone (poof!). However, all along, advice was given to do what was in the best interest of one’s company, even when suppliers had been collectively pushing back. In other words, have a well thought-out and executed 450mm plan. At this point, if you are still standing on the platform, there may be ways to catch up if that’s what you think is in your best interest. In the coming months I hope to be able to report on ways you can effectively do so. The only thing that might change the overall equation now is if 450mm somehow slips again…and EUV is available on 300mm first…hmmm…get me that box…and get me that cat…
Never doubt Frank Robertson. He is where he is now because of a career of being correct about the future of this technology. Just because one cannot see it at the beginning of a project, does not mean it won’t come to fruition in due time. The man is a genius; an utmost professional. I was glad to see this entry.
My dear, sweet daughter Chelsea may have a slight bias.
Yes, major industry transitions have a way of aligning interests, eventually. The multipliers that matter are the ones realized between Customers and Suppliers at implementation.
The train has definitely left the station…