450mm Wafers will not be Seen in Fabs

FrgMstr

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
55,602
I was reading this article over at EETimes describing that the evolution of wafer size has come to a standstill if not for the time being, but possibly for a number of years.

Between that and the semiconductor industry's modest growth rate over the past few years, there is no need for a massive expansion of capacity as there has been in the past, Hutcheson added. Without sufficient demand for the larger wafer size, building a 450mm fab would require chip makers to take 300mm fabs offline.

"450 died because it's one generation too far," Hutcheson said.

While reading this, my mind when back to visiting GlobalFoundries before its fabs came online in New York. This video showcases GloFLo's Integrated Testing and Development Center. What this is, is a mock-up of just how its simply moves it wafers around the fab and keeps up with all of those. So if you think changing over wafer size would be minimal undertaking, imagine having to retrofit, or change the "simple" materials handling portion of the equation.
 
It seems the entire foundry business in dying in terms of progress and cost reductions. Transistor cost goes up, IC design cost goes up. Companies left and right cant afford it. Node renaming madness by foundries.

Add this to the mix as well:
http://www.digitimes.com/news/a20170114PB201.html?mod=0
"Recent market response to MediaTek's 10nm products, including its Helio X30 and Helio P35 CPUs, has not been enthusiastic due to their high production costs, as well as high selling prices, Chen was quoted as saying."
Click to expand...
 
Is there any truth to Intel having exclusive rights to some of the lower node fabrication tools? If so, this alone would cause price increases to the other fabs having to rely on alternate tools and techniques.
 
My buddy got fired for dropping an entire ingot worth of wafers LOL. It was mostly an accident though. That's a LOT of money!
 
Gman1979 said:
Is there any truth to Intel having exclusive rights to some of the lower node fabrication tools? If so, this alone would cause price increases to the other fabs having to rely on alternate tools and techniques.
Click to expand...

Nopes. If anything Intel as the leader is reducing cost for the other foundries for laying the first work.
 
Wafer "lots" are usually 24-25 wafers per lot where I work. I can tell you a "lot" of 300mm wafers can get kind of heavy. I was thinking about this a couple days ago actually and was thinking they would probably have to drop down that number because it would just be such a large thing to handle. But the whole system currently revolves around 25 wafers.

But even so, you can fit a lot of chips on a 12" (300mm) wafer. It's not like we are hitting capacity limitations enough to warrant buying an entire new fabs worth of tools (which would be more expensive then normal since it's a new format).

Like everything, eventually the market settles on what is practical.

Also, with 8" we were able to make them pretty damn thin (like couple sheets of paper thin) and they would still hold their own weight when suspended by just the edges. Then we went to 12" and the wafers had to be considerably thicker...maybe 1-2mm. So they weight increased dramatically. Now you want to talk about 18" wafers? Very heavy I am thinking.
 
Gman1979 said:
Is there any truth to Intel having exclusive rights to some of the lower node fabrication tools? If so, this alone would cause price increases to the other fabs having to rely on alternate tools and techniques.
Click to expand...

Yes in some ways.
Intel designs their own processes and many of the tools they use.
Almost everyone else mostly uses processes and tools developed by IBM and partners then adds their own spin on top of it.
 
With all the costs involved in transitioning to 10nm and 7nm combined with moving to EUV lithography, it isn't surprising that the industry doesn't also want the added costs of also moving from 300mm wafers up to 450mm ones until the demand justifies spending the money to do it.

Normalizing at 300mm wafers for so long has helped keep costs down for manufacturing at process nodes that don't need to be cutting edge too. Still plenty of chips can be fabbed at 32nm-65nm and be plenty efficient across the industry beyond CPUs and GPUs that gets by fine with 300mm wafers.

Odds are , if AMD , Intel can hold off ARM taking over the desktop for a few more years, it will fit into their roadmaps to shift over to EUV and 450mm wafers eventually to stay relevant or have to hold off as they transition away from silicon to push past 7nm anyway so it is smart to plan to be on 300mm wafers for now.

It sucks for the makers of fab equipment to not be able to count on forcing the industry to upgrade everything just to stay competitive, but they knew a slowdown was either going to come from this or something else based on the way process shrinking hit a huge stumbling block at 28nm and the market downturn so they will probably just plan on upping capacity of 300mm setups and make sure they can handle demand for 3d-nand and other storage chip tech more and more since no matter what , people are using more and more storage across many sectors.
 
I was in CMP during the transition from 200 to 300mm wafers, good times. As noted above moving up in size involves not just adjusting equipment to accommodate the new size, new processes are needed as well. I remember the issues we had with consumables such as slurry and wafer inserts alone. For reference, how many fab's are still running 200mm lines(TSMC)?
 
It's interesting though.

They say the market is contracting, but at the same time, there has been a shortage of capacity in the smallest nodes for seemingly ever, which is why GPU's stayed on 28nm for so long, because mobile (Apple, Qualcomm, etc) were gobbling up all the capacity...

But yea, changing all your equipment to fit larger wafers is HUGE and hugely expensive. It will probably happen at some point, when a combination of factors occurr:

1.) Existing equipment gets old enough that keeping it running is high maintenance and difficult, so you have to spend the money on new equipment and validation anyway; and

2.) the cost savings from going to a larger size makes financial sense compared to the cost of replacing and validating all the pertinent equipment in a production line.

Until then it won't. These are extremely expensive plants, and the businesses that run them are not into frivolous spending. They will move on it when it makes financial sense, and not a moment sooner. The investment in new equipment and new validation has to pay off. Doesn't seem like that is the case today.
 
Evil Scooter said:
I was in CMP during the transition from 200 to 300mm wafers, good times. As noted above moving up in size involves not just adjusting equipment to accommodate the new size, new processes are needed as well. I remember the issues we had with consumables such as slurry and wafer inserts alone. For reference, how many fab's are still running 200mm lines(TSMC)?
Click to expand...

All the big players still have 200mm Intel, Global, Micron etc.
 
I always heard rumors that they had a better yield with a larger wafers. I have heard plenty of things which I cannot repeat. I will say that there's plenty of surprises to be had.

"“What happened is that the shrinks have slowed down the growth of the silicon.”"

This is only partially true. Say someone comes up with a novel way of thermal reduction while reducing the power requirement? We wouldn't need die shrinks to get better performance.

It's coming.
 
gobnu1 said:
All the big players still have 200mm Intel, Global, Micron etc.
Click to expand...

I dont think Intel got any 200mm left, unless its for a legacy node. TSMC uses 200mm for very old nodes too and if those nodes stop, so does 200mm.
 
undermined said:
With all the costs involved in transitioning to 10nm and 7nm combined with moving to EUV lithography, it isn't surprising that the industry doesn't also want the added costs of also moving from 300mm wafers up to 450mm ones until the demand justifies spending the money to do it.

Normalizing at 300mm wafers for so long has helped keep costs down for manufacturing at process nodes that don't need to be cutting edge too. Still plenty of chips can be fabbed at 32nm-65nm and be plenty efficient across the industry beyond CPUs and GPUs that gets by fine with 300mm wafers.

Odds are , if AMD , Intel can hold off ARM taking over the desktop for a few more years, it will fit into their roadmaps to shift over to EUV and 450mm wafers eventually to stay relevant or have to hold off as they transition away from silicon to push past 7nm anyway so it is smart to plan to be on 300mm wafers for now.

It sucks for the makers of fab equipment to not be able to count on forcing the industry to upgrade everything just to stay competitive, but they knew a slowdown was either going to come from this or something else based on the way process shrinking hit a huge stumbling block at 28nm and the market downturn so they will probably just plan on upping capacity of 300mm setups and make sure they can handle demand for 3d-nand and other storage chip tech more and more since no matter what , people are using more and more storage across many sectors.
Click to expand...

I doubt there is more than 3-4 companies that can make money at 7nm. Unless its another very renamed node into 7nm. Even MediaTek is in big trouble at "10nm".

ARM isn't coming to the desktop the next 10 years or 20. Just as x86 isn't coming to the phones in the same period.
 
westrock2000 said:
Wafer "lots" are usually 24-25 wafers per lot where I work. I can tell you a "lot" of 300mm wafers can get kind of heavy.

...

Now you want to talk about 18" wafers? Very heavy I am thinking.
Click to expand...

Will tell you that an EMPTY 450mm carrier weighs a lot; the difference between a 300 & 450 is quite staggering. The tools/handlers grew massively as well. 450 dying was very painful for a lot of companies; the one I work for included. We adapted though, as did everyone else. Would've been something to see though. I thought the change from 200>300 was pretty amazing; but seeing 450mm carriers zooming around the over-head transport tracks would've been quite a sight to behold.

BTW, not sure why on earth anyone thinks foundries aren't doing well. I mean, seriously, are you nuts?!
*edit - fixed typos*
 
Last edited:
fastgeek said:
Will tell you that an EMPTY 450mm carrier weighs a lot; the difference between a 300 & 450 is quite staggering. The tools/handlers grew massively as well. 450 dying was very painful for a lot of companies; the one I work for included. We adapted though, as did everyone else. Would've been something to see though. I thought the change from 200>300 was pretty amazing; but seeing 450mm carriers zooming around the over-head transport tracks would've been quite a sight to behold.

BTW, not sure why on earth anyone thinks foundries aren't doing well. I mean, seriously, are you nuts?!
*edit - fixed typos*
Click to expand...

Did you still use 25 wafer FOUP/Cassettes?
 
Shintai said:
I dont think Intel got any 200mm left, unless its for a legacy node. TSMC uses 200mm for very old nodes too and if those nodes stop, so does 200mm.
Click to expand...

Fab 17 was the last 200mm node for Intel. All nodes are 300mm now.
 
I suppose making larger wafer sizes would increase the number of things being make and end up reducing the end cost eventually. That is not something that most manufacturers want.
 
undermined said:
Normalizing at 300mm wafers for so long has helped keep costs down for manufacturing at process nodes that don't need to be cutting edge too. Still plenty of chips can be fabbed at 32nm-65nm and be plenty efficient across the industry beyond CPUs and GPUs that gets by fine with 300mm wafers.
Click to expand...
32-65nm?!

I do 180nm and 90nm designs all day long!
But then again, we are a mixed-signal and not big-digital company.
 
ManofGod said:
I suppose making larger wafer sizes would increase the number of things being make and end up reducing the end cost eventually. That is not something that most manufacturers want.
Click to expand...

Its something everyone wants. The real problem is volume decreases as less and less companies can afford new nodes. Also why stagnation keeps getting closer very fast.

There is also very few 450mm ready plants.

Cost example from 2013:
GF-2-640x370.png
 
westrock2000 said:
Did you still use 25 wafer FOUP/Cassettes?
Click to expand...

Sorry for the delay. We use 25 slot FOUPs for 300mm. Still need to find a 450mm unit; but am pretty sure they were 25's too. The 450mm slabs and stages were just mind blowing.

As for nm stuff? I'm reluctant to say anything, least I say something I shouldn't; but it's public knowledge that 14nm is well established and that 10nm is coming online, if not already.
 
westrock2000 said:
Looking at this list so many companies have lame names for their fabs. Fab5, Fab10, etc.......boring.
Click to expand...

That's nothing new. Very few sites use unique or interesting names I'm afraid. :(

I used to frequently visit customer sites from 1997-2002, so a lot of sites on this list are either different now, or are closed.

AMD: Sunnyvale (the old "White House" HQ)

IMEC: Leuven, Belgium

Intel: D1B, D1C, D1D, D1X, D2, Fab 5, 10, 11, 12, 15, 17 and 18.

Motorola: Phoenix, AZ and... ummm?

ST Micro: Rousset, France

This is driving me crazy... used to visit a couple sites in Houston and went to San Antonio a LOT.... but can't recall for the life of me what sites they were. Maybe VLSI, Philips or something like IMEC. Pretty sure I went to a Texas Instruments site a couple times.
 
You must log in or register to reply here.
Back
Top