[Semiwiki] Intel 450 mm wafers delayed until 2023, other SPIE News

[norseamd]

As a news site, it's just as bad, if not worse. Charlie's a douchecanoe, but he's usually right.

why are there no more free articles? also why is that copper guy threatning to ban someone who just is wondering about paying 100. they are asking a lot of money. 20 might be easy but 100? the fact is that guy had a right to wonder whether his 100 is a worthy investment? more bothered by the threatning of a ban then the cost

 

[CHADBOGA]

why are there no more free articles? also why is that copper guy threatning to ban someone who just is wondering about paying 100. they are asking a lot of money. 20 might be easy but 100? the fact is that guy had a right to wonder whether his 100 is a worthy investment? more bothered by the threatning of a ban then the cost

Because copper is an idiot who enjoys abusing what ever power he has over people on a forum.

Yes he really is that pathetic.

I'd sooner pay money to not read them, than give them a $100 a year subscription.

And it gets even better, I believe they have a $1,000 a year "Professional" subscription. LOL

 

[norseamd]

And it gets even better, I believe they have a $1,000 a year "Professional" subscription. LOL

yes i noticed. but that is for professionals who need long term computer industry outlooks. at least i think

what really pisses me off is that charlie edits his articles so the first few paragraphs will try to convince people to sign up. are a couple of free articles a month really that hard to allow? i actually enjoyed reading some of his articles that were written sometime last year. the number of articles that are available have dramatically dropped.

 

[Xpage]

i barely check out S/A anymore, no news comes from there. I thought they were going to hav eold articles expire and become available after a month or was that before the site's re-revamp.

I wonder how the daily pageviews have done.

 

[Tuna-Fish]

They don't. They just execute and pick the the correct answer later, or something like that.

You can either do this (execute both sides of branch and pick correct answer), or for branches that have effect on all lanes of a wavefront they can stall until the correct target is known (and execute other wavefronts in the meantime).

 

[Fjodor2001]

So what's the big issue with 450 mm wafers. Why is it so much harder to produce than 300 mm? And why would it be more expensive?

Also, could Intel really be holding it back due to low semiconductor chip demand in general? If they've got the technology to produce 450 mm wafers in a couple of years time, why not go ahead and use the extra capacity e.g. to produce flash memory for SSD:s or whatever, if there's not demand for CPU:s?

 

[kimmel]

So what's the big issue with 450 mm wafers. Why is it so much harder to produce than 300 mm? And why would it be more expensive?

Also, could Intel really be holding it back due to low semiconductor chip demand in general? If they've got the technology to produce 450 mm wafers in a couple of years time, why not go ahead and use the extra capacity e.g. to produce flash memory for SSD:s or whatever, if there's not demand for CPU:s?

What's the big issue? Really? If you have to ask that then you probably aren't in a position to question Intel's business decisions.

450mm won't go if all of the industry isn't supportive of the timetable. Intel could want it, but if Samsung/TSMC aren't on board it isn't going to go anywhere. Even Intel's deep pockets aren't deep enough to fund that transition alone.

 

[witeken]

So what's the big issue with 450 mm wafers. Why is it so much harder to produce than 300 mm? And why would it be more expensive?

A bigger wafer brings some technical difficulties with it, simply because... it's larger. You will need fabs and tools that are adapted for 450mm wafers. I don't think it are very big issues, but there a simply huge initial costs, also in R&D. I don't a 450mm wafer is (much) more expensive in price/area.

Also, could Intel really be holding it back due to low semiconductor chip demand in general? If they've got the technology to produce 450 mm wafers in a couple of years time, why not go ahead and use the extra capacity e.g. to produce flash memory for SSD:s or whatever, if there's not demand for CPU:s?

I really don't think it will be delayed, Intel's planning to get it ready by ~2018, so just like with new nodes, if it's ready, there's no reason to delay it, unless there are huge issues, because the benefits are very nice.

 

[norseamd]

A bigger wafer brings some technical difficulties with it, simply because... it's larger. You will need fabs and tools that are adapted for 450mm wafers. I don't think it are very big issues, but there a simply huge initial costs, also in R&D. I don't a 450mm wafer is (much) more expensive in price/area.

the 450 should be cheaper per area. the laws of economics. and that was the whole point

 

[TuxDave]

So what's the big issue with 450 mm wafers. Why is it so much harder to produce than 300 mm? And why would it be more expensive?

Also, could Intel really be holding it back due to low semiconductor chip demand in general? If they've got the technology to produce 450 mm wafers in a couple of years time, why not go ahead and use the extra capacity e.g. to produce flash memory for SSD:s or whatever, if there's not demand for CPU:s?

If you had 2 cups and I asked you to evenly fill it up halfway with water with no more than a 1mm difference in height across all cups, you probably could do a decent job. But it'll gets progressively harder if I told you to do it across 10 cups and then eventually 1,000 cups. Technically you're "doing the same thing you were doing before" except with more cups, you have to have that much more precision.

So yeah, I picture it like that. It's harder because it's bigger/more area to do fabrication across. However the goal is that when you do figure out how to do it, it'll cost you less per area.

 

[Fjodor2001]

If you had 2 cups and I asked you to evenly fill it up halfway with water with no more than a 1mm difference in height across all cups, you probably could do a decent job. But it'll gets progressively harder if I told you to do it across 10 cups and then eventually 1,000 cups. Technically you're "doing the same thing you were doing before" except with more cups, you have to have that much more precision.

So yeah, I picture it like that. It's harder because it's bigger/more area to do fabrication across. However the goal is that when you do figure out how to do it, it'll cost you less per area.

Yes, if I would wild-ass guess and make general assumptions I'd probably say the same. It would be interesting to know some actual specific technological details explaining the issues involved in moving from 300 mm to 450 mm though.

 

[Phynaz]

Yes, if I would wild-ass guess and make general assumptions I'd probably say the same. It would be interesting to know some actual specific technological details explaining the issues involved in moving from 300 mm to 450 mm though.

Dude, Tux is a CPU designer for Intel. Show a little respect for him trying to explain it to you in layman's terms. Unless you're a process engineer you aren't going to understand the technical reasons.

 

[witeken]

Yes, if I would wild-ass guess and make general assumptions I'd probably say the same. It would be interesting to know some actual specific technological details explaining the issues involved in moving from 300 mm to 450 mm though.

Here's a quote I once found when searching for information about 450mm:

Other initial technical problems in the ramp up to 300 mm included vibrational effects, gravitational bending (sag), and problems with flatness. Among the new problems in the ramp up to 450 mm are that the crystal ingots will be 3 times heavier (total weight a metric ton) and take 2-4 times longer to cool, and the process time will be double. All told, the development of 450 mm wafers require significant engineering, time, and cost to overcome.

 

[AtenRa]

At this time, the biggest obstacle for 450mm wafers is the R&D cost estimated to 25-30 Billions. It took almost 15 years to recoup the R&D cost for 300mm wafers and it will take more than 25 years for 450mm.
The industry is reluctant to spend that money for the 450mm and that was the reason Intel invested in to ASML.

If Intel will not get a big market share of the mobile/tablet market within the next 1-2 years you can say good by to 450mm wafers for this decade. It will take enormous volumes in order to recoup and be financial viable to operate 450mm wafers in the coming years. That is why Intel is trying to get a big Mobile market share now.

 

[lagokc]

Yes, if I would wild-ass guess and make general assumptions I'd probably say the same. It would be interesting to know some actual specific technological details explaining the issues involved in moving from 300 mm to 450 mm though.

The real trick is training Samurai to wield the vacuum nōdachi you need to move individual wafers around. :awe:

 

[TuxDave]

Dude, Tux is a CPU designer for Intel. Show a little respect for him trying to explain it to you in layman's terms. Unless you're a process engineer you aren't going to understand the technical reasons.

lol. Thanks but it's ok

As posted in witeken's post, a major challenge focuses on uniformity and "flatness". (also another reason why I thought my even cups of water analogy was a good one ).

For example, material like photo resist is poured on the wafer and spun out to make it a uniform thin layer across the wafer. The larger the wafer, the harder it is to control that photo resist thickness. Temperatures play a value in the rate that material can distribute into the wafer and if it cools/heats unevenly across the wafer you can get more variation. Then I guess there's etch rates, ion implantation or sputter deposition, and other things that happen across the whole wafer at once. I guess there's the basic problem of whether or not the wafer is even flat enough even before you start doing stuff to it.

Basically, several fabrication steps happen across the wafer at once and so keeping that uniformity gets progressively harder. I wouldn't call myself a process expert and so I can't really prioritize the challenges. Hell I don't even know which fabrication steps are no longer relevant or how obsolete my knowledge is.

So "real" process engineers can feel free to correct me.

 

[witeken]

At this time, the biggest obstacle for 450mm wafers is the R&D cost estimated to 25-30 Billions. It took almost 15 years to recoup the R&D cost for 300mm wafers and it will take more than 25 years for 450mm.
The industry is reluctant to spend that money for the 450mm and that was the reason Intel invested in to ASML.

If Intel will not get a big market share of the mobile/tablet market within the next 1-2 years you can say good by to 450mm wafers for this decade. It will take enormous volumes in order to recoup and be financial viable to operate 450mm wafers in the coming years. That is why Intel is trying to get a big Mobile market share now.

Source? If a chip becomes about twice as cheap to manufacture, how can it take 25 years to let it pay back the R&D costs?

 

[tarlinian]

Source? If a chip becomes about twice as cheap to manufacture, how can it take 25 years to let it pay back the R&D costs?

This is how long it took equipment manufacturers to recoup the investment they made into developing 300 mm equipment. (No one paid 2.25x the cost for a 300 mm tool, even though they needed entirely new designs in many cases and were certainly more expensive to make simply because they're bigger.)

 

[AtenRa]

Source? If a chip becomes about twice as cheap to manufacture, how can it take 25 years to let it pay back the R&D costs?

Because 450mm wafer will only bring close to 30% reduction in die cost, and that is questionable.
Also, wafer processing cost is only 15-20% of the final ASP price of a CPU. A 30% reduction will not get us the consumers any substantial price reductions. Not to mention that even if the manufacturer would get 30% reduction of the entire cost of the CPU it would not sell it lower, because they would need to recoup the massive R&D cost they have spend.

 

[AtenRa]

This is how long it took equipment manufacturers to recoup the investment they made into developing 300 mm equipment. (No one paid 2.25x the cost for a 300 mm tool, even though they needed entirely new designs in many cases and were certainly more expensive to make simply because they're bigger.)

thx i was going to say about that.

 

[mrmt]

Source? If a chip becomes about twice as cheap to manufacture, how can it take 25 years to let it pay back the R&D costs?

The issue isn't the R&D expected from the big foundries. The big three (Samsung, Intel and TSMC) can recoup that money. The problem is with the equipment manufacturers.

Given that only the big three foundries (and maybe Globalfoundries) should adopt 450mm in the initial ramp up, the equipment industry should brace for a step reduction in number of equipments sold when compared to 300mm, and that affects everything in the litography industry. Given the risks, the uncertainity in the number of orders and the necessity of keeping the 300mm R&D open for a long time after the 450mm debut, the equipment industry is really reticent of pursuing the venture.

 

[Homeles]

Because 450mm wafer will only bring close to 30% reduction in die cost, and that is questionable.
Also, wafer processing cost is only 15-20% of the final ASP price of a CPU. A 30% reduction will not get us the consumers any substantial price reductions. Not to mention that even if the manufacturer would get 30% reduction of the entire cost of the CPU it would not sell it lower, because they would need to recoup the massive R&D cost they have spend.

30% is pretty big, and I don't really see how its going to be questionable, except for the cost of the wafers themselves, since unprocessed 300mm wafers are practically free. 450mm wafers will someday as well, however, and the steps to process 450mm wafers should be the same.

30% is life or death in this industry.

 

[jdubs03]

its a real tough challenge, with the added difficulty in innovating to new transistor materials it gets even tougher. they've certainly put a lot of resources into 450 research, exemplified by CNSE; being in the hallway next to one of the clean-rooms was a unique experience.

 

[AtenRa]

http://seekingalpha.com/news/1626573-intel-others-slow-migration-to-450mm-chip-wafers?source=feed

Lithography equipment giant ASML has "paused" the development of hardware meant to work with next-gen 450mm wafers, which offer 125% more wafer space (and thus better economies of scale) than current-gen 300mm wafers. Likewise, Applied Materials (AMAT) CEO Gary Dickerson says the 450mm migration "has definitely been pushed out from a timing standpoint."

Due to ASML's move, Intel (INTC), which agreed in 2012 to pour $4.1B into the company to help finance investments in 450mm wafers and EUV lithography, has "adjusted" the pace of its payments to ASML.

Last year, Intel began constructing a $2B Oregon development fab meant to be its first 450mm facility. But it's reevaluating its timetable amid soft PC demand and concerns about its share of the bill. Spokesman Chuck Mulloy: "We still believe 450 is the right thing to do ... But we have been clear: we will not do it ourselves."

EUV, considered necessary to maintain Moore's Law long-term, has also seen delays. ASML CEO Peter Wennink recently predicted EUV will reach the stability levels required by chip manufacturers by the 2H16 or 2017.

And that confirms the 450mm delay.

 

[Nothingness]

That confirms the delay, but that doesn't confirm that Intel will delay to 2023 volume production making use of 450mm wafers