Can anyone explain why ASML has been allowed to remain pretty much a monopoly in the high end for these machines? I'd have expected major fab players to try for some vertical integration.
Take a care, scroll up to the site search box, enter "EUV" and start reading: Anandtech has done such an outstanding job at explaining the insane effort required to produce silicon structures at that small size. It takes slightly more than an erector kit, matches and Scotch tape to build an EUV scanner.
And I am glad no one has as of yet even attempted to replicate that effort, because that would make Europe a chip technology wasteland. Without ASML, no continent or nation can claim silicon supremacy and that is quite a good thing IMHO.
You seriously are saying that with ARM in your backyard???? (Arguably the most important & influential semiconductor design company in the entire world). Go home Abu, you're drunk. Europe's doing just fine.
I understand EUV is difficult. But it's possible, as ASML itself has demonstrated. Other companies like Intel surely have the R+D money to put into this area, potentially more than ASML. I'm wondering why they haven't? Even if we don't consider EUV, all the major fabs use ASML for DUV tech as well.
Also, I can't really say I agree with the second part of your comment. Competition is only ever a good thing. Another company somehow building tech superior to ASML's doesn't immediately make ASML worthless. And, as another commenter pointed out, ARM still exists.
Every company has it main focus, for pure-play companies they have to develop and maintain their manufacturing processes which by itself takes a lot of resources, developing and manufacturing Scanners will take unnecessary resources especially when there is working solution in the market, those companies will resort to make their own tools if there is no other available option in the market and no one intends and there is a clear need for the tool. And for price the tool would be much cheaper buying than developing and manufacturing and with such high demand it always pays itself.
"Other companies like Intel surely have the R+D money to put into this area, potentially more than ASML."
In 2012 Intel invested $4.1billion into ASML for a 15% stake. AFAIK they still retain that stake. Therefore effectively they did plough $4.1billion into EUV by purchasing that stake in ASML.
Why would they want to compete with a company they are heavily invested in, and receive a share of profits from?
Intel was one of the semiconductor companies that funded ASML. And you may have to look at the development cost ASML have spent, That's a LOT of money. The EUV system itself is very expensive too.
Intel, and others have investigated this. But Intel isn’t a machine design and manufacturing company. They don’t really sell machines either. The effort to produce EUV has taken a good 15 years. It’s use has been pushed out for close to a decade. Why should Intel get wrapped up in this, when it’s cheaper, and more efficient to buy it when it comes out?
Because it is business 101. Why didn't Apple buy TSMC? own the Fab, Why didn't Apple buy Qualcomm? Why Intel didn't buy AMD or Nvidia? Why should a company owns every part of their Supply Chain?
Have you consider the consequence of Intel falling to doing EUV themselves ( Cough ) while ASML did it and sell it to other Fab with higher priorities? And may even be not selling to Intel due to competitive reason.
And from 3rd part competition? There were, it was NEC, and others and they all gave up.
Semiconductor fabbing is much more than just the ASML machine. The EUV light source is from Cymer (bought by ASML, but in San Diego). The EUV optics are from Zeiss in Germany. The wafers used to (I'm guessing still are) from Japan, along with many of the other super-pure chemicals. I don't know about the current (less than ideal) resists, but the newest resists are mostly being investigated by the US as far as I can tell (but I don't follow chemistry closely). Many of the newest ideas for new transistors and how to fab them are being worked out at imec in the Netherlands.
And all these places are hiring people from all over the world.
Bottom line is: anyone stupid enough to imagine they can unilaterally take over modern fabbing (by grabbing ASML, invading Taiwan, or whatever) is going to be confronted very soon with just HOW international the whole enterprise is, and how no-one country or region has the know-how to do it all themselves. (Each could, perhaps, build up that know-how with enough time and money, but it would not be a cheap or fast process.)
There is so much cross-industry knowledge sharing as well. ASML scouts our engineers, we recruit people directly from the vendors, project managers move on to these R&D companies. This is constantly happening, there is a near infinite amount of knowledge on the subject that cannot be contained by one group.
Nikon and Canon failed in developing competing EUV scanners.. Regarding your second question, no large semiconductor manufacturer makes their own equipment (except for a few small scale research tools), they depend on 3rd party suppliers such as ASML, LAM research, Applied Materials, etc.
Fabricating microcircuits at sub-20nm levels is challenging; at sub-10nm it's an order of magnitude more difficult. More difficult = more expensive, and companies that design chips don't have enough money to spend on both design and fab, so they'd rather focus on their core competency and farm out the fabbing to an expert in the field who will do it better and cheaper.
After AMD and IBM sold their fabs off to GloFo, Intel and Samsung are the only notable companies with in-house foundries, and I mostly discount Samsung because the majority of their design and manufacture is dedicated to the relatively simple DRAM and NAND and a lot of their foundry tech is licensed from TSMC. Intel, the former king of vertical integration, has been struggling to get its in-house 10nm node working correctly for over *half a decade*, AKA an epoch in semiconductor timescales, which is a massive blow to the vertical integration argument. Particularly since the firms that walked away from vertical (AMD) or never had it (Apple et al) are doing so well.
As for ASML, it has been a market leader for over 3 decades and it simply invested more time and money into EUV tech than its competitors, which has allowed it to effectively become the only game in town for EUV. That doesn't mean there is no room for other companies that can developer faster, simpler, better, cheaper processes for fabbing on larger nodes, but what it does mean is that ASML is at the forefront of bleeding-edge semiconductor fabrication, and as such its name is naturally going to feature heavily on tech websites.
Oh, and Intel also owns a 15% stake in ASML, which should tell you just how highly esteemed the former is in tech circles. ASML is a de facto monopoly simply because it's better than everyone else; it's really difficult to compete with the best.
It's also worth noting that EUV turned out to be much harder than anyone had imagined. EUV was for a long time like practical fusion, "just 5 years away". Except it was 5 years away for 20 years.
You have no idea how difficult it was to make EUV work in high volume production, I'm surprised they actually did it and did not abandon it for some other tech. Let say the physics and complexity "just work against you" on this wavelength.
Not sure where you picked up that Samsung licenses from TSMC, but I sure never saw that news. AFAIK neither group shares IP with the other, and they are direct competitors in the foundry business.
shabby, they sold 52 machines, 45 DUV and 7 EUV, for $3.3B. My wild guess is $55 million for each DUV and $100 million for each EUV...with a little extra money coming in for services and parts.
To finally have EUV entering High Volume Manufacturing (HVM) is a key milestone, but the usage of EUV will become really significant from the node TSMC 5nm / Samsung 5nm with 10s of layers using EUV and that should start HVM in H1 2020.
I am really looking forward to H2 2020 to get my hands on the first devices manufactured with chip manufactured with extensive usage of EUV (Apple iPhone 12 with Apple A14 chip ?).
I consider the 5nm EUV node as really key milestone in humanity history as it will bring chip with 10s of billion of transistor with more than 100 million transistor per mm2.
According to Wikipedia, the first Pentium 4-branded processor to implement 64-bit was the Prescott (90 nm) (February 2004), and it seems it had ~125 millions transistors with a die size of 112mm2.
So in 2020 / 2021, thanks to lithography scaling, with 5nm EUV node, it will be possible to shrink an Intel Pentium 4 Prescott (90nm) processing power in less than 1mm2, while consuming much less energy !!! Amazing !!!
I am wondering if lithography scaling of silicon CMOS will be able to reach the next milestone of 1 billion transistor by 1mm2 : As of 2019, my guess would be that lithography scaling alone of silicon CMOS won’t allow to achieve it, and that there will be a need for new packaging (going 3D with 3D IC), and also new materials (Carbon nanotubes, spintronics,...) to achieve the equivalent of 1 billion transitors per mm2 (not necessarily silicon transistors)
We are reaching the limit of how small we can manufacture semiconductors chips and makes them work due various reasons. Density was and still is important aspect (see Moore law which had more impact in different than just mere observation) but as we reach the limit we need to find new methods to increase "compute power" either be 3d stacking to get more transistor in unit of volume or other ways like optical compute.
Thanks Anton, because of you Anandtech is finally covering more industry business news and supply chain partners that dont normally get much attention from consumers.
One of the reason I have been sceptical of Intel's EUV claims is based on the ASML numbers, their EUV shipment numbers and forecast does not bold well for Intel. All the current shipment are gong to TSMC and Samsung. Next year will be the same and leaving Intel with very little stocks, which means Intel's 2021 7nm wont have volume anywhere close to TSMC or Samsung.
There are 45 NXE:3400B total to be installed. Only 36 of these will be up on any day, across the world. That's not a lot, compared to the immersion baseline.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
46 Comments
Back to Article
Duncan Macdonald - Wednesday, October 16, 2019 - link
How does the 18 EUV tools sold so far this year plus 8 in Q4 equal 30 ?ToTTenTranz - Wednesday, October 16, 2019 - link
18 for Q3 2019 plus 8 in Q4. I'm guessing they sold another 4 in H1 2019.boeush - Wednesday, October 16, 2019 - link
"In total, the company sold 136 step-and-scan tools ****this year so far****, of which 18 were EUV tools."I guess reading comprehension is hard for some people...
Anton Shilov - Thursday, October 17, 2019 - link
30 machines was the plan the company outlined early this year.andychow - Wednesday, October 23, 2019 - link
30 shipped, not sold. They sold 18 this year, but of those shipped this year, some were sold in the previous year.skavi - Wednesday, October 16, 2019 - link
Can anyone explain why ASML has been allowed to remain pretty much a monopoly in the high end for these machines? I'd have expected major fab players to try for some vertical integration.abufrejoval - Wednesday, October 16, 2019 - link
Take a care, scroll up to the site search box, enter "EUV" and start reading: Anandtech has done such an outstanding job at explaining the insane effort required to produce silicon structures at that small size. It takes slightly more than an erector kit, matches and Scotch tape to build an EUV scanner.And I am glad no one has as of yet even attempted to replicate that effort, because that would make Europe a chip technology wasteland. Without ASML, no continent or nation can claim silicon supremacy and that is quite a good thing IMHO.
Cooe - Wednesday, October 16, 2019 - link
You seriously are saying that with ARM in your backyard???? (Arguably the most important & influential semiconductor design company in the entire world). Go home Abu, you're drunk. Europe's doing just fine.s.yu - Thursday, October 17, 2019 - link
ARM is now Japanese :)skavi - Wednesday, October 16, 2019 - link
I understand EUV is difficult. But it's possible, as ASML itself has demonstrated. Other companies like Intel surely have the R+D money to put into this area, potentially more than ASML. I'm wondering why they haven't? Even if we don't consider EUV, all the major fabs use ASML for DUV tech as well.Also, I can't really say I agree with the second part of your comment. Competition is only ever a good thing. Another company somehow building tech superior to ASML's doesn't immediately make ASML worthless. And, as another commenter pointed out, ARM still exists.
drexnx - Wednesday, October 16, 2019 - link
why sink billions into a parallel effort when ASML is amortizing that cost across every fab in the world for you?Eliadbu - Thursday, October 17, 2019 - link
Every company has it main focus, for pure-play companies they have to develop and maintain their manufacturing processes which by itself takes a lot of resources, developing and manufacturing Scanners will take unnecessary resources especially when there is working solution in the market, those companies will resort to make their own tools if there is no other available option in the market and no one intends and there is a clear need for the tool. And for price the tool would be much cheaper buying than developing and manufacturing and with such high demand it always pays itself.eldakka - Thursday, October 17, 2019 - link
"Other companies like Intel surely have the R+D money to put into this area, potentially more than ASML."In 2012 Intel invested $4.1billion into ASML for a 15% stake. AFAIK they still retain that stake. Therefore effectively they did plough $4.1billion into EUV by purchasing that stake in ASML.
Why would they want to compete with a company they are heavily invested in, and receive a share of profits from?
dropme - Thursday, October 17, 2019 - link
Intel was one of the semiconductor companies that funded ASML. And you may have to look at the development cost ASML have spent, That's a LOT of money. The EUV system itself is very expensive too.melgross - Thursday, October 17, 2019 - link
Intel, and others have investigated this. But Intel isn’t a machine design and manufacturing company. They don’t really sell machines either. The effort to produce EUV has taken a good 15 years. It’s use has been pushed out for close to a decade. Why should Intel get wrapped up in this, when it’s cheaper, and more efficient to buy it when it comes out?ksec - Thursday, October 17, 2019 - link
> I'm wondering why they haven't?Because it is business 101. Why didn't Apple buy TSMC? own the Fab, Why didn't Apple buy Qualcomm? Why Intel didn't buy AMD or Nvidia? Why should a company owns every part of their Supply Chain?
Have you consider the consequence of Intel falling to doing EUV themselves ( Cough ) while ASML did it and sell it to other Fab with higher priorities? And may even be not selling to Intel due to competitive reason.
And from 3rd part competition? There were, it was NEC, and others and they all gave up.
name99 - Thursday, October 17, 2019 - link
Semiconductor fabbing is much more than just the ASML machine.The EUV light source is from Cymer (bought by ASML, but in San Diego).
The EUV optics are from Zeiss in Germany.
The wafers used to (I'm guessing still are) from Japan, along with many of the other super-pure chemicals.
I don't know about the current (less than ideal) resists, but the newest resists are mostly being investigated by the US as far as I can tell (but I don't follow chemistry closely).
Many of the newest ideas for new transistors and how to fab them are being worked out at imec in the Netherlands.
And all these places are hiring people from all over the world.
Bottom line is: anyone stupid enough to imagine they can unilaterally take over modern fabbing (by grabbing ASML, invading Taiwan, or whatever) is going to be confronted very soon with just HOW international the whole enterprise is, and how no-one country or region has the know-how to do it all themselves.
(Each could, perhaps, build up that know-how with enough time and money, but it would not be a cheap or fast process.)
FullmetalTitan - Thursday, October 17, 2019 - link
There is so much cross-industry knowledge sharing as well. ASML scouts our engineers, we recruit people directly from the vendors, project managers move on to these R&D companies. This is constantly happening, there is a near infinite amount of knowledge on the subject that cannot be contained by one group.jvl - Friday, October 18, 2019 - link
Great insight, just minor correction: IMEC is in Belgium :-) Been to the tower, great placeArsenica - Thursday, October 17, 2019 - link
Nikon and Canon failed in developing competing EUV scanners..Regarding your second question, no large semiconductor manufacturer makes their own equipment (except for a few small scale research tools), they depend on 3rd party suppliers such as ASML, LAM research, Applied Materials, etc.
The_Assimilator - Thursday, October 17, 2019 - link
Fabricating microcircuits at sub-20nm levels is challenging; at sub-10nm it's an order of magnitude more difficult. More difficult = more expensive, and companies that design chips don't have enough money to spend on both design and fab, so they'd rather focus on their core competency and farm out the fabbing to an expert in the field who will do it better and cheaper.After AMD and IBM sold their fabs off to GloFo, Intel and Samsung are the only notable companies with in-house foundries, and I mostly discount Samsung because the majority of their design and manufacture is dedicated to the relatively simple DRAM and NAND and a lot of their foundry tech is licensed from TSMC. Intel, the former king of vertical integration, has been struggling to get its in-house 10nm node working correctly for over *half a decade*, AKA an epoch in semiconductor timescales, which is a massive blow to the vertical integration argument. Particularly since the firms that walked away from vertical (AMD) or never had it (Apple et al) are doing so well.
As for ASML, it has been a market leader for over 3 decades and it simply invested more time and money into EUV tech than its competitors, which has allowed it to effectively become the only game in town for EUV. That doesn't mean there is no room for other companies that can developer faster, simpler, better, cheaper processes for fabbing on larger nodes, but what it does mean is that ASML is at the forefront of bleeding-edge semiconductor fabrication, and as such its name is naturally going to feature heavily on tech websites.
Oh, and Intel also owns a 15% stake in ASML, which should tell you just how highly esteemed the former is in tech circles. ASML is a de facto monopoly simply because it's better than everyone else; it's really difficult to compete with the best.
s.yu - Thursday, October 17, 2019 - link
China now claims to have developed DUV that could be used for memory at ~10nm level...zmatt - Thursday, October 17, 2019 - link
It's also worth noting that EUV turned out to be much harder than anyone had imagined. EUV was for a long time like practical fusion, "just 5 years away". Except it was 5 years away for 20 years.The_Assimilator - Friday, October 18, 2019 - link
And unlike fusion, EUV actually finally happened ;)GreenReaper - Friday, October 18, 2019 - link
Well, profitable fusion has always been ten to fifteen years away.Eliadbu - Monday, October 21, 2019 - link
You have no idea how difficult it was to make EUV work in high volume production, I'm surprised they actually did it and did not abandon it for some other tech. Let say the physics and complexity "just work against you" on this wavelength.FullmetalTitan - Thursday, October 17, 2019 - link
Not sure where you picked up that Samsung licenses from TSMC, but I sure never saw that news. AFAIK neither group shares IP with the other, and they are direct competitors in the foundry business.The_Assimilator - Friday, October 18, 2019 - link
Mea culpa - Samsung licensed GloFo's tech, *not* TSMC's. Thanks for the correction!The_Assimilator - Friday, October 18, 2019 - link
Mea culpa - Samsung licensed GloFo's tech, *not* TSMC's. Thanks for the correction!Trikkiedikkie - Friday, October 18, 2019 - link
Many people did not get the bit of news that Intel sold most of the shares they had in ASML. They are now somewhere below 5 %Freeb!rd - Saturday, October 19, 2019 - link
Go empty your bank account and compete with them for business... I'm sure you won't have a problem.shabby - Wednesday, October 16, 2019 - link
So what's one these puppies go for? *kicks the side of it*Adramtech - Thursday, October 17, 2019 - link
shabby, they sold 52 machines, 45 DUV and 7 EUV, for $3.3B. My wild guess is $55 million for each DUV and $100 million for each EUV...with a little extra money coming in for services and parts.nevcairiel - Thursday, October 17, 2019 - link
Back in 2018, EUV scanners were around $145 million, not sure if prices may have shifted going into 2019.Death666Angel - Thursday, October 17, 2019 - link
Last I heard they still have no problem selling every EUV they manufacture, so I doubt the price has changed significantly. :)vladx - Thursday, October 17, 2019 - link
And to think Philips gave up on this company and its cutting edge productss.yu - Thursday, October 17, 2019 - link
Philips should first improve their toothbrushes so they don't break at a fall on the side.PeachNCream - Thursday, October 17, 2019 - link
How are people breaking toothbrushes? It's a plastic stick with some bristles on the end.melgross - Thursday, October 17, 2019 - link
Electric toothbrush. I have one. Pretty good, if you get a better model.PeachNCream - Thursday, October 17, 2019 - link
Ooh! That makes more sense. Thanks!vladx - Thursday, October 17, 2019 - link
Completely irrelevant, I was talking about Philips in the '90s when they gave up ASMLDiogene7 - Thursday, October 17, 2019 - link
To finally have EUV entering High Volume Manufacturing (HVM) is a key milestone, but the usage of EUV will become really significant from the node TSMC 5nm / Samsung 5nm with 10s of layers using EUV and that should start HVM in H1 2020.I am really looking forward to H2 2020 to get my hands on the first devices manufactured with chip manufactured with extensive usage of EUV (Apple iPhone 12 with Apple A14 chip ?).
I consider the 5nm EUV node as really key milestone in humanity history as it will bring chip with 10s of billion of transistor with more than 100 million transistor per mm2.
According to Wikipedia, the first Pentium 4-branded processor to implement 64-bit was the Prescott (90 nm) (February 2004), and it seems it had ~125 millions transistors with a die size of 112mm2.
So in 2020 / 2021, thanks to lithography scaling, with 5nm EUV node, it will be possible to shrink an Intel Pentium 4 Prescott (90nm) processing power in less than 1mm2, while consuming much less energy !!! Amazing !!!
I am wondering if lithography scaling of silicon CMOS will be able to reach the next milestone of 1 billion transistor by 1mm2 : As of 2019, my guess would be that lithography scaling alone of silicon CMOS won’t allow to achieve it, and that there will be a need for new packaging (going 3D with 3D IC), and also new materials (Carbon nanotubes, spintronics,...) to achieve the equivalent of 1 billion transitors per mm2 (not necessarily silicon transistors)
Eliadbu - Thursday, October 17, 2019 - link
We are reaching the limit of how small we can manufacture semiconductors chips and makes them work due various reasons. Density was and still is important aspect (see Moore law which had more impact in different than just mere observation) but as we reach the limit we need to find new methods to increase "compute power" either be 3d stacking to get more transistor in unit of volume or other ways like optical compute.abufrejoval - Friday, October 18, 2019 - link
To label a process node "a keystone of humanity" could be considered a slight exaggeration.ksec - Thursday, October 17, 2019 - link
Thanks Anton, because of you Anandtech is finally covering more industry business news and supply chain partners that dont normally get much attention from consumers.One of the reason I have been sceptical of Intel's EUV claims is based on the ASML numbers, their EUV shipment numbers and forecast does not bold well for Intel. All the current shipment are gong to TSMC and Samsung. Next year will be the same and leaving Intel with very little stocks, which means Intel's 2021 7nm wont have volume anywhere close to TSMC or Samsung.
Anymoore - Monday, October 21, 2019 - link
There are 45 NXE:3400B total to be installed. Only 36 of these will be up on any day, across the world. That's not a lot, compared to the immersion baseline.