I'm sure it's way more complicated than I could possibly imagine, but when I see these things I always wonder: If you have the knowledge, capability, and capitalization to make something like this, why bother selling it to Intel or GloFo or TSMC?
Like how cryptomining ASICs in the big times could be hard to actually get new because the companies that knew how to make them were just making them and then mining with them themselves.
These machines may be extremely expensive at an Airbus A320 per unit but the cost of outfitting a fab facility is even higher, at several billion dollars each. It's the same reason why Airbus makes planes but doesn't run an airline.
>y imagine, but when I see these things I always wonder: If you have the knowledge, capability, and capitalization to make something like this, why bother selling it to Intel or GloFo or TSMC?
The scanner alone isn't enough to make chips, so while they have the know how to relay near-x-ray photons, they're not going to be making their own desktop processors.
There's a lot more to running a fab than just these machines, and with next gen fabs running $5-10bn each building them is expensive even compared to the machines used to run them.
ASML won't be making their own chips for four reasons: 1. They are the dominant, but not only player in the scanner market. If they would start competing with their current customers, that revenue stream would be cut down a lot or off entirely fast. 2. They would be subject to the motherload of IP lawsuits and injunctions to stop; as part of their business, they are privy to most manufacturers processes, all covered by NDAs. They'd have to prove that they don'y violate anybody else's IP, public (patented) or just privately shared. As others have pointed out, the lithography process alone has multiple, complex steps besides the scanning. ->3. Same reason why makers of robotic (remote control) surgery equipment (Da Vinci and others) don't open their own hospitals - knowing how to build great equipment doesn't qualify you to run a hospital. and 4. Chip making is insanely capital-intensive and thus financially risky. If you invest billions of $$$ in a fab and are then prevented from getting revenue by lawsuits and injunctions, you go bankrupt quickly. So, even if they wanted to, I don't believe they could get the financing to do this.
Even if they have the knowledge to make every Machine in the fab (which they do not ) there is way more than just having the machines and running them. Those are the leading manufacturing fabs not some clothes factory. There is so much to how design and develop the technology than just running the machine especially in those small nodes. In the past there were several manufacturers that produced chip in the leading nodes now we are left with just 3. Mostly because it is very complicated and very expensive to compete in these new nodes.
30 is Inline with expectation. After years of delay we are finally reaching EUV in production. Although 30 isn't a lot, Samsung and TSBC would have taken it all. Which is why I do not believe Intel will have 7nm EUV by 2020.
2020? No, no way, that's way too early. Assuming they are going to release three generations of "fixed 10nm" (since the original 10nm node they employed for Cannon Lake is broken), one each for Sunny, Willow and Golden Cove, and also assuming that these three CPU generations are released in 2019, 2020 and 2021 respectively, with *no* delays for either, then 7nm should debut in 2022 with Ocean Cove.
That's the "brand new" CPU design Jim Keller was hired to design. Now, if Golden Cove is a new architecture rather than a (twice) optimized Sunny Cove then they might release an optimized version of Golden Cove at 7nm in 2022 and then release Ocean Cove in 2023. It's also possible that Golden Cove is (or will be) the new name of Ocean Cove.
On the other hand, they might even release Golden Cove at 7nm, in order to narrow their widening "Moore's Lag gap", though I strongly doubt that. My prediction for release of Intel's 7nm node (I mean in high volume, not a "pull the wool over the eyes of investors release", like they did with Core i3-8121U) would be 2022 at the earliest and 2023 at the latest - assuming smooth 7nm sails.
The 2020 was reference to how Intel keep stating 7nm is on schedule, and said they will jump on 7nm when it is ready. The previous Anandtech Interview already states their CPU design will no longer be node dependent.
And just to point out TSMC's 5nm is scheduled for 2020 iPhone, and early 2021 for High Power node.
Welcome news. There's been a lot of work put into this. It's not just a matter of working out how to focus EUV but also how to to make it work in a production environment. Processing 30 or 40 wafers an hour isn't good enough for the fab guys. Wonder what ASML are going to do next?
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.
Facebook
Twitter
RSS
17 Comments
Back to Article
Sahrin - Monday, January 28, 2019 - link
How much does one of these scanners cost, if they're only selling 30 tools in a year?diehardmacfan - Monday, January 28, 2019 - link
I believe they are about $119 Million, but that might be a previous generation.FullmetalTitan - Wednesday, January 30, 2019 - link
High end DUV (immersion) scanners run in the $120-150M range. EUV scanners are more like $200-300MGreenReaper - Thursday, January 31, 2019 - link
Damn . . . and I thought CanoScan were pricy! ;-pCullinaire - Monday, January 28, 2019 - link
They cost a fortune, and that's not all. Their upkeep also costs a fortune, so there's a recurring revenue stream as well.Ashinjuka - Monday, January 28, 2019 - link
I'm sure it's way more complicated than I could possibly imagine, but when I see these things I always wonder: If you have the knowledge, capability, and capitalization to make something like this, why bother selling it to Intel or GloFo or TSMC?Like how cryptomining ASICs in the big times could be hard to actually get new because the companies that knew how to make them were just making them and then mining with them themselves.
serendip - Monday, January 28, 2019 - link
These machines may be extremely expensive at an Airbus A320 per unit but the cost of outfitting a fab facility is even higher, at several billion dollars each. It's the same reason why Airbus makes planes but doesn't run an airline.saratoga4 - Tuesday, January 29, 2019 - link
>y imagine, but when I see these things I always wonder: If you have the knowledge, capability, and capitalization to make something like this, why bother selling it to Intel or GloFo or TSMC?The scanner alone isn't enough to make chips, so while they have the know how to relay near-x-ray photons, they're not going to be making their own desktop processors.
DanNeely - Tuesday, January 29, 2019 - link
There's a lot more to running a fab than just these machines, and with next gen fabs running $5-10bn each building them is expensive even compared to the machines used to run them.eastcoast_pete - Tuesday, January 29, 2019 - link
ASML won't be making their own chips for four reasons: 1. They are the dominant, but not only player in the scanner market. If they would start competing with their current customers, that revenue stream would be cut down a lot or off entirely fast. 2. They would be subject to the motherload of IP lawsuits and injunctions to stop; as part of their business, they are privy to most manufacturers processes, all covered by NDAs. They'd have to prove that they don'y violate anybody else's IP, public (patented) or just privately shared. As others have pointed out, the lithography process alone has multiple, complex steps besides the scanning. ->3. Same reason why makers of robotic (remote control) surgery equipment (Da Vinci and others) don't open their own hospitals - knowing how to build great equipment doesn't qualify you to run a hospital. and 4. Chip making is insanely capital-intensive and thus financially risky. If you invest billions of $$$ in a fab and are then prevented from getting revenue by lawsuits and injunctions, you go bankrupt quickly. So, even if they wanted to, I don't believe they could get the financing to do this.surt - Tuesday, January 29, 2019 - link
Because Intel has the knowledge, capability, and capitalization to use such a tool, and they don't.Eliadbu - Tuesday, January 29, 2019 - link
Even if they have the knowledge to make every Machine in the fab (which they do not ) there is way more than just having the machines and running them. Those are the leading manufacturing fabs not some clothes factory. There is so much to how design and develop the technology than just running the machine especially in those small nodes. In the past there were several manufacturers that produced chip in the leading nodes now we are left with just 3. Mostly because it is very complicated and very expensive to compete in these new nodes.iwod - Tuesday, January 29, 2019 - link
30 is Inline with expectation. After years of delay we are finally reaching EUV in production. Although 30 isn't a lot, Samsung and TSBC would have taken it all. Which is why I do not believe Intel will have 7nm EUV by 2020.Santoval - Tuesday, January 29, 2019 - link
2020? No, no way, that's way too early. Assuming they are going to release three generations of "fixed 10nm" (since the original 10nm node they employed for Cannon Lake is broken), one each for Sunny, Willow and Golden Cove, and also assuming that these three CPU generations are released in 2019, 2020 and 2021 respectively, with *no* delays for either, then 7nm should debut in 2022 with Ocean Cove.That's the "brand new" CPU design Jim Keller was hired to design. Now, if Golden Cove is a new architecture rather than a (twice) optimized Sunny Cove then they might release an optimized version of Golden Cove at 7nm in 2022 and then release Ocean Cove in 2023. It's also possible that Golden Cove is (or will be) the new name of Ocean Cove.
On the other hand, they might even release Golden Cove at 7nm, in order to narrow their widening "Moore's Lag gap", though I strongly doubt that. My prediction for release of Intel's 7nm node (I mean in high volume, not a "pull the wool over the eyes of investors release", like they did with Core i3-8121U) would be 2022 at the earliest and 2023 at the latest - assuming smooth 7nm sails.
iwod - Tuesday, January 29, 2019 - link
The 2020 was reference to how Intel keep stating 7nm is on schedule, and said they will jump on 7nm when it is ready. The previous Anandtech Interview already states their CPU design will no longer be node dependent.And just to point out TSMC's 5nm is scheduled for 2020 iPhone, and early 2021 for High Power node.
gizmo23 - Tuesday, January 29, 2019 - link
Welcome news. There's been a lot of work put into this. It's not just a matter of working out how to focus EUV but also how to to make it work in a production environment. Processing 30 or 40 wafers an hour isn't good enough for the fab guys. Wonder what ASML are going to do next?PeachNCream - Tuesday, January 29, 2019 - link
The company plans to make and sell more scanners next.