Yeah, I understand a lot about chip design and architecture, but I get completely lost in the fabrication and materials science side of things. What I do understand is that the engineering required to do EUV is utterly amazing.
I work in the industry and this was a very nice overview. Broad enough that most technical people should be able to follow, but still exact enough to be relevant.
That's not how language works. Something isn't "good or it isn't". It can be good enough.
However, in this case, the proper word would be "sufficiently" as in "sufficiently cost effective". Using enough is incorrect use of English, it's a minor issue, but obviously internet trolls care deeply about it.
III-V semiconductors are the material of the future - and will always stay like that. By now this saying is probably a few decades old, but III-V channels grown on Si actually offer intriguing potential.
You forgot Gallium Arsenide (GaAs). GaAs is the technology of the future, in the future.
Note: for those that don't know, GaAs is a different substrate, used instead of Silicon (Si), used in most integrated circuits today. GaAs substrate allows for *much* higher frequencies, but at the cost of power. This is currently used for power transmitter in your cell phone (the amplifier just before the antenna, that need to amplify and send the multi giga hertz signal). It has been named as a solution for the frequency wall of silicon for the past 20 years...
The complexity, novelty, and wear characterustucs of the tin droplet EUV source make me wonder why they aren't just going with a synchrotron instead - a well-studied, mature tech that can produce very well-behaved, high-power, and highly wavelength/polarization-tunable beams (and doesn't require a hyper-precise collector mirror that must be frequently replaced)...
It's because of the wavelength, synchrotrons produce mostly x-Ray's by bending the path of electrons. Those X-rays are useful for imaging because the pass easily through matter without affecting it too much. In lithography you need Ray's that affect your material to produce a big enough change to allow selective etching.
Not really. I was speaking more broadly of synchrotron radiation - not just ring accelerators in particular (which incidentally emit broad spectrum, not just x-rays). For instance, a linac feeding into a wiggler array can efficiently produce highly tuned, narrow-band emission centered around virtually any wavelength you want (you just tweak various design parameters accordingly.)
AFAIK synchrotrons can't produce light longer than 1nm, which bounds them as pure X-ray sources. To reliably manipulate anything, 6-7 nm would be required, enabling ~2-2.5 nm element size. That's the hard limit before switching to direct nanoscale manipulation. X-ray has to be skipped entirely. This article show that we have 30-40 years worth of photolitography development ahead of us. After this, either we must develop high volume electron manipulation processes. Alternative is to either switch to another material that will allow to be easily patterned, or introduce room-temperature superconductors to chips, which will allow to scale back a little on process, by limiting current leaks.
"and doesn't require a hyper-precise collector mirror that must be frequently replaced"
You cannot do away with these mirrors. They replace lenses and you would have no way of focusing the beam without them.
And unless they invent mirrors with better reflectivity at these wavelengths ablation will always take place and mirrors will have to be replaced after certain period of usage.
Focus doesn't matter if you block rays that are not going the right direction. Not matter which ways you do it, EUV lithography will waste more power (produce more heat) and be very costly to use. There is 1 layer of transistors (that may require a few masks and developing/processing) on top of that are metal layers. First the small and short close to the transistors then the fat wide ones on top for power buses and longer wires. The fat/wide wires don't need EUV. Bottom few layers will use EUV.
I think that was intentional: EUV will be used on 7 nm only if it's ready by then (otherwise, they'll make do with ridiculously convoluted multi-patterning, as a stopgap.)
Awesome article. Very interesting to see the curtain drawn back a bit on some of our society's most advanced endeavors. Rockets have a lot of glamour, but in the last 50 years our progress in advanced semiconductors is like miles compared with a few steps there. (More or less in line with the financial investments.)
also this stuff makes use of all the advanced physics we have and justifies all the fundamental research money spent decades ago.
Besides, the importance of technology can be measured by its economic impact. The economic impact of mass IC manufacturing is huge and enables everything else, including Musk's rockets.
I knew this was Anton's article as soon as I saw the title :) X-Bit labs did have some interesting articles back in the day, I appreciate pieces on contemporary theory and technical implementation aspects of computing, I can only take so many product presentations/reviews ...
I am in D&E for EUV source in ASML. It is always interesting to see the outside perspective of our work. It is very easy to loose scope as to how impressive and important these machines are for the lithography industry. The shear amount of work and people involved in developing these machines is impressive. The 3400 tool that you see for 2017 is in the proto-type state below me now, and we really do need to start selling these machines.
EUV is very delayed. Intel planned use this for 45 nm... Future is still uncertain beside progress being made. It may be that they faster develop 8x multipatterning with 500 wafers/hours scanners using 193nm light, to produce 5nm chips...
Multi patterning has its own challenges: - your chip layout is not anisotropic anymore, you need to make long lines in one direction - you now depend heavily of the etching stage and its precision (improved etching and cleaning is what allowed multi patterning and 3D flash)
One thing I've always kind of wondered is, why no intermediate wave length light source for lithography? 193nm to 13.5nm is a pretty huge jump. Is it is a question of returns? Or that they want something that they can implement and keep using until Silicon just can't physically get smaller because of quantum effects?
I would think even a change to 30-50nm could push them at least a couple of process size smaller before that couldn't cope. I'd also think something a little longer would have fewer problems with absorption and some of the other constraints that 13.5nm seems to be impossing
Around 2000 they wanted to transition to 150-something nm excimer lasers after 193 nm. I don't know why this didn't happen, but I've never seen any other "conventional" light source below 190 nm. It could be the lack of lenses.. not sure if it already applies at 150 nm, but it certainly does at 50 nm. So such a stop-gap solution wouldn't help much and they decided to go directly for full throttle.
157nm litho died because it suffered many of the same problems as EUV, and didn't offer much improvement over 193nm. Basically, 157nm light is absorbed by air and glass like EUV. F2 excimer laser couldn't hit wafer plane dose requirements without big changes to the optics, reticle, environment... if you have to change all that stuff, might as well do it for EUV. The big problem with EUV still remains... ridiculous cost of ownership. 1000 wafers/day doesn't cut it when the tool goes down hard after that and costs a fortune to recover...I just saying..
The immersion Lithography Aspect is nothing new. Intel was SVG Lithography's (in CT, USA) main customer up until SVG was acquired by ASML in about 2001-2002 time frame. Cymer at that time was a vendor for SVG. INTEL has been using Immersion techniques even back then for the 193nm/157nm (wavelength not feature size) systems. They were probably using it before that. This is probably allow them to get past what was originally going to be a 22nm (feature size) limit due to the diffraction limitation of then current I-line systems. That being said the exposure and processing have become insanely more advanced since I was in the industry.
Note that Photolithographic type methods haven't been used in almost 15 to 20 years. The defining line used to be that the Photolithographic to Micro barrier was crossed at the 1 micron feature range.
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.
38 Comments
Back to Article
Amandtec - Thursday, March 10, 2016 - link
Didn't understand a word. Fascinating none the less.r3loaded - Friday, March 11, 2016 - link
Yeah, I understand a lot about chip design and architecture, but I get completely lost in the fabrication and materials science side of things. What I do understand is that the engineering required to do EUV is utterly amazing.frenchy_2001 - Friday, March 11, 2016 - link
I work in the industry and this was a very nice overview.Broad enough that most technical people should be able to follow, but still exact enough to be relevant.
Congratz to Anton for that!
JeffFlanagan - Thursday, March 10, 2016 - link
>indicated that it will clearly utilize it once it is suitable for high-volume>production of semiconductors and is enough cost effective.
You have a stray "enough" in the first paragraph.
Wardrop - Thursday, March 10, 2016 - link
The enough was meant to be at the end of the sentence I assume.melgross - Thursday, March 10, 2016 - link
It isn't needed. It's either cost effective, or it isn't.Mondozai - Thursday, March 10, 2016 - link
That's not how language works. Something isn't "good or it isn't". It can be good enough.However, in this case, the proper word would be "sufficiently" as in "sufficiently cost effective". Using enough is incorrect use of English, it's a minor issue, but obviously internet trolls care deeply about it.
Ushio01 - Thursday, March 10, 2016 - link
EUV Lithography, Graphene, Nanotubes the 3 buzz words about the future of semiconductors since I got into PC gaming back in 2006.MrSpadge - Friday, March 11, 2016 - link
III-V semiconductors are the material of the future - and will always stay like that. By now this saying is probably a few decades old, but III-V channels grown on Si actually offer intriguing potential.frenchy_2001 - Friday, March 11, 2016 - link
You forgot Gallium Arsenide (GaAs).GaAs is the technology of the future, in the future.
Note: for those that don't know, GaAs is a different substrate, used instead of Silicon (Si), used in most integrated circuits today. GaAs substrate allows for *much* higher frequencies, but at the cost of power. This is currently used for power transmitter in your cell phone (the amplifier just before the antenna, that need to amplify and send the multi giga hertz signal).
It has been named as a solution for the frequency wall of silicon for the past 20 years...
alyarb - Thursday, March 10, 2016 - link
EUV still not ready, and you HEARD IT FIRST ON ANANDTECH, hahMondozai - Thursday, March 10, 2016 - link
Great, intelligent retort to this article. Just go away if you're not going to engage in substance.prisonerX - Friday, March 11, 2016 - link
Welcome to the internet, I hope you enjoy your stay,lol
boeush - Thursday, March 10, 2016 - link
The complexity, novelty, and wear characterustucs of the tin droplet EUV source make me wonder why they aren't just going with a synchrotron instead - a well-studied, mature tech that can produce very well-behaved, high-power, and highly wavelength/polarization-tunable beams (and doesn't require a hyper-precise collector mirror that must be frequently replaced)...jann5s - Thursday, March 10, 2016 - link
It's because of the wavelength, synchrotrons produce mostly x-Ray's by bending the path of electrons. Those X-rays are useful for imaging because the pass easily through matter without affecting it too much. In lithography you need Ray's that affect your material to produce a big enough change to allow selective etching.boeush - Thursday, March 10, 2016 - link
Not really. I was speaking more broadly of synchrotron radiation - not just ring accelerators in particular (which incidentally emit broad spectrum, not just x-rays). For instance, a linac feeding into a wiggler array can efficiently produce highly tuned, narrow-band emission centered around virtually any wavelength you want (you just tweak various design parameters accordingly.)Vatharian - Friday, March 11, 2016 - link
AFAIK synchrotrons can't produce light longer than 1nm, which bounds them as pure X-ray sources. To reliably manipulate anything, 6-7 nm would be required, enabling ~2-2.5 nm element size. That's the hard limit before switching to direct nanoscale manipulation. X-ray has to be skipped entirely. This article show that we have 30-40 years worth of photolitography development ahead of us. After this, either we must develop high volume electron manipulation processes. Alternative is to either switch to another material that will allow to be easily patterned, or introduce room-temperature superconductors to chips, which will allow to scale back a little on process, by limiting current leaks.Arnulf - Friday, March 11, 2016 - link
"and doesn't require a hyper-precise collector mirror that must be frequently replaced"You cannot do away with these mirrors. They replace lenses and you would have no way of focusing the beam without them.
And unless they invent mirrors with better reflectivity at these wavelengths ablation will always take place and mirrors will have to be replaced after certain period of usage.
tygrus - Sunday, March 13, 2016 - link
Focus doesn't matter if you block rays that are not going the right direction. Not matter which ways you do it, EUV lithography will waste more power (produce more heat) and be very costly to use. There is 1 layer of transistors (that may require a few masks and developing/processing) on top of that are metal layers. First the small and short close to the transistors then the fat wide ones on top for power buses and longer wires. The fat/wide wires don't need EUV. Bottom few layers will use EUV.nandnandnand - Thursday, March 10, 2016 - link
"will help to produce microprocessors and other chips using 5 nm and, perhaps, 7nm, technologies"Switch 5 nm and 7 nm.
boeush - Thursday, March 10, 2016 - link
I think that was intentional: EUV will be used on 7 nm only if it's ready by then (otherwise, they'll make do with ridiculously convoluted multi-patterning, as a stopgap.)By contrast, on 5 nm EUV is basically mandatory.
ABR - Friday, March 11, 2016 - link
Awesome article. Very interesting to see the curtain drawn back a bit on some of our society's most advanced endeavors. Rockets have a lot of glamour, but in the last 50 years our progress in advanced semiconductors is like miles compared with a few steps there. (More or less in line with the financial investments.)Murloc - Saturday, March 12, 2016 - link
also this stuff makes use of all the advanced physics we have and justifies all the fundamental research money spent decades ago.Besides, the importance of technology can be measured by its economic impact.
The economic impact of mass IC manufacturing is huge and enables everything else, including Musk's rockets.
Arnulf - Friday, March 11, 2016 - link
I knew this was Anton's article as soon as I saw the title :) X-Bit labs did have some interesting articles back in the day, I appreciate pieces on contemporary theory and technical implementation aspects of computing, I can only take so many product presentations/reviews ...not_just_a_username - Friday, March 11, 2016 - link
I am in D&E for EUV source in ASML. It is always interesting to see the outside perspective of our work. It is very easy to loose scope as to how impressive and important these machines are for the lithography industry. The shear amount of work and people involved in developing these machines is impressive. The 3400 tool that you see for 2017 is in the proto-type state below me now, and we really do need to start selling these machines.Murloc - Saturday, March 12, 2016 - link
*lose*sheer
*prototype
I hope you're successful, if cutting edge stuff doesn't make money soon enough, science suffers.
LeZhou - Friday, February 2, 2018 - link
Can you give me a email, I am a college student, and I have a research about EUV and hope you can give me some information about your workTristanSDX - Friday, March 11, 2016 - link
EUV is very delayed. Intel planned use this for 45 nm...Future is still uncertain beside progress being made. It may be that they faster develop 8x multipatterning with 500 wafers/hours scanners using 193nm light, to produce 5nm chips...
frenchy_2001 - Friday, March 11, 2016 - link
Multi patterning has its own challenges:- your chip layout is not anisotropic anymore, you need to make long lines in one direction
- you now depend heavily of the etching stage and its precision (improved etching and cleaning is what allowed multi patterning and 3D flash)
And those are just 2 of the top challenges...
azazel1024 - Friday, March 11, 2016 - link
One thing I've always kind of wondered is, why no intermediate wave length light source for lithography? 193nm to 13.5nm is a pretty huge jump. Is it is a question of returns? Or that they want something that they can implement and keep using until Silicon just can't physically get smaller because of quantum effects?I would think even a change to 30-50nm could push them at least a couple of process size smaller before that couldn't cope. I'd also think something a little longer would have fewer problems with absorption and some of the other constraints that 13.5nm seems to be impossing
MrSpadge - Friday, March 11, 2016 - link
Around 2000 they wanted to transition to 150-something nm excimer lasers after 193 nm. I don't know why this didn't happen, but I've never seen any other "conventional" light source below 190 nm. It could be the lack of lenses.. not sure if it already applies at 150 nm, but it certainly does at 50 nm. So such a stop-gap solution wouldn't help much and they decided to go directly for full throttle.Metro Man - Friday, March 11, 2016 - link
157nm litho died because it suffered many of the same problems as EUV, and didn't offer much improvement over 193nm. Basically, 157nm light is absorbed by air and glass like EUV. F2 excimer laser couldn't hit wafer plane dose requirements without big changes to the optics, reticle, environment... if you have to change all that stuff, might as well do it for EUV.The big problem with EUV still remains... ridiculous cost of ownership. 1000 wafers/day doesn't cut it when the tool goes down hard after that and costs a fortune to recover...I just saying..
Torashin - Sunday, March 13, 2016 - link
There's got to be a better solution than this!Beoir - Thursday, March 17, 2016 - link
The immersion Lithography Aspect is nothing new. Intel was SVG Lithography's (in CT, USA) main customer up until SVG was acquired by ASML in about 2001-2002 time frame. Cymer at that time was a vendor for SVG. INTEL has been using Immersion techniques even back then for the 193nm/157nm (wavelength not feature size) systems. They were probably using it before that. This is probably allow them to get past what was originally going to be a 22nm (feature size) limit due to the diffraction limitation of then current I-line systems.That being said the exposure and processing have become insanely more advanced since I was in the industry.
Beoir - Thursday, March 17, 2016 - link
Note that Photolithographic type methods haven't been used in almost 15 to 20 years. The defining line used to be that the Photolithographic to Micro barrier was crossed at the 1 micron feature range.Beoir - Thursday, March 17, 2016 - link
I remember back in 1999 when Tropel was kicking around the idea of an X-ray Lithography system using magnets. I wonder where they are with that now?Anymoore - Tuesday, September 13, 2016 - link
46 nm pitch can be done in one ArF exposure with a single spacer but you need another exposure to trim. This should be how their 10nm is done.Anymoore - Tuesday, May 28, 2019 - link
Wavelength is not 13.5 nm but a range covering 13.3-13.7 nm. Each wavelength has its own preferred angle. So this could be the showstopper.