Intel have also spent many years pushing DUV very hard, and successfully squeezing water from the multi-patterning stone. That institutional knowledge on how to design processes with the limitations of multi-patterning in mind means they have a good grasp of what is achievable in production and can confidence they can implement it faster than ASML could deliver new units.
Despite being a late adopter of mass-market chiplets products, with the successors of Raptor lake and Ponte Vecchio now Intel is strangely enough better prepared to the halving of the reticle field size consequence of high NA EUV than the companies still pushing for extremely large chips.
The halving of the reticle field size means that the maximum chip size will go from 26x33mm to 26x16.5mm and for end products it is the most consequential change that high NA EUV brings, yet AT in the rush of going over PR points not even mentions it in passing.
Given the increasing cost and size of all the parts of EUV (not just what's described here, but the other consequences like roughness resulting from stochastic distribution of photons, and the high energy of the electrons released from resist) at what point does it make more sense to switch to direct e-beams for manufacturing?
This is a serious question, and the reason I ask is that "we" (ie the Western semiconductor ecosystem) have evolved towards high-NA EUV one step at a time, every step being the obvious short-term maximizer. But have we dug ourselves into a hole? Suppose you were starting from scratch, knowing what's doable via EUV, knowing the costs, but with no existing infrastructure. Would it make sense, instead, to build a direct e-beam based infrastructure? Given that China (and perhaps Russia, who knows where they will be in five years?) are somewhat in the position I describe (little existing EUV infrastructure, probably little access to such infrastructure, but with the money and smarts to build an alternative e-beam infrastructure --- if that is not completely crazy in terms of the economics...) I think this is an interesting and important question.
My understanding is that ebeam has always been held back by being slower than photon lithography because the beam is not strong enough to flood fill an entire mask area, it needs to be scanned, which slows down throughput. But if the amount of money put into EUV and EUV sources were put into improving the ebeam intensity?... Coupled with if the ebeam equipment took only 1/10th the volume of the EUV equipment?...
> what point does it make more sense to switch to direct e-beams for manufacturing?
At the point that eBeam goes from 1 wafer per day to be able to compete with the 100s of wafers per hour of optical lithography.
Multiple beam eBeam litho is not going to get there any time soon, and anyway it is better suited for extremely repetitive patterns that are not what production of logic chips needs (and memory as a commoditized product is hardly ever on the frontlines of litho development).
(in theory) multi-beam lithography might get this into 10's per day for ~10000 beams to 12" wafers (each unit), on ~2012 technology level? (smallest technology target resolution was even ~1nm that time)
below 20-30nm it's getting really difficult considering noise because of high energy levels necessary (resulting into constant blur >10nm), secondary electron travel in resist, electron (back)scattering or ensuring data integrity for large areas with (additional) double patterning, what might be necessary for acceptable/improved results or low energy electrons (10-50eV) approach?
"In 2018, a thiol-ene resist was developed that features native reactive surface groups, which allows the direct functionalization of the resist surface with biomolecules."
As others have mentioned, so many changes come with High-NA EUV, it is still near-revolutionary. It is a different field size, half in fact. This does not match well with older tools in the fab, causing their productivity to drop unnecessarily. The stochastic aspects are also worsened due to targeting smaller spots. It's more likely the secondary electrons will blur out the High-NA resolution. The resists have to be thinner. And, possibly the worst culprit, the optics has obscuration, which was forbiddent in prevous scanner optics designs. High-NA was not supposed to happen when following legal optics designs.
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
7 Comments
Back to Article
edzieba - Thursday, May 26, 2022 - link
Intel have also spent many years pushing DUV very hard, and successfully squeezing water from the multi-patterning stone. That institutional knowledge on how to design processes with the limitations of multi-patterning in mind means they have a good grasp of what is achievable in production and can confidence they can implement it faster than ASML could deliver new units.Arsenica - Thursday, May 26, 2022 - link
Despite being a late adopter of mass-market chiplets products, with the successors of Raptor lake and Ponte Vecchio now Intel is strangely enough better prepared to the halving of the reticle field size consequence of high NA EUV than the companies still pushing for extremely large chips.The halving of the reticle field size means that the maximum chip size will go from 26x33mm to 26x16.5mm and for end products it is the most consequential change that high NA EUV brings, yet AT in the rush of going over PR points not even mentions it in passing.
name99 - Thursday, May 26, 2022 - link
Given the increasing cost and size of all the parts of EUV (not just what's described here, but the other consequences like roughness resulting from stochastic distribution of photons, and the high energy of the electrons released from resist) at what point does it make more sense to switch to direct e-beams for manufacturing?This is a serious question, and the reason I ask is that "we" (ie the Western semiconductor ecosystem) have evolved towards high-NA EUV one step at a time, every step being the obvious short-term maximizer. But have we dug ourselves into a hole? Suppose you were starting from scratch, knowing what's doable via EUV, knowing the costs, but with no existing infrastructure. Would it make sense, instead, to build a direct e-beam based infrastructure?
Given that China (and perhaps Russia, who knows where they will be in five years?) are somewhat in the position I describe (little existing EUV infrastructure, probably little access to such infrastructure, but with the money and smarts to build an alternative e-beam infrastructure --- if that is not completely crazy in terms of the economics...) I think this is an interesting and important question.
My understanding is that ebeam has always been held back by being slower than photon lithography because the beam is not strong enough to flood fill an entire mask area, it needs to be scanned, which slows down throughput. But if the amount of money put into EUV and EUV sources were put into improving the ebeam intensity?... Coupled with if the ebeam equipment took only 1/10th the volume of the EUV equipment?...
Arsenica - Thursday, May 26, 2022 - link
> what point does it make more sense to switch to direct e-beams for manufacturing?At the point that eBeam goes from 1 wafer per day to be able to compete with the 100s of wafers per hour of optical lithography.
Multiple beam eBeam litho is not going to get there any time soon, and anyway it is better suited for extremely repetitive patterns that are not what production of logic chips needs (and memory as a commoditized product is hardly ever on the frontlines of litho development).
back2future - Thursday, May 26, 2022 - link
(in theory) multi-beam lithography might get this into 10's per day for ~10000 beams to 12" wafers (each unit), on ~2012 technology level? (smallest technology target resolution was even ~1nm that time)below 20-30nm it's getting really difficult considering noise because of high energy levels necessary (resulting into constant blur >10nm), secondary electron travel in resist, electron (back)scattering or ensuring data integrity for large areas with (additional) double patterning, what might be necessary for acceptable/improved results or low energy electrons (10-50eV) approach?
"In 2018, a thiol-ene resist was developed that features native reactive surface groups, which allows the direct functionalization of the resist surface with biomolecules."
Anymoore - Sunday, May 29, 2022 - link
As others have mentioned, so many changes come with High-NA EUV, it is still near-revolutionary. It is a different field size, half in fact. This does not match well with older tools in the fab, causing their productivity to drop unnecessarily. The stochastic aspects are also worsened due to targeting smaller spots. It's more likely the secondary electrons will blur out the High-NA resolution. The resists have to be thinner. And, possibly the worst culprit, the optics has obscuration, which was forbiddent in prevous scanner optics designs. High-NA was not supposed to happen when following legal optics designs.kate2030 - Monday, May 30, 2022 - link
very nice..great.https://www.bestreplicawatches.is/