Also, SSMB already tested in German synchroton and generated EUV successfully.
Chinese semiconductor industry
- Thread starter Hendrik_2000
- Start date
- Status
Fine. But again, the image say hundred of millions to billions. You're clearly lying when you said just 100 million rmb. The huge price range is a clear sign that nobody knows how much it's gonna cost in the end. And again, this is just the light source, EUV machine is more than the light source, even if it's the most expensive and complex part. You're clearly not seeing the issue of cost and complexity when you're going from a machine the size of a huge truck, to something that can encircle an jumbo jet and has to be staffed with particle physicists.
Just the fact that you're gonna to need to build an entire new fab to house one is gonna bottleneck it.
The entire reason why they had to test a concept in germany was because it needed a special synchrotron. I'm assuming the ones in China weren't suitable.
It's going to need a specifically built synchrotron for the full powered version.
Anyway, a "test" doesn't mean shit, seeing as we have no details. I can test my aircraft carrier concept by launching paper aircraft from a tugboat. For all we know, they only generated peak power of a few watts and only for a few seconds. China outright built a EUV prototype before, getting mass produced version and the best version for low defects and good yield isn't easy.
I'm not saying that it won't work or a few won't be build. But again, economical matters for commercial products. DUVi can get to 5nm or 3nm with lots of multi-patterning. If SMIC don't care about cost, they could go ham and buy everything DUVi on the market, make hundreds of fabs and multi-pattern how many dozens of times to get 3nm to flood the market. Enjoy your $20,000 phone with 3nm node.
i typed t exact words,u go google translate it.
数亿到十亿
Alright, i tried google translate. The translation is wrong. You go ask people who know chinese. Maybe people here can help.
havok:This needs to be considered comprehensively. The current ASML equipment is not small, larger than a bus, and this does not include the volume of the driving laser. The use of SSMB can save the volume of the driving laser and the EUV collector, and the footprint in the FAB will be much smaller. The second is energy consumption. LPP uses a 200,000-watt carbon dioxide laser to bombard tin droplets to generate 300-watt extreme ultraviolet light, and then reaches the surface of the silicon wafer after 11 reflections through the collecting mirror, lighting mirror and projection objective lens. Lost 30% of the energy, the final energy reaching the silicon chip is only a pitiful 300 watts✕2%=6 watts, the energy utilization rate is extremely low, no wonder the wanwan is always short of power, using SSMB can reduce the number of reflections by about half. Another is the maintenance cost. LPP will produce a lot of metal debris when it bombards the tin target. Although the debris can be sucked away by some methods, a lot of debris will still fall on the collecting mirror and pollute the lens after a long time. Therefore, it is necessary to replace the collecting mirror every certain time, and the cost of replacing it is not cheap. With SSMB there is no need for collector mirrors and no debris issues.
这个需要综合考量，现在ASML的设备体积也不小，比一辆公交车还大，这还未算上驱动激光的体积。采用SSMB可以省去驱动激光和EUV收集器的体积，放在FAB内的占地就会小很多。其次是能耗，LPP用20万瓦的二氧化碳激光去轰击锡滴来产生300瓦的极紫外光，然后通过收集镜，照明反射镜和投影物镜经过11次反射后到达硅片表面，每次反射损耗30%能量，最终到达硅片的能量只有可怜的300瓦✕2%=6瓦，能量利用率极低，怪不得弯弯总是缺电，用SSMB可以减少一半左右的反射次数。再次是维护费用，LPP在轰击锡靶的时候会产生大量金属碎屑，虽然可以通过一些方法把碎屑吸走，但是时间久了仍会有不少碎屑掉落到收集镜上污染镜片，所以隔一定的时间是需要更换收集镜的，更换一次的费用可不菲。用SSMB不需要收集镜也没有碎屑问题。
这个需要综合考量，现在ASML的设备体积也不小，比一辆公交车还大，这还未算上驱动激光的体积。采用SSMB可以省去驱动激光和EUV收集器的体积，放在FAB内的占地就会小很多。其次是能耗，LPP用20万瓦的二氧化碳激光去轰击锡滴来产生300瓦的极紫外光，然后通过收集镜，照明反射镜和投影物镜经过11次反射后到达硅片表面，每次反射损耗30%能量，最终到达硅片的能量只有可怜的300瓦✕2%=6瓦，能量利用率极低，怪不得弯弯总是缺电，用SSMB可以减少一半左右的反射次数。再次是维护费用，LPP在轰击锡靶的时候会产生大量金属碎屑，虽然可以通过一些方法把碎屑吸走，但是时间久了仍会有不少碎屑掉落到收集镜上污染镜片，所以隔一定的时间是需要更换收集镜的，更换一次的费用可不菲。用SSMB不需要收集镜也没有碎屑问题。
The “fiber laser” they’re talking about is also a CO2 laser. It’s just a design that combines multiple laser outputs using fiber optics paths to couple their outputs together into one beam, allowing them to more easily scale drive laser power. If the power jumped from 400 W to 500 W it could either be from more optimal excitation of the tin drops or increase in drive laser power.
. You may be confused in that the EUV source is a tin droplet illuminated by a CO2 laser. . A 10 kW one powers the ASML tool. Its as hard as any other scaleup project. The key tech that is 100% new for EUV is the Sn droplet management system, which is very hard to deal with because it is hard to deal with liquids in UHV.
CVD is an entirely different instrument, it is completely unrelated to EUV except in that its related to the semiconductor industry is a whole. It is like saying that tires and basketballs are the same because they're made of rubber and round.
UHV chamber is a high cost contributor but one that any EUV system has to deal with regardless. What synchrotrons don't have to deal with is the tin droplet management and cleaning system or
Basically, we don't know what is actually better yet. SSMB is a latecomer but latecomer doesn't mean worse.
All they’re doing is adding points along a synchrotron beam line where they can compress electrons into bunches to more easily boost their energy and the pulse output of their emissions frequency in a smaller physical package. Just because it’s “newer” doesn’t make it more mechanically complex than an alternative technology. New needs to be validated. That does not mean it’s mechanically harder to do. You are conflating novelty for complexity.
SSMB *is* in general pretty easy compared to plasma produced light sources. But it’s also an idea that only came about in 2017. LPP and other plasma generated light sources were conceived of in the 90s and it basically took 20 years to get to a product that worked and 30 years to get to a product good enough for commercial use. If SSMB development finished in 2027 it would only have taken a third that time, and in prototyping they’re already hitting power that is greater than current and even future planned plasma
produced light sources. What matters most here for gauging development though is not “easy” in conceptual terms but the mechanical complexity of the physical device. There are *a lot* of mechanical steps and functions in plasma based light source with a lot of different moving parts which translates to a lot of interaction points that need to be working together to work well, and as a result there is a *lot* of potential fail modes. This is the sort of thing that drags out the development process in R&D, because you have to test all these different interaction points. A synchrotron is already inherently way simpler in mechanical terms and all SSMB does is add a few extra stages and parts to the storage ring which themselves have rather straightforward and simpler mechanical function.
The chart we just got suggests a device will be at most 1 billion RMB, or about 150 million USD. 数亿到10亿 translates to x00,000,000 to 1,000,000,000
But direct price to price comparisons between two devices isn’t what matters for gauging cost efficiency. Total power per dollar, and even more specifically, wafers per hour per dollar, is what determines true cost efficiency for a production machine. If they’re generating 1000 watts of light with one SSMB unit for the same price as an ASML LPP instrument generating 250 watts, it’s frankly speaking a no brainer to go with SSMB on the cost front. *IF* they are able to reduce half the light loss with simpler optics design (which would actually make a lot of sense since plasma generated light is not coherent or in phase which means you have to do a lot more focusing to control beam path and scatter profile, while a synchrotron based source would be coherent and in phase, and thus *much easier* to focus and direct beam path with less loss) this would not be a 4x improvement in capability but an 8x improvement.
Is this cost believable? Actually, it’s quite believable. Particle accelerator costs scale based on the size of the device. The bigger the device the greater the cost. This is because the mechanical principles are quite straightforward. For X amount of acceleration you can do with a particle, you will get X amount of energy. The more powerful your linear motors (magnets!) the more powerful your acceleration and the less distance you need to get to the same energy. If you want more energy before you harvest it for whatever you want to do you need to continuously accelerate it for longer distances or with more powerful linear motors, or often both. The costs scale linearly to how long you need your accelerator to be to get to your target beam energy, and non linearly to how powerful/expensive your linear motors are (drive magnets have gotten more powerful over time). The traditional particle accelerator approach is expensive and impractical for industrial production of chips *because* getting enough output out of a device that could be productive enough for a commercial industrial production process requires a very large and very powerful device. SSMB is promising *because* it substantially mitigates those size and power scaling factors. It does this by adding intermittent stages that feed the particle beam with RF/laser energy along the path of the accelerator to boost the energy of the beam, adding one additional method to feed energy to the particle beam other than the linear motors. Using a RF/laser feeding mechanism to boost particle beam power wasn’t very meaningful until they figured out the bunching technique because the particle beam doesn’t pick up that energy from the RF/Laser source very efficiently unless the beam interacts with the feed source in phase (an effect of quantum mechanics). The bunching method is what allows them to use this new power feeding method. Adding these booster stages into a smaller device is going to be *much* cheaper than building a much larger device with much greater construction and material costs, especially since the technology of the booster stages themselves aren’t mechanically complex or novel. They’re just run of the mill extra lasers/RF emitters.
That’s what’s likely behind those cost estimates. They’re believable cost reduction factors, and more to the point, there wouldn’t be funding to pursue this approach if there were no perceived benefits, and if I recall correctly cost efficiency has already been cited by the proponents of this technology as one them.
The LHC’s design for the record is not *that* complex. The engineering challenges for the LHC were from the massive physical scale of the device, not from functional complexity. To get to the energies the LHC targeted without building an even larger device which meant even more massive construction costs (literally the same kind of construction costs as building a subway line) they opted to use superconducting magnets to drive up the power of their linear motors. It *is* very expensive, but that’s because it’s a massive device spanning 27 kms. You literally need a transit system to get around its facilities. You are never going to need such a large amount of space, construction demand, or superconducting magnets and the municipality level power supply to drive them with an industrial use particle accelerator, even if it’s something new like SSMB. If you think the LHC is an appropriate reference point to make your point then you’re basing your judgements on hype, not a functional understanding of physics and engineering.
(Also, the entire if it were so simple why didn’t they already do it logic is kind of like saying wheels are simple why did it take till 4000 BC to invent them. Things can be mechanically simple and very obvious in retrospect but that doesn’t mean that they were obvious before you conceived of their idea. Half of science and research is really just about discovering the obvious).
Last edited:
- Status