EUV (Extreme Ultraviolet Lithography) has been a very long time coming to market. Initially predicted to be ready by 2004 (if not sooner), we’ve been waiting 13 years for the technology to arrive. Manufacturers like Intel, TSMC, GlobalFoundries, and Samsung have continued to advance 193nm lithography in the absence of a functional EUV alternative. But the 193nm well is running dry, despite the introduction of immersion lithography at the 45nm node (IBM, AMD) and the 32nm node (Intel). Now, ASML is claiming to have solved one of the major problems keeping EUV lithography from release, though it’s doing so in extremely carefully worded language.
One of the major problems with EUV has been hitting the 250W exposure power required for mainstream manufacturing. There are a number of reasons why it’s taken so long for EUV to move towards the mainstream; Lithography expert Chris Mack has a timeline of broken promises (presentation) on EUV introduction from 2002 to 2015, and you might notice it’s still not here yet.
Here’s ASML’s technical claim:
In EUV lithography, we have integrated an upgraded EUV source into a TWINSCAN NXE:3400B lithography system in our Veldhoven facility and achieved the throughput specification of 125 wafers per hour on this system. Now, with all key performance specifications demonstrated, we focus on achieving the availability that is required for high-volume manufacturing as well as further improving productivity.
That’s a significant step towards achieving the 125-wafers-per-hour mark that manufacturers require for integration into high volume manufacturing. But it’s not quite as solid a claim as it looks like at first glance. ASML is claiming to have achieved throughput specification, but it’s not actually manufacturing anything.
If this seems a bit confusing, imagine a conventional assembly line. You need the line to move quickly enough to build enough widgets per hour to make the assembly line model work in the first place. If that doesn’t happen, you’ll never have a profitable company. What ASML is claiming to have done is hit the assembly line speed it needs to make high volume manufacturing possible. What it hasn’t yet demonstrated is that it can hit all of the necessary specifications simultaneously.
In our hypothetical example, you can simulate how many widgets per hour you can manufacture by starting and stopping the assembly line at the appropriate workstations. But you’ll only get an accurate idea of your throughput if your estimated manufacturing time at every workstation is accurate while still meeting all product specifications. ASML has yet to demonstrate this level of proof.
That said, a 250W power source is a huge advantage over where we were in 2012, when companies were struggling to hit 10-25W.
Lithography expert Christopher Mack wrote about ASML’s EUV presentation at the 34th International Conference of Photopolymer Science and Technology in early July. When asked when its equipment would be ready for manufacturing insertion, ASML claimed it would be ready by the second half of 2018. Mack writes:
Tuesday ended with a panel discussion on EUV insertion into high volume manufacturing… As expected, the ASML presentation was the most optimistic, claiming manufacturing insertion for EUV as early as the second half of 2018. That is one year away! I laughed out loud when I heard that. When I asked if that prediction was serious, the response was ‘it depends on how you define manufacturing.’ In today’s fact-challenged world, every word is up for redefinition.
None of the cutting-edge foundries or fabs we’ve spoken to or seen roadmaps from — which include Intel, TSMC, GlobalFoundries, and Samsung — have any plans to integrate EUV that early. GlobalFoundries has said it wants to have EUV up and running in 2019, if possible. Samsung is aiming for a 2019 to 2020 introduction, while TSMC also wants to introduce EUV in the back half of 2019. All of the foundries in question have plans to move to the 7nm node with conventional DUV (Deep Ultraviolet) lithography, with second-and-third-generation products on that node transitioning to EUV, at least for critical layers. Whether they can hit these targets will depend on how quickly ASML can continue to improve its own hardware.
ASML’s successful demonstration of a 125 wafer-per-hour tool is significant. It’s a step towards the successful commercialization of EUV. But there are a lot of potential bumps in the roadmap between where we are today and what kind of specs EUV needs to hit to be considered useful, much less a full replacement for standard lithography. This announcement, while meaningful, should be taken with a grain of salt. Companies have been predicting that EUV was just a few years away for nearly two decades. Eventually, they’ll be right, but that’s not the same as saying they’ll be ready in 2018.