The last day of the conference gave the tool updates. So how is EUV progressing?
The final day of the Advanced Lithography Symposium contains what is commonly referred to as the “tool” sessions, where tool makers give updates on their latest and greatest products. As such it tends to have the most commercial presentations, with all the problems that come with commercial pressures. For the EUV conference it is also the day where technology cheerleading, and skepticism, reaches a fevered pitch.
ASML described the status of their production EUV scanner, the NXE:3300B. Eleven systems are under construction, nine of which should ship to customers by the end of this year or early next year. These systems will ship regardless of the source power available, and so the most anticipated talk was given at 9am by Cymer. To put this year’s report in perspective, one year ago Cymer was delivering a 9W (intermediate focus) EUV source to customers, claiming that a 20W source was about to be “available”, and predicting 100W by December 2012. The goal for production remains 250W. This year Cymer was “proud” to demonstrate 40W production-like performance and said sources for the 3300 had already been shipped to ASML. Let’s parse this announcement a bit.
What Cymer showed was a modified 3100 source that achieved 40W production-like performance for 6 hours one day last week, and 40W production-like performance for 6 hours one day this week. While a definite milestone, it is certainly not the same as delivering 40W performance to a customer for regular production-like use. So, in the last 12 months Cymer has doubled source power in the lab. In six months ASML will begin shipping tools with, they hope, an 80W source attached. I find it highly unlikely that this will be achieved. An 80W source should enable about 40 wafer-per-hour throughput or so, which will be fast enough to enable valuable production learning. Then Cymer will need to increase source power to 140W (and deliver that power to customers) by the end of 2014 to meet ASML’s stated goal of having EUV lithography that is producing chips at 70 wph by the end of 2014. Cymer has to beat Moore’s Law by a long shot, doubling source power twice in the next 18 months. That will be a hard job, indeed.
Next, Zeiss showed their progress and roadmap for EUV optical systems. An important question is how high can the NA go before two more mirrors are added (thus cutting the throughput in half). Their answer: 0.45. Since the NXE:3300 has an NA of 0.33, increasing the NA to 0.45 will allow the resolution to improve by a factor of about 0.73, close enough to the expected feature size reduction of 0.7 that a second generation of EUV production tools might take the technology to one more node. With off-axis illumination (and a low k1), 10-nm or 11-nm half-pitch might be possible with NA = 0.45. It will not be easy, though.
In the afternoon, the tool talks in the optical lithography conference touched on one of my pet peeves (some people claim I have too many pet peeves), so it is time to step up on the soap box again and talk about “lying with graphs”. There are many ways to lie with graphs, and most result from the practice of advocacy speech: using a graph to impress rather than inform. This is what marketing folks often do. It is not what scientists should do. Let me take as an example the topic of global warming. Hopefully everyone has seen plots of global surface temperatures that show a fairly steep rise over the last 50 years (close to 1 degree Centigrade). Now suppose you want to argue that global temperatures are not rising. One approach would be to plot the same data on a graph that has a y-axis origin of zero centigrade (or better yet, zero Kelvin). The result will be a trend that looks almost totally flat, since the variation will be hard to notice when squeezed into a few percent of the area of the graph. This technique works well whenever you want to show a flat trend in the data , regardless of the actual trend in the data.
The two laser talks, by Gigaphoton and Cymer, each displayed dozens of graphs that lie in exactly this way. Both lasers can control dose by pulse over time to within 0.2%. But to plot the data, they chose a y-axis that went from -1% to +1%, so that 80% of the y-axis range was unused. Wavelength stability, spectral bandwidth, beam position, beam profile size, divergence, and other laser metrics were plotted in the same way, sometimes using less than 10% of the y-axis range. Why even show the data if you purposely choose a y-axis that makes any data variation invisible? Obviously it is not to inform the audience. I’ve seen other talks over the years doing the same thing: wafer chuck temperature, aberrations across a slit, etc. There is a simple rule to prevent this: your data should use up 70% of the range of both the x- and y-axes. Don’t be caught lying with graphs.
And so ends another SPIE Advanced Lithography Symposium. 2012 was an interesting year in lithography. 2013 will be even more so.