SPIE – Day 3

Line-edge and linewidth roughness will be the ultimate limiter to resolution in optical lithography. We are already nearing that point.


Today I talked more than listened. I had two papers, both on stochastic effects in lithography. In one, I showed a complete model for EUV resist exposure and for predicting stochastic uncertainty in the acid concentration at the end of exposure. The second, looking at quenching effects, is unfortunately a work in progress. There is still much to learn. One important lesson is to get my papers done before I come to the conference.

But since I am on the topic of learning about line-edge roughness (LER), I have been very disappointed with the number and quality of this year’s papers on this topic. I’ve said it before and I’ll repeat it here – it is my opinion that line-edge/linewidth roughness (stochastic effects in general) will be the ultimate limiter to resolution in optical lithography. We are already nearing that point, with many at this conference talking about how great some experimental result is “except for the roughness”. From what I can tell, there has been essentially no progress in improving LER since last year, and insufficient progress in improving our understanding of LER mechanisms.

As for LER metrology, I can’t tell whether the metrology is not good enough, or whether metrology users are just unwilling to do it right. Doesn’t anyone realize that measuring noise is inherently noisy? If I measure LWR to be 5 nm, using a typical 128 measurement sample on one image, the 95% uncertainty just from random error is +/- 12%, that is 5 +/- 0.6 nm. I did not see more than one paper that put error bars or uncertainty estimates for LWR in their presentation. I saw several talks that presented LWR values to two decimal places, or showed materials that exhibited 5% lower LWR, without any discussion of the statistical significance of the numbers. Aren’t we engineers and scientists here? We can do better. Samsung gave an interesting talk showing how their measured LWR varied from 3.1 nm to 5.4 nm on the same sample just based on their measurement recipe. Systematic errors on top of the random errors.

And since I’m riled up, let’s talk about LER post-processing, AKA magic rinse. This is where a rinse or vapor or e-beams or ion beams, usually with some heat, is used to smooth out the roughness of the sidewalls of resist. I’m sorry, but this just won’t work. Low frequency LWR is a linewidth error. A rinse can’t remove linewidth errors from the resist (if it does you have even bigger problems to worry about), so it can only remove the high-frequency roughness that doesn’t really matter any way. It’s like using a rinse to remove my gray hair. It might make me prettier, but it won’t make me a nice guy.

OK, I’m off my soap box.