The shot noise issue is more about dose control and critical dimension (CD) response to dose. Higher sensitivity means CD will be more sensitive to shot noise dose variations like few %. This is more significant for smaller CD, where a few % would eat up the tolerance.
I didn't know the answer to your question so I asked the author, here's what he said: Dear Colin,
The commenter is partially right. Shot noise (statistical robustness)
is a concern for high sensitivity resists. Length of exposure is not what
is important but the total dose. I think they understand this as they put
long in quotes ad reported the right units of mJ/cm2. However, it is an
open area of research as to whether roughness and loss of resolution is a
result of shot noise or other inhomogeneities in the process. I don't think
we are at the shot noise limit yet. Also as we approach the shot noise
limit, material reorganization during the bake and development steps may
mitigate the problem by smoothing or filtering the result. In conclusion,
we need high sensitivity resists because of expense issues with the exposure
tools. Shot noise is not a killer and managing the roughness resulting from
the exposure is part of future resist design.
I am not sure how significant the penalty for negative tone on the
metal layers is. Definitely the champion resist is positive tone and there
is preference for positive tone but that preference can be displaced with
superior performance if it comes.
A "long" exposure time is needed to guarantee sufficient photon
density to define an image with statistical robustness. So 15 mJ/cm2 is
already too low.
Also if it's negative tone resist due to crosslinking it's going to be more
defect sensitive on the contact or metal layers.
Directed self-assembly is a no-brainer for HDD, for their bit-patterned media technology, since it is defect tolerant. Otherwise you might worry if your trench is a little narrow, a self-assembled line inside might get squeezed out.
There's a lot more to it than the exposure time. Researcher using commercially available tools today are producing less than 100 wafers a day while still in debugging mode. The optimists among them are hoping for 500 wafers a day by 2015 and 1,500 wafers per day at the 10 nanometer node slated by 2017.