ASIC development on SHA256 has been going on since 2011, with about 4-5 process node updates and 2-3 generation updates in between nodes.
Development for other algos is no older than ~2 years, and these all probably started at the very cheap 90nm node, so its very easy to get a 6x gain going from 90-45nm.
Its a little harder for SHA256 to see huge gains going from 16nm to 10 or 7. Bitcoin ASICs have pretty much caught up with the latest node tech, so there wont be any more crazy gains in the future.
I agree. the improvement of ASH256 relies more on the shrink of die now.
Thats not what he said at all, try re-reading his post. Asic chips are running into the same issue Intel and Amd have been dealing with for years now as the proceses mature they become more cost effective for the wider market as foundries open up smaller node tech and convert lines over to larger wafers and get higher yields but these are all Diminishing returns right now. there will be no exponential increases until new (maybe graphene) substrates are used. 3 dimensional processes will also allow larger footprints but run into thermal barriers.