I found the header for this post on the website of ASML, a company at the leading edge of lithography technology, the engine of the semiconductor industry. They have a wide range of lithographic systems including the most advanced Extreme UltraViolet Lithographic systems (EUVL) using wavelength of 13.5 nm. With this systems (they are really, really, expensive, a single machine costing in the range of 150 million $!) it is possible to obtain a 3nm “size”, that is in the order of 30 atoms side by side.
Interestingly, this high resolution requires both hardware and software, without software it wouldn’t be possible to get down to this size. Actually, thanks to software ASML expects to deliver even better systems thus sustaining the evolution towards denser and denser chips (this translates into les power consumption and faster -parallel- processing).
I spent some time surfing the ASML website (interesting one!) after reading a Wired article with the intriguing -and misleading- title: “The 150 million $ machine keeping Moore’s law alive”. It is obviously intriguing – how couldn’t it be an article on the extension of the Moore’s law?- but unfortunately it is misleading.
Moore’s law called for a squeezing transistors and a parallel decrease of cost per transistor at a rate of halving every 18 months. This prediction held for over 50 years but broke down in economic terms some five years ago and in “squeezing” terms 4 years ago. The squeezing goes on at a much smaller pace and the cost per transistor is rising at these technology edge.
Nevertheless, the article makes for an interesting reading.
Most notable, as I mentioned, is the fact that where lithographic machines are facing physical barriers software may step it to overcome, at least to a point, those barriers. For sure the semiconductor industry has already circumvented physical constraints in the 2 dimensional wafer by moving up to the third dimension (watch the clip) and researchers are busy looking into 2 dimensional materials, like graphene and molybdenum disulfide, to pack more transistors in a given space. Results, in terms of possibility, have been achieved but they are still far from industrial exploitation.