There's an analysis I have that I'd like to get you, @tidalwave, and anyone else with some knowledge in this field to weigh in on. I'm going to take x nanometre "feature" or "node" to mean a square of side length x nanometres. I'll also assume the validity of Moore's law, doubly of transistor (or "feature") density every two years or, equivalently, halving of feature area every two years. I also recall reading that the smallest silicon transistor created contains 12 silicon atoms (there are single-atom transistors, but they're made out of other materials and require liquid nitrogen cooling; they're not in the lithography discussion). Finally, note that a silicon (Si) atom has a diameter of 0.2 nm.
If we take 7nm as the starting point, we calculate that a 7nm square contains 1,225 Si atoms. Given the lower limit of 12 Si atoms, it would take approximately 6.67 doublings or generations to reach that lower limit (a feature length of roughly 0.7nm). Since a doubling occurs every two years, we have the physical limit of Moore's law reached between 13 and 14 years from now. That's all the runway TSMC, ASML and their ilk have left. After that they hit a hard physical wall that has nothing to with economics. The great thing is if they're stopped, it will take China on the order of months to completely close whatever gap remains (if any) - it's much easier to catch up to a stationary target.
@tidalwave, your idea about synchrotrons really intrigued me. The issue I see with it (beyond the logistical matter of siting the fab adjacent to the synchrotron) is that synchs produce all manner of radiation besides UV, so some kind of filter would have to be employed to screen out the undesirable components of the synch's output. I think that synchrotrons would also produce UV of a shorter wavelength than ASML's EUV machines produce, which would allow smaller node sizes. The great thing there is that only the filter or the average particle energy in the synchrotron would have to be changed to get the desired wavelength of UV light, instead of having to design a light source de novo. It really is a neat idea and I hope Chinese scientists are looking into it.
The US is investigating it:
If we take 7nm as the starting point, we calculate that a 7nm square contains 1,225 Si atoms. Given the lower limit of 12 Si atoms, it would take approximately 6.67 doublings or generations to reach that lower limit (a feature length of roughly 0.7nm). Since a doubling occurs every two years, we have the physical limit of Moore's law reached between 13 and 14 years from now. That's all the runway TSMC, ASML and their ilk have left. After that they hit a hard physical wall that has nothing to with economics. The great thing is if they're stopped, it will take China on the order of months to completely close whatever gap remains (if any) - it's much easier to catch up to a stationary target.
@tidalwave, your idea about synchrotrons really intrigued me. The issue I see with it (beyond the logistical matter of siting the fab adjacent to the synchrotron) is that synchs produce all manner of radiation besides UV, so some kind of filter would have to be employed to screen out the undesirable components of the synch's output. I think that synchrotrons would also produce UV of a shorter wavelength than ASML's EUV machines produce, which would allow smaller node sizes. The great thing there is that only the filter or the average particle energy in the synchrotron would have to be changed to get the desired wavelength of UV light, instead of having to design a light source de novo. It really is a neat idea and I hope Chinese scientists are looking into it.
The US is investigating it:
