Re: per transistor cost
It's true that a new Fab for State-of-the-art is scary expensive. However leakage and tunnelling are now serious issues. 8nm to 14nm minimum feature does need very high quality materials and is close to the limits for sensible performance. Yes, you can as lab demo make a rubbish transistor from a few atoms. You've also a problem with high speed production of chips as you'd have switch from masks to Vector graphic writing using electron beams or something.
Temperature and background radiation and device life (metals drifting) become an issue too. I've wondered is best performance/reliability around the 28 nm to 40nm device geometry.
Biting the hand that feeds IT
