So a manufacturing cost in terms of joules per processor? Curious ask.. I'm not really sure I guess.. Certainly a big factor is the huge amount of processors made in the last year compared to any previous year, which is thanks to a combination of economic scaling and technology improvements. If you ignore the newfangled chip designs, you can buy a chip today for $1 that roughly does what a chip in 1993 did for (1993 $) $600. The costs of manufacturing (energy included) are supposedly baked into the price of a processor unit...
If you want more specific, direct details you may need to look up intel's financials and do some math estimations? Could be an interesting exercise and produce some interestingly correlated graphs upon which you might be able to predict some things. Maybe a good subject for an MBA student that wants to work at a HW tech company?
Has anyone proposed a conversion of all costs on balance sheets (R&D) to energy costs? e.g. R&D is made up of humans that are ~.1kWh each...
note: Koomey's_law from above was found first in DSP space. This makes a lot of sense considering those are far more ASIC-like: the benchmark workload for a DSP, DFT, has been pretty constant and used in practice since the beginning. Whereas CPUs have general purpose workloads.