In keeping with Stigler's law, probably neither Spolsky nor Varian first identified the pattern; the question with these things is communication: will we communicate better this concept if we decide to throw all that out and start talking about "Yule's law" or "the Yule effect" and never say "commoditize your complement" again? Quite aside from the problem of all the things already named after Yule, the latter seems to already be familiar to many of the people who need to know it, and more satisfactory than starting over from scratch.
I am all for “C your C” but I sometimes wish that The Man would once again forget the concept (like IBM, popularly)—
Since Spolsky is not quite the Man but Varian is presumed to speak for Him it is probably well to credit Varian (and Yule).. in order to remind ourselves that The Man is in on it.
Looking into it more, I'm not sure Yule should be mentioned at all. The Wikipedia link says nothing of the sort (and none of the other relevant articles do either), and I've been unable to find a single mention in G/GS of Yule and Yule-Simon distributions and complements, much less one that is so clear as to merit being called "Yule's law". He gives a 1912 date at random (is it supposed to be a reference?) but that just takes me to some obscure binary correlation coefficient which is almost-but-not-exactly an odds ratio. So I'm left wondering what Yule has to do with this at all.
(And I'm a little dubious about OP's research in general after that look: as his own WP link tells you, the Gillette story is a myth so why is he repeating it, and his Apple iPod is rather thinly sourced to a search query in a book you can't read. I have not checked anything in the other laws.)
The cost of routing on a chip has been climbing for decades. If we decreased the cost of routing, the demand for transistors would go up again? satisfies #1
Making the chip less complex, satisfies #2
What if you had a chip with no routing? Just an array of Look Up Tables? (LUTs) 4x4 bits, one bit coming from each cartesian neighbor, and one going to each, clocked in the manner of checkerboard colors to prevent race conditions.
It sounds quite silly... but it has a lot of advantages, at the cost of using the LUTs to route bits in the worst case.
move signals around to get to any I/O pin
rotate/flip the placement of a software "module"
route around a bad LUT
build a secure box around any function using a wall of LUTS
compute with LUTs on data going all directions simultaneously
easily debug by inserting probes dynamically, and re-routing around them
all the gates run all the time, unlike the transistors in RAM, better utilization of silicon
make it possible to use proofs against deployed programs in the LUTs
You've basically invented the FPGA, and they're mostly good for hardware accelerating specific things. A CPU is going to be better if you don't have time to convert your problem into a hardware description language.
If you do have that time and expertise you can probably get good results on a regular GPU.
CPUs are just way easier to program for than FPGAs. Maybe with better tooling.
