In “How do computers work?”, SDSTAFF Karl makes a minor error when says,
Actually, NOT gates are superfluous.
NAND is (~(A&B)). If both inputs of the gate are connected to the same source, however, the gate will compute (~(A&A)). Since (A&A) always has the same (truth) value as (A), the output of the NAND gate is (~A). (The same trick can be with NOR gates, but NAND logic is simpler to etch, thus cheaper; the actual implementation of NOR logic does not, I think, have any corresponding advantage.)
Now, a NOT gate is certainly simpler than using a NAND gate for NOT (one transistor versus four, as Karl points out). But then it’s necessary to etch two different gates into the chip. It’s cheaper to etch one, and bite the bullet in terms of overhead.
“Kings die, and leave their crowns to their sons. Shmuel HaKatan took all the treasures in the world, and went away.”