01, The EPU
A processor named for
what it spends.
CPU, GPU, TPU. Each one is named for what it works on. An Energy Processing Unit is named for the thing every other processor leaves for someone else to pay for.
The machine answers by going quiet.
A normal processor is handed a list of steps and works through them. An EPU is handed a question and simply settles into the answer, the way water finds its level or a ball rolls to the bottom of a bowl. Gravity pulls it down and friction stops it, and we meter both. When it stops moving, you have your answer.
That turns the usual arrangement on its head. The settling is not something the machine does on the way to thinking. The settling is the thinking, and stillness is the answer. When it will not settle, it reports that instead of inventing one.
Memory and compute are the same place
In every machine you can buy today, what the system knows is stored in one place and the thinking happens in another, so it spends most of its power just moving things back and forth. In an EPU, what it knows and where it works are the same place. The knowledge never moves. Only the answer in progress does, and it is three symbols wide.
It learns where it sits
Show it what you wanted instead of what it gave you, and it writes the correction in place. Learning here means adding facts on the device, in billionths of a joule, not retraining a whole model. Nothing it learns leaves the board.
One cell, every scale
The same building block works at every size. A few of them make a sensor that thinks for itself. A panel makes a device. A wall of them makes a datacentre. Nothing about the design changes as it grows, only how many you put side by side.