The original note, as written at the time
How about this for the future of sustainable computer technology. Imagine each transistor was a nanobot on a chip. If we achieved high level precision in the behaviour of these nanobots, entire chips and motherboards could rewrite their hardware on the fly without need for new hardware due to compatibility issues and change in architecture. The bots would simply modify their structure and orientation on their own.
Ongoing development
For the future of sustainable computer technology: imagine each transistor was a nanobot on a chip.
If you achieved high precision in the behaviour of these nanobots, whole chips and motherboards could rewrite their hardware on the fly — no new silicon for a change of architecture, no compatibility problem, no replacement cycle. The bots would modify their own structure and orientation.
What makes this the sustainable version is not the cleverness. It is that it attacks the actual waste, which is not energy per operation but the throwing away of perfectly good matter. A processor is obsolete because it is the wrong shape for the current idea, and we bury it. The material was fine. The arrangement was wrong.
A substrate that can rearrange itself is a substrate that is never obsolete, only reconfigured, which converts a replacement economy into a maintenance economy — the same shift as the one from disposable to repairable goods, but at the level of the machine’s own logic.
The objections are honest and severe. A billion moving parts is a billion failure points; precision at that scale is where the whole proposal lives or dies. Heat removal becomes a different problem when you cannot assume the geometry. And there is an obvious hazard: a chip that can rewrite its own structure is a chip that can rewrite itself into something you did not intend, and the safety argument for fixed silicon is precisely that it cannot.
The reply to that last point is that fixed silicon has already been broken in every way that matters. A rigid chip has no way to route around a fault, no way to degrade gracefully, no way to become more efficient as the workload changes.
I would rather have a machine that can be maintained than one that can only be replaced. Even at the cost of a much harder engineering problem.