A computer the size of a millimetre, with a radio and a job to do.

At a square millimetre, the device is no longer a board you populate. It is a system you co-design with the physics.

A nanocomputer at this scale does three things at once: it reads a set of nanosensors, it makes a local decision or aggregates a measurement, and it gets that result off the die over a radio. There is no room for a separate sensor board, a separate MCU board, and a separate radio module. Sensing, compute, and the Bluetooth link have to share the same silicon — a system-on-chip in the literal sense.

I work this from two directions. On the device side it is nanoelectronics: laying out circuits at scales where the tidy textbook assumptions — clean voltage levels, negligible leakage, parasitics you can ignore — stop holding, and the layout itself becomes part of the circuit. On the workflow side it is a bench discipline: a microscope and a microsoldering setup, so I can fabricate, rework, and inspect parts that are below the threshold of the naked eye.

The two application targets are deliberate. One is biomedical control, where the constraint is that the device has to be small enough and quiet enough to sit close to the thing it measures. The other is industrial control that requires nanometric precision — processes where the tolerance band is narrow enough that an ordinary sensor-and-PLC loop cannot hold it. Same discipline, two very different rooms.

The data path is not a diagram you admire once and move past. It is a sequence of six steps, and every one of them can quietly ruin the measurement if you let it.

I keep the chain explicit because each hop has a different failure mode. The transducer can drift. The front-end can add more noise than it removes. The converter can sample at the wrong instant. The local logic can throw away the one sample that mattered. The radio can drain the budget faster than the source can refill it. And the relay can simply not be listening. Naming the steps is how I keep each one accountable.

Notice where the intelligence sits. The cheapest packet is the one you never send, so the controller decides locally — aggregating, thresholding, discarding — before the radio ever wakes. On a sub-millimetre node the radio is usually the largest single draw, so the whole design leans on sending less, not sending faster.

The same die answers two very different briefs. Both ask for nanometric precision; they ask for it in opposite environments.

In the biomedical case the binding constraint is proximity. The device has to sit close to the thing it measures and stay small and quiet enough not to disturb it. Size is not a marketing point here — it is the only way the measurement is honest, because a larger or louder instrument changes the very signal it is reading.

In the industrial case the binding constraint is the tolerance band. Some processes hold a window narrow enough that an ordinary sensor-and-PLC loop cannot keep up with it — by the time a coarse loop notices the drift, the part is already out of spec. A node with nanometric resolution sitting inside the process can see the drift while it is still small.

Both rooms reward the same discipline: put a precise, well-characterised measurement exactly where it is needed and report it without disturbing anything. The hardware is largely shared; what changes is the envelope it has to survive and the language the endpoint speaks.

The application changes; the discipline does not. These are the principles that survive whether the node ends up in a body, in a process line, or — in the materials crossover — in a fluid.