A chef tastes the soup after every pinch of salt and instantly decides what to add next. That's roughly how the new setup for growing thin crystalline films works — except instead of taste, it uses an image from an electron microscope, and instead of a chef, a computer program.

Usually, to get a film of the right composition, you have to try dozens of combinations of temperature, pressure, and vapor composition. That's slow and expensive. Here, the system changes conditions on the fly during growth, watches the crystal structure, and corrects the process immediately. The camera snaps a picture after every layer, and the algorithm compares it with a reference and decides where to go next.

As a result, the desired oxide phase was obtained on average 30 times faster than with a full parameter sweep. This approach can be transferred to other deposition systems and even to semiconductor manufacturing, where atomic-level precision matters.

This is how self-driving materials synthesis is born: the machine doesn't just follow commands; it searches for the best recipe on its own, like an experienced technologist with years of know-how.