Reservoir, physical and ionic computing

What it is

Reservoir computing sends a changing input, such as a radio, vibration or sensor signal, into a system that retains a short-lived trace of recent inputs. Think of dropping pebbles into a pond: each new ripple mixes with ripples that are still fading, so the current pattern contains clues about the recent sequence. The 'reservoir' is this changing state, not necessarily a body of water. It can be created in software or by an optical circuit, memristor array, magnetic device, mechanical resonator, fluid or electrochemical material. Usually only a simple output layer is trained to recognise useful patterns. Ionic devices create related effects through moving electrically charged atoms or molecules in solids, gels or liquids.

What it is not

This is not ordinary computing housed in unusual materials. The material itself changes the signal. Some reservoirs use many interacting elements; others pass a sequence through one element and sample it repeatedly. They can be useful when order and timing matter, but repeatability, calibration, speed, measurement and packaging remain substantial challenges.