Physicists develop liquid computer using chaotic particle orbits for forecasting and anomaly detection

Liquid Computing System Developed

Physicists at the Universities of Konstanz and Stuttgart have created a liquid computer using 400 microscopic particles. These particles, 3μm silica spheres with an 80nm carbon cap, are suspended in a water-lutidine mixture at 28°C. A 532nm laser heats the cap, causing the particles to orbit around target points due to overshooting during repositioning.

Computational Tasks and Performance

The array successfully predicted a chaotic Mackey-Glass series and identified anomalies that did not affect a signal's mean, variance, or short-time autocorrelation, achieving an F1 score of 0.90 for anomaly detection. Data is input by displacing the target points, and coupling between orbits is managed by hydrodynamic forces, with lattice spacing determining strength. Damping thresholds control particle swing.

Comparison to Existing Technologies

The liquid computer achieved a normalized root-mean-squared error of approximately 0.1 on the Mackey-Glass prediction benchmark. This performance is roughly 10 times less accurate than current memristor devices, which achieve 0.01 or better. The researchers note that memristor results often involve time-multiplexing and are the product of nearly a decade of development.

System Flexibility and Components

The system's parameters, such as lattice spacing and damping threshold, are adjustable during operation, allowing for significant variation in forecasting error. The accuracy remained consistent even when only 20% of the oscillators received input or when individual particles malfunctioned. The setup requires a 532nm laser, a 100 kHz acousto-optical deflector, real-time microscopy, a temperature-controlled quartz cell, and a conventional computer for output processing.