In a landmark result for the project, RadioSpin researchers have demonstrated a neuromorphic processor built from four coupled spin-torque nano-oscillators performing real-time classification of radiofrequency signals.
The demonstration exploits the rich nonlinear dynamics of coupled oscillators. When several STNOs are connected together, their collective response to an input signal can be trained to separate different classes of input — in effect, the physics of the oscillators performs the computation directly in hardware, without a conventional digital processor.
Running the classification in real time, and directly on RF signals, is a significant step toward the project's vision of ultra-efficient hardware that unifies signal generation, detection, and processing on a single spintronic platform.
While four oscillators is a modest network by the standards of software neural networks, the result validates the approach and provides a clear path toward larger arrays with correspondingly greater computational capability.