On-chip phonon-magnon reservoir for neuromorphic computing.

Area of Science:

• Physics
• Materials Science
• Computer Science

Background:

• Reservoir computing maps signals to a dynamical system's phase space for neural network recognition.
• Implementing reservoir computing on-chip requires novel physical systems capable of complex signal processing.

Purpose of the Study:

• To implement reservoir computing using a nanodevice based on phonon-magnon interactions.
• To demonstrate the separation of visual shapes encoded by laser input.

Main Methods:

• A nanodevice comprising a semiconductor phonon waveguide and a ferromagnetic layer was fabricated.
• Input signals were encoded into phonon wavepackets using a pulsed write-laser, interacting with magnons.
• Output signals were read using a second laser, sensitive to phonon-magnon mode interactions.

Main Results:

• The nanodevice reservoir successfully separated visual shapes drawn by the write-laser.
• Shape separation was achieved in an area comparable to a single pixel.
• Phonon-magnon interactions proved highly sensitive to laser positioning.

Conclusions:

• Phonon-magnon interaction is a viable hardware basis for on-chip reservoir computing.
• This work advances the development of future neuromorphic architectures.
• The demonstrated nanodevice offers efficient on-chip signal processing capabilities.