Neuromorphic Computing via Fission-based Broadband Frequency Generation.

Area of Science:

• Photonics and Neuromorphic Engineering
• Optical Computing Architectures

Background:

• Traditional computer architectures face performance limitations, driving interest in brain-inspired hardware.
• Optical solutions offer energy efficiency and speed, but integrating nonlinear nodes for neural emulation is challenging and limits scalability.

Purpose of the Study:

• To present a novel neuromorphic wave computing paradigm for energy-efficient information processing.
• To demonstrate the emulation of digital neural networks using nonlinear optical interactions in a fiber-integrated system.

Main Methods:

• Utilizing broadband frequency conversion via coherent higher-order soliton fission in a single-mode fiber.
• Employing phase encoding on femtosecond pulses and frequency selection/weighting for interpretable system states.
• Conducting experiments in a compact, fully fiber-integrated setup.

Main Results:

• Demonstrated energy-efficient emulation of digital neural networks.
• Observed enhanced computational performance with increasing system nonlinearity.
• Validated the concept in a compact, fiber-integrated experimental setup.

Conclusions:

• Broadband frequency generation in fiber offers a viable alternative to traditional node-based brain-inspired hardware.
• This approach enables energy-efficient, scalable, and dependable computing with minimal optical hardware.
• Neuromorphic wave computing challenges conventional designs by leveraging nonlinear optical dynamics.