Addressing optical modulator non-linearities for photonic neural networks

Area of Science:

• Neuromorphic Computing
• Analog Photonics
• Integrated Photonics

Background:

• Analog photonic systems offer high computational speeds and reduced power consumption for neuromorphic computing.
• Non-linear distortions and noise in analog systems limit signal resolution and overall functionality.

Purpose of the Study:

• To develop a method for analyzing and minimizing non-linearities in optical power transfer functions of modulators.
• To compare different modulator types (Mach-Zehnder interferometer, micro-ring modulator, ring-assisted Mach-Zehnder interferometer) and their suitability for analog photonic processors.
• To evaluate three analog photonic processor architectures for machine learning applications based on multiplexing techniques.

Main Methods:

• Analysis of non-linearities in the optical power transfer function of generic modulators.
• Comparative performance evaluation of Mach-Zehnder interferometer, micro-ring modulator, and ring-assisted Mach-Zehnder interferometer.
• Application of the analysis to wavelength, space, and time division multiplexing architectures for machine learning.

Main Results:

• A method for analyzing and minimizing modulator non-linearities was presented and applied.
• Mach-Zehnder interferometers, despite lower maximum resolution, demonstrated superior balance in stability and power consumption for specific architectures.
• The study identified optimal design and operation conditions for analog photonic processors.

Conclusions:

• Mach-Zehnder interferometers are a balanced choice for space and time division multiplexing analog photonic processor architectures in machine learning.
• Minimizing non-linearities is crucial for enhancing the performance and reliability of analog photonic systems.
• The presented analysis method aids in the informed design of next-generation neuromorphic computing hardware.