Nonlinear optical vector processing using linear silicon photonic circuits for 50 Gb/s memory and string similarity...

Area of Science:

• Photonics and optical computing
• Silicon photonics for advanced computation
• Nonlinear optical signal processing

Background:

• Conventional electronics face limitations in speed and energy efficiency for machine learning and communications.
• Photonics offer potential for high-speed, low-power computational tasks, including linear and nonlinear optical transformations.

Purpose of the Study:

• To demonstrate nonlinear optical vector processing using silicon photonic circuits.
• To enable high-speed pattern matching and look-up operations for machine learning and communications.

Main Methods:

• Implementation of a 4x4 crossbar architecture with silicon germanium modulators.
• Utilizing nonlinear optical vector processing for Hamming distance calculations between 2-bit optical vectors.
• Employing space-wavelength multiplexing with a wavelength-division multiplexing cell for scalability.

Main Results:

• Achieved high-speed nonlinear optical vector processing at 50 Gb/s.
• Demonstrated Hamming distance calculations and content addressable memory functionality.
• Obtained error rates of approximately 10⁻³ in binary/ternary content matching, exceeding previous speeds by over 2.5 times.
• Experimentally validated scalability using wavelength-division multiplexing, showing reduced losses and power consumption.

Conclusions:

• Silicon photonic processors can perform complex nonlinear optical vector operations at high speeds.
• The demonstrated processor enables efficient pattern matching and look-up operations crucial for machine learning and communications.
• Scalable photonic architectures offer a pathway to enhanced computational capacity with improved energy efficiency.