Demonstration of reconfigurable and tunable all-optical matrix-vector multiplication using nonlinear wave mixing

Area of Science:

• Photonics and Optical Computing
• Nonlinear Optics
• Integrated Photonics

Background:

• Growing interest in optical domain processing for matrix operations due to advantages in bandwidth, latency, and parallelism.
• Existing optical matrix-vector multiplication (MVM) implementations still partially rely on the electronic domain.

Purpose of the Study:

• To demonstrate a fully optical matrix-vector multiplication (MVM) using nonlinear wave mixing.
• To perform matrix operations entirely within the optical domain, overcoming limitations of hybrid approaches.

Main Methods:

• Modulating matrix columns onto different subcarriers, serialized in time.
• Encoding vector elements onto separate continuous-wave (CW) tones at different wavelengths.
• Performing MVM via nonlinear wave mixing in a periodically poled lithium niobate (PPLN) waveguide.

Main Results:

• Successful demonstration of MVM at 3 GHz and 5 GHz clock rates.
• Output vectors achieved error standard deviations between 2.89% and 3.55%.
• Demonstrated precision of 4.8–5.1 bits for the computed output vectors.

Conclusions:

• Nonlinear wave mixing enables fully optical MVM, a significant step towards optical computing.
• The PPLN waveguide approach provides a viable platform for high-speed optical matrix operations.
• This method holds promise for future optical processors with enhanced speed and efficiency.