Outer product processor using polarization encoding

Area of Science:

• Optical engineering
• Computer architecture
• Artificial intelligence

Background:

• Traditional architectures face limitations in parallel processing and spatial capacity.
• Evaluating complex functions like second-order polynomials requires specialized hardware.

Purpose of the Study:

• To present a novel optical architecture for outer product operations and element weighting.
• To enhance spatial capacity and computational versatility for applications like quadratic neural networks.

Main Methods:

• Utilizing a polarization encoding technique to increase spatial capacity.
• Employing space-integrating photodiodes for evaluating second-order polynomials.
• Implementing spatial multiplexing for parallel operations.

Main Results:

• Achieved a fourfold increase in spatial capacity compared to other techniques.
• Demonstrated the capability to evaluate second-order polynomials.
• Enabled versatile parallel processing for complex computations.

Conclusions:

• The proposed architecture offers significant advantages in spatial capacity and parallel processing.
• It is well-suited for implementing advanced computational models such as quadratic neural networks.