On-chip photonic convolution by phase-change in-memory computing cells with quasi-continuous tuning.

Area of Science:

• Photonics
• Optical Computing
• Materials Science

Background:

• On-chip photonic integrated circuits (PICs) offer speed and bandwidth advantages for matrix multiplication.
• Phase-change materials enable optical storage and computing, outperforming electrical methods.
• Integrating Mach-Zehnder interferometers (MZIs) and micro-ring resonators (MRRs) presents design challenges.

Purpose of the Study:

• To propose and design an on-chip photonic convolution architecture for optical in-memory computing.
• To integrate phase-change chalcogenide (GSST) into an asymmetric directional coupler for in-memory computing cells.
• To leverage the benefits of MZIs and MRRs for enhanced computing capabilities.

Main Methods:

• Designed an in-memory computing cell by integrating GSST into an asymmetric directional coupler.
• Utilized quasi-continuous electro-thermal tuning for GSST phase transition.
• Performed numerical calculations to analyze optical and electro-thermal behaviors.
• Applied the optical convolutional kernel to an image edge detection task.

Main Results:

• Confirmed the tunability of programmable elements within [-1, 1] through GSST phase transition.
• Demonstrated the feasibility of on-chip photonic convolution.
• Achieved accuracy comparable to digital computer implementations for image edge detection.

Conclusions:

• The proposed GSST-integrated photonic integrated circuit is a viable scheme for optical in-memory computing.
• This approach offers a promising solution for high-performance, on-chip image processing.
• The design marries the advantages of MZIs and MRRs for efficient optical computing.