In-memory computing on a photonic platform.

Area of Science:

• Optoelectronics
• Materials Science
• Computer Science

Background:

• Biological computing systems, like the brain, utilize collocated data processing and storage.
• Advancements in hardware drive exploration of computational paradigms beyond traditional von Neumann architectures.
• Integrated photonic circuits offer high speed and bandwidth, minimizing electro-optical conversions for on-chip computing.

Purpose of the Study:

• To demonstrate all-photonic in-memory computations by integrating optics with collocated data storage and processing.
• To explore the use of nonvolatile photonic elements for efficient optical computing.
• To develop a novel approach for direct light-matter interaction-based computation.

Main Methods:

• Utilizing integrated photonic circuits with nonvolatile phase-change material (Ge$_{2}$Sb$_{2}$Te$_{5}$) elements.
• Implementing collocated data storage and processing directly within the optical domain.
• Achieving scalar and matrix-vector multiplication through light-matter interactions.

Main Results:

• Successful demonstration of all-photonic in-memory computations.
• Achieved direct scalar and matrix-vector multiplication using Ge$_{2}$Sb$_{2}$Te$_{5}$ photonic elements.
• Introduced a novel single-shot Write/Erase mechanism with a drift-free process.

Conclusions:

• The developed all-optical approach enables direct in-memory computations, eliminating electro-optical conversions.
• The method is novel, easy to fabricate, and operate, offering a drift-free solution.
• This work lays the foundation for the development of entirely photonic computers.