Integrated photonic computing-in-memory with 2D-addressable VCSELs arrays for phase change materials actuation

Area of Science:

• Photonics
• Optical Computing
• Materials Science

Background:

• The von Neumann bottleneck limits computing performance.
• Scalability and integration complexity hinder photonic computing-in-memory (PCIM) deployment.
• External optical devices for phase-change material (PCM) programming pose integration challenges.

Purpose of the Study:

• To propose and demonstrate an integrated PCIM module overcoming current limitations.
• To enable scalable and high-density integration for practical PCIM deployment.
• To showcase parallel signal and image processing capabilities.

Main Methods:

• Heterogeneous integration of a 2D-addressable vertical-cavity surface-emitting laser (VCSEL) array with a SiN-Silicon-on-Insulator (SiN-SOI) phase-change material (PCM) memory bank using flip-chip technology.
• Development of a 32 × 32 VCSEL array integrated with a GST cell crossbar memory bank.
• Execution of discrete Fourier transform (DFT), discrete cosine transform (DCT), and convolution operations using wavelength-division multiplexing (WDM).

Main Results:

• Elimination of discrete fiber-optic components for PCM control.
• Reduction of electrical interconnects from O(N^2) to O(N) for N × N arrays.
• Demonstration of over 70 distinct multilevel states (>6 bits) per cell within an 11 dB optical dynamic range.
• High-fidelity parallel execution of DFT, DCT, and convolution operations.

Conclusions:

• The integrated PCIM module offers a scalable and high-density pathway for advanced optical computing.
• This approach addresses critical barriers to the practical deployment of PCIM.
• The demonstrated technology paves the way for future high-performance optical computing systems.