Perspective on 3D vertically-integrated photonic neural networks based on VCSEL arrays

Area of Science:

• Photonics and Artificial Intelligence
• Integrated Optics and Neural Computing

Background:

• Artificial intelligence advancements drive interest in novel photonic neural network designs.
• Diffractive neural networks (DNNs) excel at high-speed 2D optical data processing but face miniaturization challenges due to centimeter-scale input devices.
• Current DNNs lack on-chip integration capabilities, hindering practical applications.

Purpose of the Study:

• To propose vertical-cavity surface-emitting laser (VCSEL) arrays as a solution for compact, active DNNs.
• To explore VCSEL arrays as an integrated platform for next-generation on-chip photonic neural networks.
• To analyze future directions and challenges for 3D photonic chips based on VCSEL arrays.

Main Methods:

• Conceptualizing the use of addressable VCSEL arrays for data input in DNNs.
• Investigating the potential for micron-scale 3D photonic chips.
• Analyzing the modulation bandwidth capabilities of VCSEL-based photonic chips.

Main Results:

• VCSEL arrays can serve as compact, active data input devices for DNNs.
• Micron-scale 3D photonic chips with tens of GHz modulation bandwidth are feasible using VCSEL arrays.
• This approach facilitates the development of on-chip vertical-stacked photonic neural networks.

Conclusions:

• VCSEL arrays represent a promising technology for miniaturizing and integrating DNNs.
• The proposed platform enables the creation of high-performance, compact photonic neural networks.
• Further research is needed to address the challenges and realize the full potential of 3D photonic chips.