Literature review demonstrates progress in optical neural network architectures, suggesting optical computing can overcome electronic hardware bottlenecks in artificial intelligence.
Key Points
The review evaluates the principles, hardware configurations, and performance metrics of optical neural networks designed to overcome electronic computing bottlenecks in speed and energy efficiency.
Examined design frameworks and operating principles across various optical components used in computing hardware.
Compared non-integrated systems built with free-space volume optics against integrated systems fabricated with photonic on-chip devices.
Assessed core system metrics including optical nonlinearity, computational density, hardware scalability, and operational constraints.
Optical neural networks provide sub-nanosecond processing latency, reduced thermal dissipation, and high parallelism compared to traditional electronic architectures.
Integrated on-chip photonic designs enhance system miniaturization, though scaling remains constrained by fabrication tolerances and footprint demands.
Practical deployment requires resolving foundational challenges in reconfigurable optical nonlinearity, energy-efficient optical-to-electrical interfaces, and scalable manufacturing.