Micro-ring resonator assisted spiking neural network for efficient object detection.

Area of Science:

• Photonics
• Neuromorphic Computing
• Artificial Intelligence

Background:

• Spiking neural networks (SNNs) and optical computing offer high parallelism and low energy consumption.
• Current photonic SNN implementations face challenges in large-scale integration and energy efficiency due to active devices or nonlinear effects.
• Existing optical SNN applications are primarily limited to basic image classification.

Purpose of the Study:

• To propose an optical-assisted SNN model utilizing passive add-drop micro-ring resonators (ADMRRs).
• To simulate neuronal membrane potential accumulation via optical temporal integration.
• To enable complex image processing tasks like object detection using photonic SNNs.

Main Methods:

• Developed an SNN model based on passive add-drop micro-ring resonators (ADMRRs).
• Simulated membrane potential accumulation using optical temporal integration.
• Implemented a spiking version of the modified YOLO algorithm with ADMRR-based neurons for object detection.

Main Results:

• The proposed photonic SNN achieved over 98% of the performance of computer-based SNNs on the PASCAL VOC dataset.
• The model successfully performed system-level object detection on a dataset with 11,530 images across 20 categories.
• Demonstrated effective emulation of neuronal leakage and integration dynamics.

Conclusions:

• The passive ADMRR-based photonic SNN offers simplicity, enhanced parallelism, and ease of large-scale integration.
• This approach overcomes limitations of current photonic SNNs, enabling more complex image processing.
• Paves the way for widespread adoption of photonic SNNs in advanced AI applications.