Photonic convolution accelerator based on a hybrid integrated multi-wavelength laser array by photonic wire bonding...

Area of Science:

• Optoelectronics
• Artificial Intelligence
• Integrated Photonics

Background:

• Convolutional neural networks (CNNs) are crucial for image classification.
• Optical implementations of CNNs offer potential for high speed and low power consumption.
• Hybrid integration platforms are key for complex photonic systems.

Purpose of the Study:

• To propose and demonstrate a compact and efficient photonic convolution accelerator.
• To leverage hybrid integration of a multi-wavelength DFB laser array for optical computing.
• To achieve real-time image classification using the developed photonic accelerator.

Main Methods:

• Utilized photonic wire bonding for hybrid integration of a DFB laser array.
• Designed and implemented a photonic convolution accelerator architecture.
• Operated the accelerator at 60.12 GOPS for 3x3 kernel operations.
• Performed real-time image classification on the MNIST dataset.

Main Results:

• Demonstrated a compact and efficient photonic convolution accelerator.
• Achieved real-time image classification of 500 images.
• Attained a prediction accuracy of 93.86% on the MNIST handwritten digit database.
• Showcased the viability of the hybrid integration platform for optical CNNs.

Conclusions:

• The proposed photonic convolution accelerator is compact and efficient.
• Hybrid integration using photonic wire bonding enables advanced optical CNNs.
• This technology offers a promising pathway for high-performance optical computing and AI acceleration.