.png){kind=icon}
.png){kind=icon}
Return
Photonic edge computing based on directly modulated laser array
.png){kind=icon}
PRE
AI
DOI：10.1016/j.optlastec.2025.113651.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Advanced artificial intelligence (AI) models are becoming increasingly powerful. However, their immense parameter size makes it challenging to deploy these models on small-scale devices such as smartphones, sensors, and other edge terminals. To address this issue, we propose a solution using an array of Directly Modulated Lasers (DML) as cloud-based transmitters. These transmitters efficiently deliver the required weight bank to application terminals. In our implementation, we achieved edge computing capabilities over 10 km transmission distance by utilizing an 8-wavelength DML array under 10 Gb/s per wavelength. This setup delivered a dot-product precision of 4.3 bits, along with classification accuracies: 97 % for the MNIST dataset and 92.4 % for the FashionMNIST dataset. This approach demonstrates the feasibility of leveraging optical communication technologies to overcome deployment limitations of large AI models on resource-constrained devices.
Keywords:
AI deployment
edge computing
optical communication
Directly Modulated Lasers
neural network inference
Journal
O
IF:
5
Papers:
1.9K
Citations:
3.5W