.png){kind=icon}
.png){kind=icon}
Return
Integrated quantum dot lasers for parallelized photonic edge computing
.png){kind=icon}
PRE
AI
Z
H
W
O
W
L
S
X
L
W
S
B
W
DOI：10.1117/1.AP.8.2.026017.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Scalable photonic computing for edge artificial intelligence remains limited by the inefficiency of integrated light sources. We report a photonic computing unit architecture powered by a quantum dot mode-locked laser (QD-MLL) heterogeneously integrated on a foundry 300 mm silicon photonics platform. Our approach bridges experimentally validated device- and link-level characteristics with system-level evaluations, providing a measurement-anchored assessment of photonic computing unit operation and scalability. Compared with current hybrid light sources, the isolator-free QD-MLL reduces front-end optical loss by 2 to 6 dB. When combined with a codesigned computing architecture, the link-level evaluation indicates over 10 dB improvement in signal-to-noise ratio and more than 50% reduction in thermally induced weight drift. Under realistic laser structure and power constraints, the system projects 1.7 & times; higher effective scalability, corresponding to >40% enhancement in computational density and similar to 30% gain in energy efficiency. To further enable robust deployment at the edge, we introduce a hardware-aware training framework that strengthens inference reliability while minimizing hardware overhead. Collectively, we establish a cross-layer framework-from device to architecture to algorithm-toward a unified hardware-software strategy for scalable photonic edge computing.
Keywords:
quantum dot laser
photonic computing
heterogeneous integration
edge device
Journal
IF:
18.8
Papers:
961
Citations:
3.6K
Organization
Cited Papers