Post-weld-shift compensation strategy with improved U-net edge detection for reducing coupling loss of 25Gbps TOSA

Area of Science:

• Opto-electronic device packaging
• Deep learning applications in engineering
• Optical coupling

Background:

• Post-weld-shift is a critical issue in opto-electronic device packaging, causing significant power attenuation.
• Accurate recognition of weld point features is essential for effective post-weld-shift compensation.
• Existing methods struggle with the precise identification of local weld point features.

Purpose of the Study:

• To propose a novel compensation method for post-weld-shift in opto-electronic devices.
• To enhance the accuracy of weld point feature recognition using deep learning.
• To improve optical power compensation through precise alignment and laser energy modulation.

Main Methods:

• Implementation of a U-net based deep learning architecture, termed MPFG-net.
• Integration of CG-MLP and MIF modules to improve local weld point feature recognition.
• Development of a compensation strategy combining MPFG-net based alignment with laser energy modulation.

Main Results:

• The MPFG-net achieved an exceptional weld-point edge recognition accuracy of 78%.
• The proposed network outperformed leading mainstream networks in image recognition performance.
• The compensation strategy resulted in a 15% improvement in optical power compensation.

Conclusions:

• The MPFG-net architecture effectively addresses the challenge of post-weld-shift in opto-electronic devices.
• Deep learning-based image recognition significantly enhances the precision of weld point alignment.
• The proposed method offers a substantial improvement in optical power compensation for opto-electronic packaging.