DFB Laser Chip Defect Detection Based on Successive Subspace Learning

2020 10th Annual Computing and Communication Workshop and Conference (CCWC)
Date of Conference: 06-08 January 2020
Conference Location: Las Vegas, NV, USA

Transcript

1. 1 DFB Laser Chip Defect Detection Based on Successive Subspace Learning Tuo Liu, Xiaobin Zhang, Yuxi Wang, Yen-Ting Pan Yuanjie Semiconductor Technology Dennis Hou, Rutgers University Janpu Hou, Applied Data Research January 6, 2020
2. 2 After the acceptance criteria been defined between distributed feedback (DFB) laser chip manufacturer and laser module assembly house, there is a need of reliable tool for quality engineers to carry out the acceptance sampling plan. There are variations of DFB lasers with different wavelength, power, threshold, operating temperature requirement etc. To make acceptance sampling reliable for semiconductor laser chip producer who need to control the specificity and for laser module assembly factories who need to control the sensitivity, a classifier with on-device training capability is needed. In this paper, an embedded automated distributed feedback (DFB) lasers chip defect classifier based on successive subspace learning (SSL) is developed for such application to classify active, surface, edge and coating defects generated by manufacturing processes. The designed classifier was verified with accuracy=97.43%, precision=96.80%, sensitivity=91.32%, specificity=97.66% and F1 score=0.968. Abstract
3. 3 Outlines: Introduction • Laser Chips and Laser Modules • Quality control of Contract Manufacturing Signal Methodology • Successive Subspace Learning • Ensemble Classifier Results and Discussion
5. 5 Laser Chips and Laser Modules
6. 6 Wafer Fab Chip Fab “partial” outsourcing Consignment Manufacturers third-party partner Contract Manufacturers In-house manufacturing AQL- 1 AQL- 2 AQL- 3 Laser Chip Defect Inspection for ACCEPTANCE SAMPLING PLANS
9. 9 Outlines: Introduction • Laser Chips and Laser Modules • Quality control of Contract Manufacturing Signal Methodology • Successive Subspace Learning • Ensemble Classifier Results and Discussion
10. 10 A Standard CNN Architecture • Convolution • Max pool • Non-linear Activation Filter Weights to be Learned from D ata LeCun
11. 11 A Saak Transform Architecture 64x64x3 32x32x8 16x16x12 8x8x16 4x4x24 2x2x32 • Saak Transform: Convolution • Remove low variance: Max pool • ReLU + Augmented Kernels : Non- linear Activation Data-driven Filter Weights
14. 14 Outlines: Introduction • Laser Chips and Laser Modules • Quality control of Contract Manufacturing Signal Methodology • Successive Subspace Learning • Ensemble Classifier Results and Discussion
16. 16 SVM Classifier
17. 17 RF Classifier
18. 18 KNN Classifier
19. 19 Global Model Local Model Local Model Local Model M … M1 M2 Mk M = 1 𝑘 σ𝑛=1 𝑘 𝑴𝑘
20. 20 Ensemble Classifier
21. 21 Summary: Laser Chip Defect Detector Simple Model • Semi-supervised Subspace Learning applied to automatic optical inspection of laser chips Simple Hardware • Implemented and tested on Low cost ARM platform Easy to Use • Point and click