Rethinking Fully Convolutional Networks for the Analysis of Photoluminescence Wafer Images
The manufacturing of light-emitting diodes is a complex semiconductor-manufacturing process, interspersed with different measurements. Among the employed measurements, photoluminescence imaging has several advantages, namely being a non-destructive, fast and thus cost-effective measurement. On a pho...
Saved in:
Main Authors | , , |
---|---|
Format | Journal Article |
Language | English |
Published |
01.03.2020
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | The manufacturing of light-emitting diodes is a complex
semiconductor-manufacturing process, interspersed with different measurements.
Among the employed measurements, photoluminescence imaging has several
advantages, namely being a non-destructive, fast and thus cost-effective
measurement. On a photoluminescence measurement image of an LED wafer, every
pixel corresponds to an LED chip's brightness after photo-excitation, revealing
chip performance information. However, generating a chip-fine defect map of the
LED wafer, based on photoluminescence images, proves challenging for multiple
reasons: on the one hand, the measured brightness values vary from image to
image, in addition to local spots of differing brightness. On the other hand,
certain defect structures may assume multiple shapes, sizes and brightness
gradients, where salient brightness values may correspond to defective LED
chips, measurement artefacts or non-defective structures. In this work, we
revisit the creation of chip-fine defect maps using fully convolutional
networks and show that the problem of segmenting objects at multiple scales can
be improved by the incorporation of densely connected convolutional blocks and
atrous spatial pyramid pooling modules. We also share implementation details
and our experiences with training networks with small datasets of measurement
images. The proposed architecture significantly improves the segmentation
accuracy of highly variable defect structures over our previous version. |
---|---|
DOI: | 10.48550/arxiv.2003.00594 |