Defect characterization of GaN-based materials’ surface using Python
In the modern era, the increasing demand for electronic devices and advancements in technology have led to the growth of the semiconductor industry. With increasing demand for higher performance and compactness, companies and researchers alike have directed their attention to gallium nitride (GaN...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Final Year Project |
Language: | English |
Published: |
Nanyang Technological University
2023
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/167612 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | In the modern era, the increasing demand for electronic devices and advancements in
technology have led to the growth of the semiconductor industry. With increasing demand for
higher performance and compactness, companies and researchers alike have directed their
attention to gallium nitride (GaN) semiconductor. Although GaN is great, it has some reliability
issue. During the epitaxial growth process of GaN-based semiconductor devices such as the
high electron mobility transistors (HEMTs), large number of defects can form on the surface.
These defects can negatively impact the performance of the devices, so it is important to detect
and control the surface morphology.
In the given atomic force microscopy (AFM) images of the GaN epitaxial layer, the defects are
depicted as either small round and dark dots or bright white circles. This difference was caused
by the use of different epitaxial deposition techniques. For this project, the aim is to develop a
program using Python to detect these defects in the AFM images provided and estimate the
defect densities. Neural network was chosen as the preferred method and U-Net architecture
was used to accomplish this task. Data augmentation was also used to supplement the lack of
training samples.
Two models were trained, one for each type of epitaxial deposition technique used. One of the
models achieved relatively good training results while the other did not do as well. Inference
using the trained models were able to get acceptable output masks for the object detection
algorithm to locate the defects. A comparison between the threading dislocation densities
(TDDs) calculated from X-ray diffraction (XRD) data and the defect densities obtained from
the program was performed. The comparison outcome shows that using AFM images along
with the developed program is an acceptable alternative to using XRD and transmission
electron microscopy (TEM). This is beneficial as AFM is more cost-effective as compared to
the other two methods while also requiring less preparation work to complete. Another benefit
is that the program will not have any requirement on the user unlike with the expensive
machines used for XRD and TEM. |
---|