Using machine learning with optical profilometry for GaN wafer screening

Abstract

To improve the manufacturing process of GaN wafers, inexpensive wafer screening techniques are required to both provide feedback to the manufacturing process and prevent fabrication on low quality or defective wafers, thus reducing costs resulting from wasted processing effort. Many of the wafer scale characterization techniques—including optical profilometry—produce difficult to interpret results, while models using classical programming techniques require laborious translation of the human-generated data interpretation methodology. Alternatively, machine learning techniques are effective at producing such models if sufficient data is available. For this research project, we fabricated over 6000 vertical PiN GaN diodes across 10 wafers. Using low resolution wafer scale optical profilometry data taken before fabrication, we successfully trained four different machine learning models. All models predict device pass and fail with 70–75% accuracy, and the wafer yield can be predicted within 15% error on the majority of wafers.

Breakdown characteristics analysis of kV-class vertical GaN PIN rectifiers by wafer-level sub-bandgap photoluminescence mapping

“…Various non-destructive methods have been developed to characterize substrate imperfections, such as Raman spectroscopy, 11,12 Nomarski imaging, 7 and optical profilometry. 13,14 Photoluminescence (PL) mapping as a contactless, non-destructive, high-sensitivity inspection method has been implemented in recent years for the classification of a wide range of substrate-and epitaxy-related defects in GaN devices. 15,16 Compared to other characterization methods, PL mapping can detect signals related to radiative microstructures of the sample and perform quantitative analysis of specific defect types over a long range.…”

Detecting defects that reduce breakdown voltage using machine learning and optical profilometry

“…ML has also been used to predict the quality of GaN Ohmic contacts from the fabrication recipe 9 . Previous work has shown that we can predict low voltage properties of vertical GaN diodes with over 75% accuracy from optical profilometry data using both convolutional neural networks 10 and simpler models 11 (logistic regression 12 , decision tree 13 , and K nearest neighbor 14 (KNN)). This work expands on this by using high voltage breakdown as the quality metric and discusses the possible defects causing reduced diode performance.…”

“…Many types of defects affect the crystal structure in GaN, with a significant proportion discernable via observable morphological defects on the surface and small defects in the substrate that often extend through the epitaxial layers during growth. Hence, optical profilometry provides a data-rich input for developing ML models for wafer screening 11,[19][20][21][22] . This study involved the collection of optical profilometry data from seven GaN PiN diode wafers by employing a combination of data pre-processing and supervised machine learning algorithms to develop predictive models for different targeted performance metrics.…”

“…The input data for this program consisted of four parameters: the root mean squared (RMS) roughness, the number of bumps and pits, the size of the devices, and the distance from the center of the wafer. The first two parameters are extracted from optical profilometry in the same manner as they were in our previous research 11,23 .…”

“…1 for diagram): The p-layer was isolated by N implantation. A guard ring structure was fabricated using the design optimized in our previous research 11,25,26 . An outer trench isolation layer was produced using Cl 2 etching.…”