HP INCFollow

Abstract

The wafer defects are outcomes of wafer manufacturing process. It is important for the

business to detect them at a very early stages, so that the corrective action can be taken

immediately. Any defect which are undetected will pass through the further steps of manufacturing

and would add additional production cost. Currently the wafer defect detection

is being done manually with help of operators. Six to Seven operators are deployed, to do

a visual inspection. As the volume of images that need to be examined are much higher

(order of 1000s per day), this process is prone to human errors. Furthermore, addition of

human resources incurring extra cost to the manufacturing process. Furthermore, the wafer

defect detection process is challenging as the defect span is in the order of microns. The

image acquisition at this zoom level could introduce grains and focusing artifacts. This

makes the detecting task complicated for the traditional computer vision algorithms.

We have developed a deep neural network (Hybrid network, Resnet and Modified Yolo V5)

based algorithm to detect the defective Micro Fluidic Devices and localize the defect the

within the wafer. Our network will first detect the defective Micro Fluidic Devices, and then

classify the detected one in to one of the three categories (For the POC, we addressed only

good, contamination and blocked device categories). Our network provided a precision close

to 99%, and a Recall close to 97.6%. Our data sample set consists of 500 images. We

developed an open CV based labeling algorithm to label ground truth. Our architecture

consists of two independent networks, one for detection, and second one for reviewing the

outcome of first network. Both networks were independently trained with 500 samples

(Each sample contains approximately 10 MicroFluidic Device), for 1000 epochs and the

model which gave best results were extracted.

Keywords: Defect Detection, Defect Classification, Object Detection, Defect Binning

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution-Share Alike 4.0 License.

Download

Share

COinS