Defect size measurement using deep learning methods
A system has detectors configured to receive a beam of light reflected from a wafer. For example, three detectors may be used. Each of the detectors is a different channel. Images from the detectors are combined into a pseudo-color RGB image. A convolutional neural network unit (CNN) can receive the pseudo-color RGB image and determine a size of a defect in the pseudo-color RGB image. The CNN also can classify the defect into a size category.
1. A system comprising:
a light source configured to generate a beam of light;
at least two detectors configured to receive the beam of light reflected from a wafer, wherein each of the detectors is a different channel; and
a processor in electronic communication with the detectors, wherein the processor is configured to:
assign colors to images of each of the channels of the detectors based on an intensity value of the images, wherein the channels include a red channel, a green channel, and a blue channel;
align the images from the detectors with respect to each other whereby the images correspond to a same location on a surface of the wafer; and
combine the images from the detectors into a pseudo-color RGB image using image overlay or image subtraction; and
a convolutional neural network unit configured to:
receive the pseudo-color RGB image from the processor; and
determine a size of a defect in the pseudo-color RGB image using a defect position within each of the channels and a number of pixels that the defect covers in each of the channels, wherein the size of the defect is less than 1 micron.
2. The system of claim 1 , wherein the at least two detectors include three detectors.
3. The system of claim 1 , wherein the convolutional neural network unit is further configured to classify the defect into one of a plurality of size categories.
4. The system of claim 1 , wherein the light source is a laser.
5. The system of claim 1 , wherein the processor is configured to add design images to the pseudo-color RGB image.
6. A method comprising:
receiving a beam of light reflected from a wafer at a plurality of detectors, wherein each of the detectors is a different channel;
assigning, using a processor, colors to images of each of the channels of the detectors based on an intensity value of the images, wherein the channels include a red channel, a green channel, and a blue channel;
aligning, using the processor, the images from the detectors with respect to each other whereby the images correspond to a same location on a surface of the wafer;
combining, using the processor, the images from the detectors into a pseudo-color RGB image using image overlay or image subtraction; and
determining a size of a defect in the pseudo-color RGB image using a convolutional neural network unit that uses a defect position within each of the channels and a number of pixels that the defect covers in each of the channels, wherein the size of the defect is less than 1 micron.
7. The method of claim 6 , wherein the plurality of detectors includes three detectors.
8. The method of claim 6 , further comprising classifying the defect into one of a plurality of size categories using the convolution neural network unit.
9. The method of claim 6 , further comprising generating the beam of light with a laser.
10. The method of claim 6 , further comprising adding, using the processor, design images or mask images to the pseudo-color RGB image.
11. A non-transitory computer-readable storage medium, comprising one or more programs for executing the following steps on one or more computing devices:
assigning colors to images of channels from a plurality of detectors based on an intensity value of the images, wherein each of the detectors is a different one of the channels, and wherein the channels include a red channel, a green channel, and a blue channel;
aligning the images from the detectors with respect to each other whereby the images correspond to a same location on a surface of the wafer;
combining the images of a wafer from a plurality of detectors into a pseudo-color RGB image using image overlay or image subtraction; and
determining a size of a defect in the pseudo-color RGB image using a convolutional neural network unit that uses a defect position within each of the channels and a number of pixels that the defect covers in each of the channels, wherein the size of the defect is less than 1 micron.
12. The non-transitory computer-readable storage medium of claim 11 , wherein the steps further include classifying the defect into one of a plurality of size categories using the convolution neural network unit.
13. The non-transitory computer-readable storage medium of claim 11 , wherein the plurality of detectors includes three detectors.
14. The non-transitory computer-readable storage medium of claim 11 , wherein the steps further include adding design images or mask images to the pseudo-color RGB image.
15. The system of claim 1 , wherein the images from the detectors are combined using image overlay.
16. The system of claim 1 , wherein the images from the detectors are combined using image subtraction.
17. The system of claim 1 , wherein the detectors are non-imaging detectors, and wherein the processor is further configured to generate the images of each of the channels from non-imaging output of the detectors.
18. The system of claim 2 , wherein the beam of light for two of the three detectors has an oblique angle of incidence relative to a surface of the wafer.
19. The system of claim 1 , wherein the processor is configured to add mask images to the pseudo-color RGB image.
20. The system of claim 1 , wherein the convolutional neural network unit configured to determine the size of the defect in the pseudo-color RGB image further using differences in a relative intensity value in each of the channels.