Neural network training device, system and method

1. A device, comprising:

image generation circuitry, which, in operation, generates a digital image representation of a wafer defect map (WDM); and

convolutional-neural-network (CNN) circuitry, which, in operation, generates a defect classification associated with the WDM based on the digital image representation of the WDM and a data-driven model generated using training data consisting of an artificial wafer defect digital image (AWDI) data set associating AWDIs with classes of a defined set of classes of wafer defects, wherein images of the AWDI data set are computer-generated using Bezier images.

2. The device of claim 1 wherein the CNN circuitry, in operation,

associates, based on the digital image representation of the WDM and the data-driven model, one or more labels and one or more tags with the WDM which are associated with a defect cause.

3. The device of claim 2 wherein a tag identifies a machine associated with the defect cause.

4. The device of claim 1 wherein the CNN circuitry includes one or more convolutional layers.

5. The device of claim 1 wherein the CNN circuitry includes one or more layers which, in operation, introduce a non-linearity.

6. The device of claim 1 wherein the CNN circuitry includes one or more pooling layers.

7. The device of claim 1 wherein the CNN circuitry includes one or more fully connected layers.

8. The device of claim 1 wherein, in a training mode of operation, the CNN circuitry generates the data-driven model using the AWDI data set.

9. The device of claim 8 , comprising:

artificial image generation circuitry, which, in operation, generates the images of the AWDI data set using;

a graphical user interface;

python language; or

a graphical user interface and python language.

10. The device of claim 8 wherein the generating the data-driven model includes providing layer of inspection information, type of inspection information, or both, to a fully connected layer of the CNN circuitry.

11. The device of claim 8 wherein the AWDI data set includes, for each defined class of the set of classes of wafer defects, a same number N of AWDI images.

12. The device of claim 1 wherein the data driven model associates AWDIs with root causes of wafer defects and the CNN circuitry includes an activation function, which, in operation, generates a label identifying a class of the defined set of classes associated with the WDM and a tag identifying a root cause associated with the WDM.

13. A system, comprising:

one or more memories; and

wafer-defect-map (WDM) classification circuitry coupled to the one or more memories, and which, in operation,

generates a defect classification associated with a WDM based on a digital image representation of the WDM and a data-driven model generated using training data consisting of an artificial wafer defect digital image (AWDI) data set associating AWDIs with classes of a defined set of classes of wafer defects, wherein images of the AWDI data set are computer-generated using Bezier images.

14. The system of claim 13 wherein the WDM classification circuitry includes one or more convolutional layers, one or more pooling layers and one or more fully connected layers.

15. The system of claim 13 wherein, in a training mode of operation, the WDM classification circuitry generates the data-driven model using the AWDI data set.

16. The system of claim 15 , comprising:

artificial image generation circuitry, which, in operation, generates the AWDI data set using;

a graphical user interface;

python language; or

a graphical user interface and python language.

17. The system of claim 15 wherein the AWDI data set includes, for each defined class of the set of classes of wafer defects, a same number N of AWDI images.

18. The system of claim 13 wherein the data driven model associates AWDIs with root causes of wafer defects and the WDM classification circuitry includes an activation function, which, in operation, generates a label identifying a class of the defined set of classes associated with the WDM and a tag identifying a root cause associated with the WDM.

19. The system of claim 13 wherein the WDM classification circuitry, in operation, generates one or more control signals to control a wafer-production system based on defect classifications associated with one or more WDMs.

20. The system of claim 13 , wherein the WDM classification circuitry generates the defect classification associated with a WDM based on layer of inspection information, type of inspection information, or both, provided to a fully connected layer of the WDM classification circuitry.

21. A method, comprising:

generating a digital image representation of a wafer defect map (WDM); and

generating a defect classification associated with the WDM based on the digital image representation of the WDM and a data-driven model generated using training data consisting of an artificial wafer defect digital image (AWDI) data set associating AWDIs with classes of a defined set of classes of wafer defects, wherein images of the AWDI data set are computer-generated using Bezier images.

22. The method of claim 21 , comprising generating the data-driven model using the AWDI data set.

23. The method of claim 22 , comprising:

generating the AWDI data set.

24. The method of claim 23 wherein the AWDI data set includes, for each defined class of the set of classes of wafer defects, a same number N of AWDI images.

25. The method of claim 21 wherein the data driven model associates AWDIs with root causes of wafer defects and the generating the defect classification associated with the WDM comprises generating a label identifying a class of the defined set of classes associated with the WDM and a tag identifying a root cause associated with the WDM.

26. The method of claim 21 , comprising:

generating one or more control signals to control a wafer-production process based on defect classifications associated with one or more WDMs.

27. A non-transitory computer-readable medium having contents which configure a wafer defect map (WDM) classification system to perform a method, the method comprising:

generating a digital image representation of a wafer defect map (WDM); and

generating a defect classification associated with the WDM based on the digital image representation of the WDM and a data-driven model generated using training data consisting of an artificial wafer defect digital image (AWDI) data set associating AWDIs with classes of a defined set of classes of wafer defects, wherein images of the AWDI data set are computer-generated using Bezier images.

28. The non-transitory computer-readable medium of claim 27 , wherein the method comprises generating the data-driven model using the AWDI data set.

29. The non-transitory computer-readable medium of claim 28 , wherein the method comprises generating the AWDI data set.

30. The non-transitory computer-readable medium of claim 27 , wherein the AWDI data set includes, for each defined class of the set of classes of wafer defects, a same number N of AWDI images.

31. The non-transitory computer-readable medium of claim 27 , wherein the contents comprise parameters of the data-driven model.

32. The non-transitory computer-readable medium of claim 27 , wherein the data driven model associates AWDIs with root causes of wafer defects and the generating the defect classification associated with the WDM comprises generating a label identifying a class of the defined set of classes associated with the WDM and a tag identifying a root cause associated with the WDM.

33. The non-transitory computer-readable medium of claim 27 , wherein the method comprises:

generating one or more control signals to control a wafer-production process based on defect classifications associated with one or more WDMs.