CORRELATING SEM AND OPTICAL IMAGES FOR WAFER NOISE NUISANCE IDENTIFICATION
원문보기
IPC분류정보
국가/구분
United States(US) Patent
공개
국제특허분류(IPC7판)
G01N-021/95
G06T-007/00
G06T-007/174
G06T-007/11
G01N-023/2251
G01N-021/956
출원번호
16886255
(2020-05-28)
공개번호
20200292468
(2020-09-17)
발명자
/ 주소
Zhang, Qiang
Chen, Grace H.
출원인 / 주소
KLA-Tencor Corporation
인용정보
피인용 횟수 :
0인용 특허 :
0
초록▼
Disclosed are apparatus and methods for inspecting a sample. Locations corresponding to candidate defect events on a sample are provided from an inspector operable to acquire optical images from which such candidate defect events are detected at their corresponding locations across the sample. High-
Disclosed are apparatus and methods for inspecting a sample. Locations corresponding to candidate defect events on a sample are provided from an inspector operable to acquire optical images from which such candidate defect events are detected at their corresponding locations across the sample. High-resolution images are acquired from a high-resolution inspector of the candidate defect events at their corresponding locations on the sample. Each of a set of modelled optical images, which have been modeled from a set of the acquired high-resolution images, is correlated with corresponding ones of a set of the acquired optical images, to identify surface noise events, as shown in the set of high-resolution images, as sources for the corresponding candidate events in the set of acquired optical images. Otherwise, a subsurface event is identified as a likely source for a corresponding candidate defect event.
대표청구항▼
1. A method of inspecting a sample, the method comprising: providing a plurality of locations corresponding to a plurality of candidate defect events on a sample from an inspector operable to acquire a plurality of acquired optical images from which such candidate defect events are detected at their
1. A method of inspecting a sample, the method comprising: providing a plurality of locations corresponding to a plurality of candidate defect events on a sample from an inspector operable to acquire a plurality of acquired optical images from which such candidate defect events are detected at their corresponding locations across the sample;acquiring high-resolution images from a high-resolution inspector operable to acquire such high-resolution images of the plurality of candidate defect events at their corresponding locations on the sample; andcorrelating each of a first set of modelled optical images, which have been modeled from a first subset of the acquired high-resolution images, with a corresponding one of a first set of the acquired optical images, to identify a plurality of surface noise events, as shown in the first set of high-resolution images, as sources for the corresponding candidate defect events in the first set of acquired optical images, wherein the correlating of each of the first set of modelled optical images results in identification of the surface noise events, rather than a subsurface defect event, as sources if the corresponding modelled and acquired optical images are substantially identical. 2. The method of claim 1, wherein each candidate event represents a surface defect event, one or more noise events, or a subsurface event present on the sample. 3. The method of claim 1, further comprising: prior to correlating the first set of modelled optical images with their corresponding first set of acquired optical images, analyzing the high-resolution images to classify the candidate events into ambiguous and unambiguous events, wherein each high-resolution image in the first set of high-resolution images is associated with a classified ambiguous event, wherein each unambiguous event is a bridge, break, protrusion, intrusion, or other known defect types, wherein the ambiguous events were unclassifiable as a known defect type. 4. The method of claim 3, further comprising: training a near field (NF) model to model a plurality of NF images from corresponding acquired high-resolution images, wherein the NF model is trained with a set of training high-resolution images and acquired optical images that correspond to unambiguous and classified events;modeling the first set of modelled optical images by modeling a plurality of corresponding NF images from the first set of acquired high-resolution images using the trained NF model and modeling the first set of modelled optical images from the corresponding NF images using a tool model for the inspector. 5. The method of claim 4, wherein the NF model is configured to simulate light reflected and scattered, with a plurality of light characteristic parameters, from a wafer pattern, having a set of pattern characteristic parameters, that is represented in the corresponding high-resolution images, wherein the NF model is trained by: inputting the training high-resolution images into the NF model to model corresponding training NF images based on the light and pattern characteristic metrics;inputting the training NF images that were modelled from the training high-resolution images into the tool model to model corresponding training modelled optical images;correlating the training modelled optical images with their corresponding acquired optical images; andadjusting the light and pattern characteristic parameters and repeating the operations for inputting the training high-resolution images into the NF model, inputting the training NF images into the tool model, and correlating the training modelled optical images until such correlating operation results in a maximum correlation between the training modelled optical images and their corresponding acquired optical images. 6. The method of claim 5, wherein modeling the first set of modelled optical images is performed with respect to the first set of high-resolution images after they have been smoothed to remove noise introduced by the high-resolution inspector and have been binarized by a normalization process, and the correlating of each of the first set of modelled optical images with corresponding acquired optical images is performed after such first modelled optical images are down-sampled so that their resolution and/or size are the same as a resolution and size of the corresponding acquired optical images. 7. The method of claim 5, further comprising: shifting each modelled optical image with respect to its corresponding acquired image by an offset that is determined by aligning a one of the training modelled optical images with a corresponding one of the acquired optical images, wherein the shifting results in one or more noise events in a high-resolution image from the first set of high-resolution images being accurately correlated with a corresponding candidate event in the corresponding acquired image. 8. The method of claim 3, further comprising: after correlating the first set of modelled optical images, determining whether the sample is to be processed further with or without repair or discarded based on review of the high-resolution images, the classified unambiguous events, and the identified noise and subsurface events if any. 9. The method of claim 1, wherein the high-resolution inspector is a scanning electron (SEM) microscope. 10. A high-resolution inspector system for inspecting a sample, comprising at least one processor and memory that are operable for performing the following operations: providing a plurality of locations corresponding to a plurality of candidate defect events on a sample from an inspector operable to acquire a plurality of acquired optical images from which such candidate defect events are detected at their corresponding locations across the sample;acquiring high-resolution images of the plurality of candidate defect events at their corresponding locations on the sample; andcorrelating each of a first set of modelled optical images, which have been modeled from a first subset of the acquired high-resolution images, with a corresponding one of a first set of the acquired optical images, to identify a plurality of surface noise events, as shown in the first set of high-resolution images, as sources for the corresponding candidate defect events in the first set of acquired optical images, wherein the correlating of each of the first set of modelled optical images results in identification of the surface noise events as sources if the corresponding modelled and acquired optical images are substantially identical. 11. The system of claim 10, wherein each candidate event represents a surface defect event, one or more noise events, or a subsurface event present on the sample. 12. The system of claim 10, wherein the at least one processor and memory are further operable for, prior to correlating the first set of modelled optical images with their corresponding first set of acquired optical images, analyzing the high-resolution images to classify the candidate events into ambiguous and unambiguous events, wherein each high-resolution image in the first set of high-resolution images is associated with a classified ambiguous event, wherein each unambiguous event is a bridge, break, protrusion, intrusion, or other known defect types, wherein the ambiguous events were unclassifiable as a known defect type. 13. The system of claim 12, wherein the at least one processor and memory are further operable for: training a near field (NF) model to model a plurality of NF images from corresponding acquired high-resolution images, wherein the NF model is trained with a set of training high-resolution images that correspond to unambiguous and classified events;modeling the first set of modelled optical images by modeling a plurality of corresponding NF images from the first set of acquired high-resolution images using the trained NF model and modeling the first set of modelled optical images from the corresponding NF images using a tool model for the inspector. 14. The system of claim 13, wherein the NF model is configured to simulate light reflected and scattered, with a plurality of light characteristic parameters, from a wafer pattern, having a set of pattern characteristic parameters, that is represented in the corresponding high-resolution images, wherein the NF model is trained by: inputting the training high-resolution images into the NF model to model corresponding training NF images based on the light and pattern characteristic metrics;inputting the training NF images that were modelled from the training high-resolution images into the tool model to model corresponding training modelled optical images;correlating the training modelled optical images with their corresponding acquired optical images; andadjusting the light and pattern characteristic parameters and repeating the operations for inputting the training high-resolution images into the NF model, inputting the training NF images into the tool model, and correlating the training modelled optical images until such correlating operation results in a maximum correlation between the training modelled optical images and their corresponding acquired optical images. 15. The system of claim 13, wherein modeling the first set of modelled optical images is performed with respect to the first set of high-resolution images after they have been smoothed to remove noise introduced by the high-resolution inspector and have been binarized by a normalization process, and the correlating of each of the first set of modelled optical images with corresponding acquired optical images is performed after such first modelled optical images are down-sampled so that their resolution and/or size are the same as a resolution and size of the corresponding acquired optical images. 16. The system of claim 13, wherein the at least one processor and memory are further operable for: shifting each modelled optical image with respect to its corresponding acquired image by an offset that is determined by aligning a one of the training modelled optical images with a corresponding one of the acquired optical images, wherein the shifting results in one or more noise events in a high-resolution image from the first set of high-resolution images being accurately correlated with a corresponding candidate event in the corresponding acquired image. 17. The system of claim 12, wherein the at least one processor and memory are further operable for: after correlating the first set of modelled optical images, determining whether the sample is to be processed further with or without repair or discarded based on review of the high-resolution images, the classified unambiguous events, and the identified noise and subsurface events if any. 18. The system of claim 10, comprising a SEM system.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.