WO2005013199A1

WO2005013199A1 - Color variation compensation algorithm for bright field wafer inspection

Info

Publication number: WO2005013199A1
Application number: PCT/US2004/023571
Authority: WO
Inventors: Avishai Bartov
Original assignee: Applied Materials Israel, Ltd.; Applied Materials, Inc.
Priority date: 2003-07-25
Filing date: 2004-07-21
Publication date: 2005-02-10

Abstract

Gray level variations between images of dies produced using a bright filed inspection system may be measured to produce measured statistics; and one or more of those measured statistics applied as a correction factor for a difference image produced by a comparison of two of the images of dies. The measured statistics may be average gray level variations across one or more of the images of dies

Description

COLOR VARIATION COMPENSATION ALGORITHM FOR BRIGHT FIELD WAFER INSPECTION

RELATED APPLICATIONS

[0001] The present application incorporates by reference and claims the priority benefit of U.S. provisional Application 60/490,323, filed 25 July 2003.

BACKGROUND

[0002] Chemical mechanical planarization (CMP) is a semiconductor manufacturing process used primarily in connection with the polishing of inter-metal dielectric layers and other steps in the fabrication of integrated circuits. Because CMP can introduce defects in a semiconductor wafer these processes must be closely monitored using sophisticated wafer inspection tools, for example inspection systems based on bright-field digital image-processing technology. Such bright field inspection devices are widely regarded as the most sensitive and comprehensive defect inspection tools in the industry and have always provided a high rate of defect capture. [0003] Nevertheless, the sensitivity of bright field inspection tools is limited, in part because of oxide film thickness variations caused during the CMP process. Non-uniform film thicknesses interfere with reflected light, resulting in color variations in the images obtained during wafer inspections. In gray-scale representations, which are used by bright field inspection tools for die- to-die processing, this color variation translates into a variation in gray-scale levels, creating "noise" that can reduce the sensitivity of the inspection system.

SUMMARY OF THE INVENTION

[0004] Gray level variations between images of dies produced using a bright filed inspection system may be measured to produce measured statistics; and one or more of those measured statistics applied as a correction factor for a difference image produced by a comparison of two of the images of dies. The measured statistics may be average gray level variations across one or more of the images of dies.

DETAILED DESCRIPTION

[0005] Described herein are methods for compensating for color variation in images obtained by bright field inspection systems. By compensating for such color variations, sensitivities of the inspection tools can be improved. The examples discussed below, however, should not be read as limiting the scope of the present invention and should be understood as being merely illustrative thereof.

[0006] As indicated above, CMP processes cause oxide thickness variations across a wafer and those variations appear as image color (or gray level (GL)) variations when inspected by a bright field inspection system. This creates a problem when conventional die-to-die detection schemes are employed in order to detect defect.

[0007] Die-to-die detection is based on a comparison of scanned images of adjacent dies. Essentially, images of consecutive dies are subtracted from one another and the differences examined. Any differences between the images are regarded as potential defects (e.g., due to improper processing of the wafer from which the dies were sliced) and thresholding is often used in order to permit only a reasonable number of false positives. The problem arises in that if color variations due to oxide (or other) level thicknesses are present in the images, the thresholds must be set rather high (i.e., the sensitivity of the detector must be lowered) in order to ensure that the number of false positives does not unduly compromise the throughput of the system. [0008] One can interpret the above-described detection process as subtracting the gray level value of a pixel in a current die image from its best-known predictor. In the conventional die-to- die inspection process, the predictor is simply the corresponding gray level of the previous die image; however, color variations tend to deteriorate the quality of this gray level predictor. For example, suppose that gray level variations of some specific layer vary in range of +/-10 GL. Using this predictor then will result in adding 10 GL to the threshold of the detector in order to accommodate the color variations at the selected rate of false positives. [0009] It should thus be apparent that improving the predictor for the die-to-die comparison process would result in lower useable thresholds in a bright field inspection system allowing for improved defect detection due to improved sensitivity. In one embodiment of the present invention, the predictor is "improved" through the use of global statistics regarding a number of images used in a bright field inspection process. That is, in an embodiment of the present invention variation in gray levels, on average, is measured across a number of images of dies and this measured value is then subtracted from each difference image involved in a die-to-die comparison. This procedure provides good results because the color variation described above is a relatively "slow" phenomenon; that is, in practice actual GL variation is typically less than a few GLs in the range of a few nanometers on a wafer. Note, however, that pixels near an edge ⁽ may need to be treated separately from pixels not near an edge; also, pixel mappings need not necessarily be one-to-one.

[0010] In accordance with the present invention then, the average mapping of gray levels between two successive dies in a slice denoted as Current and Previous can be measured. The measurement will generate the following statistics: a. Diff_AvPre[256]: Diff_AvPre[i] = The average Difference Image value when Previous image GL = i. b. Diff_StdPre[256] : Diff_StdPre[i] = The Standard variation of the Difference Image value when Previous image GL = i. c. Diff_AvCur[256]: Diff_AvCur[i] = The average Difference Image value when Current image GL = i. d. Diff_StdCur[256] : Diff_StdCur [i] = The Standard variation of the Difference Image value when Current image GL = i; where Difference Image = Current - Previous. Statistics may be gathered (e.g., in parallel) for edgy (i.e., pixels near an edge) and non-edgy (i.e., pixels not near an edge) parts (e.g., 8 vectors in total).

[0011] The correction process (to improve the predictor) may then make use of the measured statistics. In the case where die-to-die (or image-to-image) mapping is not one-to-one, one might expect the measured "standard" to be higher when a histogram pick on one die is mapped to more than one on the other die. For example, suppose that both gray levels (GLs) 10 and 40 of the Previous image are mapped to gray level 20 in the current image. In this case we expect the value Diff_StdCur[20] to be significantly larger than Diff_StdPre[10] or Diff_StdPre[40]. The color correction will modify the difference image accordingly.

[0012] By way of example, suppose Diffljp] = Currfp] - Prev[p], where Curr[p] and Prevfp] are the gray levels of corresponding pixels in the "p" position. a. Apply an edge detector on the current image (e.g., the same edge detector that was used for color variation measurement) to classify the pixel as edge or non- edge and use the correction vectors accordingly. b. Compare Diff_StdPre[Pre[p]] with Diff_StdCur[Cur[p]], use the smaller value to chose the correction vector. For example: if Diff_StdCur[Cur[p]] > Diff_StdPre[Pre[p]], use Diff_AvPre[Pre[p]] for correction: Diff[p] = Difffp] - Diff AvPre[Pre[p]]. [0013] Thus, methods for compensating for color variation in images obtained by bright field inspection systems have been described. The examples discussed above should not be read as limiting the scope of the present invention, which should be measured only in terms of the following claims.

Claims

CLAIMSWhat is claimed is:

1. A method, comprising measuring gray level variations between images of dies produced using a bright filed inspection system to produce measured statistics; and applying one or more of said measured statistics as a correction factor for a difference image produced by a comparison of two of the images of dies.

2. The method of claim 1, wherein the measured statistics comprise average gray level variations across one or more of the images of dies.