JP4060558B2 - Defect inspection method and apparatus - Google Patents

Description

以上に述べたように、本発明では２枚の画像（検出画像と参照画像）を比較し、その差分値から欠陥を検出する検査において、明るさむらや位置ずれにより画像間の差が大きくなる領域に対し、あらかじめ明るさの合わせ込みを行う。また、欠陥としての特徴をもつ領域では、明るさの合わせ込みを弱くする、もしくは明るさの合わせ込みを行わない。図１２（ｄ）は、図３で正しい位置ずれ量が検出されなかったときの本発明による明るさ合わせ込み後の差分値の波形である。図１３（ｄ）は、図４で明るさむらがあったときの本発明による明るさ合わせ込み後の差分値の波形である。どちらも本来、欠陥として検出したくない領域に対して明るさの合わせ込みが行われ、差分値が小さくなる。これに対し、欠陥部分では明るさの合わせ込みを弱くしているため、差分値はさほど小さくならない。このため、従来、しきい値を全領域に対しＴＨ２に設定する、もしくはＴＨとＴＨ２の２つのしきい値を設定して虚報の発生を避けていたが、本発明により、感度を落とさずに位置ずれや明るさむらによる虚報の発生を避け、かつ、唯一の低しきい値ＴＨ３での高感度検査と容易な感度調整を可能とする。
【００４８】
更に、本発明によれば、光学式外観検査装置では通常、検出感度１００ｎｍが達成可能となる。更に本発明を用いれば、画質が良好な場合には５０ｎｍの検出感度を出すことも可能である。
【００４９】
以上、本発明の一実施例を、半導体ウェハを対象とした光学式外観検査装置における比較検査画像を例にとって説明したが、電子線式パターン検査やＤＵＶを光源とした外観検査にも適用可能である。この場合、検出感度３０〜７０ｎｍが達成可能である。また、被検査対象は、半導体ウェハに限られるわけではなく、画像の比較により欠陥検出が行われているものであれば、例えばＴＦＴ基板、ホトマスク、プリント板などでも適用可能である。
【００５０】
【発明の効果】
以上に説明したごとく本発明によれば、特に位置ずれや画像歪みによる虚報の発生しやすい高コントラスト部分や明るさむらに対して明るさの合わせ込みを行うことにより虚報の発生を低減することができる効果を奏する。
【００５１】
また、本発明によれば、明るさの合わせ込みを行うことにより結果として低しきい値の設定が可能となり、高感度検査を実現することができる効果を奏する。
【００５２】
また、本発明によれば、唯一のしきい値設定で虚報の発生低減と欠陥の検出が両立し、感度の調整が容易にできる効果を奏する。
【００５３】
また、本発明によれば、光学式外観検査装置における比較検査に適用することにより、検出感度５０ｎｍを達成できる。また、電子線式パターン検査やＤＵＶを光源とした外観検査に適用することにより、検出感度３０〜７０ｎｍが達成可能である。
【図面の簡単な説明】
【図１】被検査対象の検出画像とその時の輝度波形の一例を示す図である。
【図２】位置ずれ量が正しく算出された時の従来のしきい値設定方法の一例を示す図である。
【図３】位置ずれ量が誤って算出された時の従来のしきい値設定方法の一例を示す図である。
【図４】比較チップ間に明るさむらがあった時の被検査対象の検出画像と従来のしきい値設定方法の一例を示す図である。
【図５】本発明に係る欠陥検査装置の一実施例を示す構成図である。
【図６】被検査対象となる半導体ウェハを示す図である。
【図７】図５に示す画像処理部の画像比較部における処理フローを示す図である。
【図８】本発明に係る各画素での特徴量演算である着目画素（ｉ，ｊ）でのコントラスト演算方法の一実施例を説明する図である。
【図９】本発明に係る各画素での特徴量演算である着目画素（ｉ，ｊ）でのコントラスト演算方法の他の実施例を説明する図である。
【図１０】本発明に係る散布図からの近似直線演算の一実施例を説明するための図である。
【図１１】本発明に係る画素単位位置合わせ後の検出画像ｆと参照画像ｇとの位置関係を示す図である。
【図１２】本発明に係る位置ずれ量が誤って算出された時のしきい値設定方法の一実施例を説明するための図である。
【図１３】本発明に係る比較チップ間に明るさむらがあった時のしきい値設定方法の一実施例を説明するための図である。
【図１４】本発明に係る散布図の直線近似による明るさの合わせ込みの挙動の一実施例を説明するための図である。
【図１５】本発明に係る散布図のスリム化の一実施例を説明するための図である。
【図１６】本発明に係る散布図のスリム化の別の実施例を説明するための図である。
【図１７】本発明に係る頻度によるカテゴリ判別方法の一実施例を説明するための図である。
【図１８】本発明に係る頻度しきい値の演算方法の一実施例を説明するための図である。
【図１９】本発明に係る正常カテゴリと非正常カテゴリの補正の違いを示す図である。
【図２０】本発明に係るコントラスト毎の散布図とコントラストによる直線近似の実施例を説明するための図である。
【図２１】本発明に係るコントラストによる頻度しきい値設定方法の一実施例を説明するための図である。
【符号の説明】
１１…検出画像、１２…参照画像、５１…試料、５２…ステージ、５３…検出部、５０１…光源、５０２…照明光学系、５０３…対物レンズ、５０４…イメージセンサ、５５…画像処理部、５４…ＡＤ変換部、５０５…前処理部、５０６…遅延メモリ、５０７…位置ずれ量検出部、５０８…画像比較部、５０９…特徴抽出部、５６…全体制御部、５１０…ユーザインターフェース部、５１１…記憶装置、５１２…メカニカルコントローラ、１０１…近似直線。[0001]
BACKGROUND OF THE INVENTION
The present invention compares images of objects such as semiconductor wafers, TFTs, photomasks, etc. obtained using light or laser with reference images stored in advance, and inspects fine pattern defects and foreign materials from the differences. The present invention relates to a defect inspection method and apparatus therefor. In particular, the present invention relates to a defect inspection method and apparatus suitable for visual inspection of a semiconductor wafer.
[0002]
[Prior art]
As a conventional technique for detecting a defect by comparing a detection image to be inspected with a reference image, a method described in Japanese Patent Laid-Open No. 05-264467 is known.
[0003]
In this method, a sample to be inspected in which repetitive patterns are regularly arranged is sequentially imaged by a line sensor, compared with an image with a time delay corresponding to the repetitive pattern pitch, and the mismatched portion is detected as a pattern defect. . However, in reality, there are stage vibrations, tilt of the target, etc., and the positions of the two images are not necessarily aligned. Therefore, the image captured by the sensor and the image delayed by the repeated pattern pitch It is necessary to determine the amount of displacement. Then, after aligning the two images based on the obtained misregistration amount, the difference between the images is taken, and when the difference is larger than the prescribed threshold value, it is judged as defective, and when it is smaller, it is judged as non-defect. judge.
[0004]
[Problems to be solved by the invention]
As described above, in the conventional technique, the defect detection is performed by the procedure of the positional deviation detection of the detected image to be inspected and the reference image, the alignment, the image comparison, that is, the difference calculation between the two images, and the defect extraction procedure. If the correct amount of misalignment is not found when misalignment is detected, alignment at an incorrect position is performed. In this case, in the image to be aligned, the difference in the image to be aligned does not increase in an area where the amount of change in brightness is small, but the difference increases in an area where the amount of change in brightness is large. For example, 11 in FIG. 1 is an example of a detected image, 12 is an example of a reference image, 1a is a uniformly bright background region, and 1b is a region having a dark pattern on a bright background. The detected image 11 has a defect 1c. In the image of this example, the waveform of the luminance value at the position 1D-1D ′ is as shown in FIG. Here, when the amount of misalignment between 11 and 12 is obtained correctly, the difference image after the alignment between 11 and 12 is as shown in FIG. The difference image is an image displayed as a difference in density according to a difference value between corresponding positions of the detected image and the reference image. If a portion where the difference value is equal to or greater than a specific threshold value TH is defined as a defect, only the defect 1c in the detected image 11 is detected in FIG. However, when the misregistration amounts of 11 and 12 are calculated incorrectly, the difference image after alignment is as shown in FIG. 3, and in an area where the change in luminance value is large, such as the edge of the pattern in the area 1b, Even with a slight positional deviation, the difference value becomes large and is detected as a defect. This should not be detected as a defect. In other words, it is a false report. Conventionally, the threshold value TH is increased as one method for avoiding the generation of the false alarm as shown in FIG. This lowers the sensitivity, and a defect having a difference value less than or equal to the edge portion cannot be detected. As another method, the threshold value TH is set high in a high-contrast portion where false information is likely to occur, and the threshold value TH is set low in a low-contrast portion where false information is unlikely to occur. Sensitivity adjustment becomes complicated.
[0005]
Further, when the object to be inspected is a semiconductor wafer, even if the positional deviation amount is correctly obtained, if there is a difference in the film thickness within the wafer, as shown in FIGS. A difference in brightness occurs in the same pattern with the reference image 12, and the difference value becomes large as in 4c. This is also false information, and the threshold value TH must be increased in order to prevent detection. Alternatively, threshold values must be set separately for areas with uneven brightness and areas without brightness.
[0006]
An object of the present invention is to solve such problems of the prior art, in a comparison inspection in which a detected image of an inspection object is compared with a reference image and a defect is detected from the difference, color unevenness, positional deviation, image distortion It is an object of the present invention to provide a defect inspection method and an apparatus for realizing high-sensitivity defect inspection while reducing the false information caused by the above.
[0007]
Another object of the present invention is to provide a defect inspection method that realizes high-sensitivity defect inspection by reducing false information due to uneven brightness of a pattern caused by a difference in film thickness in inspection for a semiconductor wafer. To provide an apparatus.
[0008]
[Means for Solving the Problems]
In order to achieve the above object, the present invention is characterized in that brightness is adjusted at a non-defect portion in an inspection in which a defect is detected from a difference between two images by comparison. One effective configuration is characterized by detecting a defect with high sensitivity at a low threshold even if there is a minute positional shift. As another configuration, when the inspection target is a semiconductor wafer, even if a difference in brightness occurs in the same pattern between images due to a difference in film thickness in the wafer, the brightness is adjusted by adjusting the brightness in advance. Regardless of the presence or absence of unevenness, a high-sensitivity defect inspection is performed with a low threshold value.
[0009]
Further, in an inspection in which a portion where the difference in brightness between two images is equal to or greater than a specific threshold is detected as a defect, a correction that reduces the difference between the detected image to be inspected and the reference image is applied to the image contrast. It is performed according to the feature. As an effective configuration, even if there is a minute misalignment or image distortion, the difference becomes large, and the brightness between images is reduced so that the difference is reduced in advance in high-contrast areas where false detections (false reports) are likely to occur. In addition, in the final defect detection process, a high-sensitivity defect inspection is performed with a low threshold.
[0010]
By providing these configurations, it is possible to realize a high-sensitivity defect inspection method and defect inspection apparatus that do not generate false information even if the threshold value is low for all inspection target regions.
[0011]
Further, the present invention provides a positional deviation amount detection process for detecting a positional deviation amount between a detected image to be inspected and a reference image serving as a reference, and the detection based on the positional deviation amount calculated in the positional deviation amount detection process. In each of the alignment process of performing the alignment process of the image and the reference image on a pixel basis, and the detected image and the reference image aligned in the alignment process, the feature amount of each pixel is calculated. A feature amount calculation process and an image for decomposing the image into a plurality of categories according to a correlation at a corresponding pixel in a feature amount at each pixel calculated in each image in the feature amount calculation step of each thumbtack. A category decomposition process, a correction coefficient calculation process for calculating a correction coefficient for matching the brightness of the detected image and the reference image for each category decomposed in the category decomposition process of the image, and a calculation in the correction coefficient calculation process Brightness correction process in which brightness is adjusted by correcting the luminance value of each pixel in at least one image using the correction coefficient for each category, and brightness adjustment is performed in the brightness correction process. A defect inspection method comprising: a defect detection step of comparing a detected image and a reference image performed to calculate a difference value and detecting a defect or a defect candidate with a predetermined determination threshold value.
[0012]
Further, the present invention is characterized in that, in the correction coefficient calculation process, a correction coefficient for adjusting brightness is calculated using a scatter diagram showing a relationship between a luminance value of a detected image and a luminance value of a reference image. To do.
[0013]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described in detail with reference to FIGS.
[0014]
As an embodiment, taking a defect inspection method in an optical appearance inspection apparatus for a semiconductor wafer as an example, FIG. 5 shows an example of the structure of the apparatus. The appearance inspection apparatus includes a stage 52 on which a sample (inspection object such as a semiconductor wafer) 51 is mounted and moved, a light source 501, and illumination that illuminates the sample 51 as illumination light by collecting light emitted from the light source 501. A detection unit having an optical system 502, an objective lens 503 that forms an optical image obtained by reflection from the sample 51, and an image sensor 504 that receives the formed optical image and converts it into an image signal corresponding to brightness. 53. The image processing unit 55 calculates defect candidates in the wafer as a sample from the image detected by the detection unit 53. The image processing unit 55 includes an AD conversion unit 54 that converts an input signal from the detection unit 53 into a digital signal, a preprocessing unit 505 that performs image correction such as shading correction and dark level correction from the digital signal, and a digital to be compared. A delay memory 506 that stores a signal as a reference image signal, and a positional deviation amount detection unit 507 that detects a positional deviation amount between the digital signal (detected image signal) detected by the detection unit 53 and the reference image signal of the delay memory 506. Using the calculated misregistration amount, the misregistration between the detected image signal and the reference image signal is corrected and the two image signals are compared, and a portion where the difference value is larger than a specific threshold is determined as a defect candidate. And a feature extraction unit 509 that calculates coordinates and feature amounts of defect candidates. The overall control unit 56 receives a change in inspection parameters (such as a threshold value used in image comparison) from a user, and displays a detected defect information and a user interface unit 510 having an input unit, It comprises a storage device 511 that stores the feature quantities and images of detected defect candidates, and a CPU that performs various controls. The mechanical controller 512 controls the drive of the stage 52 based on a control command from the overall control unit 56. Note that the image processing unit 55, the detection unit 53, and the like are also drive-controlled based on commands from the overall control unit 56.
[0015]
As shown in FIG. 6, a semiconductor wafer 51 to be inspected is regularly arranged with a large number of chips having the same pattern. In the inspection apparatus shown in FIG. 5, the images at the same position of two adjacent chips, for example, the regions 61 and 62 in FIG. 6 are compared, and the difference is detected as a defect.
[0016]
Explaining the operation, the overall control unit 56 controls the mechanical controller 512 to continuously move the semiconductor wafer 51 as a sample by the stage 52. In synchronization with this, images of the chip are sequentially taken from the image sensor 504 of the detection unit 53. The image sensor 504 of the detection unit 53 outputs the input signal to the image processing unit 55.
[0017]
In the image processing unit 55, first, the input analog signal is converted into a digital signal by the AD conversion unit 54, and the preprocessing unit 505 performs shading correction, dark level correction, and the like. In addition, the preprocessing unit 505 performs processing for improving S / N by removing noise and edge enhancement as necessary. However, the image quality improvement processing by this S / N improvement can also be performed later.
[0018]
The misregistration detection unit 507 includes an image signal (detection image signal) of the inspection target chip output from the preprocessing unit 505 and a reference output from the delay memory 506 that is delayed by a time that the stage moves by the chip interval. An image signal, that is, an image signal (reference image signal) of a chip immediately before the inspection target chip is input as a set. These two-chip image signals that are sequentially input in synchronization with the movement of the stage do not become signals at exactly the same location when the stage is vibrated or the wafer set on the stage is tilted. For this reason, the misregistration detection unit 507 calculates the misregistration amount between two images that are continuously input. At this time, the detected image signal and the reference image signal are continuously input, but the calculation of the positional shift amount is sequentially performed for each processing unit with a specific length as one processing unit. The following processing is also performed for each processing unit.
[0019]
The image comparison unit 508 performs image alignment using the calculated misregistration amount, compares the detected image signal with the reference image signal, and outputs a region whose difference value is larger than a specific threshold value as a defect candidate. To do.
[0020]
The feature extraction unit 509 edits each of a plurality of defect candidates, such as deleting a small one as noise or merging neighboring defect candidates as one defect, so that the position, area, size, etc. in the wafer Is calculated and output as the final defect. These pieces of information are stored in the storage device 511. Also, it is presented to the user via the user interface unit 510.
[0021]
Here, when the image comparison unit 508 obtains defect candidates from simple difference values, they are not all true defects. An example of this will be described below.
[0022]
Both the detection image 11 and the reference image 12 shown in FIG. 1 have a dark cross pattern in a uniformly bright background. Further, only the detected image 11 has the defect 1c. The graph of FIG. 1B is a waveform of the luminance value at the position 1D-1D ′ of the detected image 11. As shown in this graph, the image has a region 1a having a high luminance and little change in luminance, that is, a low contrast region 1a, and a region 1b having a large luminance change in the edge portion of the pattern, ie, a high contrast region 1b. FIG. 2A shows an image of a difference value at each corresponding position when a correct misregistration amount is calculated by the misregistration detection unit 507 and alignment is performed on these images, that is, the difference is calculated. This is an image in which a small portion is dark and a large portion is brightly displayed. FIG. 2B shows a waveform of the difference value at the position 1D-1D ′. If a region having a difference value equal to or greater than the threshold value TH is defined as a defect, only the region of the defect 1c is detected in this case.
[0023]
On the other hand, FIGS. 3A and 3B are an image of a difference value and a waveform of the difference value when the misregistration amount is erroneously calculated and incorrect alignment is performed. In the low contrast region where the amount of change in brightness is small as in the region 1a, the difference value does not increase even if the positions of the detected image 11 and the reference image 12 are slightly shifted, but the amount of change in brightness is large as in the region 1b. In the high contrast region, the difference value becomes large even with a slight positional deviation. In the case where an area that is equal to or greater than the threshold value TH is set as a defect candidate, a place where the difference value becomes large due to such a positional deviation is also detected. These should not be detected by nature. Hereinafter, what is not a true defect is described as false information.
[0024]
Conventionally, in order to avoid detection of false information due to misalignment, as one method, the threshold value is increased from TH to TH2 as shown in FIG. 3B, but this cannot detect a defect having a small difference value. In other words, in order to avoid the generation of false information, the inspection is performed with reduced sensitivity. As another method, as shown in FIG. 3C, the threshold value is set to TH2 in the high contrast region and the threshold value is set to TH in the low contrast region. Therefore, the sensitivity adjustment is complicated for the user.
[0025]
Further, when the film thickness of the semiconductor wafer 51 is not uniform, a difference in brightness occurs between the inspection target image and the reference image. For example, three crosses 4a and 4b in FIG. 4 are corresponding patterns in the inspection object image 11 and the reference image 12, but the brightness differs greatly due to the difference in film thickness (hereinafter, brightness In addition, the defect 1c is present only in the detected image 11. In Fig. 4A, a correct misregistration amount is calculated by the misregistration detection unit 507 and alignment is performed on these images. 4B is an image of the difference value at each corresponding position at the time, but the difference value becomes large in a portion with uneven brightness even in the same pattern, FIG. 4B shows the difference at the position 1D-1D ′. If the region where the difference value is greater than or equal to the threshold value TH is determined to be a defect, in addition to the defect 1c, a cross pattern in which the difference value increases due to uneven brightness is also detected. There is false detection by such uneven brightness. In order to avoid this, the threshold value is increased from TH to TH2 and the entire inspection is performed with low sensitivity, or the threshold value is set to TH2 when there is uneven brightness. In a portion where there is no brightness unevenness, a method of performing inspection by adjusting the sensitivity with a plurality of threshold values such as setting the threshold value to TH has been adopted.
[0026]
On the other hand, in the present invention, the image comparison unit 508 performs brightness adjustment between images before calculating the difference between the detected image and the reference image. FIG. 7 shows an embodiment of the outline of processing. First, alignment of the inspection image F (i, j) and the reference image G (i, j) in pixel units is performed based on the amount of displacement calculated by the displacement detection unit 507 (S70). The feature amount of each pixel is calculated for the image after the alignment (S71), and the target image is decomposed into a plurality of pieces according to the feature amount (S72). The group of pixels after decomposition is hereinafter referred to as a category. Next, a correction coefficient for adjusting the brightness (luminance value: gradation value) F (i, j) and G (i, j) of the detected image and the reference image is calculated for each category (73). Using this correction coefficient, brightness adjustment is performed by correcting the luminance value of one image to be close to the luminance value of the other image for each category (74). Then, a difference is calculated at each pixel (i, j) corresponding to the detected image F ′ (i, j) after correction and the reference image G (i, j) (S75), and a threshold value calculated for each pixel. A larger one is extracted as a defect candidate (S76).
[0027]
Next, an example of the brightness adjustment processing procedure from S71 to S74 will be described in detail. Here, the luminance value (brightness) is corrected for one of the detected images F (i, j) and the reference image G (i, j) to be compared. First, the feature amount of each pixel (i, j) is calculated using the detected image F (i, j) and the reference image G (i, j), which are aligned in pixel units in S70 (S71). There are many feature quantities, such as brightness, contrast, brightness difference (darkness difference) between the detected image and the reference image, and frequency. Hereinafter, a case where contrast is used as the feature quantity will be described as an example. First, contrast is calculated for each of all the pixels in the target area (S71). There are various operators in the contrast calculation, and one of them is a range filter. As shown in FIG. 8, the contrast C (i, j) at the coordinate position (i, j) in the target area is expressed as the maximum brightness value Max (A, B, C, D) in the vicinity area. The difference is the minimum value Min (A, B, C, D). When the filter size is 2 × 2, if the luminance value at (i, j) is A and the brightness in the vicinity is B, C, or D, the calculation formula is (Formula 1).
[0028]
C (i, j) = Max (A, B, C, D) −Min (A, B, C, D) (Formula 1)
Also, a percentile filter that reduces the influence of noise may be used instead of a range filter depending on the image quality. Further, the contrast C (i, j) at the coordinate position (i, j) in the target area may be calculated with a secondary differential value. In this case, as shown in FIG. 9, the luminance values A to I in the vicinity of 9 are used and the calculation according to (Expression 2) is performed.
[0029]
Dx = B + H−2 × E,
Dy = D + F-2 × E
C (i, j) = Max (Dx, Dy) (Formula 2)
In addition to this, various calculation methods can be used to obtain the luminance change amount in the vicinity. In this way, the contrast FC (i, j) of each pixel (i, j) in the detected image F (i, j) and the contrast GC of each pixel (i, j) in the reference image G (i, j) (I, j) is calculated, and the average (expression 3) or the difference (expression 4) at the corresponding pixel (i, j) of the detected image F (i, j) and the reference image G (i, j) Alternatively, the contrast of the two images is integrated by taking the larger one (Equation 5), and the contrast value at each pixel (i, j) is uniquely determined. Then, according to the uniquely determined contrast value C (i, j), the image is decomposed into several stages as a contrast category (S72). Hereinafter, what was decomposed in several stages is described as a contrast category. As a result, from the portion where the brightness is uniform (low contrast region) such as the region 1a to the portion where the brightness is sharply changed (high contrast region) such as the pattern edge portion of the region 1b, there are stages as the contrast category. Is broken down.
[0030]
C (i, j) = (FC (i, j) + GC (i, j)) / 2 (Formula 3)
C (i, j) = | FC (i, j) −GC (i, j) | (Formula 4)
C (i, j) = Max (FC (i, j), GC (i, j)) (Formula 5)
Next, a correction coefficient for adjusting the brightness is calculated for each contrast category (S73). An example thereof will be described with reference to FIG. 10. First, for pixels belonging to the same contrast category, a scatter diagram is created with the horizontal axis representing the luminance value X in the detected image and the vertical axis representing the corresponding luminance value Y in the reference image. Then, an approximate straight line is obtained from the scatter diagram. 101 shown in FIG. 10 is an approximate straight line obtained from a scatter diagram of pixels belonging to a certain contrast category. There are various methods for calculating the approximate straight line. As an example, there is a least square approximation (a method for obtaining a straight line that minimizes the sum of the distances from each point). Then, the calculated slope a and Y intercept b of the approximate straight line are correction coefficients for the contrast category (S73).
[0031]
The brightness value F (i, j) of the detected image is corrected using the correction coefficient thus calculated, and brightness adjustment is performed (S74). Actually, assuming that the luminance value of the detected image is F (i, j), the corrected detected image F ′ (i, j) is calculated from the inclination a of the approximate straight line and the Y intercept b using (Equation 6). . This (Equation 6) is brightness adjustment by correcting the luminance value of one image so as to be close to the luminance value of the other image for each contrast category, which is a feature of the present invention.
[0032]
F ′ (i, j) = a × F (i, j) + b (Formula 6)
Then, the difference between the brightness value F ′ (i, j) after correction of the detected image for which the brightness adjustment has been performed and the brightness value G (i, j) of the reference image is represented by the difference D (i) in (Equation 7). , J) (S75), a portion where the difference value is larger than the set threshold value TH is set as a defect candidate (S76).
[0033]
D (i, j) = F ′ (i, j) −G (i, j) (Expression 7)
As shown in (Equation 6), the brightness of the detected image is corrected by rotating the luminance value of each pixel in the scatter diagram (rotation amount: gain) in order to place the scatter diagram on a straight line with an inclination of 45 degrees and a y-intercept of 0. ) And shift (shift amount: offset). FIG. 14 shows the operation. The difference value D (i, j) is equivalent to the distance from the converted straight line. This means that the closer the distance from the straight line is, the smaller the corrected difference value is, that is, a stronger correction is applied. Further, the threshold value TH for defect detection is set outside the scatter diagram after conversion (FIG. 14A). For this reason, in order to set the threshold value TH low and perform a highly sensitive test, it is necessary to make the scatter diagram after conversion slim as shown in FIG.
[0034]
Therefore, in the present invention, the scatter diagram for each contrast category is decomposed into finer categories, and an approximate straight line is obtained for each decomposed small category. One example thereof will be described with reference to FIG. 151 is a scatter diagram in a certain contrast category. This is decomposed with a feature amount different from contrast. Here, an example in which the difference in density between the detected image before correction and the reference image is used as the feature amount will be described. First, for the pixel of interest
Light / dark difference: F (i, j) -G (i, j)
And the scatter diagram area is further decomposed into several levels of shade difference categories in accordance with the set shade difference increments. This is equivalent to slicing the scatter diagram parallel to a straight line having an inclination of 45 degrees, such as 152. Next, approximate straight lines 152-1 to 152-N are calculated for each sliced small category. And as above-mentioned, it correct | amends by the inclination and y intercept of each approximate straight line 152-1-152-N. Since the inclination a and the y-intercept b, which are correction coefficients, are determined depending on which small category the pixel of interest belongs to, the correction coefficients at F (i, j) are a (i, j) and b (i, j). Then, the corrected F ′ (i, j) is expressed by (Equation 8).
[0035]
F ′ (i, j) = a (i, j) × F (i, j) + b (i, j) (Formula 8)
In this way, by further disassembling the image using another feature amount and making the scatter diagram slim, the corrected scatter diagram can also be made slim like 153 and the threshold value TH can be set low. It becomes.
[0036]
Here, the difference in density is used as a feature amount for further slicing the scatter diagram. However, another feature amount such as brightness information of the detected image and the reference image may be used. Further, instead of slicing the scatter diagram, a plurality of straight lines may be calculated simultaneously from one scatter diagram using Hough transform or the like as shown in FIG. Further, the correction coefficient in each category may be calculated not by linear approximation but by polynomial curve approximation.
[0037]
Here, if all the pixels are corrected, even the defect is corrected. Therefore, in the present invention, in order to detect normal areas such as color unevenness with strong brightness adjustment and detect abnormalities (defect candidates) without much correction, another feature amount is used in the scatter diagram. Distinguish between normal and non-normal parts and change the brightness adjustment method. For example, there are various feature quantities for identifying defects and color unevenness, such as texture information of inspection images and reference images, density differences, contrast, frequency, etc. Here, identification using frequency (number of pixels) is an example. explain. As a general feature, the color unevenness (normal region) has a wide frequency due to features such as wide range, repetitive occurrence, and occurrence on the entire surface of a certain pattern. In contrast, defects (non-normal areas) are less frequent. This is used to identify defects and uneven color and change the method of matching brightness. 171 shown in FIG. 17 is a scatter diagram obtained by slicing a scatter diagram of a certain contrast category, and 172 is a graph (hereinafter referred to as frequency distribution) showing the number of pixels for each category after slicing. Is likely to be a defect or uneven color in a normal part and a part with a large contrast. And as the difference in shade increases, the frequency decreases. Therefore, it is identified whether it is normal or abnormal depending on the frequency. For example, as shown at 173, even if the difference in density is the same, if the frequency is greater than the threshold value THN, it is a normal category such as color unevenness, and if the frequency is less than THN, it is set as an abnormal category. The threshold value THN is set in advance as a parameter. In addition, as shown in FIG. 18, the standard deviation σ can be calculated from the frequency distribution, and the place where N × σ can be set as the threshold value.
[0038]
In this way, after the category is determined to be normal or abnormal, correction (brightness adjustment) is performed according to the determination result. This method is shown in FIG. First, for a category determined to be normal, an approximate straight line 101 is obtained using data in the category as in 191. As a result, the brightness in the data in this category is adjusted (192). In addition, straight line approximation is not performed in the category determined to be abnormal, and brightness adjustment is performed using the approximate straight line 101 of the nearest normal category (193). As a result, in the abnormal category, the distance to the approximate straight line increases, so that the brightness adjustment becomes weak, and the corrected differential value remains uncorrected to some extent (194). Further, if it is clear that the defect category, it is possible not to perform brightness adjustment at all. In this case, gain = 1 and offset = 0 may be set for the defect category. In this way, in the present invention, it is possible to detect a minute defect while keeping the noise signal at a low level.
[0039]
Furthermore, these methods can be varied for each contrast category. As an example, as shown in FIG. 20, in the low contrast category, the scatter diagram tends to vary as compared to the high contrast category. In particular, most of the color unevenness between chips caused by the difference in film thickness is included in the low contrast category. However, when the color unevenness is particularly strong, slices into the same small category with all contrasts and normal / non-normal discrimination (201), the brightness cannot be adjusted in the low contrast area and remains as false information There is. In such a case, it is possible to increase the number of approximate lines by finely slicing in the low contrast portion, thereby strengthening the correction more than in the high contrast category and reducing the false alarm (202).
[0040]
As another example, the pattern edge portion is included in the high contrast category as described above, but the difference value becomes large even for a slight misalignment or image distortion in such a place, It is easy to become false information. Therefore, as shown in FIG. 21, in the high contrast region, the frequency threshold value THN for determining normal / unnormal can be set smaller than that in the low contrast category. As a result, the number of categories that are determined to be normal increases, and matching is performed in a wide range.
[0041]
As described above, sensitivity adjustment according to contrast can be performed by several methods such as changing the slice method and the frequency threshold value THN. These may be used properly depending on the sample, or may be used depending on the image quality and the feature amount of the image. Of course, the sensitivity adjustment can be switched on / off. Furthermore, it is possible to constantly monitor the accuracy of misregistration detection, and to increase the brightness adjustment more than usual in a high contrast region only when the correct misregistration amount is not calculated.
[0042]
Here, the type of the contrast calculation filter, the filter size, the number of divisions into the contrast category, the step width, the number of slices due to the difference in gradation, the step width, and the like can be flexibly changed by defining them in the lookup table.
[0043]
As described above, the example in which the image is decomposed based on the contrast and the contrast difference and the scatter diagram is slimmed has been described. Alternatively, image decomposition may be performed based on a variance value of brightness. In short, it is within the scope of the present invention if the image can be decomposed for each region having the same characteristics and the scatter diagram can be made slim.
[0044]
Moreover, although the example which decomposes | disassembles an image using two feature-values and changes the degree of fitting using frequency further was shown, you may slim down a scatter diagram from three or more feature-values. For example, as described above, in a high-contrast region such as a pattern edge, even if there is a slight misalignment, the shading difference becomes large. In other words, the scatter diagram expands when a positional shift occurs. Therefore, the spread of the scatter diagram data is checked, and the amount of positional deviation that minimizes the spread is calculated in subpixel units, and the brightness is adjusted using this scatter diagram. That is, optimization is performed so that the comparative image can be inspected with the highest sensitivity by simultaneously performing the brightness adjustment and the position shift detection in units of sub-pixels.
[0045]
Hereinafter, the positional deviation amount detection is naturally performed in two dimensions, but for the sake of simplicity, the case where the positional deviation amount detection is performed in one dimension (X-axis direction) will be described. As shown in FIG. 11, it is assumed that two images aligned in pixel units are shifted by α in sub-pixel units (−0.5 <α <0.5 pixels). At this time, in order to align the detected image f and the reference image g by ½ × α and perform alignment in units of sub-pixels, an image F (x shown in (Expression 9) and (Expression 10) by linear interpolation is used. ), G (x).
[0046]
F (x) = f (x) + α / 2 · (f (x) −f (x−1)) (Equation 9)
G (x) = g (x) + α / 2 · (g (x + 1) −g (x)) (Formula 10)
The detected image F ′ (x) after the brightness adjustment is expressed by (Equation 11).
[0047]
F ′ (x) = gain (x) · F (x) + offset (x) (Formula 11)
Here, in order to calculate the sub-pixel positional deviation amount and the brightness correction coefficient of the detected image and the reference image at the same time, the sum of the luminance differences between the two images after the position correction and after the brightness correction. What is necessary is just to obtain | require the coefficient which minimizes. That is, it is only necessary to solve the three formulas (Formula 12) to (Formula 14).

As described above, in the present invention, two images (detected image and reference image) are compared, and in the inspection for detecting a defect from the difference value, the difference between the images increases due to uneven brightness or misalignment. The brightness is adjusted in advance for the area. Further, in the area having the feature as a defect, the brightness adjustment is weakened or the brightness adjustment is not performed. FIG. 12D shows a waveform of the difference value after brightness adjustment according to the present invention when the correct positional deviation amount is not detected in FIG. FIG. 13D shows a waveform of the difference value after brightness adjustment according to the present invention when there is uneven brightness in FIG. In both cases, brightness adjustment is performed on an area that is not originally detected as a defect, and the difference value becomes small. On the other hand, since the brightness adjustment is weakened in the defective portion, the difference value is not so small. For this reason, conventionally, the threshold value is set to TH2 for the entire region, or two threshold values of TH and TH2 are set to avoid the generation of false alarms. However, according to the present invention, the sensitivity is not lowered. The generation of false information due to misalignment or unevenness of brightness is avoided, and high sensitivity inspection with a low threshold TH3 and easy sensitivity adjustment are possible.
[0048]
Furthermore, according to the present invention, a detection sensitivity of 100 nm can usually be achieved in an optical appearance inspection apparatus. Furthermore, if the present invention is used, it is possible to obtain a detection sensitivity of 50 nm when the image quality is good.
[0049]
As described above, the embodiment of the present invention has been described by taking the comparative inspection image in the optical appearance inspection apparatus for the semiconductor wafer as an example. However, it can be applied to the appearance inspection using the electron beam pattern inspection or the DUV as the light source. is there. In this case, a detection sensitivity of 30 to 70 nm can be achieved. In addition, the inspection target is not limited to a semiconductor wafer, and a TFT substrate, a photomask, a printed board, or the like can be applied as long as defect detection is performed by image comparison.
[0050]
【The invention's effect】
As described above, according to the present invention, it is possible to reduce the occurrence of false information by performing brightness adjustment especially on high-contrast portions and uneven brightness that are likely to generate false information due to misalignment and image distortion. There is an effect that can be done.
[0051]
Further, according to the present invention, by adjusting the brightness, a low threshold value can be set as a result, and an effect of realizing a high-sensitivity inspection can be achieved.
[0052]
In addition, according to the present invention, it is possible to reduce the generation of false alarms and detect a defect with only one threshold setting, and to easily adjust the sensitivity.
[0053]
In addition, according to the present invention, a detection sensitivity of 50 nm can be achieved by applying to a comparative inspection in an optical appearance inspection apparatus. In addition, detection sensitivity of 30 to 70 nm can be achieved by applying to electron beam pattern inspection or appearance inspection using DUV as a light source.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating an example of a detected image of an inspection target and a luminance waveform at that time.
FIG. 2 is a diagram illustrating an example of a conventional threshold setting method when a positional deviation amount is correctly calculated.
FIG. 3 is a diagram illustrating an example of a conventional threshold setting method when a misregistration amount is erroneously calculated.
FIG. 4 is a diagram illustrating an example of a detection image to be inspected when there is uneven brightness between comparison chips and an example of a conventional threshold setting method.
FIG. 5 is a block diagram showing an embodiment of a defect inspection apparatus according to the present invention.
FIG. 6 is a view showing a semiconductor wafer to be inspected.
7 is a diagram showing a processing flow in an image comparison unit of the image processing unit shown in FIG. 5;
FIG. 8 is a diagram for explaining an embodiment of a contrast calculation method for a pixel of interest (i, j), which is a feature amount calculation for each pixel according to the present invention.
FIG. 9 is a diagram for explaining another embodiment of a contrast calculation method for a target pixel (i, j), which is a feature amount calculation for each pixel according to the present invention.
FIG. 10 is a diagram for explaining an embodiment of approximate line calculation from a scatter diagram according to the present invention.
FIG. 11 is a diagram showing a positional relationship between a detected image f and a reference image g after pixel unit alignment according to the present invention.
FIG. 12 is a diagram for explaining an embodiment of a threshold value setting method when a misregistration amount according to the present invention is erroneously calculated.
FIG. 13 is a diagram for explaining an embodiment of a threshold setting method when there is uneven brightness between comparison chips according to the present invention.
FIG. 14 is a diagram for explaining an embodiment of brightness matching behavior by linear approximation of a scatter diagram according to the present invention.
FIG. 15 is a diagram for explaining an embodiment of slimming of the scatter diagram according to the present invention.
FIG. 16 is a diagram for explaining another embodiment of slimming of the scatter diagram according to the present invention.
FIG. 17 is a diagram for explaining an embodiment of a method for discriminating categories by frequency according to the present invention.
FIG. 18 is a diagram for explaining an embodiment of a frequency threshold calculation method according to the present invention.
FIG. 19 is a diagram showing a difference in correction between a normal category and an abnormal category according to the present invention.
FIG. 20 is a diagram for explaining an example of a scatter diagram for each contrast and a linear approximation by contrast according to the present invention.
FIG. 21 is a diagram for explaining an embodiment of a frequency threshold setting method using contrast according to the present invention.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 11 ... Detection image, 12 ... Reference image, 51 ... Sample, 52 ... Stage, 53 ... Detection part, 501 ... Light source, 502 ... Illumination optical system, 503 ... Objective lens, 504 ... Image sensor, 55 ... Image processing part, 54 ... AD conversion unit, 505 ... pre-processing unit, 506 ... delay memory, 507 ... misregistration amount detection unit, 508 ... image comparison unit, 509 ... feature extraction unit, 56 ... overall control unit, 510 ... user interface unit, 511 ... Storage device, 512... Mechanical controller, 101.

Claims (2)

A displacement amount detection process for detecting a displacement amount between a detected image to be inspected and a reference image serving as a reference;
The detected image and the reference image are aligned based on the displacement amount calculated in the displacement amount detection process, and detection is performed at each pixel of interest (i, j) in the aligned inspection target. An integrated contrast value C (i, j) in two images consisting of the image F (i, j) and the reference image G (i, j) is determined by calculation, and the determined contrast value C According to (i, j), each image is decomposed into several levels as a contrast category, and both images F (i, j), G (in the decomposed low-contrast category are compared with the decomposed high-contrast category. gain a (i, j) for finely slicing the lightness difference category according to the lightness difference (F (i, j) −G (i, j)) between i, j) and adjusting the brightness. And a correction coefficient consisting of an offset b (i, j) When the brightness of at least one image is adjusted for each contrast category so that the difference in brightness between the aligned images is reduced and the number of approximate straight lines is increased, the high contrast is applied in the low contrast category. the performed strengthened correction is intended to adjust the brightness by the correction coefficient, the difference is compared with the reference image and the detection image in which the brightness has been incorporated combined determined by the approximate line increased the number as compared with the contrast category And a defect detection process for detecting a defect or a defect candidate from the defect inspection method. A stage on which a sample to be inspected is mounted;
An irradiation optical system for irradiating the sample with a beam;
A detector that detects the beam from the sample and converts it into a signal;
An AD conversion circuit for converting the output signal of the detector into a detected digital image;
A memory for storing the detected digital image ;
A misregistration amount detection unit for detecting a misregistration amount between the detected digital image and a reference digital image serving as a reference, and a detected digital image and a reference digital image based on the misregistration amount detected by the misregistration amount detection unit. Alignment unit for alignment , detected digital image F (i, j) and reference digital image G (i, j) at each pixel of interest (i, j) in the inspection object aligned by the alignment unit The integrated contrast value C (i, j) in the two images consisting of is calculated and determined, and according to the determined contrast value C (i, j), each image is counted as a contrast category. In the decomposed low-contrast category, the gray level difference (F (i, j) between both images F (i, j) and G (i, j) is compared with the decomposed high-contrast category. -G (i j)) to increase the number of approximate straight lines from which a correction coefficient consisting of gain a (i, j) and offset b (i, j) for finely slicing the brightness difference category and adjusting the brightness is obtained. When the brightness of at least one image is adjusted for each of the contrast categories so that the difference in brightness between the aligned images is small, the number of the low contrast category is higher than the number of the high contrast category. A brightness correction unit that performs enhancement by adjusting the brightness by the correction coefficient determined by each approximate straight line that is increased , and a detection that is aligned by the alignment unit and the brightness is corrected by the brightness correction unit An image processing unit including an image comparison unit that detects a defect or a defect candidate by comparing a digital image with a reference digital image;
A defect inspection apparatus comprising: an output monitor that displays a defect detection result detected by the image processing unit.

Images