JP2006284617A - Pattern inspection method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a defect inspection method with high sensitivity while suppressing influences by fluctuation in an optical system. <P>SOLUTION: The defect detection method in a pattern such as an exposure mask includes: a variation similarity calculating step of calculating similarity in variation between an optical image in other pixels than inspection target pixels and a reference image based on the variation between an optical image in the inspection target pixels and the reference image; a priority determining step of calculating the priority as comparing members based on the variation similarity; and a mutual comparing step of mutually comparing patterns in an identical design present in the optical image to detect a defect. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention is an inspection method for a pattern formed on a reticle, and is a pattern inspection method for detecting a defect in a pattern.

The reticle is used in a process of forming a required pattern on the surface of a semiconductor wafer by a photolithography method in a manufacturing process of a semiconductor integrated circuit device. In general, the production of LSIs requires a great deal of cost, and thus improvement in yield is indispensable.
In a semiconductor device manufacturing process, one of the factors that reduce the yield is a pattern defect of a reticle used when a fine pattern is exposed and transferred onto a semiconductor wafer by a lithography technique. That is, if the reticle itself has a defect, the defect is transferred to a large number of semiconductor devices, and the yield is greatly reduced.
In recent years, with the miniaturization of LSI pattern dimensions, the minimum dimensions of defects that must be detected have also been miniaturized. Therefore, it is necessary to increase the accuracy of the pattern inspection apparatus that inspects the defects of the reticle.

  Conventionally, as a method for inspecting such a reticle defect, an optical image is formed by irradiating an inspection substrate with a light beam, condensing light transmitted through the substrate or reflected light, and forming an image on a sensor surface; It is known to detect a pattern defect by generating a reference image serving as a pattern reference from CAD data designed with this reticle and comparing the optical image with the reference image (Patent Document 1).

By the way, with the recent miniaturization of the pattern on the reticle due to the high integration of semiconductor devices, it has become necessary to detect defects that are buried in sensing noise due to pixel position shifts between images to be compared, image expansion / contraction and waviness. Yes.
However, the invention described in Patent Document 1 has not been able to sufficiently meet the recent high demands.

  Further, conventional methods for inspecting the presence or absence of pattern defects are roughly classified into die-to-die comparison (Die to Die comparison) and die-to-database comparison (Die to Database comparison). Die to Die comparison (DD comparison) is a method of detecting defects by comparing two dies on a reticle, and Die to Database comparison (DB comparison) is a database generated from CAD data for die and LSI design. This is a method for detecting defects by comparing them.

  By the way, in general, the comparison between the die and the die has a higher defect detection sensitivity than the comparison between the die and the database. This is because the reference image created from the database does not have a defect in the image itself, whereas the optical image has a distortion-free pattern caused by the optical system. This is because there is a possibility that the captured optical image and the reference image corresponding to the captured optical image do not match, and the optical image and the reference image cannot be strictly compared.

On the other hand, in the die-to-die comparison, the problems caused by the optical system are canceled by using the same optical system, and there is a possibility that the above-mentioned problems will occur in the die-to-die comparison. Low. However, the sample to be inspected does not necessarily have the same die, and there is a limitation in using the die-to-die comparison for defect inspection.
JP-A-8-76359

  As a method that can be applied even when the material to be inspected does not have the same die, the same partial pattern in design is searched from the optical image of one die using the reference image, and the corresponding optical image A method of detecting defects by comparing the upper partial patterns with each other is conceivable.

  However, the conventional method has the following problems. That is, since the optical system is subject to various variations, even if the partial pattern is the same in design, a difference may occur on the optical image. Since this is a noise component, it causes the S / N ratio to deteriorate when compared with each other.

  Accordingly, the present invention has been made to solve the above problems in the defect detection technique, and an object of the present invention is to realize a highly sensitive defect inspection method by suppressing the influence of fluctuations in the optical system.

The present invention is a pattern inspection method for comparing and inspecting using an optical image formed by imaging a reticle pattern to be inspected and a reference image formed based on the design data of the pattern,
Variation between any inspection target pixel region and the pixel region of the reference image corresponding to the inspection target pixel region, and variation between the optical image in the pixel region other than the inspection target pixel region and the corresponding reference image A variable dissimilarity calculating step for calculating dissimilarity;
Priority for setting a priority order as a comparison target pixel region in another pixel region in the optical image to be compared with the inspection target pixel region based on the variation dissimilarity data obtained in the variation dissimilarity calculation step A rank setting step;
A pixel area having the same design pattern as the inspection target pixel area is selected from the comparison target pixel area of the optical image, and an optical image of the inspection target pixel area and an optical of the selected comparison pixel area are selected. A defect detection method comprising: a comparison inspection step of detecting defects by comparing images with each other according to the priority order determined in the step.

In the present invention, the step of calculating the variation dissimilarity includes a simultaneous equation generating step of generating simultaneous equations that describe an input / output relationship using a two-dimensional linear prediction model for an optical image and a reference image;
The simultaneous equation describing the input / output relationship may be estimated by a least square method to solve the simultaneous equation solving step for obtaining a parameter of the simultaneous equation.

  In the present invention, in the priority setting step, when the variation dissimilarity of any other comparison target region exceeds a predetermined level, the region can be set as an image not to be compared.

  Rather than simply comparing patterns that are identical in design, they are compared with patterns that are identical in design and subjected to similar variations with the pattern of interest. The S / N ratio in the comparison can be improved.

  Hereinafter, embodiments of the present invention will be described. As shown in the process block diagram of FIG. 1, the defect detection method according to the present embodiment includes an optical image of a sample such as a mask on which a pattern to be a sample is formed, and a database such as CAD data designed with this pattern. From the reference image formed from the information of the specific pixel to be inspected, the step of calculating the variation dissimilarity between the pixels (S1), a plurality of pixels in consideration of the variation dissimilarity calculated in the step, A priority order setting step (S2) for determining the priority order of comparison, and a comparison inspection step (S3) for sequentially comparing a predetermined number of images from the high priority pixels determined in S2.

[Inspection equipment]
Hereinafter, an inspection apparatus that can be used in this embodiment will be described.
A mask defect inspection apparatus that can be used in the present embodiment will be described below with reference to FIG.
The mask defect inspection apparatus includes an arithmetic control unit 200 centering on a host computer and an observation data generation unit 210 that captures a pattern image of a mask serving as a sample.

  The arithmetic control unit 200 includes a host computer 201, a signal transmission path 202 such as an address bus and a data bus connected thereto, a stage control unit 203 connected to the signal transmission path 202, a data memory 204, A data expansion circuit 205, a reference circuit 206, and a comparison circuit 207 are provided.

  The observation data generation unit 210 also includes a light source 211, an illumination optical system 212 that collects light emitted from the light source 211 and irradiates the sample, a sample 213, a stage 214 on which the sample 213 is placed, and a pattern of the sample 213. Are composed of a magnifying optical system 215 and a sensor circuit 216 for capturing an optical image corresponding to the above. A driving system 208 for driving the stage 214 is connected to the stage 214, and the driving system 208 is controlled by the stage control unit 203.

Hereinafter, the operation of the mask defect inspection apparatus according to the present embodiment will be described.
A sample 213 such as a mask is automatically conveyed onto the stage 214 by an unillustrated autoloader mechanism, and is automatically discharged after the inspection is completed.
The light beam emitted from the light source 211 disposed above the stage 214 irradiates the sample 213 via the illumination optical system 212. A magnifying optical system 215 and a sensor circuit 216 are arranged below the sample 213, and transmitted light that has passed through the sample 213 such as an exposure mask forms an image on the sensor surface of the sensor circuit 216 via the magnifying optical system 215. Is done. The magnifying optical system 215 may be automatically focused by an unillustrated autofocus mechanism.

The stage 214 is controlled by a stage control unit 203 that has received a command from the host computer 201 and can be moved by a drive system 216 such as a three-axis (XY-θ) motor that drives in the X, Y, and θ directions. It has become. Step motors can be used as these X motor, Y motor, and θ motor.
The sensor circuit 216 is provided with a sensor such as a TDI sensor. The TDI sensor images the pattern of the sample 213 by continuously moving the stage 214 in the X-axis direction. This imaging data is sent to the comparison circuit 207 as inspected pattern image data. The pattern image data to be inspected is, for example, 8-bit unsigned data, and represents the brightness gradation of each pixel.

On the other hand, a mask pattern image is generated by a data development circuit 205 and a reference circuit 206 from a database such as mask design data stored in the data memory 204, and this information is transmitted to the comparison circuit 207.
The comparison circuit 207 takes in the inspection standard pattern image generated by the data development circuit 205 and the reference circuit 206 and the inspection pattern image generated by the sensor circuit 216 with respect to the transmission image obtained from the sample 213, and inspects the inspection standard pattern. After correcting the image, a comparison is made according to a predetermined algorithm to determine the presence or absence of a defect.

Note that the stage control unit 203, the data expansion circuit 205, the reference circuit 206, and the comparison circuit 207 that constitute the arithmetic control unit may be configured by electrical circuits or can be processed by the host computer 201. It may be realized as software. Moreover, you may implement | achieve with the combination of an electrical circuit and software.
In the above-described apparatus, an example in which a transmission image is acquired as an image obtained from a sample has been shown. In such an apparatus, in FIG. 2, a beam splitter is arranged between the illumination optical system 212 and the sample 213, and the light derived from the beam splitter is condensed by the magnifying optical system and converted into an image by the sensor circuit. Can be realized.

  In the following embodiment, an example will be described in which it is determined whether or not a designated pixel region (= inspection target pixel region) on an optical image has a defect.

(1) Fluctuation similarity calculation step First, an image is appropriately divided into regions for convenience of processing. For example, as shown in FIG. Further, it is assumed that the pixel to be inspected exists in the region (6, 4) in FIG. Here, the fluctuation amount between the optical image and the reference image is obtained as a parameter for each region.

  Hereinafter, an example of the calculation method of the fluctuation amount will be specifically described.

(Simultaneous equation generation step)
In this step, first, a two-dimensional input / output linear prediction model is set by regarding an optical image as two-dimensional input data and a reference image as two-dimensional output data. This method will be described. Here, a 5 × 5 order two-dimensional linear prediction model using a 5 × 5 pixel region will be described as an example.

  Let two-dimensional input data and two-dimensional output data be u (i, j) and y (i, j), respectively. The suffix of the pixel of interest is i, j, and the suffixes of a total of 25 pixels around 2 rows and 2 columns surrounding this pixel are set as shown in Table 1.

The following relational expression (1) is set for y that is two-dimensional output data and u that is two-dimensional input data for a set of pixel data in a 5 × 5 region.

In equation (1), b _{00 to} b ₄₄ are model parameters to be identified, and ε (i, j) is noise.

The expression (1) means that one pixel data y _k = y (i, j) having a pattern of the optical image is linear of 5 × 5 pixel data surrounding one pixel of the corresponding reference image pattern. It can be expressed as a bond.

(Steps for solving simultaneous equations)
The next step is a step of obtaining the parameter b by solving the simultaneous equations set in the previous step (identification of model parameters).
That is, when Expression (1) is expressed as a vector,

It is put together so that. Therefore, the model parameter b can be identified by scanning the coordinates i and j of the reference image and the optical image and combining 25 sets of data.
Actually, from a statistical point of view, n (> 25) sets of data are prepared as in Equation (3), and a 25-dimensional simultaneous equation is solved based on the following least-squares method to identify α. .

Where A = [x ₁ , x ₂ ,... X _n ] ^T , y = [y ₁ , y ₂ ,... Y _n ] ^T , x _k ^T α = y _k (k = 1, 2,... N) It is.
For example, if the reference image and the optical image are each 512 × 512 pixels, the periphery of the image is reduced by two pixels by scanning the 5 × 5 model,
Thus, a sufficient number of data can be secured from a statistical point of view.

  By such a method, the variation amount is obtained as a 5 × 5 = 25-dimensional vector (hereinafter referred to as a variation vector) for each region. Here, the variation vector in the region (x, y) is expressed as P (x, y). Using this variation vector, the variation similarity Q (x, y) between (6, 4) and the region (x, y) is defined as follows.

Here, as a reference for evaluating the fluctuation amount, a method other than the above method can be used.
For example, a difference image between the reference image and the optical image is obtained, and each pixel value of the difference image is arranged one-dimensionally to obtain a variation vector P (x, y).
Further, as a definition of the variation similarity Q, it is also possible to use a method other than the equation (5), that is, a method such as a Mahalanobis distance between P (6,4) and P (x, y).

(2) Priority Order Determination Step Q (x, y) obtained in the above step is arranged in ascending order and priority is assigned to each area in ascending order. However, an area where Q (x, y) is larger than a threshold value prepared in advance is not suitable for comparison and is not assigned a priority (see FIG. 4). As a result, the processing time can be shortened. Here, x and y are the numbers in the row direction and the column direction of the lattice-like region (8 × 8 region in FIG. 3) shown in FIG. 3, and in FIG. 3, x, y = 8. It is.

(3) Comparative Inspection Step First, a maximum of k graphic patterns that are the same as the vicinity of the inspection target pixel are searched for. In this step, a pixel having the same pattern data as the pattern data of the inspection target pixel is obtained from design data such as CAD, and the comparison target region corresponding to this region is determined in the order determined in the above-described priority order determination step. , Search each area. Here, in order to search the same graphic pattern as the vicinity of the inspection target pixel from the region (x, y), for example, there is the following method. All points belonging to the region (x, y) are listed in units of 1/8 pixel. Let this point sequence be a point sequence of interest {Cn}. A rectangular area of 15 × 15 pixels centered on Ci (i = 0,... N) is extracted from the reference image, and between the rectangular area of 15 × 15 pixels extracted from the reference image centering on the pixel to be inspected Then, the cumulative square difference between the pixels belonging to both rectangles is calculated, and if the cumulative square difference is equal to or less than a predetermined threshold, they are regarded as the same pattern and added to the point sequence {Dm}. If the points Di and Dj (i <j) exist such that Di and Dj are 7 pixels or less at the city block distance in the point sequence {Dm}, Dj is deleted from the point sequence {Dm}. Let the point sequence finally obtained in this way be {Em}. The points included in {Em} are the center coordinates of the same graphic pattern as the vicinity of the pixel to be inspected, searched from the region (x, y). Furthermore, the elements after k + 1 from the top of {Em} are deleted. The center point sequence of the graphic pattern obtained in this way is defined as {Fl} (l ≦ k).

A 15 × 15 pixel rectangular region centered at Fi (i = 0,..., L) is extracted from the optical image, and a 15 × 15 pixel rectangular region extracted from the optical image with the pixel to be inspected as the center. The cumulative square difference between the pixels belonging to both rectangles is calculated as Gi. Here, H = (G0 + G1 +... + Gl) / l is set as the defect determination value. However, when l is 0, H = 0.
If H is equal to or greater than a predetermined threshold value, it is determined as a defect, and if not, it is determined as a non-defect. This threshold value can be determined by the user as needed depending on how many defects are desired to be detected.

The method described above is suitable for application to, for example, a small region inspection method in which the quality of a single contact hole pattern is inspected with high sensitivity. Inspecting the entire mask having a pattern such as a large number of contact holes requires a considerable inspection time. Therefore, it is practical to perform simple inspections such as die and database comparison inspections in advance, extract areas that may be defective images, and convert them into high-sensitivity inspections of limited areas. Is.
In the above method, the optical image to be inspected may be a transmission image or a reflection image. Further, the pixel area may be an area corresponding to the pattern pitch or may be a frame area of the sensor.

The block diagram which shows the process in this Embodiment. The schematic block diagram of the test | inspection apparatus which can be used in this Embodiment. FIG. 6 is a diagram for illustrating a pixel region in this embodiment. The figure for demonstrating the priority setting step of this implementation. The figure for demonstrating the effect of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 201 ... Processing unit 202 ... Signal transmission path 203 ... Stage control circuit 204 ... Data memory 205 ... Data expansion circuit 206 ... Reference circuit 207 ... Comparison circuit 210 ... Driving device 211 ... Light source 212 ... Illumination optical system 213 ... Sample 214 ... Stage 215: Magnifying optical system 216: Sensor circuit

Claims (3)

A pattern inspection method for comparing and inspecting an optical image formed by imaging a pattern of a reticle to be inspected and a reference image formed based on design data of the pattern,
Variation between any inspection target pixel region and the pixel region of the reference image corresponding to the inspection target pixel region, and variation between the optical image in the pixel region other than the inspection target pixel region and the corresponding reference image A variable dissimilarity calculating step for calculating dissimilarity;
Priority for setting a priority order as a comparison target pixel region in another pixel region in the optical image to be compared with the inspection target pixel region based on the variation dissimilarity data obtained in the variation dissimilarity calculation step A rank setting step;
A pixel area having the same design pattern as the inspection target pixel area is selected from the comparison target pixel area of the optical image, and an optical image of the inspection target pixel area and an optical of the selected comparison pixel area are selected. A defect detection method comprising: a comparison inspection step of detecting defects by comparing images with each other according to the priority order determined in the step. The step of calculating the variation dissimilarity includes a simultaneous equation generation step of generating simultaneous equations describing an input / output relationship using a two-dimensional linear prediction model with respect to an optical image and a reference image;
2. The defect detection method according to claim 1, further comprising a simultaneous equation solving step of estimating the simultaneous equations describing the input / output relationship by a least square method to obtain parameters of the simultaneous equations. 2. In the priority setting step, when the variation dissimilarity of any other comparison target region exceeds a predetermined level, the region is set as an image that is not a comparison target. The defect detection method according to claim 2.