JP2016145887A - Inspection device and method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an inspection device and an inspection method capable of inspecting with high reliability, reducing a ratio of a pseudo defect, and improving convenience of inspection.SOLUTION: An inspection device 100 comprises: a table 101 for mounting a mask Ma thereon; an imaging part 104 for imaging an area to be inspected of the mask Ma for every stripe, for creating optical image data; a reference image creation part 116 for performing filter processing to the image data created from design data, for creating reference image data; a reference image creation function calculation part 115 for determining a function of the filter processing; a comparison part 117 for comparing optical frame data and reference frame data, for counting a number of defects for every stripe; and a defect analysis part 118 for, analyzing a factor of occurrence of many defects on the stripe in which a number of detected defects exceeds a permission value, by using defect information output from the comparison part 117.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an inspection apparatus and an inspection method.

  In a semiconductor device manufacturing process, an original image pattern (a mask or a reticle, which will be collectively referred to as a mask hereinafter) on which a circuit pattern is formed is exposed and transferred onto a wafer by a reduction projection exposure apparatus called a stepper or a scanner. . For LSIs that require a large amount of manufacturing cost, it is essential to improve the yield in the manufacturing process. Since the defect of the mask pattern is a major factor that reduces the yield of the semiconductor element, an inspection process for detecting the defect is important when manufacturing the mask.

  With the recent high integration and large capacity of large scale integrated circuits (LSIs), circuit dimensions required for semiconductor elements are becoming increasingly narrow. For example, a typical logic device is required to form a pattern with a line width of several tens of nanometers. As the dimensions of the circuit pattern formed on the wafer are miniaturized, the defect of the mask pattern is also miniaturized. In addition, by increasing the dimensional accuracy of the mask, there have been attempts to absorb variations in process conditions. In the mask inspection process, it is necessary to detect even a very small pattern defect. For this reason, development of inspection apparatuses capable of detecting minute defects has been underway.

  By the way, in recent mask patterns, the dimensions of defects to be pointed out as shape defects and line width fluctuations are about the same as the line width fluctuations (line width error distribution) on the entire mask surface. Therefore, the number of pseudo defects that originally do not need to be detected as defects is increasing.

  The relationship between the line width variation on the entire mask surface and the pseudo defect will be described with reference to an inspection by a die-to-database comparison method as an example. In this inspection, the coefficient of the reference image generation function is learned at a typical pattern portion of the mask, and reference image data similar to the optical image data is generated. That is, the reference image data has a line width tendency that follows the pattern line width of the learned region. Therefore, if there is an error distribution in the line width dimension on the entire mask surface, optical image data and reference image data will be displayed when trying to inspect a finished area with a pattern line width different from the pattern line width of the learned area. Are compared with each other with a pattern line width bias (deviation). As a result, a shape or line width error that does not actually cause a problem is often detected as a defect and is a pseudo defect.

  Therefore, in addition to being able to detect minute variations in pattern shape and line width, there is a need for an inspection apparatus that can perform highly reliable inspection by reducing the proportion of pseudo defects.

  As an inspection apparatus capable of reducing pseudo defects, for example, there is one described in Patent Document 1. In this inspection device, the state when the image is displayed on the display device to be inspected is photographed, and using this photographed image, the presence or absence of the inhibition factor that inhibits the defect inspection and the type of the inhibition factor are determined, According to the determined obstruction factor, the inspection area where defect detection is automatically performed and the non-inspection area where defect detection is not automatically performed are divided into two areas. I do.

  In addition, in Patent Document 2, when inspecting a defect of a pattern formed on a mask, an area where an important pattern is formed is inspected for a defect with a strict judgment threshold, and a pattern that is not so important is formed. There has been described an inspection apparatus that prevents the occurrence of pseudo defects by not performing defect inspection on an unnecessarily strict judgment threshold for a given region.

  Further, in Patent Document 3, when detecting defects by comparing the optical image data of the inspection substrate to be inspected with the reference image data, the optical image data of the inspection target is detected by the time delay integration sensor. By converting the position deviation caused by stage speed unevenness or yawing using stage coordinates to create reference image data, and comparing the detected image data with the reference image data to detect defects There is described a defect inspection apparatus that improves the sensitivity of defect detection and makes it possible to prevent detection of pseudo defects.

JP 2012-107912 A JP 2011-145263 A Japanese Patent Laid-Open No. 2003-90717

  The inspection apparatus of Patent Document 1 reduces pseudo defects by visual inspection of the inspected area, so that although it is used in combination with automatic inspection, the time required for inspection becomes long and the load on the worker is large. . Moreover, in the inspection apparatus of patent document 2 or patent document 3, when the number of defects suddenly increases during the inspection, whether the rapid increase in the number of defects is due to a true defect or a pseudo defect. I can't judge. For this reason, it is necessary to restart the inspection from the beginning, and there arises a problem that all the inspections up to that point are wasted.

  The present invention has been made in view of the above points. That is, an inspection apparatus is required that can not only detect fine defects but also reduce the proportion of pseudo defects and perform highly reliable inspection. There is a need for an inspection device that can be improved. Specifically, if the number of detected defects increases rapidly during the inspection, if these defects are analyzed and appropriate measures are taken, the inspection is continued. In addition to preventing frequent defects, it is not necessary to waste inspection results before the number of defects suddenly increases, so that the convenience of inspection can be greatly improved. Furthermore, if the inspection apparatus has a function for estimating the cause and a function for determining an appropriate treatment according to the function, the burden on the worker can be reduced. In view of the above, an object of the present invention is to provide an inspection apparatus and an inspection method capable of performing a highly reliable inspection by reducing the ratio of pseudo defects and improving the convenience of inspection. .

A first aspect of the present invention is an inspection apparatus for inspecting a defect of a pattern formed on a substrate,
A table on which the substrate is placed;
An imaging unit that images the inspection region for each stripe that virtually divides the inspection region on which the pattern of the substrate is formed, and generates optical image data;
A reference image generation unit that generates a reference image data by performing filter processing using a function on image data generated from pattern design data;
A reference image generation function calculation unit for obtaining the above function;
By comparing the optical frame data obtained by dividing the optical image data into a predetermined size and the reference frame data obtained by dividing the reference image data corresponding to the optical image data into the same size, defects in the optical image data are detected, and stripes are detected. A comparison unit that counts the number of detected defects for each,
For a stripe whose number of detected defects counted by the comparison unit exceeds the allowable value that permits the inspection to continue, whether the frequent occurrence of defects in this stripe is due to the occurrence of true defects or the reference image data corresponding to the optical image data Defect analysis that analyzes whether the function that generates the error is due to an incompatible function or due to insufficient alignment accuracy between the pattern and the table using the defect information in this stripe output from the comparison unit Part.

  In the first aspect of the present invention, the defect analysis unit has at least one of the luminance value of the defect of the stripe exceeding the allowable value allowing the inspection to continue and the line width error value of the pattern constituting the defect. Is less than or equal to the first threshold value, and if defects are distributed in a region other than the edge of the pattern, the continuation of the inspection is instructed on the assumption that frequent defects are due to frequent true defects. It is preferable to output a signal.

In the first aspect of the present invention, the defect analysis unit includes:
The density of defects detected in stripes where the number of detected defects exceeds an allowable value that allows continuation of inspection exceeds the second threshold, and the value of x in the position coordinates (x, y) of the defect or y If the value is ubiquitous in one or more specific numerical ranges,
Compared with this stripe, the stripes arranged in succession, and the tendency of the pattern deviation between this stripe and 1 to 10 stripes that have been processed in the comparison part prior to this stripe,
The number of defects in the stripes where the number of detected defects exceeds the allowable value that allows the inspection to continue is caused by a function that is incompatible with the function that generates the reference image data corresponding to the optical image data, or the alignment accuracy Analyzing whether it is due to lack of
When the function that generates the reference image data corresponding to the optical image data is based on an incompatible function, a signal that instructs to re-determine the function is output,
In the case where the accuracy of alignment is insufficient, it is preferable to output a signal instructing re-alignment.

  In the first aspect of the present invention, the reference image generation function calculation unit machine-learns pattern shape data included in a stripe whose number of detected defects exceeds an allowable value permitting continuation of inspection to obtain optical image data. It is preferable to re-determine a function that generates corresponding reference image data.

According to a first aspect of the present invention, there is provided a display device for displaying an optical image of a stripe in which the number of detected defects exceeds an allowable value that allows continuation of inspection,
When an optical image is displayed on the display device, an input screen that allows the operator to instruct pattern shape data to be machine-learned is also displayed on the display device, and pattern shape data that the operator should treat as a defect on the input screen. Appropriateness of adoption of this data is judged when instructed, and when it is not appropriate, information recommending the instruction of other data is displayed on the input screen, and when appropriate, this pattern shape data is displayed on the machine. It is preferable to have a data input support unit that instructs the reference image generation function calculation unit to learn and recalculate the function.

A second aspect of the present invention is an inspection method for inspecting a defect of a pattern formed on a substrate,
Placing the substrate on the table;
A step of aligning the pattern and the table;
Imaging the inspected area for each stripe that virtually divides the inspected area in which the pattern of the substrate is formed, and generating optical image data;
A step of obtaining a function for filtering the image data generated from the pattern design data;
Generating reference image data from image data using the above function;
By comparing the optical frame data obtained by dividing the optical image data into a predetermined size and the reference frame data obtained by dividing the reference image data corresponding to the optical image data into the same size, defects in the optical image data are detected, and stripes are detected. A step of counting the number of detected defects every time,
For a stripe whose number of detected defects counted by the comparison unit exceeds the allowable value that permits the inspection to continue, whether the frequent occurrence of defects in this stripe is due to the occurrence of true defects or the reference image data corresponding to the optical image data Analyzing whether or not the function for generating is based on an incompatible function or insufficient alignment accuracy using information on defects in the stripe output from the comparison unit, and
The value obtained from at least one of the brightness value of the stripe defect and the line width error value of the pattern constituting the defect exceeds the allowable value that allows the inspection to be continued, and is not more than the first threshold value. , If the defects are distributed in areas other than the edges of the pattern, continue the inspection, assuming that frequent defects are due to frequent true defects,
The density of defects detected in stripes where the number of detected defects exceeds an allowable value that allows continuation of inspection exceeds the second threshold, and the value of x in the position coordinates (x, y) of the defect or y If the value is ubiquitous in one or more specific numerical ranges,
Compared with this stripe, the stripes arranged in succession, and the tendency of the pattern deviation between this stripe and 1 to 10 stripes that have been processed in the comparison part prior to this stripe,
The number of defects in the stripes where the number of detected defects exceeds the allowable value that allows the inspection to continue is caused by a function that is incompatible with the function that generates the reference image data corresponding to the optical image data, or the alignment accuracy Analyzing whether it is due to lack of
If the function that generates the reference image data corresponding to the optical image data is based on an incompatible function, the function is obtained again,
In the case where the accuracy of alignment is insufficient, the alignment is performed again.

  In the second aspect of the present invention, machine learning is performed on pattern shape data included in stripes in which the number of detected defects exceeds an allowable value that permits continuation of inspection, and reference image data corresponding to optical image data is generated. It is preferable to re-determine the function to perform.

  According to the inspection apparatus of the first aspect of the present invention, with respect to a stripe in which the number of detected defects counted by the comparison unit exceeds an allowable value that allows continuation of inspection, frequent occurrence of defects in this stripe The comparison unit outputs whether it is due to frequent occurrence, whether the function that generates the reference image data corresponding to the optical image data is due to an incompatible function, or due to insufficient accuracy in alignment between the pattern and the table. In addition, since the defect analysis unit that analyzes using the information of defects in the stripes is provided, it is possible to prevent the occurrence of pseudo defects and to use the inspection results before the defects increase rapidly. In addition, the burden on workers can be reduced.

  According to the inspection method of the second aspect of the present invention, with respect to a stripe in which the number of detected defects counted by the comparison unit exceeds an allowable value that allows continuation of inspection, frequent occurrence of defects in this stripe The comparison unit outputs whether it is due to frequent occurrence, whether the function that generates the reference image data corresponding to the optical image data is due to an incompatible function, or due to insufficient accuracy in alignment between the pattern and the table. Therefore, not only the occurrence of pseudo defects is prevented, but also the inspection results before the number of defects rapidly increase can be saved. Therefore, the convenience of inspection can be improved.

1 is a schematic configuration diagram of an inspection apparatus according to Embodiment 1. FIG. 3 is an example of a flowchart illustrating an inspection method according to the first embodiment. It is a conceptual diagram for demonstrating the relationship between a mask, a stripe, and a frame. It is a conceptual diagram which shows the flow of the data and instruction | indication in the inspection apparatus of FIG. It is a flowchart which shows roughly the flow of the process performed in a defect analysis part. It is a schematic block diagram of the inspection apparatus in Embodiment 2. It is a top view which shows the typical example of a learning point. It is a top view which shows the typical example of a learning point. It is a top view which shows the typical example of a learning point. This is an example in which the shape of the corner of the fine pattern does not match between the optical image and the reference image. This is an example in which the luminance amplitude of a part of the optical image is insufficient compared to the luminance amplitude of the reference image. This is an example in which an optical image includes an Inret-shaped pattern. It is a figure which shows an example of in-plane distribution of the line | wire width fluctuation value of a fine pattern.

Embodiment 1 FIG.
FIG. 1 is a schematic configuration diagram of an inspection apparatus according to the present embodiment. In FIG. 1, constituent parts necessary for the present embodiment are illustrated, but other known constituent parts necessary for inspection may be included. In addition, in this specification, what is described as “to part” can be configured by a computer-operable program, but not only a program that becomes software but also a combination of hardware and software or firmware. It may be implemented by a combination. When configured by a program, the program is recorded in a recording device such as a magnetic disk device. For example, the autoloader control unit 112 may be configured by an electric circuit, and may be realized as software that can be processed by the control computer 110. Further, it may be realized by a combination of an electric circuit and software.

  As shown in FIG. 1, an inspection apparatus 100 includes a configuration unit A that acquires optical image data of an inspection target, and a configuration unit that performs processing necessary for inspection using the optical image data acquired by the configuration unit A. B. The component A captures an optical image of the mask Ma and outputs optical image data corresponding to the optical image. Here, the mask Ma has a configuration in which a pattern as an object to be inspected is formed on the main surface of a base material such as a transparent glass substrate, and the configuration unit A outputs optical image data of the pattern. To do. In this specification, the optical image data of the pattern in the mask Ma may be simply referred to as the optical image data of the mask Ma. On the other hand, the configuration unit B uses the reference image data generated by using the design pattern data of the mask Ma and a predetermined reference image generation function, and the optical image data output from the configuration unit A, and the pattern in the mask Ma. Detect defects.

  The component A includes a table 101 that can be driven in the horizontal direction (X-axis direction, Y-axis direction) and the rotation direction (θ-axis direction), a laser length measurement system 102 that measures the position of the table 101, and a table 101. An illumination optical system 103 that illuminates the placed mask Ma, an imaging unit 104 that generates optical image data of the mask Ma, and an autoloader 105 are included.

  The table 101 is driven by an X-axis motor M1, a Y-axis motor M2, and a θ-axis motor M3 controlled by the table control unit 114. For example, an air slider and a linear motor or a step motor can be used in combination for these drive mechanisms.

  Although detailed illustration is omitted, the laser measurement system 102 includes a laser interferometer such as a heterodyne interferometer. The laser interferometer measures the position coordinates of the table 101 by irradiating and receiving laser light between X-axis and Y-axis mirrors provided on the table 101. Data measured by the laser length measurement system 102 is sent to the position information unit 113. Note that the method of measuring the position coordinates of the table 101 is not limited to using a laser interferometer, and may use a magnetic or optical linear encoder.

  The illumination optical system 103 includes a light source 103a and an expansion optical system 103b, and a predetermined beam is irradiated from the illumination optical system 103 onto the mask Ma. The magnifying optical system 103b is a means for dividing the illumination light beam emitted from the light source 103a into an optical path for transmitting and illuminating the mask Ma and an optical path for reflecting and illuminating, as necessary, and means for changing the light into circularly polarized light, linearly polarized light, etc. And means for changing to a light source shape such as a point light source or a ring zone.

  The imaging unit 104 focuses the illumination light that has passed or reflected through the mask Ma to form an optical image of the pattern of the mask Ma, a photodiode array 104b that photoelectrically converts this optical image, and a photodiode array. And a sensor circuit 104c that converts an analog signal output from the digital signal 104b into a digital signal as optical image data and outputs the digital signal. For the imaging unit 104, for example, a TDI (Time Delay Integration) image sensor can be used. Note that the imaging unit 104 may be configured to automatically perform focus adjustment by an automatic focus mechanism (not shown).

  The configuration unit B includes a control computer 110 that controls the entire inspection apparatus 100, a bus 111 that serves as a data transmission path, an autoloader control unit 112 that is connected to the control computer 110 via the bus 111, a position information unit 113, and a table. A control unit 114, a reference image generation function calculation unit 115, a reference image generation unit 116, a comparison unit 117, a defect analysis unit 118, a magnetic disk device 119 as an example of a main storage device, and a magnetic tape device 120 as an example of an auxiliary storage device. And a flexible disk device 121 as another example of the auxiliary storage device, a monitor display 122 as an example of the display device, a microscope pattern monitor 123 by an ITV camera as another example of the display device, and a printer 124.

  Next, an example of a method for inspecting the mask Ma using the inspection apparatus 100 of FIG. 1 will be described.

  FIG. 2 is an example of a flowchart showing the inspection method of the present embodiment. As shown in this figure, the inspection method according to the present embodiment includes an optical image data acquisition step S1, a reference image data generation step S2, a defect detection step S3, a defect analysis step S4, and a recovery step S5.

<Optical image data acquisition step S1>
In the optical image data acquisition step S1 of FIG. 2, an optical image of the mask Ma is acquired, and then this optical image is digitally converted to obtain optical image data used for inspection.

  First, the mask Ma is placed on the table 101 by the autoloader 105. Here, the autoloader 105 is driven by the autoloader control unit 112, and the autoloader control unit 112 is controlled by the control computer 110. The mask Ma is fixed to the table 101 by means such as a vacuum chuck.

  When the mask Ma is placed on the table 101, the illumination optical system 103 irradiates the pattern formed on the mask Ma. Specifically, the light beam emitted from the light source 103a is applied to the mask Ma via the magnifying optical system 103b. The illumination light transmitted or reflected through the mask Ma is focused to obtain an optical image of the pattern of the mask Ma.

  In order to obtain an accurate inspection result, a pattern to be inspected (on the mask Ma) (hereinafter also referred to as an inspection pattern) needs to be aligned at a predetermined position on the table 101. Therefore, for example, an alignment mark for alignment is formed on the mask Ma, the alignment mark is actually photographed by the imaging unit 104, and the inspection pattern of the mask Ma is aligned on the table 101. In the present embodiment, the relative alignment between the inspection pattern of the mask Ma and the table 101 may be referred to as plate alignment.

  For example, it is assumed that cross-shaped mask alignment marks MA are formed at positions corresponding to the vertices of a rectangle near the four corners of the pattern to be inspected of the mask Ma. Further, it is assumed that a plurality of chip patterns are formed on the mask Ma, and chip alignment marks CA are also formed on each chip. On the other hand, the table 101 is configured by an XY table that moves in the horizontal direction and a θ table that is arranged on the XY table and moves in the rotation direction. In this case, the alignment step is specifically a step of aligning the X axis and Y axis of the pattern to be inspected with the traveling axis of the XY table in a state where the mask Ma is placed on the table 101.

  First, of the mask alignment marks MA provided at four locations, two mask alignment marks MA having a small Y coordinate value are photographed, and the θ table is rotated so that both marks have exactly the same Y coordinate. Thus, fine adjustment of the rotation direction of the mask Ma is performed. At this time, the distance between the mask alignment marks MA is also accurately measured. Next, two mask alignment marks MA having a large Y coordinate value are photographed. Thereby, the coordinates of all four mask alignment marks MA are accurately measured.

  From the above measurement, it is assumed that the other two mask alignment marks MA are positioned at the vertices of a quadrangle having two mask alignment marks MA having small Y-coordinate values at both ends of the base. Here, the other two alignment marks that should be located at the respective vertex coordinates of the rectangle are distorted by shifting to the measured coordinates, and between the measured alignment marks Since the distance is stretched with respect to the design coordinate distance, it is estimated that the pattern area to be inspected is also distorted and stretched similar to the above-mentioned rectangle, and corrections reflecting this are corrected. This is performed when the reference image generation unit 116 generates reference image data.

  Note that the mask Ma may not have the mask alignment mark MA. In that case, alignment can be performed using the corner vertices and edge pattern sides that are as close as possible to the outer periphery of the mask Ma and have the same XY coordinates in the pattern of the mask Ma.

  The inspection area of the mask Ma (area where the inspection pattern is formed) is virtually divided into a plurality of strip-shaped areas. This strip-like region is called a stripe. Each stripe can be, for example, a region having a width of several hundred μm and a length of about 100 mm corresponding to the entire length in the X direction of the region to be inspected of the mask Ma.

  Further, in each stripe, a plurality of imaging target units (hereinafter, each imaging target unit is referred to as “frame”) divided in a lattice shape is virtually set. The size of each frame is suitably a stripe width or a square that is obtained by dividing the stripe width into four.

  FIG. 3 is a conceptual diagram for explaining the relationship between the region to be inspected in the mask Ma and the stripes and frames. In this example, the inspection area in the mask Ma is virtually divided by four stripes St1 to St4, and 45 frames F are virtually set in each of the stripes St1 to St4. Yes.

  The stripes St1 to St4 are aligned in the Y-axis direction. On the other hand, each frame has a rectangular shape of, for example, about a dozen μm □. Here, in order to prevent image capturing omission, between two adjacent frames, the edge of one frame and the edge of the other frame are set to overlap each other with a predetermined width. The predetermined width can be, for example, about 20 pixels when the pixel size of the photodiode array 104b is used as a reference. The same applies to the stripes, and the edges of adjacent stripes are set to overlap each other.

  The optical image of the mask Ma is captured for each stripe. That is, when an optical image is captured in the example of FIG. 3, the operation of the table 101 is controlled so that each stripe St1, St2, St3, St4 is continuously scanned. Specifically, first, the optical image of the stripe St1 is sequentially taken in the X direction while the table 101 moves in the -X direction in FIG. Then, an optical image is continuously input to the photodiode array 104b in FIG. When the imaging of the optical image of the stripe St1 is completed, the optical image of the stripe St2 is captured. At this time, the table 101 moves stepwise in the −Y direction, and then moves in the direction (X direction) opposite to the direction (−X direction) at the time of acquisition of the optical image in the stripe St1. The captured optical image of the stripe St2 is also continuously input to the photodiode array 104b. When acquiring the optical image of the stripe St3, after the table 101 is moved stepwise in the -Y direction, the optical image of the stripe St1 is obtained in the direction opposite to the direction (X direction) in which the optical image of the stripe St2 is acquired. The table 101 moves in the acquired direction (−X direction). Similarly, an optical image of the stripe St4 is also taken.

  The light that has passed through the mask Ma is imaged as an optical image of the pattern of the mask Ma by the imaging unit 104, and is then A / D (analog-digital) converted and output as optical image data. Specifically, the photodiode array 104b captures an optical image of the mask Ma, and sequentially outputs analog signals corresponding to the optical image to the sensor circuit 104c. The sensor circuit 104c converts each analog signal output from the photodiode array 104b into a digital signal, which is optical image data, and outputs the digital signal.

  The optical image data is input to a digital amplifier (not shown) provided in the sensor circuit 104c and capable of adjusting offset / gain for each pixel. The gain for each pixel of the digital amplifier is determined by calibration. For example, in the calibration for transmitted light, the black level is determined during photographing of the light-shielding area of the mask Ma that is sufficiently wide with respect to the area captured by the imaging unit 104. Next, the white level is determined during imaging of the transmitted light region of the mask Ma that is sufficiently wide with respect to the area captured by the imaging unit 104. At this time, in anticipation of light quantity fluctuation during inspection, for example, the offset of each pixel is set so that the amplitude of the white level and the black level is distributed from 10% to 240% corresponding to about 4% to about 94% of the 8-bit gradation data. And adjust the gain.

<Reference Image Data Generation Step S2>
In the reference image data generation step S2 of FIG. 2, reference image data is generated based on the design pattern data of the mask Ma. The reference image data is a standard for determining defects in the optical image data in the inspection by the die-to-database comparison method.

  FIG. 4 is a conceptual diagram showing the flow of data and instructions in the inspection apparatus 100 of FIG. The reference image data generation step S2 will be described with reference to FIGS.

  The design pattern data of the mask Ma is stored in the magnetic disk device 119 of the inspection apparatus 100. The design pattern data is read from the magnetic disk device 119 by the control computer 110 and sent to the reference image generator 116.

  The reference image generation unit 116 includes a development circuit 116a and a reference circuit 116b, and the design pattern data read from the magnetic disk device 119 is converted into binary or multi-value image data by the development circuit 116a. .

  The image data is sent from the expansion circuit 116a to the reference circuit 116b. In the reference circuit 116b, appropriate filter processing is performed on the image data. The reason is as follows.

  In general, the roundness of the corners and the finished dimensions of the line width are adjusted in the manufacturing process of the mask pattern, and the mask pattern does not completely match the design pattern. Further, for example, the optical image data obtained by the imaging unit 104 in FIG. 1 is blurred due to the resolution characteristic of the magnifying optical system 103b, the aperture effect of the photodiode array 104b, or the like, in other words, a spatial low-pass filter. It is in an acted state.

  Therefore, based on the mask design pattern data and optical image data, a reference image generation function simulating changes due to the mask manufacturing process and the optical system of the inspection apparatus is determined. This decision is made prior to inspection. The inspection apparatus applies a two-dimensional digital filter to the image data using the reference image generation function. In this way, the process of making the reference image data resemble the optical image data is performed.

  In determining the reference image generation function, the inspection apparatus 100 is machine-learned using a typical mask pattern. The typical pattern can be, for example, a pattern specified in a procedure manual distributed along with the mask in the manufacturing process. Further, for example, a pattern selected by an operator based on experience such as a large pattern and a fine pattern may be used. On the other hand, the mask used for learning may be a reference mask determined in the manufacturing process, or a mask (a mask Ma in the present embodiment) on which a pattern to be inspected is formed. Also good.

  In the present embodiment, the reference image generation function calculation unit 115 determines a reference image generation function. Specifically, the design pattern data read from the magnetic disk device 119 by the control computer 110 and the optical image data output from the imaging unit 104 are sent to the reference image generation function calculation unit 115. Based on the design pattern data and the optical image data, the reference image generation function calculation unit 115 determines a reference image generation function that simulates a change in the mask manufacturing process and the optical system of the inspection apparatus.

  Next, the reference image generation function is sent from the reference image generation function calculation unit 115 to the reference image generation unit 116. Then, in the reference circuit 116b in the reference image generation unit 116, the image data output from the expansion circuit 116a is subjected to filter processing using a reference image generation function, and reference image data is generated.

<Defect detection step S3>
The defect detection of the mask Ma is performed in the defect detection step S3 using the optical image data obtained in the optical image data acquisition step S1 and the reference image data obtained in the reference image data generation step S2.

  First, optical image data from the imaging unit 104 and reference image data from the reference image generation unit 116 are sent to the comparison unit 117, respectively. In addition, position data obtained by measuring the position coordinates of the table 101 is sent from the position information unit 113 to the comparison unit 117.

  According to the inspection apparatus 100 of FIG. 1, the imaging unit 104 converges the illumination light that has passed through the mask Ma to obtain an optical image of the mask Ma. Therefore, in the comparison unit 117, transmission optical image data and reference image data are obtained. Will be compared. If the inspection apparatus is configured to focus the illumination light reflected by the mask Ma and obtain an optical image of the mask Ma, the reflected optical image data and the reference image data are compared.

  In the area division processing unit 117a, the optical image data output from the imaging unit 104 is divided into data of a predetermined size, in this embodiment, for each frame. The reference image data output from the reference image generation unit 116 is also divided into data for each frame corresponding to the optical image data. Hereinafter, each of the optical image data divided for each frame is referred to as “optical frame data”, and each of the reference image data divided for each frame is referred to as “reference frame data”.

  The comparison unit 117 detects a defect in the optical frame data by comparing the optical frame data with the reference frame data. Further, the position coordinate data of the defect is created using the position data sent from the position information unit 113.

  The comparison unit 117 is equipped with several tens of comparison units. Thereby, a plurality of optical frame data are simultaneously processed in parallel with the corresponding reference frame data. Each comparison unit includes a frame alignment unit, an algorithm comparison processing unit, and a defect registration unit. Then, when the processing of one optical frame data is completed, each comparison unit captures unprocessed optical frame data and reference frame data corresponding thereto. In this way, a large number of optical frame data are sequentially processed to detect defects.

  Specifically, the processing in the comparison unit is performed as follows.

  First, optical frame data and reference frame data corresponding to the optical frame data are set and output from the region division processing unit 117a to the frame alignment units (117b1, 117b2,...) Of each comparison unit. . The frame alignment unit performs alignment (frame alignment) between the optical frame data and the reference frame data. At this time, in addition to the parallel shift in units of pixels (of the photodiode array 104b) so that the edge positions of the pattern and the peak positions of the brightness are aligned, the luminance values of neighboring pixels are proportionally distributed. Matching is also performed. In this embodiment, the displacement amount and the displacement direction of the coordinate system of the optical frame data or the coordinate system of the reference frame data corresponding to the optical frame data when aligning the optical frame data and the reference frame data are respectively set to “frame”. It is called “alignment amount” and “frame alignment direction”.

  After the alignment between the optical frame data and the reference frame data is completed, the algorithm comparison processing unit (117c1, 117c2,...) Performs defect detection according to an appropriate comparison algorithm. For example, the level difference for each pixel between the optical frame data and the reference frame data is evaluated, and the differential values of the pixels in the pattern edge direction are compared. When the difference between the optical image data and the reference image data exceeds a predetermined threshold value, the location is determined as a defect.

  For example, the threshold value when registered as a line width defect is specified in units of a dimension difference (nm) and a dimension ratio (%) of the line width (CD: Critical Dimension) between the optical image data and the reference image data. For example, two threshold values are designated such that the dimensional difference in line width is 16 nm and the dimensional ratio is 8%. When the pattern of the optical image data has a line width of 200 nm, if the dimensional difference from the reference image data is 20 nm, the pattern has a defect because it is larger than both the dimensional difference threshold and the dimensional ratio threshold. It is determined.

  Note that the defect determination threshold can be specified separately for a case where the line width is larger than that of the reference image data and a case where the line width is thinner than the reference image data. Further, instead of the line width, a threshold value may be specified for each of the case where the space width between lines (distance between patterns) is thicker and thinner than the reference image data. Furthermore, for a hole-shaped pattern, a hole diameter dimension and a threshold ratio of the diameter ratio can be designated. In this case, the threshold value can be specified for each of the X-direction cross section and the Y-direction cross section of the hole.

  In addition to the above, algorithms used for defect detection include, for example, a level comparison method and a differential comparison method. In the level comparison method, for example, the luminance value of each pixel in the optical frame data, that is, the luminance value of the region corresponding to the pixel of the photodiode array 104b is calculated. Then, the defect is detected by comparing the calculated luminance value with the luminance value in the reference frame data. In the differential comparison method, the amount of change in luminance value in units of pixels in the direction along the edge of the fine pattern on the optical frame data, for example, the direction along the edge of the line pattern is obtained by differentiation. A defect is detected by comparing the amount of change with the amount of change in luminance value in the reference frame data. Note that a physical quantity value such as a line width dimensional difference, a dimensional ratio, a luminance value, and a luminance value change amount used for defect detection by a comparison algorithm is referred to as a “defect reaction value”.

  When the algorithm comparison processing unit in FIG. 4 determines that the optical frame data has a defect, defect information such as the optical frame data, the defect position coordinate data, and the compared reference frame data is stored in the defect registration unit (117d1). , 117d2... Specifically, the position coordinate data of the defect, the defect reaction value that is the basis for the defect detection, the optical frame data and the reference frame data that are the basis for the defect detection, and the coordinate system at the time of alignment in the frame alignment unit The type of frame data, the amount of frame alignment, and the frame alignment direction (hereinafter referred to as “frame alignment information”) are registered for each frame.

  The comparison unit 117 performs a series of comparisons of frame data alignment, defect detection, and totalization of the number of detected defects for each set of optical frame data and corresponding reference frame data and for each comparison algorithm. The determination operation is performed a plurality of times while changing the alignment condition of the frame data, and the defect detection result in the comparison determination operation with the smallest number of defect detections can be registered in the defect registration unit.

  For example, in the first comparison / determination operation, alignment is performed by placing a weight on an area corresponding to the upper right side of the optical image corresponding to the optical frame data, and in the second comparison / determination operation, the optical image corresponding to the optical frame data is aligned. Alignment is performed by placing weight on the area corresponding to the lower right part of the inside, and in the third comparison judgment operation, alignment is performed by placing weight on the area corresponding to the lower left part in the optical image corresponding to the optical frame data. In the fourth comparison / determination operation, alignment is performed by placing emphasis on a region corresponding to the upper left portion of the optical image corresponding to the optical frame data. Then, the number of detected defects in each of the four comparison determination operations can be compared, and the defect detection result in the comparison determination operation with the smallest number of defect detections can be registered.

  The specific number of times when the comparison determination operation is performed a plurality of times while changing the alignment condition of the frame data is not limited to the above four times, and can be a desired number of times of two or more. In addition, the condition for aligning the frame data is not limited to the above. For example, it is possible to determine which region in the optical image corresponding to the optical frame data is to be weighted by performing individual comparison and determination. It can be set appropriately according to the operation.

  As described above, the optical image data and the reference image data are sequentially taken into the comparison unit 117 and subjected to the comparison process, whereby defect detection in the optical image data is performed. Here, in the same mask Ma, when the number of defects registered in the defect registration unit suddenly increases during the inspection, the number of true defects may have increased, but the number of pseudo defects has increased. There is a possibility. In the latter case, one of the causes is an abnormal status of the inspection apparatus 100. For example, fluctuations in the output of the light source 103a, meandering of the table 101, or an abnormality in the autofocus mechanism of the imaging unit 104.

  When the above-described status abnormality occurs in the inspection apparatus 100, an accurate inspection result cannot be obtained. For example, when the output of the light source 103a changes and the amount of light varies, the output of the photodiode array 104b also varies. On the other hand, the gain of the digital amplifier determined by the calibration described above is adjusted before the output of the light source 103a changes. Therefore, since it is different from the actual gain, it is easy to induce a pseudo defect.

  When a status abnormality occurs in the inspection apparatus 100, the inspection apparatus 100 automatically detects such abnormality and stops or terminates the inspection, so that a lot of pseudo defects are included in the inspection result. Inspection can be prevented in advance. For example, when the output of the light source 103a changes and the amount of light changes, the amount of change can be detected by providing a light amount sensor (not shown) in the illumination optical system 103. When the fluctuation amount detected by the light quantity sensor exceeds a predetermined allowable range, the inspection apparatus 100 automatically stops or ends the inspection. Also, when the table 101 meanders, or when the autofocus mechanism does not operate normally and the focus position cannot be determined, the inspection apparatus 100 automatically stops the inspection by quantifying those abnormalities. Can be terminated.

  On the other hand, even if the status changes to such an extent that the inspection apparatus 100 does not automatically stop or end the inspection, a pseudo defect may occur. This means that the size of defects that should be pointed out as shape defects and line width fluctuations is comparable to line width fluctuations (line width distribution) and pattern positional error (positional error distribution) over the entire mask. This is particularly a problem with recent mask patterns.

If the allowable range of the variation amount of the status of the inspection apparatus 100 is narrowed, the inspection apparatus 100 can automatically stop or end the inspection even when there is a small variation amount. Since it becomes an unnecessary operation, it is not realistic. Further, in addition to the status change of the inspection apparatus 100, the causes of the pseudo defect include a case where the reference image generation function is not appropriate and a case where the plate alignment is not appropriate. The plate alignment is performed immediately after the mask is loaded on the inspection apparatus and before the start of the inspection. However, due to temperature differences inside and outside the inspection device, changes in the temperature of the mask itself caused by the light irradiated to the mask during inspection, and changes in atmospheric pressure, the laser measurement system may cause measurement errors or the optical system If changes occur, the plate alignment may change over time. Therefore, a process of re-executing plate alignment every predetermined time is also incorporated into the inspection. However, even in such a case, a situation in which the number of detected defects suddenly changes before plate alignment is reexecuted is unavoidable.

  Therefore, in the present embodiment, when the number of defects detected in the middle of the inspection increases rapidly (as will be described later), the inspection apparatus 100 analyzes the defects in the defect analysis step S4 of FIG. In the recovery step S5, calibration is re-executed, the reference image generation function is re-determined, and plate alignment is re-executed.

  A variation in the number of defects detected by the inspection is grasped by the comparison unit 117. Specifically, in addition to registering the above-described defect detection results, the defect registration unit adds up the number of detected defects as needed in stripe units. When the total number of detections increases compared to the number of detections in the previous stripe, specifically, when the inspection continuation allowable value is exceeded, the inspection apparatus 100 analyzes the defect and determines the action to be taken. .

  The inspection continuation allowable value can be defined by, for example, the number of defects per stripe or the rate of increase with respect to the number of detected defects in the inspected stripe. When the inspection continuation allowable value is defined by the number of defects per stripe, the value can be in the range of about 50 to about 100, for example. In addition, when the inspection continuation allowable value is defined by the increasing rate with respect to the number of detected defects in the inspected stripe, it can be within a range of about 50 times or less. The inspection continuation allowable value is further determined within this range according to experiments and empirical rules, and stored in the magnetic disk device 119. Hereinafter, stripes in which the number of detected defects exceeds the inspection continuation allowable value are referred to as “frequently occurring stripes”.

  When the number of detected defects counted by the defect registration unit is equal to or smaller than the inspection continuation allowable value, the defect detection result registered in the defect registration unit is sent to the magnetic disk device 119. Further, the comparison unit 117 outputs a signal (inspection continuation instruction) instructing continuation of the inspection. Then, the inspection apparatus 100 starts the defect detection operation for the stripe in which the number of detected defects is tabulated, that is, the stripe adjacent to the stripe for which the comparison determination operation in the comparison unit 117 is completed. This flow is indicated by arrows from the defect detection step S3 to the continuation of inspection in FIG.

  On the other hand, when the number of detected defects counted by the defect registration unit exceeds the inspection continuation allowable value, the comparison unit 117 sends a signal (defect analysis instruction) instructing the start of defect analysis to the defect analysis unit 118. In addition, a defect detection result corresponding to a stripe whose detected number exceeds the inspection continuation allowable value is sent to the defect analysis unit 118. Subsequently, the defect analysis unit 118 performs a defect analysis step S4.

<Defect analysis step S4>
FIG. 5 is a flowchart schematically showing the flow of processing performed by the defect analysis unit 118.

  As already described, when a status abnormality exceeding a predetermined allowable range occurs in the light source 103a or the table 101 of the inspection apparatus 100, the inspection apparatus 100 automatically stops or ends the inspection. This step is indicated by S10 in FIG.

  On the other hand, a pseudo defect may occur even if the inspection apparatus 100 does not automatically stop or end the inspection. Specifically, the values for the status set in the inspection apparatus 100 are all within the allowable range, but the number of defects registered in the defect registration unit with the same mask Ma increases rapidly during the inspection. It is. In FIG. 5, this corresponds to the case where no abnormal status of the inspection apparatus 100 is detected in S10 and a defect-prone stripe occurs in S11. In such a case, the defect analysis unit 118 analyzes the defect whose detection number has increased rapidly and takes an appropriate action.

First, the defect analysis unit 118 analyzes the defect, identifies the cause causing the increase in the number of defects exceeding the inspection continuation allowable value, and classifies the cause into any of the following (1) to (4). Subsequently, the defect analysis unit 118 takes measures according to the causes classified above.
(1) Frequent occurrence of true defects (2) Status fluctuation of inspection equipment (3) Inappropriate reference image generation function (4) Insufficient accuracy of alignment (plate alignment)

  With reference to FIG. 4, each part constituting the defect analysis unit 118 will be described.

  The defect analysis unit 118 includes a defect reaction value trend statistics unit 118a, a line width error determination unit 118b, a coordinate feature determination unit 118c, and a frame alignment integrated statistics unit 118d. The cause of the sudden increase is identified.

  The defect reaction value trend statistics unit 118a analyzes the defect reaction value tendency and the defect distribution tendency for each defect detected in the defect-prone stripe. For example, when the defect reaction value is high and the area of the defect is large, it is determined as “large defect”. Further, for example, when defects are concentrated in a region other than the edge of the pattern, there is a possibility of a “haze defect”. A haze defect is a growth defect that is caused by foreign matter adhering to glass and increases by ultraviolet rays. The haze defect has a lower defect reaction value than other defects. Therefore, when defects are distributed in a region other than the edge of the pattern and the defect reaction value is low, it can be determined as a “haze defect”.

  Both “large defects” and “haze defects” are determined to be due to “(1) frequent occurrence of true defects” and are not pseudo defects. Note that the defect reaction value used as a reference for determining a “large defect” or “haze defect” is the luminance value of the defect in the defect-prone stripe and the line of the pattern constituting the defect (in the case of a line width defect). A value obtained from at least one of the width error values, for example, a luminance value, a change amount of the luminance value, a line width dimensional difference, and a dimensional ratio. The threshold value (first threshold value) of the defect reaction value determined to be a haze defect is determined to an appropriate value based on an empirical rule including experiments and stored in the magnetic disk device 119.

  The line width error determination unit 118b analyzes the trend of the actual size error of the line width or the error ratio of the line width in each defect detected by the defect-prone stripe. For defects whose actual width error or line width error ratio exceeds the allowable range, it is checked whether the ratio of such defects to the entire defect is equal to or less than a predetermined threshold value (third threshold value). If the number of defects exceeds the third threshold value, it is determined that the defect is a pseudo defect due to the light amount fluctuation of the light source 103a in the inspection apparatus 100, that is, “(2) fluctuation in the status of the inspection apparatus”.

  For example, it is assumed that the allowable range of the dimensional difference between the actual line width and the design value is ± 16 nm and the allowable range of the dimensional ratio is ± 8%. When the line width of a pattern detected as a defect in a defect-prone stripe is 200 nm and the dimensional difference from the reference image data is +20 nm, it can be seen that the line width of this pattern is larger than the design value. If the ratio of such defects to the entire defect exceeds a predetermined third threshold, it is determined that the defect is a pseudo defect caused by an increase in the light amount of the light source 103a in the inspection apparatus 100. On the other hand, when the ratio of the defect having a line width smaller than the design value to the entire defect exceeds the third threshold, it is determined that the defect is a pseudo defect caused by a decrease in the light amount of the light source 103a. The third threshold value is determined to be a reasonable value to continue the inspection based on empirical rules including experiments, and is stored in the magnetic disk device 119.

  Even when the reference image data cannot be generated due to an abnormality in the circuit or processing system of the reference image generation unit 116, a frequent defect stripe occurs. Such an abnormality can be grasped by performing a die-to-die comparison of optical image data of a defect-prone stripe. If the number of defects detected by the die-to-die comparison method is equal to or smaller than the inspection continuation allowable value, it is determined that there is an abnormality in the reference image data generation step S2, and the inspection is stopped or terminated.

  The coordinate feature determination unit 118c analyzes the tendency of the defect reaction value using the position coordinate data, and calculates the density of the defects detected in the defect frequent stripes. Then, it is checked whether or not the obtained density is equal to or less than a predetermined threshold value (second threshold value). When the density exceeds the second threshold value, the position coordinates (x, y) of each defect are further increased. It is examined whether the value of x or the value of y is unevenly distributed within a range of one or more specific numerical values. In the case of uneven distribution, the defect in the stripe that is continuously arranged in the target frequent defect stripe, and the defect in the stripe that has been processed in the comparison unit 117 prior to the frequent defect stripe and its surrounding optical image, If there is no tendency of pattern deviation similar to the optical image of the defect in the defect-prone stripe, it is determined that the defect is a pseudo defect induced by “(3) inappropriate reference image generation function”. Note that the number of stripes that have been processed in the comparison unit 117 first can be 1 or more and 10 or less, and preferably 2 or more and 5 or less. The defect information of these stripes is read from the defect detection result stored in the magnetic disk device 119. The second threshold value and the uneven distribution threshold value are determined to be appropriate values based on empirical rules including experiments, and stored in the magnetic disk device 119.

  Many of the pseudo defects induced by “(3) inappropriate reference image generation function” have a low defect response value. In addition, the pattern in which such a pseudo defect occurs includes an auxiliary pattern (SRAF: Sub-Resolution Assist Feature) for increasing the resolution of the pattern drawn on the mask and an intricate shape (Inret shape) pattern. There are many. The coordinate feature determination unit 118c preferably determines with reference to information based on such empirical rules.

  The frame alignment integrated statistics unit 118d uses the defect detection result for the defect frequent stripe sent from the comparison unit 117 and the defect detection result stored in the magnetic disk device 119 to detect the defect frequent stripe and the defect frequent occurrence. The frame alignment amount of the frame data in which the coordinate system is displaced at the time of frame alignment between the stripes arranged in succession to the stripes and the stripes that have been processed by the comparison unit 117 prior to the defect-prone stripes Analyze whether there is similarity in the frame alignment direction. Further, it is checked whether or not the density of defects detected in the defect-prone stripe is equal to or lower than a predetermined second threshold value, and the x value or y value of the position coordinates (x, y) of each defect is determined. Check if it is unevenly distributed within one or more specific numerical ranges.

  In the frame alignment accumulation statistics unit 118d, the number of stripes that have been processed in the comparison unit 117 can be 1 or more and 10 or less, and preferably 2 or more and 5 or less. Also, the second threshold value and the uneven distribution threshold value are determined to be appropriate values based on empirical rules including experiments, and stored in the magnetic disk device 119. These can be the same as the coordinate feature determination unit 118c.

  The defect density exceeds the second threshold, and the x value or y value of the position coordinates (x, y) of each defect is unevenly distributed within one or more specific numerical ranges. Further, if there is a similarity between the frame alignment amount and the frame alignment direction, it is determined that the defect is a pseudo defect induced by “(4) insufficient accuracy of alignment (plate alignment)”.

  The coordinate feature determination unit 118c and the frame alignment accumulation statistics unit 118d calculate the density of the defects detected in the frequent defect stripes, and check whether the obtained density is equal to or lower than a predetermined second threshold value. The process is common to the process of examining whether or not the x value or y value of the position coordinates (x, y) of each defect is unevenly distributed within one or more specific numerical ranges. Therefore, one of the coordinate feature determination unit 118c and the frame alignment accumulation statistics unit 118d can perform these processes, and the other can use the result.

  According to FIG. 5, the defect analysis unit 118 analyzes the defect and identifies the cause of the rapid increase in each of the defect reaction value trend statistics unit 118a, the line width error determination unit 118b, the coordinate feature determination unit 118c, and the frame alignment integration statistics unit 118d. And take appropriate action according to the cause. Note that the order of processing in the defect analysis unit 118 is not limited to the example of FIG. 5, and the order can be appropriately changed, and the processing may be performed in parallel.

  When a defect frequent stripe occurs in S11 of FIG. 5, the defect reaction value trend statistics unit 118a analyzes the defect reaction value tendency and the defect distribution tendency of each defect detected in the defect frequent stripe in S12. As a result, if it is determined that the cause of the rapid increase in the number of defects is due to “(1) frequent occurrence of true defects” and not a pseudo defect, the defect detection results of these defects are sent to the magnetic disk device 119. It is done. Further, the defect analysis unit 118 outputs a signal (inspection continuation instruction) instructing continuation of the inspection. Then, for example, the inspection apparatus 100 starts a defect detection operation for a stripe adjacent to the frequent defect stripe. This flow is indicated by an arrow from S12 to continuation of the inspection in FIG.

  If it is determined by the analysis of S12 that there is no characteristic tendency of “large defect” or “haze defect” in each defect detected in the frequent defect stripe, the process proceeds to S13 in FIG. Then, in the line width error determination unit 118b, the actual line width error or the tendency of the line width error ratio in each defect detected in the defect-prone stripe is analyzed. As a result, for the defect whose line width actual size error or line width error ratio exceeds the allowable range, the ratio of the defect to the entire defect exceeds a predetermined third threshold value. It is determined that this is a pseudo defect due to the light amount variation of the light source 103a, that is, “(2) variation in status of the inspection apparatus”. Next, the defect analysis unit 118 outputs a signal (calibration re-execution instruction) instructing re-execution of calibration in the sensor circuit 104c.

  In S13, the ratio of the actual line width error or the line width error ratio of each defect detected in the defect-existing stripe exceeding the allowable range to the entire defect exceeds the predetermined third threshold value. If not, the process proceeds to S14 in FIG. Then, the coordinate feature determination unit 118c analyzes the tendency of the defect reaction value using the position coordinate data, and calculates the density of the defects detected in the frequent defect stripes. The obtained density exceeds a predetermined second threshold, and the x value or y value of the position coordinates (x, y) of each defect is unevenly distributed within one or more specific numerical ranges. If yes, the process proceeds to S15.

  On the other hand, in the coordinate feature determination unit 118c in S14, when the density of defects detected in the defect-prone stripe is equal to or lower than the second threshold, or the value x of the position coordinates (x, y) of each defect or y If the value is not unevenly distributed within one or a plurality of specific numerical ranges, it is determined that it is due to “(1) frequent occurrence of true defects” and is not a pseudo defect. The defect detection results of these defects are sent to the magnetic disk device 119. Further, the defect analysis unit 118 outputs a signal (inspection continuation instruction) instructing continuation of the inspection. Then, the inspection apparatus 100 starts a defect detection operation for a stripe adjacent to the frequent defect stripe. This flow is indicated by an arrow from S14 to continuation of the inspection in FIG.

  In S15, the frame alignment integrated statistics unit 118d uses the defect detection result for the defect frequent stripe sent from the comparison unit 117 and the defect detection result stored in the magnetic disk device 119, and A frame of frame data in which the coordinate system is displaced during frame alignment between the stripes arranged successively to the frequent defect stripes and the stripes that have been processed by the comparison unit 117 prior to the frequent defect stripes. It is analyzed whether there is similarity between the alignment amount and the frame alignment direction.

  In addition, the frame alignment accumulation statistics unit 118d determines that “the density of the defects detected in the defect-prone stripe exceeds the predetermined allowable range, and the value x or y of the position coordinates (x, y) of each defect. "Is distributed unevenly within one or more specific numerical ranges" from the coordinate feature determination unit 118c.

  In S15, when there is no similarity between the frame alignment amount and the frame alignment direction of the frame data, the frame alignment integrated statistics unit 118d determines the defect detected by the defect-prone stripe by “(3) inappropriate reference image generation function”. It is determined as an induced pseudo defect. Then, the defect analysis unit 118 outputs a signal (function redetermination instruction) instructing redetermination of the reference image generation function.

  On the other hand, when there is a similarity between the frame alignment amount of the frame data and the frame alignment direction, the frame alignment integrated statistical unit 118d determines that the defect detected by the defect-prone stripe is “(4) insufficient alignment (plate alignment) accuracy”. Is determined to be a fake defect induced by. Next, the defect analysis unit 118 outputs a signal (plate alignment re-execution instruction) instructing re-execution of plate alignment.

<Recovery process S5>
When any one of the calibration re-execution instruction, the function re-determination instruction, and the plate alignment re-execution instruction is output from the defect analysis unit 118, the recovery step S5 in FIG. 2 is performed. The recovery process S5 includes processes corresponding to each signal, that is, a calibration re-execution process, a function re-determination process, and a plate alignment re-execution process.

  In the calibration re-execution step, for example, the control computer 110 adjusts the white level of the sensor circuit 104c to lower the contrast of the optical image corresponding to the optical image data, and for the region where the brightness was relatively low when the mask Ma was captured. Perform light intensity calibration to increase resolution. Alternatively, the control computer 110 adjusts the black level of the sensor circuit 104c to increase the contrast of the optical image corresponding to the optical image data, and performs light amount calibration to reduce the resolution for the area that was relatively bright when the mask Ma was captured. Do.

  In the function redetermination step, the control computer 110 reads out the defect detection result for the frequent defect stripe from the magnetic disk device 119 and sends it to the reference image generation function calculation unit 115. The reference image generation function calculation unit 115 automatically extracts machine learning learning points using optical image data of defect-prone stripes, and generates a reference image generation function so that the learning points are treated as not defective. The learning point is data of a pattern shape included in the frequent defect stripes and is data to be machine-learned.

  The automatic extraction of machine learning learning points is performed by, for example, obtaining the median value xm of the x value and the median value ym of the y value with respect to the XY coordinates (x, y) of a region where defects are unevenly distributed in the defect-prone stripe. Calculate and extract the defect closest to the coordinate (xm, ym) (including the case where there is a defect on the coordinate (xm, ym)) and the shape of the fine pattern around the defect. Can do. It should be noted that it is possible to appropriately select how wide the shape of the fine pattern around the defect is to be extracted.

  In the automatic extraction of machine learning learning points, each defect detected in a defect-prone stripe is classified based on the shape of a fine pattern around the defect, and one defect and this defect are arranged in descending order of appearance frequency. It is also possible to extract the shape of a fine pattern around the machine as a learning point for machine learning. In this case, the reference image generation function calculation unit 115 can be configured to automatically extract one defect having the highest appearance frequency and the shape of a fine pattern around the defect as a learning point for machine learning. The reference image generation function calculation unit 115 is configured to automatically extract the defect and the shape of the fine pattern around the defect as the learning points of machine learning for two or more desired numbers of defects in order of highest appearance frequency. You can also

  In the plate alignment re-execution step, the above-described plate alignment is performed again, and alignment is performed so that the pattern to be inspected of the mask Ma is accurately imaged at a predetermined position on the table 101.

  When the recovery step S5 is completed, the inspection apparatus 100 restarts the inspection with a predetermined number of stripes in the inspection order. That is, among the stripes that have been processed by the comparison unit 117 prior to the frequent defect stripes, the inspection is resumed by returning from the stripe that has been inspected immediately before the frequent defect stripes to the predetermined stripe in the inspection order. The number of stripes that can be traced back can be set to a desired value of 1 or more, and about 5% of the total number of stripes in the mask Ma can be set as the upper limit value. When the inspection is resumed, the optical image data acquisition step S1, the reference image data generation step S2, and the defect detection step S3 in FIG. 2 are sequentially performed.

  As described above, according to the inspection method of the present embodiment, when the number of detected defects increases rapidly during the inspection, these defects are analyzed and appropriate measures are taken. In addition, since the inspection is continued, not only the occurrence of pseudo defects is prevented, but also the inspection results before the number of defects rapidly increase are not wasted. Therefore, the convenience of inspection can be improved.

  In addition, according to the inspection apparatus of the present embodiment, since it has a function of estimating the cause of the rapid increase in the number of defects and a function of determining an appropriate treatment according to the function, it is possible to prevent the occurrence of pseudo defects and It is possible to use the inspection results before the surge increases, and to reduce the burden on workers.

Embodiment 2. FIG.
FIG. 6 is a schematic configuration diagram of the inspection apparatus according to the present embodiment. The inspection apparatus 100A in FIG. 6 is provided with a reference image generation function calculation unit 115A instead of the reference image generation function calculation unit 115 illustrated in FIG. 1 and a learning point input support unit 126, respectively. The configuration is the same as that of the inspection apparatus 100 shown in FIG. Among the constituent elements shown in FIG. 6, the remaining constituent elements excluding the reference image generation function calculation unit 115A and the learning point input support unit 126 are denoted by the same reference numerals as those used in FIG. The description is omitted.

  Similar to the reference image generation function calculation unit 115 shown in FIG. 1, the reference image generation function calculation unit 115A that constitutes the inspection apparatus 100A automatically extracts machine learning learning points using optical image data of defective stripes. Thus, the reference image generation function is re-determined so that the learning point is treated as not defective. Further, the reference image generation function calculation unit 115A treats the machine learning learning point pointed out by the operator as a defect using the input assistance function of the learning point input assistance unit 126, or treats the learning point as not defective. And a function of re-determining the reference image generation function.

  The inspection method of the present embodiment implemented using the inspection apparatus 100A includes the optical image data acquisition step S1, the reference image data generation step S2, the defect detection step S3, and the defect analysis step S4 described in the first embodiment. . After the defect analysis step S4, a recovery step is also performed in the present embodiment, but after the signal (function redecision instruction) for instructing the re-determination of the reference image generation function is output from the defect analysis unit 118. It differs in the function redetermination process. The other processes in the recovery process, that is, the calibration re-execution process and the plate alignment re-execution process are the same as those in the first embodiment.

  Hereinafter, a method for re-determining the reference image generation function according to the present embodiment will be described.

  When the function redetermination instruction is output from the defect analysis unit 118, the learning point input support unit 126 causes (1) the reference image generation function calculation unit 115A to automatically extract learning points for machine learning, or (2) an operator. Outputs a signal (selection screen display instruction) for instructing the CRT 122 to display, for example, a screen for selecting whether machine learning learning points are manually extracted. When the manual extraction of (2) is selected and a signal indicating that (manual extraction instruction) is input from the input device 125, the learning point input support unit 126 displays the learning point input screen on the pattern monitor 123, for example. A signal (input screen display instruction) instructing display is output. Here, the learning point input screen refers to an input screen on which an operator can point out a learning point to be treated as a defect or a learning point to be treated as not a defect on an optical image of a defect-prone stripe.

  When the operator points out a learning point on the learning point input screen, the learning point input support unit 126 determines whether it is appropriate to employ the learning point. If it is not appropriate, a signal (input support instruction) for instructing to display information for recommending other learning points on the learning point input screen is output. On the other hand, when the learning point pointed out by the operator is appropriate, the learning point input support unit 126 instructs the reference image generation function calculation unit 115A to re-determine the reference image generation function with this learning point as a defect (or not a defect). Instruct.

  The suitability of adoption as a learning point is determined, for example, by determining the number of pattern vertices, vertex density, pattern density, etc. using layout analysis software, and whether these match the recommended conditions. Further, the image data of the screen for selecting the above (1) or (2) and the image data of the learning point input screen are stored in the magnetic disk device 119 in advance. Furthermore, typical data of learning points can be stored in the magnetic disk device 119 in advance. In this case, it is preferable to store data of appropriate learning points pointed out by the operator in the magnetic disk device 119 as well.

  7 to 9 are plan views showing typical examples of learning points. 10 to 12 are plan views showing examples of learning points pointed out by the operator. In these drawings, the white area indicates a space pattern on the mask Ma, and the other area indicates a line pattern on the mask Ma.

  A learning point LP1 shown in FIG. 7 is an example of a figure including a large pattern edge. A learning point LP2 shown in FIG. 8 is another example of a figure including a large pattern edge. Furthermore, the learning point LP3 shown in FIG. 9 is an example of a typical fine pattern in the mask Ma.

  FIG. 10 shows an example in which the shape of the corners of the fine pattern does not match between the optical image corresponding to the optical image data and the reference image corresponding to the reference image data. In this figure, each corner of the fine pattern is set as learning points LP11 to LP14. In FIG. 10, the solid line represents the contour of the optical image, and the dotted line represents the contour of the reference image. However, the solid line may represent the contour of the reference image, and the dotted line may represent the contour of the optical image. .

  FIG. 11 shows an example in which the luminance amplitude is insufficient compared to the luminance amplitude in the reference image corresponding to the reference image data in a part of the region in the optical image corresponding to the optical image data. In this figure, the areas where the luminance amplitude value is insufficient are the learning points LP21 and LP22. Note that the upper diagram in FIG. 11 is a plan view showing a partial region in the optical image corresponding to the optical image data of the mask Ma, and the lower diagram is a line AA in the upper diagram. It is a graph which shows the luminance amplitude in a cross section.

  FIG. 12 shows an example in which an optical image corresponding to the optical image data includes a pattern having an intricate shape (Inret shape), and this shape is used as a learning point LP31. In addition, when the auxiliary pattern (SRAF: Sub-Resolution Assist Feature) is included in the optical image, this auxiliary pattern can be used as a learning point.

  Since the inspection apparatus 100A according to the present embodiment has the function of estimating the cause of the rapid increase in defects and the function of determining appropriate measures according to the same as the inspection apparatus 100 of the first embodiment. In addition to preventing frequent occurrences of pseudo defects, it is possible to utilize inspection results before the number of defects increases, and to reduce the burden on workers.

  Furthermore, in the inspection apparatus 100A, the operator can point out the learning points for machine learning on the learning point input screen when re-deciding the reference image generation function, and thus has only a function for automatically extracting learning points for machine learning. Compared to the inspection apparatus 100, the degree of freedom in selecting learning points is high. Therefore, according to the inspection apparatus 100A, a reference image generation function more appropriate than the inspection apparatus 100 can be determined by the reference image generation function calculation unit 115A.

  For example, let us consider a case where the line width of a fine pattern formed on the mask Ma changes in the manufacturing process. The variation value of the line width varies depending on the position where the fine pattern is formed, and it is assumed that distribution is generated in the variation value of the line width when viewed over the entire surface of the mask Ma. In such a case, prior to starting the inspection of the mask Ma, a plurality of representative fine patterns on the mask Ma are extracted, and a reference image generation function is generated by the reference image generation function calculation unit 115A using these optical image data. In the conventional example, the reference image generation function may not be able to properly detect a defect in a region other than the extracted portion, resulting in a pseudo defect.

  FIG. 13 is a diagram illustrating an example of the in-plane distribution of the variation value of the line width of the fine pattern in the mask Ma. The upper diagram in FIG. 13 is a map of fluctuation values of the line width, and the lower diagram shows the fluctuation values of the line width of the fine pattern in the BB line cross section in the upper diagram.

  According to the inspection apparatus 100A of the present embodiment, the reference image generation function is re-determined by the reference image generation function calculation unit 115A using the optical image data of the machine learning learning point pointed out by the operator. Even if the distribution as described above is generated in the variation value of the line width of the pattern, the reference image data can be generated using an appropriate reference image generation function for each region having a different variation value. Therefore, it is possible to perform an accurate inspection while suppressing pseudo defects.

  Further, in the inspection apparatus 100A, when an operator points out a learning point that is not suitable as a machine learning learning point on the learning point input screen, information for recommending other learning points is displayed on the pattern monitor 123, for example. . This makes it easy for an operator who is not familiar with the selection of learning points to generate an appropriate reference image generation function in the reference image generation function calculation unit 115A.

  The embodiments of the inspection apparatus and the inspection method of the present invention have been described above, but the present invention is not limited to the inspection apparatus and the inspection method described in the embodiments. Various changes, improvements, combinations and the like are possible for the present invention. All inspection apparatuses and inspection methods that include elements of the present invention and that can be appropriately modified by those skilled in the art are included in the scope of the present invention.

100, 100A Inspection apparatus 101 Table 104 Imaging unit 113 Position information unit 114 Table control unit 115, 115A Reference image generation function calculation unit 116 Reference image generation unit 117 Comparison unit 118 Defect analysis unit 118a Defect reaction value trend statistics unit 118b Line width error Determination unit 118c Coordinate feature determination unit 118d Frame alignment integrated statistics unit 126 Learning point input support unit Ma mask St stripe F frame S1 optical image data acquisition step S2 reference image data generation step S3 defect detection step S4 defect analysis step S5 recovery step

Claims (7)

An inspection apparatus for inspecting a defect of a pattern formed on a substrate,
A table on which the substrate is placed;
An imaging unit that images the inspected area for each stripe that virtually divides the inspected area in which the pattern of the base material is formed; and generates optical image data;
A reference image generation unit that generates a reference image data by performing a filtering process using a function on the image data generated from the design data of the pattern;
A reference image generation function calculation unit for obtaining the function;
Optical frame data obtained by dividing the optical image data into a predetermined size and reference frame data obtained by dividing the reference image data corresponding to the optical image data into the size are compared to detect defects in the optical image data. And a comparator for counting the number of detected defects for each stripe,
For stripes in which the number of detected defects counted by the comparison unit exceeds an allowable value that permits continuation of inspection, whether the number of defects in the stripe is due to the occurrence of true defects or the optical image data corresponding to the optical image data Whether the function that generates the reference image data is a function that is incompatible or due to a lack of alignment accuracy between the pattern and the table, information on defects in the stripe output from the comparison unit is used. An inspection apparatus comprising: a defect analysis unit that analyzes using the defect analysis unit.   The defect analysis unit has a value obtained from at least one of a luminance value of a stripe defect exceeding the allowable value and a line width error value of the pattern constituting the defect that is equal to or less than a first threshold value, and the defect 2 is output in a signal instructing continuation of inspection, assuming that the frequent occurrence of the defect is caused by the frequent occurrence of the true defect. Inspection equipment. The defect analysis unit
The density of defects detected in stripes exceeding the tolerance exceeds a second threshold, and the x value or y value of the position coordinates (x, y) of the defect is one or more specified. If it is unevenly distributed in the numerical range,
Compared with the stripes arranged successively to the stripes, the tendency of the pattern shift between the stripes of 1 to 10 stripes that have been processed in the comparison unit prior to the stripes,
Analyzes whether the number of defects in the stripe exceeding the allowable value is due to a function that is incompatible with the function that generates the reference image data corresponding to the optical image data, or due to insufficient alignment accuracy And
If the function that generates the reference image data corresponding to the optical image data is based on an incompatible function, a signal that instructs to re-determine the function is output,
3. The inspection apparatus according to claim 1, wherein a signal instructing re-execution of the alignment is output when the alignment accuracy is insufficient. 4.   The inspection apparatus according to claim 3, wherein the reference image generation function calculation unit re-determines the function by machine learning of pattern shape data included in a stripe exceeding the allowable value. A display device for displaying an optical image of a stripe exceeding the allowable value;
When the optical image is displayed on the display device, an input screen capable of instructing the pattern shape data to be machine-learned by an operator is also displayed on the display device, and the operator treats it as a defect on the input screen. When the data of the pattern shape to be specified is specified, whether or not the data is adopted is determined. If the data is not appropriate, information recommending other data is displayed on the input screen. The inspection apparatus according to claim 4, further comprising: a data input support unit that instructs the reference image generation function calculation unit to re-determine the function by machine learning of the data. An inspection method for inspecting a defect of a pattern formed on a substrate,
Placing the substrate on a table;
Aligning the pattern with the table;
Imaging the inspected area for each stripe that virtually divides the inspected area in which the pattern of the substrate is formed, and generating optical image data;
Obtaining a function for performing filtering on the image data generated from the design data of the pattern;
Generating reference image data from the image data using the function;
Optical frame data obtained by dividing the optical image data into a predetermined size and reference frame data obtained by dividing the reference image data corresponding to the optical image data into the size are compared to detect defects in the optical image data. And counting the number of detected defects for each stripe,
For stripes in which the number of detected defects counted by the comparison unit exceeds an allowable value that permits continuation of inspection, whether the frequent occurrence of defects in the stripe is due to frequent true defects or the reference corresponding to the optical image data Analyzing whether the function for generating image data is a function that is incompatible or due to a lack of alignment accuracy, using information on defects in the stripe output from the comparison unit; Have
A value obtained from at least one of a luminance value of a stripe defect exceeding the allowable value and a line width error value of the pattern constituting the defect is equal to or less than a first threshold value, and the defect is other than an edge of the pattern If the defect is distributed in the region, the inspection is continued as the frequent occurrence of the defect is due to the occurrence of the true defect,
The density of defects detected in stripes exceeding the tolerance exceeds a second threshold, and the x value or y value of the position coordinates (x, y) of the defect is one or more specified. If it is unevenly distributed in the numerical range,
Compared with the stripes arranged successively to the stripes, the tendency of the pattern shift between the stripes of 1 to 10 stripes that have been processed in the comparison unit prior to the stripes,
Analyzing whether the number of defects in the stripe exceeding the allowable value is due to a function that is incompatible with the function that generates the reference image data corresponding to the optical image data or due to the lack of accuracy of the alignment. ,
If the function that generates the reference image data corresponding to the optical image data is due to an incompatible function, the function is obtained again,
An inspection method characterized by re-adjusting the alignment if the alignment accuracy is insufficient.   The inspection method according to claim 6, wherein the function is recalculated by machine learning of pattern shape data included in a stripe exceeding the allowable value.

Images