DE3714011A1 - Method and arrangement for image comparison - Google Patents

Abstract

The aim of the invention is to create a method and arrangement for exact and effective checking of templates or intermediate templates, which are required for the production of integrated circuits. The way this aim is achieved is that a serial, binary actual image is generated from the templates or intermediate templates, that a serial, binary "target" image is generated from the data set which is used as the pattern for production of these templates or intermediate templates, and describes the structure of the templates, that deviations between the "target" and "actual" images are recognised as defects of the "actual" image, that the parameters which are required for classification into types, that is tone value, number of determined contour sections and presence of undetermined contour sections, are successively determined for the defects, and that the defects are uniquely classified into new types corresponding to the named parameters. It is essential here that the whole process, including the determination of the specific defect parameters, progresses in real time, independently of the density of the structure and defects. It is also important that data which is used to acquire the "target" image takes into account determined, device-dependent pseudo-deviations and locally oriented structure tolerances.

Description

The invention relates to a method and arrangement for exact Comparison of an objectified, two-dimensional Image with the template for producing the image serving CAD data set.

The invention comes in particular in the control of stencil products, for the manufacture of integrated circuits are required to apply. More applications are part of control processes on circuit boards, printing, Textile, cartographic and Ä. Products and optical and optoelectronic imaging devices possible.

A method is described in the published patent application DE 33 36 471 A1 and described an arrangement with which deviations an image of target data can be determined and classified. This method and this arrangement are opposite The following disadvantages of the present invention:

• - The determination of the parameters required for classification runs in a two-stage process. Faulty Place the object to be assessed for the actual Analysis can be scanned a second time.  
• - The classification-specific parameters are based on determined a process that is not real-time capable.
• - Only six types of defects are distinguished.
• - The arrangement does not allow that during a control run such target data are used for comparison be both positive and negative Tolerances, especially locally oriented tolerances consider.
• - Too high a defect density leads to incorrect classifications.

A method is described in the published patent application DE 33 36 470 A1 and described an arrangement with differences a first image with respect to a second image can be characterized in that the differences suppressed along existing edges in the first image will.

This procedure specifically takes into account of tolerances strictly based on edges. Locally oriented Tolerances affecting the relationship of neighboring structural elements can not take into account with this procedure are processed.

The aim of the invention is a method and an arrangement, which enable exact defect determination in real time operation, regardless of the structure of the image and the frequency the defects, and thus the good yield at Increase integrated circuit manufacturing.  

The invention has for its object the described To avoid disadvantages of the prior art.

This task is carried out in the characterizing part of the Claims 1 and 6 given teaching solved. Preferred further training are described in the subclaims.

The objectified image and the data leading to it Manufacturing have been made comparable, by optoelectronically making the objectified image sampled and a binary by means of threshold operation Actual image is generated and on the other hand according to the Creation of the object-serving image data with the help of a computer or an electronic circuit a binary target image is generated. Thus stand for the exact comparison 2 data records are available, the pictures, which consist of structural elements with a tonal value of one and put together zero, describe.

A target image adapted to the actual image is generated, to suppress the determined pseudo deviations. This is achieved in that on the one hand the determined, pictorial changes caused by the devices caused to produce the objectified Images and scanning required for control and on the other hand for binaryization the actual value defined threshold value are taken into account.

The line-by-line oriented pixels of the actual and target images are provided serially and synchronously for defect detection. Whether the pixel belongs to a defect is determined determined by antivalence analysis.

On the one hand, delay elements are used in addition to the defect information the current point that of the neighboring points in the current and previous line and on the other In addition to the target image information of the current point, the 8 neighboring points provided at the same time.  

With the help of this information, a log. circuit the parameters describing a defect. It is essential that in addition to the area, the perimeter and the extremal coordinates the parameters tone value information, number of determined Contour sections and presence of undetermined Contour sections, the starting points for a Type classification are to be determined.

The tone value information is separated into tone value one and Tone value determined zero, which gives the indication of mixed Tonal values is made possible. The tone value information of the defect is based on the actual picture.

A determined contour section of a defect is identified by defines a contour of a structure in the target image. Corresponding to the tonal value of the defect in the considered Contour point is a determined contour section identical or adjacent to the target contour.

The number of determined contour sections is about the number of changes between determined and undetermined Contour sections won. The number of Alternation gives twice the number of determined contour sections except for the case where there is no undetermined Contour section is present. To solve this Problems arise from the presence of undetermined Contour points determined.

The change between a determined and an undetermined Contour section is recognized in that the contour points, which is characterized by the presence of at least one mark neighboring pixels that do not belong to the defect, be detected and that the status regarding determinacy neighboring contour points is evaluated. A Defect contour point is determined if and only if dependent its tone value is either the current target pixel is a contour point or one of the 8 target image point neighbors.  

The named log. Circuit also has controlling Function. It organizes the clear summary of partial results of a defect, so that each defect represented by a resulting parameter set becomes. It is essential that the parameter determination for a defect for the pixel clocks for which in the current or previous line without interruption, a defect is displayed, that is then activated the determined parameters with possibly already existing parameters belonging to this defect, can be summarized that the problems unite Defect parts by combining the parameter sets and branching defects can be solved by assignment and that the end of the parameter determination for a defect by zero indication of a count associated with the reopening a defect begins at one, with each branch is increased by one and with each partial Is reduced by one.

The intermediate and final results of a defect are shown in stored in a defect memory. A defect becomes one fixed memory address assigned. Free memory addresses are taken from an availability store. Become defective parts united, the memory address that has become free made available in the availability store again. In the Defects processed on a line are identified by their memory addresses registered in a map memory. A second allocation memory contains the memory addresses, of the defects processed in the previous line.

If a defect is completely recorded, the log divides. circuit its address via a transmission memory to a computer with, which then takes the parameter set from the defect memory can retrieve and the address in the availability store again available. The log. circuit suppresses this process if the defect is a predetermined one Limit for the area falls below and represents the previously specified address in the availability memory available again.  

The computer divides the defects depending on the parameters Tone value information, number of determined contour sections and presence of undetermined contour sections in 9 classes:

The following shows an example of a division into 9 classes Table:

The invention is based on exemplary embodiments explained in more detail. The pictures show in detail:

Fig. 1: Arrangement for the detection and classification of deviations of objectified images compared to the data set used in the production of these images and defining the image content.

Fig. 2: arrangement for detecting and classifying deviations of materialized images compared to the serving and in the production of these images as a template to the image content data defining locally oriented taking into account permissible tolerances.

Fig. 3a: Example of a target structure

FIG. 3b caused by the transfer function of the equipment used in the preparation of objectified images and their scan for check-determined deviation from the in Fig Sturktur shown. 3a.

FIG. 4a: Example of a target image

Fig. 4b: Possible actual image according to the target structure shown in Fig. 4a

Fig. 4c: Extracted defect

Fig. 5a: Determined contour section of a defect with tone value Φ

Fig. 5b: deterministic contour portion of a defect with tone 1

Fig. 6a: target image

Fig. 6b: Possible actual image according to the target structure shown in Fig. 6a

Fig. 6c: Extracted defect with specification of essential processing lines and columns

Fig. 7a: target image

FIG. 7b: actual image with failures of the "hole"

Fig. 8a: Should image

FIG. 8b: actual image with failures of the "indentation"

Fig. 9a: target image

FIG. 9b: actual image with failures of the "interrupt"

FIG. 10a: target image

Fig. 10b: actual image with failures of the "missing structural element with one tone"

FIG. 11a: target image

Fig. 11b: actual image with failures of the "additional, non-defined structure element"

Fig. 12a: target image

Fig. 12b: actual image with failures of the "bulge"

FIG. 13a: target image

Fig. 13b: actual image with failures of the "bridge"

FIG. 14a: target image

Fig. 14b: actual image with failures of the type "missing structural element having tone value zero"

FIG. 15a: target image

Fig. 15b: actual image with failures of the "global defect"

Fig. 16: Target structure with representation of the locally oriented, permissible tolerance ranges

In Fig. 1, an arrangement for detecting and classifying deviations of materialized images compared to the serving and in the production of these images as a template image content defining data record is displayed.

A device 1 scans the objectified image optoelectronically, eliminates scanning-related errors, carries out binarization of the pixel-characteristic information by means of a threshold value and outputs the pixel information serially.

A device 2 consisting of a computer with a corresponding data memory or a log. Circuitry with a corresponding data memory or a combination of both generates a binary target image adapted to the actual image and outputs the pixel information serially.

The target image deviates meaningfully from the data set used to produce the object-like image in accordance with the determined visual changes which are caused on the one hand by the production process and on the other hand by the control process. Fig. 3a shows a section of serving as a template image. In Fig. 3b, the resulting visual change can be seen that, as is known, can be determined by convolution of the output image to the respective apparatus function. The determined visual changes also include the positional deviations between the structure edges displayed in the binaryized actual image and actually present in the objectified image.

The positional deviations depend on the binaryization of the Actual image predetermined threshold, the transmission parameters the scanning device and the geometric relationships from.

While the generation of the target image must take place online in real time, the transformation, which generates a target image data record adapted to the actual image in accordance with the determined visual changes from the data record used to produce the object-related image, can advantageously take place offline. Through a logic circuit, e.g. B. antivalence circuit 3 , a difference between the respective target and actual image point is recognized.

A delay circuit 4 , implemented by shift registers, provides at 6 outputs the information determined by the antivalence circuit 3 of a pixel to be designated as current, its left and right neighboring pixel in the line and the neighbors to these pixels in the previous line at the same time.

A delay circuit 5 , also implemented by shift registers, provides the target image information of a pixel to be designated as current and its 8 surrounding neighbors at 9 outputs.

A logic circuit 6 processes the information offered by delay circuits 4 and 5 and determines the parameters area, circumference, extreme coordinates, tone value information, number of determined contour sections and existence of undetermined contour sections for each defect.

The log. Circuit 6 is connected to a memory 11 which stores the intermediate and final results for the defects.

There is a memory location for each defect. When contacting a defect for the first time, the address of the memory location is taken from a memory 7 which contains all the unoccupied addresses of the memory 11 . A memory 8 records the addresses of the defects processed in the current line in chronological order. In order to solve the assignment problem, the addresses of the defects processed in the previous line are stored in a memory 9 in chronological order. Lines 8 and 9 change their function line by line.

The addresses of the fully detected defects are stored in a memory 10 if the area does not fall below a limit value specified in register 12 . If the limit is undershot, the log. Switch the address, which was provided for the detected defect, again by storing it in memory 7 again.

A computer 13 takes the addresses of the completely detected defects from the memory 10 , accordingly takes the final parameters of a defect from the memory 11 , carries out a typing and a classification. The computer 13 outputs the result data and generates control data for a repair station if necessary. A clock generation unit 14 controls the working regimes of the logical processing units.

In one example, a defect is sufficiently descriptive Parameters explained.

FIG. 4c shows a defect 21 which results from the comparison between the target image in FIG. 4a and the actual image in FIG. 4b. The area of the defect is equivalent to the number of its pixels. The extent of the defect is determined by the sum of the lengths of lines 22, 23, 24 and 25 . The extremal coordinates of the defect are given by the points 26, 27, 28 and 29 on the coordinate axes 30 and 31 .

The tone value information results from the tone value of the section 32 of the actual image that is congruent with the defect, FIG. 4b. The tone value is therefore one. The defect has two determined contour sections 24 and 25 , which are defined by the structural edges 33 and 34 . With the contour sections 22 and 23 there are undetermined contour sections.

FIG. 5a demonstrates the relationship between a deterministic contour section 35 of the defect 36 with the tone value zero, and the desired contour 37th The determined contour section and the corresponding section of the target contour are congruent.

Fig. 5b demonstrates the analogous relationship for a defect 38 to the tone value of one. The determined contour section 39 and the corresponding section of the target contour 40 are adjacent.

The function of the logic circuit 6 will be explained using the example of a defect 41 , FIG. 6c. The defect results from the comparison between a target image Fig. 6a and an actual image Fig. 6b.

The pixel information including the above Environmental information is displayed line by line and within the Row column by column through the upstream, above. Items provided.

Contacting the defect begins in a row 43 and column 58 , determined by the fact that there was no defect in the previous column 57 . The registers required for determining parameters across a contact are deleted. The column value assigned to column 58 is recorded as the minimum and maximum column value of the current contact.

The area is increased by one. The contour length does not increase since there is no contour relationship with either the previous line 42 or the previous column 57 . The tonal value results in one because the pixel projected into the actual image has the tonal value one. There is an undetermined contour point because the image point projected into the target image has no contour point as a neighbor.

With the next cycle, the pixel information assigned to row 43 and column 59 , including the above-mentioned environmental information, is made available. Since the pixel information shows a defect, the contact is retained. The area is increased again by one and the column value is recorded as the maximum column value. The contour length is also increased by one because there is a contour relationship with the previous column 58 . There is no new information for the tonal value. This applies to the entire defect. There is also an undetermined contour point. A change between a densely determined and a determined contour point is not counted.

The pixel information assigned to row 43 and column 60 repeats the foregoing, but the determinance analysis reveals that a determined contour point is present, since the pixel projected into the target image has at least one contour point as a neighbor. A change between an undetermined and determined contour point is counted.

Column 61 is reached with the next bar. The defect is no longer contacted. Since no connection to defect points of the previous line 42 was registered in the course of contacting, the line value assigned to line 43 is recorded as the minimum line value, a branching parameter for defect detection is set to one and a new defect-related set of features is placed in memory 11 , FIG. 1. For this purpose, an address is taken from the memory 7 , assigned to the defect and the intermediate result is stored in the memory 11 at the address provided. In addition, the address is stored in a memory 8 in order to enable recognition in the next line.

The next contact is made in line 44 from column 57 . The registers are deleted again. The column value assigned to column 57 is recorded as the minimum and maximum column value of the current contact. The area is increased by one. The contour length is extended by √ because there is an adjacent contour point in column 58 in the previous contact in line 43 . There is an undetermined contour point. There was no change in the determinance analysis based on the contour point just quoted.

The next two points of contacting only increase the area value and update the maximum column value. Editing column 58 also includes recognizing the previous contact in line 43 and providing the address from memory 8 . The intermediate result associated with this contacting is taken from the memory 11 in accordance with the address provided, in order to summarize it with the result of the current contacting in column 61 , here the end of the current contacting is signaled, and to store it again in the memory 11 under the same address.

The point of column 60 of the current contact provides an increase in the area by one and a new maximum column value. The contour length is also increased by one because there is an adjacent contour point in column 60 in the previous contact in line 43 . There is a determined contour point. This also applies to the contour point just quoted. This meant that there was no change between determined and undetermined contour points.

The next cycle signals that contacting has ended. The intermediate result of the contacting from line 43 is combined with the last one obtained by determining the absolute extreme values, adding the surface parts, adding the contour length, taking over the tone value information, adding the number of changes between determined and undetermined contour points, taking over the presence of undetermined contour points and the branch parameter is added. As already mentioned, this summarized intermediate result is stored in the memory 11 .

The next contact results from row 44 in column 66 . At this point in time it cannot be seen that the points belong to the same defect. As in the first contact, a new defect-related feature set with its own storage space in the memory 11 is created. A change between determined and non-determined contour points was registered for this contact.

The processing of the contacting of line 45 in the first section is analogous to the first contacting of line 44 . From column 60 , determined contour points are registered. A change between determined and non-determined contour points does not take place, however, since the contour section in column 66 opens into a determined contour point of the previous line.

From column 66 , the current contacting overlaps with a further contacting of the previous line 44 . The repeated overlap signals that this contacting is not an independent defect. The interim results are summarized in the manner already specified. In addition, the resulting branching parameter is reduced by one, which now shows the value one again. The address of the second defect is made available again in the memory 7 . The intermediate result of the second defect is thus ignored.

The next lines of the defect are described in Edited way.

In line 50 there is a repeated overlap with a previous contact. Column 61 summarizes the first contact with the previous interim result. Column 68 summarizes the second contact with the new interim result. In addition, the branching parameter is increased by one and now shows the value two. In order to guarantee a clear defect association in the next line to be processed, the address assigned to the defect, in the meantime the memory 9 , is created both after the first contacting and the second contacting the line 50 .

In line 53 there is no contact up to column 61 . However, the contour length, the number of changes between determined and non-determined contour points and the presence of undetermined contour points are updated in accordance with the previous contact. The branch parameter is reduced by one and now shows the value one.

Line 55 completes the defect, since the branching parameter shows the value zero after being reduced by one. This causes the logic circuit to perform the area significance test already mentioned. The log. Circuit 6 can work in such a way that the information is processed completely per pixel clock, the log. Circuit is controlled with a clock frequency that corresponds to an integer multiple of the pixel repetition rate, as well as in such a way that the processing takes place over at least two pixel clocks (pipelining).

A clock generation unit 14 generates control clocks which ensure a synchronous sequence of image scanning, image generation, pixel delay and pixel processing.

The classification of the defects can e.g. B. be made in nine types. Examples of 9 types are shown in FIGS . 7 to 15 by comparing the target and actual images.

In FIG. 2, an arrangement is shown having two channels A and B for defect detection. The comparison between the target image and actual image data is carried out in such a way that channel A only processes the defects with tone value one and channel B those with tone value zero. This has the advantage that two target images can be generated, both of which correspond to the data record used to produce the objectified images and describe the image content, but which differ on the one hand by thickened and on the other hand thinned structures according to permissible tolerances, the value for the thickening or thinning is not constant, but rather depends on the local conditions.

In Fig. 16, two inter-related structural elements are shown, the edges of which are determined by the components used to manufacture the objectified images and the image content data 69 and 70 defining. On the one hand, due to possible tolerances, a target image can be generated which corresponds to thickened structures according to edge positions 71 and 72 , and on the other hand one which corresponds to thinned structures according to edge positions 73 and 74 .

The arrangement shown in FIG. 2 is essentially composed of the components of the arrangement shown in FIG. 1.

The binary actual image is generated by a device 1 in the manner described and output pixel by pixel.

The channel A uses a gate 19 to link the direct actual image information conjunctively with the desired image information negated by means of a gate 17 , which corresponds to the thickened structures and is generated by a device 15 . The delay devices 4 a and 5 a provide the required environmental information about the current pixel and the log. Circuit 6 a in conjunction with the memories 7 a , 8 a , 9 a , 10 a and 11 a and register 12 a determines the parameters describing a defect. The tonal value can be omitted as a parameter, since only defects with the tonal value one are processed by this channel and the memory 11 a consequently only contains these sets of features.

Channel B works in the same way. Here, the actual image information negated with the aid of a gate 18 is conjunctively linked with the direct target image information, which corresponds to the thinned structures and is generated by a device 16 , by means of a gate 20 . The delay devices 4 b and 5 b provide the required environmental information about the current pixel and the log. Circuit 6 b, in conjunction with memories 7 b , 8 b , 9 b , 10 b and 11 b and register 12 b, determines the parameters describing a defect. In this case, can be omitted also the tone value as a parameter, since only defects are processed with the tone value zero of that channel and the memory 11 b thus includes only those feature sets.

The computer 13 takes the addresses 10 a and 10 b from the addresses of the completely detected defects and takes the parameter sets describing the defects from the addresses 11 a and 11 b and processes them in the manner specified.

Clock controller 14 synchronizes processing in channels A and B.

Claims (6)

1. Method for exact image comparison of an objectified image, which consists of image points with tonal values one and zero, which form structures with contours, with a CAD data set used to produce this image, whereby a binary actual image is generated by clock-controlled scanning of the objectified image is synchronized with the generation of a TARGET image adapted to the ACTUAL image from the CAD data record and corresponding image points of TARGET and ACTUAL image are compared with one another with regard to their tonal values, and thus defect information is determined that defects in the ACTUAL image determined, which are composed of pixels with opposite tonal values to the corresponding points of the TARGET image, the coordinates of each defect point are detected and pixel fields are built up and stored, which on the one hand include a TARGET image section and on the other hand corresponding pixel-related defect information, and that pseudo-error ler are suppressed, as are defects if they fall below a predetermined limit value and that sampling, generation and pixel processing are controlled via a pixel sequence clock, characterized in that

• - That an image field containing defect information is formed from 6 image points, corresponding to an actual image point to be considered as current as well as its two immediate neighboring points in the line to be considered current and corresponding to the 3 immediately adjacent in the previous line to these 3 points Points and
• - That an image field of the TARGET image is formed from 9 pixels by a TARGET pixel corresponding to the ACTUAL pixel to be considered as current, as well as its immediate neighboring points and surrounding it
• - That the contour points of a defect in the current or previous line are determined by comparing the defect information of the image point to be designated as current of the current or previous line with its neighboring points of the defined image field and
• - That the determinateness of a contour point of a defect is determined by comparing the corresponding TARGET image point with its immediate TARGET image neighboring points of the defined image field and that there is determinance if, depending on the defect tone value, either the TARGET image point corresponding to the contour point of the defect or at least one of them immediate TARGET image neighboring points is a contour point in the TARGET image, and
• - that contiguous contour sections are detected, which differ in their determinateness and
• - That the number of changes between different contour sections of a defect, which correspond to the number of determined contour sections, is determined and the presence of undetermined contour sections is ascertained and the defects are classified with these two parameters and the tonal value information.

2. The method according to claim 1, characterized in that two TARGET images are generated from the CAD data set in accordance with permitted global upper and lower error limits, a TARGET image taking into account the lower error limit and thus a diluted TARGET image and a TARGET Image takes into account the upper error limit and thus a thickened TARGET image arises and that on the one hand the ACTUAL image data are linked in a data channel with the negated TARGET image data of the thickened TARGET image and only defects with tone value 1 are processed in this data channel and that, on the other hand, the negated actual image data are linked with the target image data of the diluted target image in a second data channel and only defects with the tone value Φ are processed further in this data channel. 3. The method according to claim 2, characterized in that the error limits are locally oriented. 4. The method according to claim 1, 2 and 3, characterized in that that the pixel processing in at least two time stages takes place and the time stages are fixed by a control cycle are whose clock sequence is an integer, the number of Time multiples corresponding to the pixel sequence clock is. 5. The method according to claim 1, 2 and 3, characterized in that that the pixel processing in at least two time stages takes place, each comprising exactly one pixel sequence clock and counted starting with the current pixel sequence clock will. 6. Arrangement for image comparison with a scanning and at least a TARGET image generation device and at least a data processing channel made up of logical computing elements and storing, characterized in that the actual image is recorded by a scanner, the output of which opens into at least one processing channel, by one output each of the scanner's input is a logic circuit, the second input with connected to an output of a target image generation device is, like the input of a first delay circuit, their 9 outputs in a logical processing unit lead and where the output of the logic circuit forms the input of a second delay circuit, the 6 outputs also in the logic processing circuit open, connected to the memory and a limit register and that these arrangement elements of a processing channel controlled by a computer and a clock generation unit will.