Compensation for reference misalignment during part inspection

By defining the tolerance band and measurement vector, self-learning automation algorithm extracts measurement values ​​from CAD data, realizing submicron-level line width and spacing measurements of precision parts, solving the problem of low detection efficiency in the prior art and meeting high Throughput production demand.

CN113454445BActive Publication Date: 2024-11-12RAPID IMAGE PROCESSING LLC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN201980082521.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2019-03-19
Filing Date
2019-10-09
Publication Date
2024-11-12
Estimated Expiration
2039-10-09

AI Technical Summary

Technical Problem

When detecting defects of precision parts, the prior art cannot measure the conductor line width and spacing at the submicron level due to the alignment accuracy of the scanned image and the reference image. The traditional method requires the fineness of the scanned image and the reference image. Alignment results in low detection efficiency and cannot meet high throughput requirements.

Method used

By defining tolerance bands and measurement vectors, searching for feature dimensions along the perimeter of the part, self-learning automated algorithms extract measurement values ​​from CAD data, allowing manufacturing tolerances, measuring conductor line widths and spacing in real time, and calculating statistics, automatically determining Optimal measurement position to achieve measurement of subcamera pixels and submicron accuracy.

Benefits of technology

Real-time, submicron accurate conductor line width and spacing measurements of parts at production speeds are achieved, which improves detection efficiency and can perform metering measurements and defect detection simultaneously, suitable for high-throughput production of large-scale parts.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113454445B_ABST
    Figure CN113454445B_ABST
Patent Text Reader

Abstract

Methods and apparatus for inspecting electronic parts such as semiconductor wafers, high density circuit boards, multi-layer substrates, chrome-on-glass masks, and other fine line products that compensate for reference misalignment. Conductor line width and spacing measurements are performed in real time along the entire length of each conductor with sub-camera pixel, sub-micron accuracy as the part is scanned. Billions of metrology measurements are also performed in real time with sub-micron accuracy. A self-learning automated method is described that extracts measurements from a computer-aided design (CAD) file. The method automatically determines where to perform these billions of operations within defined allowable manufacturing tolerances.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit of co-pending U.S. Provisional Patent Application Serial No. 62 / 744,746, filed on October 12, 2018, entitled “SUB MICRON INSPECTION,” which is hereby incorporated by reference in its entirety. Technical Field

[0003] This patent relates to optical inspection and / or metrology of components such as semiconductor wafers, high density circuit boards, multi-layer substrates, inner substrates, chrome masks on glass, and other precision parts. Background Art

[0004] Consider an example precision part such as a multi-layer semiconductor or circuit board with multiple dies, such as Figure 1A Such parts are typically manufactured by laminating together multiple layers, each layer containing metal conductors. Each layer is manufactured by first coating the layer with a resist, then optically projecting an image containing one or more dies at different locations or sections of the layer, and then etching the metal to form the desired metal circuit conductors. The procedure is often referred to as a project, step, and repeat process. Due to current manufacturing limitations, the placement of the projected pattern is known to vary slightly between the same layer design manufactured at different times, or even between layers in the same manufacturing batch. This is particularly true for large circuit boards manufactured on thin, flexible organic substrates, some of which are currently on the order of 500 mm by 500 mm in size.

[0005] To facilitate fabrication and optical inspection of all the dies on a substrate, alignment marks are often placed in each individual die pattern. During the inspection process, the substrate is typically scanned by mechanically moving the substrate under an inspection camera or by moving a camera over the substrate, or some combination thereof. Alignment marks or unique conductor patterns within each die or die sub-segment are located in the scanned image and are used to align the scanned image with a stored reference image of good dies or die sub-segments. Once aligned, the images are overlaid and defects are reported in those locations where the stored reference and scanned camera images differ. As an example, Figure 1B A reference image of a conductor is shown, shown in black. Figure 1C A scanning camera image of a conductor with a hole 101 , shown in grey, is illustrated. Figure 1D The diagram shows the placement Figure 1B On the reference image Figure 1C A perfect overlay of the scanning camera images of FIG. 100 is shown in FIG. 101 . Hole 101 is detected as a defect where the overlay images are different. Summary of the invention

[0006] Problem Identification

[0007] Defect detection by overlaying scanned images on reference images has many limitations and is now used Figures 2A to 2D , Figure 3A to Figure 3D and Figure 4 To describe the limitation. As a first example, Figure 2A A reference pattern 201 of a known good circuit is shown (shown in black), and Figure 2B A scanning camera image of this pattern 202 (shown in grey) is shown superimposed on a reference. Inaccuracies in the scanned image caused by limitations in the manufacturing process, non-linearities in the imaging optics, stage vibrations, and stage non-linearities limit the accuracy of this comparison and therefore the size of the smallest detectable defect. As a result, when the scanned image and the reference image are overlaid, there are areas of imperfect match. To accommodate such mismatches, prior art systems define one or more error tolerance bands such that only pattern differences greater than a minimum size are reported. Such tolerance bands 203 are defined by FIG. 2A to FIG. 2E The dashed perimeter boundary line in the figure is shown. Only when the scanned pattern falls outside the tolerance band, such as Figure 2C As shown in , defect 204 is reported, such as Figure 2D as shown in .

[0008] Camera pixel quantization is another factor that limits how accurately the scanned image and the reference pattern can be compared, thereby limiting the size of the minimum detectable defect. In many prior art systems, such as in U.S. Patent 6,427,024 to Bishop, entitled "Apparatus for and method of automatic optical inspection of electronic circuit boards, wafers and the like for defects, using skeletal reference inspection and programmable alignment tolerance and detection parameters", the inspection process involves first classifying each camera pixel in the scanned image as a conductor or insulator, and then processing the classified pixels to enable comparison with the reference image. As an example of camera pixel quantization, FIG3A shows an optical image of a conductor 301, whose width is equal to 2.7 camera pixels, projected onto an imaging camera. Individual camera pixels are represented by a 2-dimensional grid. In this example, if fifty percent or more of the pixel (P ≥ 50%) is illuminated by the intensity characteristic of the conductor, the pixel is classified as a conductor. If less than fifty percent of the pixel is illuminated with that intensity (P < 50%), the pixel is classified as a substrate. In FIG3A , the left edge of conductor 301 covers one hundred percent (100%) of pixel 1 (the leftmost pixel in the row), one hundred percent (100%) of pixel 2 (the next pixel to the right), and seventy percent (70%) of pixel 3 (the pixel in the third column). Therefore, using the 50% threshold rule, conductor 301 is classified as being three (3) camera pixels wide, as shown by classified conductor 302 in FIG3B . However, if the conductor pattern projected onto the camera is shifted to the right by only a fraction of a pixel, as shown in FIG3B , the conductor 301 may be classified as being three (3) camera pixels wide. Figure 3C 303 in FIG, the same conductor will now be classified as having a width of only two (2) camera pixels, as shown in FIG. Figure 3D The reduction in line width occurs because the Figure 3C , the image of conductor 303 now covers thirty-five percent (35%) of camera pixel 1, one hundred percent of camera pixels 2 and 3, and thirty-five percent (35%) of camera pixel 4, leaving only pixels two (2) and three (3) with more than fifty percent of the pixels illuminated. Figure 3D As shown, the location of the conductor relative to the pixels in the camera can significantly alter the measured width of the conductor across the camera pixel, further contributing to differences when comparing the scanned image to the reference image.

[0009] To minimize the number of false calls caused by such differences and misalignments of the scanned and reference images, as Figures 2A to 2D As shown in the figure, the prior art system increases the size of the tolerance band to reduce the sensitivity of the comparison. Unfortunately, when doing so, true defects that appear within the tolerance band will be missed. As an example, both Figure 2E Both the small notch defect 204 and the small protrusion defect 205 illustrated in FIG. 2 will not be detected. Missing such defects is a major limitation of prior art systems.

[0010] Another limitation of detecting defects by overlaying the scanned image and the reference image is that the conductor line width cannot be checked or verified with a precision finer than the alignment accuracy of the scanned image and the reference image.

[0011] As an example, consider the Figure 4 4 shows a circuit with two micrometer (2µm) lines represented by conductors 401 and 402 in FIG. 4 , where both conductors are in the same camera field of view. In the scanned image, conductor 401 is aligned with conductor 405 in the reference to within 0.1µm, as indicated by misalignment arrow 406, and conductor 402 in the same scanned image is aligned with conductor 407 in the same reference image to within 1 µm, as indicated by misalignment arrow 408. In this example, the scanned image meets manufacturing requirements because the line width of both conductors 401 and 402 is corrected to within + / - 0.1 µm, which is required to control high frequency impedance and meet electrical requirements, and the spatial misalignment between the scanned image and the reference image is corrected to within 0.1 µm, which is required to ensure circuit alignment between different layers of the product.

[0012] However, using the prior art inspection method, if the tolerance band is placed within + / - 0.1 µm of reference conductors 405 and 407 to check for proper line width, a 0.1 µm misalignment of scanned image conductor 402 relative to reference conductor 407 will be incorrectly reported as an error. If the tolerance band is placed 0.1 µm away from reference conductors 405 and 407 to check for conductor misalignment, a line width error of less than 0.1 µm of conductors 401 and 402 will not be detected. Therefore, as described above, the prior art system that overlays the scanned image and the reference image cannot verify the conductor line width with a precision finer than the alignment accuracy of the scanned image and the reference image.

[0013] When the number of conductors to be measured is small, automated systems have been developed to measure line widths and critical dimensions at “sample locations” on the part. Examples of such systems are the Orbotech® Discovery II, Orbotech Fusion, and Orbotech Ultra Fusion with 2D metrology capabilities available from Orbotech Ltd. of Yavne, Israel. (Orbotech is a registered trademark of Orbotech Ltd.) However, unless the defect is expected to occur at exactly one of the predetermined sample locations, the defect will not be detected. Since the location of the defect is unknown, such sample-based systems are generally intended for process control and are not suitable or intended for the detection of random defects. For example, a 300 mm round wafer scanned using a 0.7 µm camera pixel contains over 1.4×10 11 Pixels. Using such prior art techniques to measure conductor width and critical dimensions at every location on such a part to detect defects would take many hours, if not days. Therefore, such prior art techniques cannot meet today's inspection throughput requirements.

[0014] In contrast, the methods and systems described herein can inspect parts, perform metrology measurements, and calculate feature size statistics by measuring conductor line width and spacing with sub-micron accuracy along the entire length or perimeter of each conductor in the part in real time as the part is scanned at production speeds, and do not require fine alignment of the scanned image with a stored reference pattern.

[0015] Due to the large number of measurements performed, a self-learning automated algorithm is also described for extracting measurements from a reference file, increasing manufacturing tolerances, and determining the best location to perform critical dimension measurements.

[0016] Overview of the Preferred Embodiments

[0017] In one embodiment, a method or system for inspecting features in a part for defects includes locating reference data and then defining (one or more) tolerance bands that specify how much one or more features can deviate from the reference data in space. So-called measurement vectors are then created at fixed spatial locations around the perimeter of each feature. Each measurement vector represents the feature size in the reference data from one point on the perimeter to another point on the perimeter at a specified angle. Linear search ranges are also defined around each fixed spatial position so that the starting and ending points of each measurement vector can be found. Each linear search range has a search axis that is oriented at the same angular orientation as the measurement vector to which it corresponds.

[0018] Next, for each feature to be inspected, the feature is moved to multiple positions within the feature's tolerance band. For each such position of the feature, a point is searched along the perimeter of the feature within a specified search range at each previously created spatially fixed measurement position. When a point is found on the perimeter, the vector size is measured at the angle specified by the corresponding measurement vector. The resulting size is stored as allowable. When a vector measurement exceeds the maximum measurable value, or when no point is found on the perimeter, the result is also stored as allowable. For each spatially fixed measurement position, the allowable minimum and maximum vector sizes are stored, corresponding to multiple positions of the feature within its tolerance band. These minimum and maximum vector sizes, or representations thereof, are then modified to allow for acceptable process variations during part inspection.

[0019] The process of inspecting the part from the scanned image then continues as follows. Next, for each measurement vector, and within the search range for that vector, a point is searched along the perimeter of the feature. When a point is not located, an indication is generated indicating that the point was not found on the perimeter. When a point is located, the vector dimension is measured at the angle specified for that vector, and an indication is generated indicating when the dimension exceeds the maximum measurable value. Then, for vector dimensions that are less than the minimum allowable dimension (or its representation) or greater than the maximum allowable dimension (or its representation), a defect is reported.

[0020] In another embodiment, a method for measuring features in a part begins by locating reference data and defining tolerance band(s) that specify how much one or more features may deviate in space from the reference data. Measurement vectors are created at fixed spatial locations around the perimeter of each feature, where each measurement vector represents the feature dimension in the reference data from one point on the perimeter to another point on the perimeter at a specified angle. Linear search ranges are also defined around each fixed spatial location so that a starting point and an ending point for each measurement vector can be found. Each linear search range has a search axis that is oriented at the same angular orientation of the measurement vector to which it corresponds.

[0021] Next, for each feature to be inspected, the feature is moved to multiple positions within the feature's tolerance band. For each such position of the feature, and at each previously created spatially fixed measurement position, a point is searched along the perimeter of the feature within the specified search range. When a point is found on the perimeter, the vector size is measured at the angle specified by the corresponding measurement vector. The resulting size is stored as allowable. When a vector measurement exceeds the maximum measurable value, or when no point is found on the perimeter, the result is also stored as allowable. For each spatially fixed measurement position, the allowable minimum and maximum vector sizes are stored corresponding to multiple positions of the feature within its tolerance band.

[0022] As each feature moves within its corresponding tolerance band, the measurement vectors whose dimensions do not change or change only within the defined minimum critical dimension tolerance are stored as critical dimension vectors.

[0023] The process of measuring the part from the scanned image then continues by defining locations along the perimeter of the feature within the search range of that vector for each critical dimension vector. At each such perimeter point located, the vector dimension is measured at the same angle as its corresponding measurement vector.

[0024] Statistics are then calculated from the vector size measurements as a function of feature type, wherein the statistics are calculated over the entire part, or for a section of the part, or for different dies on the part or sections within different dies on the part, where a given feature type is represented by features of the same size in the reference data. Statistics may be calculated for feature types as at least one of minimum, maximum, average, and standard deviation statistics. Finally, the statistics are reported for use in characterizing the part or in process control. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The following description refers to the accompanying drawings, in which:

[0026] Figure 1A is an illustration of a wafer of a circuit board substrate having a plurality of dies;

[0027] Figure 1B is an illustration of a conductor reference image;

[0028] Figure 1C is an illustration of a scanning camera image of a conductor;

[0029] Figure 1D is an illustration of a perfect overlay of the scanning camera image on the reference image showing the defect;

[0030] FIG. 2A to FIG. 2E illustrates the limitations of pattern matching used by prior art inspection systems;

[0031] Figure 3A to Figure 3D shows how grayscale camera pixels are classified as conductors or insulators;

[0032] Figure 4 A scanned image subsection compared to CAD data is shown;

[0033] Figure 5 is a diagram with vectors showing the conductor width and the spacing between conductors measured along the entire length of the conductor;

[0034] Figure 6 is a diagram showing that the measurement is not affected by motion within the tolerance band;

[0035] Fig. 7A shows vector measurements calculated from reference data;

[0036] Figure 7B is a table listing the measurements generated from the reference data;

[0037] Figure 8 shows the search range for finding edges in a scanned image;

[0038] Fig. 9 is an illustration in which the reference image is shifted horizontally to the right and vertically downward;

[0039] Fig. 10A is a diagram showing feature edges found in a shifted reference;

[0040] Fig. 10B is a table listing the measurements generated from the shifted reference;

[0041] Fig.11A A bounding box is shown around each vector, indicating the region of uncertainty, and showing the range of valid vector values;

[0042] Fig. 11B is a table with a list of check parameters for conductor edge locations, measurement angles, and allowable vector lengths;

[0043] Fig. 12A is an illustration of a scanned image containing narrow line defects and wide line defects;

[0044] Fig. 12B Shows a list of inspection parameters and inspection results;

[0045] Fig.13 is a graphical representation of vector measurements exceeding the maximum measurable distance;

[0046] Fig.14A is a diagram showing how measurements can be performed over the entire perimeter of a feature with a measurement spacing close to one pixel;

[0047] Fig. 14B is shown as Fig.14A Table containing a list of check parameters for feature calculations in;

[0048] Fig.15A is a graphical representation showing that uncertainty in scan image placement does not affect or compromise the ability of defect protrusions, gaps, and defects that are smaller than the uncertainty region;

[0049] Fig. 15B is shown for checking Fig.15A A list of inspection parameters for the features and a table of inspection results;

[0050] Fig. 15C is a list of inspection parameters (where manufacturing tolerances are added to inspect Fig.15D Features in the table) and inspection results;

[0051] Fig.15D is a diagram showing a feature containing a defect plus a protrusion and a notch within manufacturing tolerances;

[0052] Fig.16A is a diagram showing reference data of an insulator between two conductors;

[0053] Fig. 16B It is shown Fig.16A Table with a list of inspection parameters for the measured insulator edge position, the measuring angle and the permissible vector length;

[0054] Fig.17A is an illustration of a scanned image of an insulator between two conductors;

[0055] Fig. 17B It is used to check Fig.16A List of inspection parameters for features and inspection results in ;

[0056] Figure 18A to Fig.18E Signals and images used to find the edges of conductors and insulators with sub-pixel accuracy are shown;

[0057] Fig.19A is a grayscale image of chrome on glass scanned from test target 1;

[0058] Fig.19B is a table showing the results of measuring the line width on the test target 1 using the Sobel gradient algorithm;

[0059] Fig. 20A A method for converting CAD data into reference grayscale pixels is shown;

[0060] Fig. 20B is a plot of reference pixel intensity versus reference pixel position across the sampling line;

[0061] Fig.21A is the scanning camera image of target 2;

[0062] Fig.21B is the grayscale-converted CAD data of target 2 convolved with the camera lens blur function;

[0063] Fig. 21C is a table that lists the CAD data convolved with the lens blur function ( Fig.21B ) and scanned camera images ( Fig.21A ) is the line width difference measured between

[0064] Fig.22A is a scanned fluorescent image of a metal conductor, which appears black on a bright fluorescent substrate;

[0065] Fig. 22B Shown in Fig.22A Linewidth violations detected in the fluorescence images of;

[0066] Fig. 22C Shown in Fig.22A Spacing violations detected in the fluorescence images of;

[0067] Fig.23 is a diagram showing additional metal defects in the reference data;

[0068] Fig.24 is a diagram showing additional metal defects in a large area insulator;

[0069] Fig.25 is a diagram showing a number of additional metal defects;

[0070] Fig.26 is a diagram showing a hole defect in a large area of ​​metal;

[0071] Fig.27A is a diagram with a bounding box around each vector indicating the region of uncertainty and showing the range of the vector values;

[0072] Fig.27B yes Fig.27A A list of measurement parameters for the features shown in , which indicates a critical width vector;

[0073] Fig.28A is an illustration of a scanned image of a part showing the found feature edges and the calculated critical width vector;

[0074] Fig.28B Is listed according to Fig.28A A table of critical width vector measurements calculated from features in the scanned image;

[0075] Fig.28C It lists the Fig.28A A table of statistics calculated from the critical width vector measurements of the features in ;

[0076] Fig.29 is an illustration of a scanned image compared to reference data and wherein measurement vectors extend to opposite sides of a feature;

[0077] Fig.30 is a diagram showing reference data of elliptical pad edge pixels and vectors;

[0078] Fig.31is a diagram of reference data showing elliptical pad edge pixels and search range;

[0079] Fig.32 is a graphical representation of reference data showing expected edge pixel locations and search ranges for locating an ellipse in scan data;

[0080] FIG33 is an illustration of a scanned image of a defective pad superimposed on an expected edge pixel position with a search range;

[0081] FIG. 34 is a diagram of a scanned image of a defective pad having edge pixels located within a search range,

[0082] Fig.35 is a diagram of a feature shape processor;

[0083] Fig.36 is the system block diagram;

[0084] Fig.37 is a process diagram for example inspection; and

[0085] Fig.38 is a process diagram for example measurements. DETAILED DESCRIPTION

[0086] As mentioned above, this patent relates to improved inspection of precision parts such as wafers, high density circuit boards, multi-layer substrates, inner substrates, chrome masks on glass, and other fine line products. The method performs billions of two-dimensional conductor width and spacing metrology measurements within sub-camera pixel, sub-micron accuracy as the parts are scanned in real time at production speeds.

[0087] Using this approach, the accuracy of sub-camera pixel, sub-micron width and spacing metrology measurements is not limited by conductor placement accuracy. As a result, variations in conductor placement in a scanned image of a part due to limitations in the manufacturing process or limitations in the inspection scan process do not limit or degrade the accuracy of conductor width or spacing measurements.

[0088] Billions of metrology measurements can be performed at sub-camera pixel, sub-micron accuracy at every pixel location along the entire length or perimeter of every conductor on the part, with all calculations performed in real time. Our new method of verifying correct conductor width and spacing requirements at sub-micron accuracy does so without the need for elaborate alignment comparisons of scanned images to stored reference patterns.

[0089] In short, to inspect a part, the method or apparatus first searches the scanned image to locate each feature, and once located, measures the size of each feature and calculates feature size statistics. Metrology measurements and defect detection can be performed simultaneously in real time as the part is scanned. Examples of different feature types are wires and pads of different widths and diameters, such as, for example, 2 µm wires, 5 µm wires, 20 µm diameter pads, 50 µm diameter pads, etc.

[0090] Since billions of measurements must typically be performed across a part, a self-learning automated method or device extracts the measurements from a computer-aided design (CAD) file or reference data, which includes a way to define allowable manufacturing tolerances.

[0091] In addition, statistics can be calculated, such as, but not limited to, minimum, maximum, average, and standard deviation of feature size, line width, and spacing measurements calculated for the entire part, for a section of the part, for each die on the part, or for a section of each die, and for up to 100 percent of the features on the part. In addition to generating statistics based on billions of locations on the part, the best locations for performing measurements used to calculate statistics can be determined as the part is scanned in real time. Such an automatic setup is advantageous for factories with a large volume of different parts.

[0092] More specifically, conductor width is measured in real time along the entire length of each conductor line in the die and the spacing between conductors for all dies on the part as the substrate is scanned at production speeds. Such conductor width and spacing measurements are performed with sub-camera pixel, sub-micron accuracy. As the part is scanned, the method first searches to locate each feature and then measures its size once located. Measurements can be performed for every feature on the part. Thus, billions of calculations to measure line width, spacing, and critical dimensions can all be performed in real time as the part is scanned.

[0093] More specifically, Figure 5 501 is an example where the grey arrows in the black conductors represent line width measurements and the black arrows in the white areas between the conductors represent line spacing measurements. The grey double-sided arrow 501 is an example of a line width measurement and the black double-sided arrow 502 is an example of a line spacing measurement. Both line width and spacing measurements are performed along the entire length of each conductor and for each conductor in the circuit. In addition, measurements are preferably calculated at each pixel position along the entire length or perimeter of each conductor. These measurements are then compared to a list of expected or correct measurements previously calculated from CAD data or from scans of known good parts. In addition, such measurements are performed with sub-pixel, sub-micron accuracy without being compromised by variations in feature placement on the part.

[0094] For example, Figure 6Illustrated is how conductors 601 and 602 can be independently varied in spatial position within a specified tolerance band 603 (indicated by the dashed lines around each conductor) and still be measured with sub-pixel, sub-micron accuracy - as indicated by the double-sided arrow vector lines within and between the black conductors. All vectors extend from one edge of the conductor to the other edge of the conductor, bypassing the tolerance band. Vector 604 is an example of a vector used to measure conductor line width, and vector 605 is an example of a vector used to measure conductor spacing.

[0095] Fig. 7A Reference should now be made to Figure 11 in conjunction with the steps described below, which explain how a list of acceptable width measurements is generated from reference data and then used to inspect a part.

[0096] a. Fig. 7A As shown in , CAD data or a scan of a golden known good part is identified and used as reference data 701 for what is considered a perfect part.

[0097] b. Based on the reference data 701, for all features or at least a subset of features in the reference data, edge (boundary) pixels are located along the entire perimeter of each feature. Fig. 7A In FIG. 1 , points 1A, 1B, 2A, 2B, 3A, 3B, 4A, 4B and 5A, 5B are examples of edge pixels. The width of each feature is measured along the perimeter of each feature, and the set of starting and ending edge pixel positions used to calculate the width and angle of each measurement is recorded. The measurement can generally be taken perpendicular to the corresponding edge pixel or at some other predetermined angle. Other ways of representing the measurement are also possible. Fig. 7A as well as Figure 7B Vectors V1 to V5 in Table 7 of show examples of such measurements and their corresponding angles.

[0098] exist Fig. 7A In the example reference data of Figure 7B As shown in Table 7, point 1 A is the starting vector position, point 1 B is the ending vector position, the angle measured by the vector is zero (0) degrees relative to the horizontal, and R1 is equal to the length or distance of the measured vector V1. As a second example, point 3 A is the starting vector position, point 3B is the ending vector position, the angle measured by the vector is forty-five (45) degrees relative to the horizontal, and R3 is equal to the length or distance of the measured vector V3. As a third example, point 4 A is the starting vector position, point 4 B is the ending vector position, the angle measured by the vector is ninety (90) degrees relative to the horizontal, and R4 is equal to the length or distance of the measured vector V4.

[0099] c. The next step is to define uncertainty or tolerance bands around features in the reference data, which indicate the maximum offset or spatial deviation allowed between the reference data and the scanned image when inspected. Such tolerance bands 702 are defined by Fig. 7A The dashed boundary line in the figure indicates the

[0100] d. For each starting and ending pixel position in the reference data, a "linear search range" is then defined that will be used to search for edges in the scanned image when inspecting. Each linear search range is oriented along the same measured angle as the vector defined by the starting and ending pixel positions in the reference data. The search range is determined by Figure 8 As an example, the linear search range around pixel position 5b is Sr. Each linear search range should extend beyond or at least to the edge of the tolerance band, as indicated by Figure 8 This is indicated by the dotted lines in the figure to ensure that during inspection, the edges are found in the scanned image when the conductor moves within the uncertainty area or tolerance band.

[0101] e. To simulate how the vector length measurement will be affected by such scanned image placement uncertainty at inspection, the reference data is moved within a tolerance band to simulate such motion, and the resulting vector length measurement is stored for each position of the reference data.

[0102] As an example, Fig. 9 Shows relative to Fig. 7A The reference data is shifted down by half a grid box and to the right by half a grid box. To simulate the measurement process that would be performed on a scanned image during real-life part inspection, edges 1A, 1B to 5A, 5B are searched for within a linear search range around each pixel, as shown in FIG. Fig. 9 As an example, Fig. 9 Sr 902 in FIG. 9 indicates the linear search range around edge pixel 5B. Once the edge of the shift reference is found through this linear search, Fig. 10A The length of each vector is measured at the position indicated by the circle (o). As an example, circle (o) 1001 corresponds to an edge pixel found at position 2BF. Fig. 10B Table 10 in the table lists the Fig. 10A Each result found in the example gives the location of the edge pixel, as well as the corresponding measurement angle and the length of the measurement vector.

[0103] like Fig. 10A As shown in Table 10, pattern motion in the direction along the vector measurement axis does not affect the measured length of the vector. Furthermore, in regions of constant feature width, motion beyond the measurement axis does not affect the reported length of the vector. As an example, Fig. 10A and Fig.11A Vectors V2, V3, and V4 in are located in regions of constant feature width because the values ​​of these vectors do not change when the reference data is shifted within the tolerance band. These regions of constant width are defined by Fig.11A Solid rectangles 1102, 1103 and 1104 drawn around vectors V2, V3 and V4 are graphically illustrated in FIG.

[0104] As shown in Table 10, Fig. 10B As shown in , the values ​​of vectors V2, V3 and V4 are therefore not affected by such pattern shifting and are consistent with Figure 7B The values ​​listed in Table 7 are consistent with those measured before the reference data was shifted. However, when the reference data is shifted, vectors V1 and V5 in the fast feature width transition region are affected. For example, the length of vector 1 changes from Figure 7B R1 in Table 7 is reduced to Fig. 10B The zeros in Table 10 of , because when the reference data is shifted, no edge 1 is found at the very end of the feature A and 1 B ,like Fig. 10A Similarly, the width of vector V5 is calculated from R5 (in Figure 7B 10) to R5+D (in Fig. 10B 10), because this vector is also in the region of changing feature width.

[0105] Respectively indicated by arrows 1101, 1102, 1103, 1104 and 1105 Fig.11A The rectangle surrounding vectors V1 to V4 and the quadrilateral surrounding vector V5 graphically indicate the range of measurement values ​​for each vector that will be measured when the reference data is moved back and forth within the tolerance band. Fig. 11B Table 11 in is an example of an inspection list for inspecting a part from a scanned image. The list contains the expected conductor edge start position, expected conductor edge end position, vector measurement angle, linear search range length, minimum allowable vector length, and maximum allowable vector length for each vector.

[0106] After this, the process continues as follows:

[0107] f. To create a list of valid width measurement ranges for each vector prior to inspection, the reference data is “moved” within a tolerance band to simulate the motion of the scanned image during inspection (according to Fig. 9 , Fig. 10A , Fig.11A The vector length of each position of the reference data is measured by searching and locating the edges of the feature within each specified linear search range, and then measuring the length between the edges, such as the measurement taken perpendicular to the edge direction ( Fig. 9 , Fig. 10A , Figure 11).

[0108] g. Use the same edge finding and measurement procedure on the reference data image to generate a check list, as will be used later on the scanned image to measure vector lengths during inspection. Then create an inspection parameter list containing all the vectors to be measured ( Fig. 11B ). For each vector so measured, the entry in the parameter list is checked to contain the expected starting edge position of the vector, the expected ending edge position, the measurement angle, the length of the linear search area, the minimum allowable vector length, and the maximum allowable vector length.

[0109] It should be noted that given the starting edge position, measurement angle, minimum vector length and maximum vector length are included in the inspection parameter list, the inclusion of the ending edge position in the list is optional as it provides redundant information.

[0110] h. Once the inspection parameter list is created, use it during the inspection process to measure the line width of all critical features and detect defects, such as using Fig. 12A and Fig. 12B Described.

[0111] As an example, Fig. 12A A scan feature with a reduced line width (notch) defect 1202 (vector V3) and a wide line protrusion defect 1203 (vector V4) is shown, where both defects fall within the allowed tolerance band. Figure 2E As shown in , when these types of defects (wide line protrusions and narrow line notches) fall within the tolerance band, they are not detectable using prior art inspection techniques.

[0112] More specifically, in our improved method, during the inspection of the scanned image, Fig. 12B Each starting pixel position listed in Table 12 is 1 A to 5 A The linear search range Sr around Figure 8 ) to search for feature edges. Fig. 12A As indicated in , somewhere within the specified search range, the conductor edge is at the actual position 2 AF to 5 AF For each edge found, perpendicular to the actual starting edge 2 AF to 5 AF Measures feature width where vector measurements V2 to V5 end at opposite edges 2 BF to 5 BF .like Fig. 12AAs shown in , the actual starting and ending edge positions found in the scanned image are marked with circles (o) at the beginning and end of vectors V2 to V5. As an example, circle (o) 1204 corresponds to an edge pixel found at position 2BF.

[0113] i. Note that Fig. 12A The downward shifted scan image does not find the corresponding table 12 Fig. 12B Position 1 in A and 1 B , so the vector V1 in Table 12 has a measured value of zero (V1 = 0). Therefore, in the next step, the vector lengths V1 to V5 generated during the check are compared with the corresponding minimum and maximum permissible values ​​created before the check. As an example, the measured vector length V1 equal to zero (V1 = 0) is compared with Fig. 12B The minimum allowable vector length zero (0) and the maximum allowable vector length for R1 listed in Table 12 are compared. Since the zero vector length (V1 = 0) measured during the scan part inspection is within the allowable range listed, no defect is reported. However, when the vector V3 measurement (R3-D1) is compared to the minimum allowable value for R3 listed in Table 12, the defect is reported. Fig. 12B The narrow conductor defect measurement (R3-D1) is reported in the Classification column of because (R3-D1) < R3. Similarly, when the vector V4 measurement R4+D2 during part inspection is compared to the maximum allowable length of R4 listed in Table 12, a wide conductor defect measurement of length (R4+D2) is reported as indicated in the Classification column of Table 12.

[0114] In order to implement such an inspection system, one must recognize that there may always be some features whose width may exceed the maximum measurable distance capability of the inspection system. Fig.13 The vector V3 in FIG. 130 indicates that it exceeds the maximum measurable distance (D ) indicated by the double-headed arrow 1301. MAX ). When this happens, D MAX or its equivalent indicator is entered into a check list of allowable vector lengths.

[0115] Figure 7 to Fig.13 The vector measurement procedure described can be performed at every location on the part to ensure that all defects are detected, as the location of the defects is unknown prior to inspection. This is another major difference compared to prior art systems, which only perform sample measurements at predetermined locations on the part.

[0116] As an example, FIG. 14 shows a preferred implementation of the algorithm, in which each pixel position along the edge of each feature in the reference data is measured to create the inspection parameter list of Table 14, Fig. 14B The example list shown in Table 14 includes, for each measured vector: its starting conductor edge position, ending conductor edge position, measurement angle, search range length, minimum allowable vector length, and maximum allowable vector length.

[0117] For example, in Table 14, vector H1 has an expected starting conductor edge location of x=4, y=11, an expected ending conductor edge location of x=6, y=11, a measurement angle of zero (0) degrees relative to horizontal, a linear search range in the scanned image to find an edge of + / - ½ measurement unit around the starting and ending expected conductor edge locations, a minimum allowable measurement vector length of 0, and a maximum allowable vector length of W1.

[0118] Created from reference data Fig. 14B The inspection parameter list of Table 14 is then used to inspect the scanned part, look for defects, and measure the conductor line width along the entire length of the feature in the part, such as Fig.15A as shown in .

[0119] Fig.15A The high density of spatial measurements shown in provides redundancy to greatly improve defect detection. For example, Fig.15A The pinhole defect at the intersection of vectors H2 and V2 is found by both vectors H2 and V2. Fig. 15B The vector H2 measurement ½ W1 in the "Measurement Vector Length" column of Table 15-1 is less than the minimum allowable value of W1, and the vector V2 measurement L1-D2 is less than the minimum allowable value of L1.

[0120] The reduction in characteristic width (gap) at the intersection of vectors H5 and V6 illustrates why such redundancy is necessary. Fig. 15B The vector H5 measurement L2-D3 in Table 15-1 definitely intersects the defect, but the measurement does not fall within Fig. 15B If the scanned image is shifted down to the minimum permissible vector length value of W1 listed in Table 15-1, Fig.14A This can occur if the tolerance band shown in the figure indicates the lowest position. As a result, the vector length H5 equal to L2-D3 does not mark the defect. However, due to redundancy, the vector V6 during the inspection measures W2-D5 ( Fig.15A and Fig. 15B Table 15-1) is less than the minimum allowable vector length of W2, so the defect is detected by vector V6 and reported as a reduced line width defect with width W2-D5, such as Fig. 15B As indicated in Table 15-1.

[0121] The reduced line width (notch) defect at the intersection of vectors H6 and V10 further demonstrates another example of why such redundancy is beneficial.

[0122] Although in Fig.15A and Fig. 15B The vector V10 measurement W2-D6 in Table 15-1 definitely intersects the defect, but the measurement does not fall within the minimum allowable vector length value of zero (0) listed in Table 15-1. If the scan image is shifted to Fig.14A This can occur if the tolerance band shown in the figure is at the far left position. As a result, vector V10 equal to W2-D6 does not mark the defect. However, due to redundancy, vector H6 during inspection measures L2-D4 ( Fig.15A and Fig. 15B Table 15-1) is less than the minimum allowable vector length of L2, so the defect is detected by vector H6 and reported as a reduced line width defect with a width of L2-D4, such as Fig. 15B Indicated in the "Measurement Vector Length" column in Table 15-1.

[0123] In another example, now we find Fig.15A The wide line protrusion defect at the end of the midpoint vector V1 is Fig. 15B Table 15-1 lists a minimum allowable length of zero (0) and a maximum allowable length of L1, and the measured length of V1, which is equal to L1+D1, also exceeds the listed maximum allowable length of L1.

[0124] In production applications, some conductor size tolerance is often allowed relative to the dimensions specified in the CAD or reference data to allow for variance in the manufacturing process. Such manufacturing variances can be entered into the checklist. As an example, Fig.14A The features shown in Fig. 14B The corresponding checklist provided in Table 14 contains two conductor widths W1 and W2 and two conductor lengths L1 and L2. If the allowable manufacturing variance of the conductor width is , and the allowable manufacturing variance of the conductor length is , the checklist in Table 14 can be modified to include these allowable production variances, such as Fig. 15C The modifications are entered into the columns entitled "Minimum Vector Length" and "Maximum Vector Length" of Table 15-2. Use the variance modification parameters to check Fig.15D The features shown in Fig. 15C Table 15-2, titled “ Fig.15D The following information is generated during the inspection. The results are provided in the columns of Table 15-2. Referring to Table 15-2, vector V1 measures (L1+D0) due to a small protrusion of length D0 at the upper left corner of the feature. Given ,but (See Table 15-2), so now this manufacturing anomaly within the manufacturing f variance is not marked as a defect. Similarly, vector V6 measures (W2-D8), because there is a small notch with a depth of D8 near the arrow of this vector. Given ,but , so this other manufacturing anomaly is not marked as a defect in Table 15-2. However, as indicated in Table 15-2, real defects such as the pinhole at the intersection of vectors H2 and V2 and the large gap at the end of the arrow of vector H6 are marked as defects.

[0125] In addition to finding defects by measuring the width of features in a part, it is also possible to simultaneously measure the spacing between features, such as Figure 5 , 6 , 16 and 17. Figure 5 and Figure 6 , the black double-headed arrows in the white area represent the vector spacing measurement between features as shown by the black conductors.

[0126] FIG. 16A to FIG. 18E It will now be explained how these spacing measurement vectors are generated and how the spacing is measured during the inspection process.

[0127] j.Let Fig.16A The reference data 1601 in shows a section of an insulator shown in white, surrounded by a conductive metal shown in grey. Let the dashed boundary line around the insulator represent a tolerance band 1602 or uncertainty in the placement of the scanned image relative to the reference data during inspection.

[0128] k. To inspect the insulator, create an inspection parameter list based on the reference data before the inspection, using the same procedure previously described for creating the inspection list for the conductor, but now using the insulator as the feature of interest.

[0129] As an example, Fig.16A is reference data for a part containing two conductors shown in gray, separated by an insulator shown in white. To create an inspection parameter list for the insulator inspection, a vector width measurement is performed at each pixel edge location along the entire length of the insulator, as shown in Fig.16A To simulate possible misalignment of the scanned image relative to the reference data during inspection, the reference data was aligned at Fig.16A A vector width measurement is performed for each possible position within the tolerance band 1602 indicated in. For each vector measurement, a list of check parameters includes: vector start edge position, end edge position, measurement angle, search range length, minimum allowable vector length, and maximum allowable vector length.

[0130] like Fig. 16B As illustrated in Table 16, vector H1 has an expected insulator starting edge position of x=6, y=15, an expected insulator ending edge position of x=8, y=15, a measurement angle of zero (0) degrees relative to the horizontal, a linear search range of + / - ½ measurement unit around the starting and ending expected edge positions in the scanned image, a minimum allowable measurement vector length of W3, and a maximum allowable vector length of W3. As another example, vector H11 in Table 16 has an expected starting edge position of x=6, y=6, and a vector length equal to "D MAX ” to indicate that the end of the insulator is not found in the reference data within the maximum possible measurable vector distance when searching at an angle of zero (0) degrees from the starting edge at position x=6, y=6. However, if the image is shifted down one pixel position within the tolerance band, H10 will measure W3. Therefore, the minimum allowable length of vector H10 in Table 16 is set equal to W3, and the maximum allowable value is set equal to D MAX , to indicate that the edge or end of the insulator cannot be found or cannot exist within the measurable range in the reference data of this spatial position of the scanned image.

[0131] The inspection parameter list is used in the next step to inspect the scanned part, see Fig.17A and Fig. 17B During this inspection step, the vector length measurements from the scanned part are compared to a previously generated list of expected values. As an example, the left side of Table 17 contains the vector length measurements from Fig.16A The expected vector length values ​​of Table 16 generated by the reference data 1601 in , and the right side contains the vector length values ​​obtained from Fig.17A The vector lengths measured in the scanned image 1701 shown in FIG.

[0132] During this checking step, Fig.17A As shown in , the high density of spatial measurements performed provides redundancy to greatly improve the detection of defects in insulators. As a first example, Fig.17A The additional conductor defect 1702 in is found by both vectors H11 and V9. The vector H11 measurement L1 in the "Measurement Vector Length" column of Table 17 is less than D MAX The minimum allowable value of , and vector V9 measures W4-S1 which is less than the minimum allowable value of W4.

[0133] As a second example of how such redundancy increases defect detection, consider the reduction in insulator width at the intersection of vectors H10 and V7 in Figure 17a. Although the vector H10 measurement (W3+P2) in Table 17 certainly intersects the defect, the measurement does not fall below the minimum allowable vector length value for W3 listed in Table 17. Fig.16A This situation may occur if the defect is at the lowest position indicated by the tolerance band shown in . As a result, the vector length H10, which is equal to W3+P2, does not flag the defect. However, due to redundancy, the vector V7 measurement W4-P1 (in FIG. 17a and Table 17) during the inspection is less than the minimum allowable vector length of W4, so the defect is detected by vector V7 and reported as a narrow insulator defect of width W4-P1, as indicated in Table 17.

[0134] In the third example, the wide insulator defect at the end of vector H6 in FIG. 17 a is found because the measured length of H6 is equal to W3 + N1 , which exceeds the listed maximum allowable width of W3 .

[0135] It can be measured that the accuracy of conductor line width measurement is proportional to the precision with which the conductor edge position can be measured. Although the exact position of each edge is very accurately known when extracted from CAD data, when such edge positions are extracted from scanned images, camera quantization introduces position uncertainty, as previously shown in Figures 3A to 4B. Figure 3D Such pixel quantization can introduce an error of up to one camera pixel in the measured length of the vector.

[0136] To achieve finer vector measurement accuracy and measure vector lengths in scanned images with sub-pixel, sub-micron resolution, an additional two-step process can be performed.

[0137] 1. First, the grayscale camera signal is passed through a two-dimensional edge filter to identify edge transition areas.

[0138] m. A polynomial curve is then fit to the output of the edge filter to identify the precise location of the edge, which usually exists somewhere between camera pixels.

[0139] With reference to Figures 18A to Fig.18E The two-step process is explained in detail. FIG18A is an actual scanned fluorescent image of a thin flexible circuit board with metal conductors on an organic insulator. In FIG18A, the non-fluorescent metal conductor appears black, and the fluorescent insulator appears bright. FIG18A and Fig.18B Both show a scan line 1801 that intersects a metal conductor having a starting edge position X and an ending edge position Y, as indicated in both figures. Fig.18B and Fig.18C As indicated in , scan line 1801 extends from camera pixel A to camera pixel B.

[0140] Fig.18B The output image of a two-dimensional edge filter operating on the image of FIG. 18A is shown. The output of the edge filter at the left edge of the conductor trace (with Fig.18B The output of the filter at the right edge of the conductor along scan line 1801 at position "y" appears bright, indicating that this is a transition from a dark conductor to a bright insulator. The output of the two-dimensional edge filter also indicates the orientation of each edge.

[0141] Fig.18C is a plot of both the camera pixel intensity 1802 across the scan line 1801 in Figure 18A and the corresponding output signal from the edge filter 1803. Each camera pixel intensity value is indicated by a black dot on the solid black curve 1802, which has a corresponding edge filter output value indicated by a gray dot on the solid gray curve 1803.

[0142] Figure 18A and Fig.18B The approximate positions of the left and right edges x and y in correspond to Fig.18C The spatial locations of the most positive and most negative data points in the edge filter output curve 1803 shown in FIG. Since the output of the camera and edge filter only provide intensity values ​​at discrete spatial pixel locations, Fig.18C The black and grey dots in the figure indicate the actual conductor edge location, so the actual conductor edge location can easily exist anywhere between these discrete locations.

[0143] To estimate the edge position with sub-camera pixel accuracy, second order polynomials (parabolas) 1804 and 1805 are fit to the edge filter output data in the regions around the most negative and most positive peaks of the edge filter output data, respectively. The derivatives are then calculated for the best fit polynomials 1804 and 1805, and the precise edge position is defined as the coordinate where the slope or derivative of the polynomial (parabola) is equal to zero (0).

[0144] As an example, in Fig.18C In the figure, two best-fit parabolas 1804 and 1805 (plotted as black dashed lines) are independently fitted to correspond to Fig.18B and Fig.18C The edges in x and y are the surrounding edge filter data points. Fig.18D and Fig.18E Zooming in on the expanded view around the peaks of the best fit parabolas 1804 and 1805, Fig.18C As shown in Fig.18D and Fig.18EIn both, the conductor edge is defined as the point where the derivative of the best fit polynomial is zero, corresponding to its peak.

[0145] As an example, Fig.18D The edge filter output values ​​corresponding to camera pixel locations 20 and 21 are shown, where the negative peak of the best fit parabola occurs between these discrete values ​​at camera pixel location x = 21.53474 pixels. Similarly, Fig.18E The edge filter output values ​​corresponding to camera pixel locations 33 and 34 are shown, with the positive peak of the best fit parabola occurring between these discrete values ​​at camera pixel location x = 33.6488 pixels.

[0146] FIG. 18B to FIG. 18D The images and data shown in are generated using a Sobel type edge filter known in the industry, and two best fit polynomials to the edge filter output data are generated using second order parabolic polynomials. There are many two-dimensional edge filters known in the industry, such as, for example, Sobel, Canney, and Prewitt. These filters and other such similar edge filters may also be incorporated for use in this application. In addition, although second order parabolic polynomials are used in this example to best fit the edge filter output data, higher order polynomials may also be used instead for this application.

[0147] Fig.19A is a scanned image of a chrome on glass calibration target (Target 1) showing two test patterns. One test pattern contains 203 lines / mm and the other contains 228 lines / mm. The 203 lines / mm pattern has line and space widths equal to 2.46µm and the 246 lines / mm pattern has line and space widths equal to 2.20µm. When the two test patterns are measured using a Sobel edge filter and the edge positions are estimated as the peaks of the best fit parabolic curves where the derivative is zero (0), the measured line widths for both patterns are correct to within + / - 0.035 um of the known correct values ​​as shown in Table 19 Fig.19B Indicated in.

[0148] n. In order to be able to check and verify the correct measurements, the vector lengths measured in the scanned image must agree within a specified tolerance with the vector lengths generated from the pre-check reference data. Generate a list of inspection parameters from reference data ( Fig. 14B Form 14 and Fig. 15CThere are at least three ways to calculate the measurement vectors in the example shown in Table 15-2 of FIG. 15A : Scan a known good perfect "golden section" to generate a reference image. Based on this scanned grayscale reference image, the edge filter best fit polynomial procedure described in conjunction with FIG. 18A-18D is used to calculate the required measurement vectors with sub-pixel accuracy.

[0149] If the reference image is generated from CAD data, the format of the data may already be in a vector format, which allows direct calculation of these vector lengths with sub-camera pixel accuracy. If so, convert the CAD data to a grayscale reference image to simulate the output of the inspection camera, such as with the aid of Fig. 20A The measurement vector is then calculated from this grayscale reference image using the edge filter best fit polynomial procedure described in Figures 18A-18D.

[0150] In a preferred method of converting CAD data into a grayscale reference image, the intensity value of each pixel of the grayscale reference image is calculated as follows:

[0151] i. Create a reference pixel grid, where the size of each reference pixel is equal to the size of the camera pixel, such as Fig. 20A This is illustrated by the example 6×8 grid pointed to by arrow 2001.

[0152] ii. Overlaying the reference grid 2001 on CAD data of equal resolution or higher resolution, so that each reference pixel is usually composed of or contains multiple CAD data points, such as Fig. 20A As shown in Fig. 20A In the example of , each reference pixel contains 25 CAD data points, where arrow 2002 points to a single CAD data point.

[0153] iii. Fig. 20A The grayscale intensity value of each reference pixel in (reference pixel intensity (Ref Pix Intensity)) is given by:

[0154] Reference pixel intensity = reference pixel area containing CAD conductor data / total area of ​​reference pixels.

[0155] As an example, Fig. 20A Each large square of the 6×8 grid in 2001 represents the size of a reference pixel, which is equal to the size of a camera pixel. The small squares within each large square represent the resolution of the CAD data. In this example, each reference pixel contains 25 CAD data points. The grayscale intensity value is also listed in each reference pixel and is equal to the area of ​​the reference pixel covered by the conductor divided by the total area of ​​the reference pixels. In other words, Fig. 20A The numerical value written into each reference pixel in represents the grayscale intensity value of the pixel. Fig. 20B Instructions along Fig. 20A Grayscale intensity values ​​of eight reference pixels of scan line 2003 are plotted in FIG.

[0156] Each measurement vector in the inspection parameter list measurement vectors is then calculated based on the intensity values ​​in the grayscale reference image using the edge filter best fit polynomial curve procedure as described above in conjunction with Figures 18A-18D.

[0157] Another method to achieve an even closer match between the vector measurements computed from the scanned camera image and the vector measurements in the inspection list is to first convolve the grayscale reference image with the blur function of the camera imaging lens so that the resulting reference data better simulates the true output of the inspection camera, and then use an edge filter best fit polynomial operation on this convolved image to compute the inspection list of measurement vectors.

[0158] As an example, Fig.21A is a scanning camera image of a calibration test target referred to as Target 2, which contains 1.95 µm and 2.19 µm lines / spaces as indicated in the figure. Fig.21B The result of convolving the grayscale converted CAD data of target 2 with the blur function of the camera lens used to image target 2 onto the camera is shown. Note that Fig.21B The convolution reference image is Fig.21A How consistent is the actual scan camera image. Fig.21A and Fig.21B Intensity waveforms 2102 and 2104 for scan lines 2101 and 2103, respectively, are provided in FIG. 5 , which illustrate how closely the convolved reference intensity waveform and the scanned camera intensity waveform agree.

[0159] Then from Fig.21B The convolution of the reference image calculates the vector length measurement to create the inspection Fig.21A A check list of the required measurement vectors for the scanned camera image. Fig. 21C Table 21 of Target 2 provides the known line / space widths from the CAD data for Target 2 and the known line / space widths from the CAD data for Target 2. Fig.21B The convolution reference image and the Fig.21A The difference between the vector lengths calculated from the scanned camera images is used for comparison.

[0160] In order to perform meaningful inspections on electronic parts, the inspection system must provide some way for the manufacturer to specify acceptable tolerances. There are two types of tolerance parameters associated with the inspection techniques described in this article:

[0161] The tolerance used to specify the search range for feature edges in the scanned camera image, such as Figure 8 The arrow Sr indicates, and

[0162] A tolerance is specified for the acceptable range of lengths for each vector measurement found in the scanned image, such as Fig. 15C The tolerance values ​​in Table 15 and Indicated.

[0163] As illustrated in Table 15-2, different tolerance values ​​can be set for different types of features. For example, for a 2 µm line / space, the acceptable vector length measurement tolerance can be set to + / -0.2 m and the search range (Sr) set equal to + / -1.0 µm, while for a 10 µm line / space, the measurement length tolerance can be set equal to + / -1.0 µm and the search range set equal to + / -3 µm.

[0164] Fig.22A , Fig. 22B and Fig. 22C An example taken from actual inspection of a thin flexible fluorescent organic substrate shows the results of inspecting its segment with 2 µm line / space, where the search range (Sr) was set to 2.1 µm and the measured vector length tolerance was set equal to 0.2 µm. Fig.22A is a scanned image of a thin flexible fluorescent organic substrate, where the metal conductor is black and the fluorescent substrate is bright. Fig. 22B shows conductor line width violations, and Fig. 22C Shows spacing violations. Locations where the line and spacing deviate by more than the specified vector length tolerance (0.2 µm) but less than twice that tolerance (0.4 µm) are marked with the number 1. Locations where the line and spacing deviate by more than twice the specified vector length tolerance (> 0.4 µm) but are found within the specified search range (2.1 µm) are marked with the number 2. Locations where the line and spacing are not found within the specified search range (Sr = 2.1 µm) are marked with the number 3.

[0165] Fig.22A , Fig. 22B and Fig. 22C It is clearly illustrated how the technique described in this paper can be used to detect defects that violate the parameters of the check parameter list. However, defects in the form of additional metal (conductor) in the middle of a large area of ​​insulator, such as Fig.23 Defects in the form of additional insulators (holes) in the middle of large areas of metal, as shown by arrows 2301 in FIG. 1 , or are not detected because the reference data has no features in this area and, therefore, no measurement vectors are created in the inspection parameter list to inspect these areas.

[0166] In order to enable the detection of defects in the form of such additional features in the scanned image, the scanned data is also compared with an image of the reference data. However, such a comparison can be performed without requiring a fine alignment of the scanned image and the reference image. As an example, consider Fig.24 Additional metal defects in the scanned image 2401, and Fig.25 Additional metal defects 2501 in the scanned image.

[0167] Such detection can be performed at the time of inspection by searching the reference data for any metal within a neighborhood around the same coordinates in the reference data as the coordinates of the additional metal features found in the scanned image. Fig.24 The dashed rectangle 2402 and Fig.25 Within the bounds of the search area defined by the dashed rectangle 2502 in FIG. 25, no metal is present in the reference data, and thus additional metal features in the scanned image are detected and classified as defects. However, if such additional features in the scanned image are close to features in the reference data (e.g., smaller than Fig.13 The distance Dmax defined in Fig.25 If there is defect 2503 in , the additional metal defect 2503 is detected as a spacing violation because vector 2505 is shorter than Dmax.

[0168] Holes in the scanned image that expose insulators in large areas of metal, such as those in the scanned image, are detected in a similar manner by searching the reference data within the search area 2602 for insulating material around coordinates in the reference data that are identical to the coordinates of the holes in the scanned image when inspecting. Fig.26 If no hole is found in the reference data within the search area 2602, then Fig.26 The holes in the scanned images are classified as defects.

[0169] Although Fig.23 , Fig.24 and Fig.25 The presence of a conductor (black) on an insulating substrate (white) is illustrated graphically, but the methods described herein are applicable to any part where the conductor can be distinguished from the insulator. Fig.22A , Fig. 22B , Fig. 22C A metal conductor is shown appearing black on a fluorescent substrate, which appears bright. Fig.19A A chrome conductor is shown appearing bright on a glass insulating substrate that appears dark. Other examples are bright fluorescent resist patterns on metal, and bright reflective metal conductors on non-reflective or minimally reflective substrates. As long as the conductor can be distinguished from the insulator in the scanned camera image, the method of inspecting the part described herein is applicable.

[0170] Another application is to perform automated metrology measurements on electronic parts (such as but not limited to wafers, circuit boards, substrates, inner layers, and masks) to obtain statistical data such as but not limited to minimum, maximum, average, and standard deviation feature size, line width, and spacing measurements calculated for the entire part, a section of the part, for each die on the part, or for a section of each die, and for up to 100% of the features on the part. In addition to generating statistics based on billions of locations on the part, the methods described herein automatically determine the best locations for performing measurements used to calculate the statistics as the part is scanned in real time. Such an automatic setup is essential for factories with a large volume of different parts.

[0171] To explain how such an automatic setup can be performed, consider Fig.27A The features 2701 and Fig.27B Table 27 lists the measurement parameters in the measurement parameters table. Table 27 lists the conductor measurement edge position, vector measurement angle, and when the reference data about feature 2701 is in Fig.27A The minimum and maximum allowable vector lengths generated when moving within the uncertainty region defined by the tolerance band 2702 shown in FIG. 27A. In order to perform accurate metrology measurements and calculate the exact feature width when scanning the part, the system is preferably designed to ensure that the reported vector measurements are not adversely affected or incorrectly reported due to inaccuracies in the system caused by, for example, but not limited to, vibrations in the stage, non-linearities in the stage, and limitations in stage repeatability.

[0172] The locations on the part that are automatically selected for measuring the data used to calculate the statistics are called critical width vector locations. They correspond to those vectors called critical width vectors that do not change magnitude or only change within a small defined minimum critical width tolerance when the feature moves within its tolerance band. For example, a feature may move + / - 1 µm in different directions within its tolerance band, but the critical width tolerance for the conductor width may be set to 0.2 µm so that the conductor's vectors meet the requirements of critical width vectors. As an example, in Table 27, vectors V2, V3, V4, and V5 remain constant when the reference data moves within the tolerance band, and therefore these vectors can be used to measure the critical width. On the other hand, vectors V1 and V6 do not meet the requirements of valid vectors for measuring critical width because vector V1, which ranges between 0 and R1, and vector V6, which ranges between R6-D and R6+D, change magnitude when the reference data moves within the tolerance band. Because vectors V1 and V6 can change value when the part is scanned due to system uncertainties, they are not used to calculate statistics.

[0173] Fig.27BTable 27 in lists examples of such vectors and indicates which are critical width vectors that can be used to calculate statistical measures.

[0174] Fig.28A represents a scanned camera image where the conductor edges of the critical width vectors are found at positions 2AF, 2BF, 3AF, 3BF, 4AF, 4BF and 5AF, 5BF. These positions are Fig.28A Such critical width vectors found in the scanned image are marked with the letters "CW" before the vector name. For example, reference vectors V2 = CW-V2, V3 = CW-V3, V4 = CW-V4, and V5 = CW-V5.

[0175] To calculate the statistics of the scanned part, all features of the same design size in the reference data, such as same-sized lines, same-sized spaces, and same-sized pads, can be grouped together. As an example, all critical width vectors (see Fig.27A and Fig.27B ) can be grouped together to compute statistics for feature size or line width W1. Similarly, all critical width vectors whose minimum and maximum allowable magnitudes are W2 can be grouped together to compute statistics for feature size or line width W2, and so on.

[0176] This grouping results in billions of critical width measurements being performed as the part is scanned in real time. Using all the information provided in the inspection parameter list, including which vectors constitute the critical width vector, e.g. Fig.27B As indicated in the last column of Table 27, this enables simultaneous calculation of critical width statistics based on billions of sub-micron resolution measurements while simultaneously finding defects on parts also measured at sub-micron resolution.

[0177] Fig.28C Table 28-2 in the Fig.28A Table 28-2 shows the minimum, average, maximum, and standard deviation values ​​for the results for conductors W1 and W2 calculated from the critical width vectors measured across the feature. In general, such statistics can be calculated for the entire part, for a section of the part, for each die on the part, or for a section of each die, and for up to 100 percent of the features on the part.

[0178] When scanning a part during inspection, there are two preferred methods for calculating the vector length from the scanned image. The two methods are defined as follows:

[0179] Vector Length Method 1 defines the vector length of a measurement from a scanned image as the length between one of the edges found within a specified search range and the next material change along the vector measurement axis. As an example, Fig.15A , the vector H2 calculated during inspection extends from one edge of the feature to the pinhole (represented by the change in material) located at the center of the feature.

[0180] Vector Length Method 2 defines the vector length of a measurement from a scanned image as the square root of the sum of the squares of the two edge pixel locations found on opposite sides of the feature along the vector measurement axis within a specified search range. As an example, Fig.28A Neutralization Form 28 ( Fig.28B ) vector CW-V 2F The length of is given by:

[0181]

[0182] One edge of the feature is located at pixel position (2AF x , 2AF y ), and the other edge of the feature is at pixel position (2BF x , 2BF y ).

[0183] Vector Length Method 1 has the advantage that it will detect defects in the middle of a feature, but requires that all pixels along the length of the vector need to be analyzed. In contrast, Method 2 is less computationally involved because only edge pixel locations are processed to calculate the vector length across the entire feature width, but Vector Length Method 2 alone will not detect defects in the middle of a feature, such as pinholes. However, if the inspection system incorporates Method 2 to measure feature width, combined with comparing the scanned image to reference data to find additional and missing features (such as in Fig.24 , Fig.25 and Fig.26 ), then pinholes can be detected as additional insulator defects in the middle of the line, given a pattern matching search area (also as shown in Fig.24 , Fig.25 and Fig.26 ) is reduced to fall within the boundaries of the feature, as determined by Fig.29 The pattern matching search area 2901 is indicated at the intersection of vectors H2 and V2.

[0184] When the inspection goal is to find all defects and collect statistical line width and line spacing data at the same time; then, for those products when the line / space size of the conductor is close to the inaccuracy in the system, Vector Length Method 1 (which can detect defects in the middle of the conductor) is preferred because it may not be possible to create a pattern matching search area that remains within the conductor boundaries, such as Fig.29However, if the line / space width is large enough to ensure that the pattern matching search area is placed within the conductor line, then Vector Length Method 2 is applicable. Also, in those applications where only statistical line width / space data is of interest, Vector Length Method 2 is also applicable.

[0185] While linear measurements are the preferred means for characterizing conductor lines and spacing, circular and elliptical measurements are more suitable for characterizing vias and pads. When generating reference data from CAD data, features such as vias and pads are often either provided on a separate design file layer or uniquely identified in the CAD file data. This information allows the inspection system to know where such features will be located on the part, so the feature shape processor can be used to analyze such features.

[0186] For example, Fig.30 3001 points to one of the edge pixels. For each edge pixel on the oval in the reference data, a vector is created, usually perpendicular to the edge, pointing to and extending to the opposite edge of the oval, using the same techniques as in FIG. 7 and FIG. Figure 8 These vectors are generated using the same procedure illustrated in FIG. 10 for the linear conductor and spacing features.

[0187] For each edge pixel of the ellipse, Fig.31 As shown in the figure, use Figure 8 The search range Sr 3101 is defined using the same procedure illustrated in FIG. 3 for linear edge features. Fig.32 The diagram illustrates the location of edge pixels to be sought in a scanned image during inspection in physical space. FIG. 33 is a scanned image of a defective pad with a defect 3301 at the edge of the pad. FIG. 33 indicates a search range Sr 3101 within which the system expects to find edge pixels in the scanned image. FIG. 34 shows the location of edge pixels found in the scanned image, which are then presented to the feature shape processor.

[0188] Fig.35 35 shows a feature shape processor 3501 that is optimized for analysis of elliptical and circular pads. Note that a circular pad is an ellipse whose major and minor axes are equal. In this example, the feature shape processor 3501 calculates a best fit ellipse (BFE) 3502 for edge pixels 3503 of the scan data, such as Fig.35 The feature shape processor 3501 then calculates the standard deviation (STD) between the scanned data pixels 3503 and the BFE 3502, and how much each edge pixel differs from the BFE. If the difference of the scanned edge pixels exceeds a predetermined value, an error is reported, as in Fig.35As illustrated in FIG, edge pixel 3504 is a distance “D” away from BFE 3502 as indicated.

[0189] Other measurements that can be calculated by the Feature Shape Processor are, for example, but not limited to: the major and minor axes, and the center of an ellipse. Statistics can be calculated for all ellipses and circles of the same size on a part, such as, but not limited to, the STD of the major and minor axes of an elliptical pad, and the STD from the center of an elliptical pad and a circular pad. Furthermore, additional Feature Shape Processors can be added to the system, each optimized for analyzing a different shape, for example, one Feature Shape Processor can be optimized for a triangular pad, and another Feature Shape Processor optimized for a teardrop pad, and so on. In general, Feature Shape Processors can be incorporated when feature-related measurements need to be performed.

[0190] As previously mentioned, in a preferred embodiment, all calculations are performed in real time as the part is scanned, with the feature shape processor performing its processing for both pads and vias while comparing to reference data, and such processing is also performed simultaneously with any length and width measurements of every line and space on the part along the entire length of every feature on the part.

[0191] Fig.36 is an example of a block diagram of a preferred system that can provide such real-time processing. The system includes a translation stage 3601 that scans a part 3602 beneath a camera 3603, which in turn transmits an image to a frame grabber 3604. The frame grabber 3604 then makes the image available to a field programmable gate array 1 (FPGA 1) 3605 for processing and analysis, with optional additional processing and analysis performed by FPGA 2 3606. Processed data from the FPGA is transmitted to a graphics processing unit 1 (GPU-1) 3607, and optionally to GPU-2 3608, for further processing and analysis. Data from the FPGA and GPU is transmitted to a CPU 3609 for additional processing and analysis. All FPGAs, GPUs, and CPUs are connected to a computer bus 3610.

[0192] To achieve the data rates and computational power required to perform billions of measurements in real time while scanning a part, the following hardware modules may be incorporated into the implementation. Camera 3603 is manufactured by Vieworks model VT-9K7X0H250 and is an 8912 pixel x 128 row TDI sensor with a row rate of 250KHz and a data rate of 2.23 Gigapixels / second. (Vieworks is a trademark of Vieworks Co., Ltd. of Gyeonggi Province, South Korea.) Frame grabber (3604) is implemented on a board manufactured by Gidel model Proc10A, which contains multiple Altera FPGA chips (Gidel is a trademark of Gidel International of Or Akiva, Israel, and Altera is a trademark of Altera Corporation of San Jose, California). Frame grabber 3606 is implemented in one of the Altera FPGA chips, and Fig.36 The FPGA-1 3605 in the FPGA is implemented in another chip on the same board. For example, the Gidel Proc 10A—which contains multiple Altera FPGA boards—is capable of processing approximately five (5) trillion (5 x 10 12 bytes) of data.

[0193] The architecture also supports multiple FPGAs and multiple Gidel Proc 10A processors, such as Fig.36 FPGA-2, indicated by 3606. High-speed data transmission lines 3611 connect signals between frame grabber 3604, FPGA-1 3605, FPGA-2 3606, GPU-1 3607, and GPU-2 3608. GPU-1 3607 and GPU2 3608 are implemented using NVIDIA Tesla P100 and or NVIDIA Tesla V100 processing accelerator cards. NVIDIA Tesla P100 provides more than 21 teraflops of 16-bit floating point (FP16) performance," and at the time of filing this application, the NVIDIA Tesla V100 Tensor Core is the most advanced data center GPU ever built to accelerate AI (artificial intelligence), high performance computing (HPC), and graphics. It is powered by NVIDIA's Volta architecture, comes in 16 and 32GB configurations, and provides the performance of up to 100 CPUs in a single GPU.

[0194] It has incredible computing power, including up to 100 CPUs in a single GPU, capable of processing up to approximately five (5) trillion (5 x 10 12 Bytes) of data and a camera scanning at a data rate of 2.23 Gigapixels / second, which enables the processes and processors described in this article to be implemented in real time.

[0195] All high-speed hardware modules (camera, FPGA, and GPU) communicate with a general-purpose CPU 3609 via a PCIE computer bus 3610. The CPU 3609 is implemented using a Xeon processor manufactured by Intel running a Windows 10 operating system manufactured by Microsoft.

[0196] It is important to emphasize that even with this extremely high speed processing hardware, pre-processing of the reference data prior to inspection to generate the inspection parameter list and measurement parameter list is an essential component, which, combined with such high speed hardware, enables billions of metrology measurements to be performed in real time as the part is scanned.

[0197] Now you can understand, Fig.36 The system architecture shown in can be used to implement one or more of the above-mentioned inspection and / or measurement techniques.

[0198] As an example, Fig.37 is an execution flow of a method for inspecting features in a part to find defects. The first step (step 3701) locates reference data and then creates (3702) (one or more) tolerance bands that specify how much one or more features can deviate from the reference data in space. Then, measurement vectors are created (3703) at fixed spatial locations around the perimeter of each feature. Each measurement vector represents the feature dimension in the reference data from one point on the perimeter to another point on the perimeter at a specified angle. Linear search ranges are also defined (3704) around each fixed spatial location so that the starting point and ending point of each measurement vector can be found. Each linear search range has a search axis that is oriented at the same angular orientation as its corresponding measurement vector.

[0199] Next, for each feature to be inspected, the feature is moved (3705) to multiple positions within the tolerance band of the feature. For each position of the feature, a point is searched (3706) along the perimeter of the feature within a specified search range at each previously created spatially fixed measurement position. When a point is found on the perimeter, the vector size is measured (3707) at the angle specified by the corresponding measurement vector. The resulting size is stored (3708) as allowable. When the vector measurement exceeds the maximum measurable value, or when no point is found on the perimeter, the result is also stored (3708) as allowable. For each spatially fixed measurement position, corresponding to multiple positions of the feature within its tolerance band, the allowable minimum and maximum vector sizes are stored (3708). These minimum and maximum vector sizes, or representations thereof, are then modified (3709) to allow for acceptable process variations during part inspection.

[0200] The process of inspecting the part from the scanned image then continues (step 3710) by searching (3711) for points along the perimeter of the feature for each measurement vector within the search range of the vector. When a point is not located, an indication is generated (3711) indicating that the point was not found on the perimeter. When a point is located, the vector dimension is measured (3713) at the angle specified for the vector, and an indication is generated (3713) indicating when the dimension exceeds a maximum measurable value. A defect is then reported (3714) for a vector dimension that is less than the minimum allowable dimension or a representation thereof or greater than the maximum allowable dimension or a representation thereof.

[0201] exist Fig.38 In another example process depicted in , a method for measuring features can be performed. The process also begins (3801) locating reference data and defining (3802) (one or more) tolerance bands, which specify how much one or more features can deviate spatially from the reference data. Measurement vectors are created (3803) at fixed spatial locations around the perimeter of each feature, where each measurement vector represents the dimension of the feature in the reference data from one point on the perimeter to another point on the perimeter at a specified angle. Linear search ranges around each fixed spatial location are also defined (3804) so ​​that a starting point and an ending point of each measurement vector can be found. Each linear search range has a search axis that is oriented at the same angular orientation as its corresponding measurement vector. For each feature to be measured, the feature is moved (3805) to multiple locations within the tolerance band of the feature. For each location of the feature, a point is searched (3806) along the perimeter of the feature within the specified search range at each previously created spatially fixed measurement location. When a point is found along the perimeter, the vector size is measured (3807) at the angle specified by the corresponding measurement vector, and the vector size or a representation indicating when the measured size exceeds the maximum measurable value is stored (3808) as allowable. When no point is found along the perimeter, a representation of the result is also stored (3808) as allowable. Then, for each spatially fixed measurement position, the allowable minimum and maximum vector sizes and their allowable representations corresponding to multiple positions of the feature within its tolerance band are stored (3808). As each feature moves within its corresponding tolerance band, those measurement vectors whose sizes do not change or change only within a small defined minimum critical size tolerance are stored and labeled (3809) as critical size vectors.

[0202] The process (3810) of measuring a part from a scanned image then searches (3811) for points along the perimeter of the feature for each critical dimension vector (3811) and within the search range of that vector (3811). At each such perimeter point located, the vector dimension is measured (3812) at the same angle as its corresponding measurement vector.

[0203] Statistics are then calculated (3813) based on the vector size measurements as a function of the feature type, wherein the statistics are calculated over the entire part, or for a section of the part, or for different dies on the part or sections within different dies on the part, where a given feature type is represented by features of the same size in the reference data. Statistics can be calculated for the feature type as at least one of minimum, maximum, mean, and standard deviation statistics. Finally, the statistics are reported (3814) for use in characterizing the part or for process control.

[0204] in conclusion

[0205] In summary, the above method performs billions of two-dimensional conductor width and spacing metrology measurements on electronic parts such as wafers and high-density circuit boards within sub-camera pixel, sub-micron accuracy, calculating such measurements at every pixel location along the entire length or perimeter of every conductor and feature on the part.

[0206] The accuracy of sub-camera pixel, sub-micron linewidth and spacing metrology measurements is not limited by the accuracy of conductor placement on the part or in a sub-section of the part being inspected.

[0207] To inspect a part, the method first searches the scanned image to locate each feature, and once located measures its size and calculates feature size statistics. Metrology measurements and defect detection are performed simultaneously in real time while the part is scanned.

[0208] Since billions of measurements typically have to be performed across the entire part, self-learning automated methods that extract measurements from computer-aided design (CAD) files or reference data adapt to the allowed manufacturing tolerances.

[0209] In addition, the technology can calculate statistics such as, but not limited to, minimum, maximum, average, and standard deviation of feature size, line width, and space measurements calculated for the entire part, a section of the part, for each die on the part, or for a section of each die, and for up to 100% of the features on the part. In addition to generating statistics based on billions of locations on the part, since the part is scanned in real time, the method also automatically determines the best location for performing the measurements used to calculate the statistics. Such an automatic setup is preferred for factories with a large volume of different parts.

[0210] The foregoing description of exemplary embodiments illustrates and describes systems and methods for implementing inspection and metrology tools, but is not intended to be exhaustive or to be limited to the precise form disclosed.

[0211] Certain portions may be implemented as "logic" that performs one or more functions. The logic may include hardware, such as hardwired logic, an application specific integrated circuit, a field programmable gate array, a graphics processing unit, or may also include, in whole or in part, a processor that executes software instructions. Thus, some or all of the logic may be stored in one or more tangible, non-transitory computer-readable storage media, and may include computer-executable instructions that may be executed by a computer, a data processing system, an application specific integrated circuit, a programmable gate array, or any other state machine. Computer-executable instructions may include instructions that implement one or more embodiments described herein.

[0212] It should also be understood that block diagrams and process flow charts may include more or fewer elements, which are arranged differently or represented differently. For example, although a series of steps have been described above about the flow chart, the order of the steps can be modified to obtain the same result. In addition, steps, operations and steps can be performed by additional or other hardware or software modules or entities, and the modules or entities can be combined or separated to form other modules or entities. For example, although a series of steps have been described about some figures, in other consistent implementations, the order of the steps can be modified. In addition, unrelated steps can be performed in parallel. In addition, the disclosed implementation may not be limited to any particular combination of hardware or software.

[0213] Therefore, it will now be apparent that other variations and modifications may be made to the described embodiments while achieving some or all of their advantages. Therefore, it is the object of the appended claims to cover all such variations and modifications as are within the true spirit and scope of the disclosure herein and their equivalents.

Claims

1. A computer-implemented method for creating optical measurements of parts and reference data measurement tolerances, Features The method comprises: a. generating reference data by using computer-aided design (CAD) data or by combining scans of one or more known good parts, the reference data being used to generate parameters for inspecting the part, b. creating one or more tolerance bands in the reference data around each of the one or more features to be inspected, so that when a part with acceptable tolerances is scanned for inspection, each scanned feature falls within the tolerance band defined by the corresponding tolerance band in the reference data, c. locate edge pixels along the perimeter of each feature in the reference data, d. for each edge pixel so located, creating a measurement vector originating from the edge pixel and pointing to the opposite side of the feature, the vector being oriented perpendicular to the perimeter or at a predetermined angle relative to the perimeter, thereby also indicating the edge orientation, and whose vector length is the distance to the edge pixel on the opposite side of the feature, e. storing information representing the vector length, vector orientation, starting edge pixel, and ending edge pixel coordinates of each measurement vector, the stored information defining a measurement location to be used to locate the edge of a feature when the part is subsequently inspected, f. define a linear search range for locating each starting and ending edge pixel of each measurement vector when subsequently inspecting the part, i. wherein each linear search range has a search axis oriented at the same angular orientation of its corresponding measurement vector, ii. a search limit in one direction along a search axis extending away from each edge pixel, wherein the limit in said direction extends to or beyond a boundary defined by a tolerance band closest to said edge pixel, and iii. a limit in the other direction along the search axis that, for the same edge pixel, is equal to or greater than the closest distance between an edge pixel along the measurement axis and its closest tolerance band located on the other side of the feature, g. storing the corresponding search range for each expected starting edge pixel and ending edge pixel position, h. Move each feature in the reference data to all positions within its tolerance band, and i. for each position of the feature at each previously stored start and end pixel measurement position, search for edge pixels of the feature within the specified search range, and i. When an edge pixel is found on a feature, A. Calculate the length of the measurement vector as the length from the found edge pixel to the edge pixel on the opposite side of the feature, said length being determined at the previously stored measurement angle, and B. storing the calculated length of the measurement vector associated with each measurement position as an allowable result, and C. When the length of the measurement vector exceeds the maximum measurable value, store a representation of the result as an allowable result, ii. When no edge pixel is found on the feature, a representation of the result is stored as an allowable result, j. storing, for each spatially fixed measurement position, the minimum and maximum allowable vector sizes and their allowable representations corresponding to a plurality of positions of the feature within its tolerance band, k. Create an inspection parameter list storing parameters for inspecting features on a part defined by reference data, the list containing information for each measurement vector representing: i. the minimum and maximum allowable vector lengths, or a representation of the vector length that indicates when the measured length exceeds the maximum measurable value, or no edge pixel is found, ii. Vector orientation, iii. Search scope, and iv. Starting edge pixel position and ending edge pixel position, l. Modify each minimum and maximum allowable vector length in the inspection parameter list to allow for acceptable process variations during part inspection, m. For each measurement location in the inspection parameter list, inspect the part from the scanned camera image by: i. Search for the edge of the feature within a search range around the expected start edge pixel and end edge pixel locations specified in the inspection parameter list, A. When edges are found in the scanned camera image, vector measurements are performed at the measurement angles specified in the inspection parameter list, B. When the length of the measurement vector exceeds the maximum measurable value, store a representation of the result, C. When no edge pixels are found in the scanned camera image, a representation of that result is stored, ii. compare the vector measurement length to the permissible minimum and maximum length values ​​or their permissible representations specified in the check parameter list, and iii. In the check parameter list, a vector length that is less than the minimum allowable length or its representation or greater than the maximum allowable length or its representation is reported as a defect.

2. The method according to claim 1, further comprising: include: Use edge-finding filters to compute edge pixel locations with sub-pixel accuracy.

3. The method according to claim 1, additionally include: The scanned camera image of the part is compared to a reference image to look for defects.

4. The method according to claim 1, additionally include: a. Convert CAD data into grayscale images, and b. Convolve or process the grayscale image through a two-dimensional filter to simulate the blur of the imaging optics so that the reference data better matches the scanned camera image.

5. The method of claim 1, wherein the part comprises a conductor and an insulator, the conductor and the insulator being distinguishable from one another in the scanned camera image.

6. The method according to claim 1, additionally include: a. During inspection of a part from a scanned camera image, for a plurality of measurement locations corresponding to feature locations in the reference data, where feature-related measurements are to be calculated, i. searching for edge pixels of the feature in the scanned camera image within a search range around the expected edge pixel locations specified in the reference data, ii. present it to the feature shape processor for analysis, find the location of the edge pixels associated with each feature, and then b. use the feature shape processor for further i. measuring one or more feature dimensions and measuring the distance of the found edge pixels to a predetermined expected shape, or at least one of the widest dimension, narrowest dimension, or center of the feature, and i. measuring the distance of the found edge pixels to a predetermined expected shape, or at least one of the widest dimension, narrowest dimension, or center of the feature, and ii. Report measured feature sizes that exceed a predetermined value as defects.

7. The method according to claim 1, additionally include: a. storing and indicating additional vectors in the inspection parameter list, the length of which does not change or changes only within a small defined minimum critical width tolerance when each feature moves within its corresponding tolerance band in the reference data, such vectors are called critical width vectors, b. When measuring a feature on a part from a scanned camera image, for each critical width vector, search for edge pixels of the feature within a search range around the specified starting and ending edge pixel measurement locations and calculate a vector length measurement defined as the length between those two found edges, c. from those critical width vector length measurements, calculating statistics as a function of feature type, wherein the statistics are calculated over the entire part, or for a section of a part, or for different dies on a part, or sections within different dies on a part, where a given feature type is represented by features of the same size in the reference data, d. the statistic calculated for the feature type is at least one of the minimum, maximum, mean, and standard deviation statistics, and e. Report the statistical data to characterize the part or for process control.

8. The method of claim 2, wherein the edge-finding filter is a Sobel or Pruitt type filter.

9. A computer-implemented method for creating optical measurements of parts and reference data measurement tolerances, Features The method comprises: Perform automated metrology measurements on electronic parts to obtain statistical data; a. Generate reference data by using computer-aided design (CAD) data or by combining scans of one or more known good parts, the reference data being used to generate parameters for measuring the part, b. creating one or more tolerance bands around each of the one or more features to be measured in the reference data so that when scanning a part with acceptable tolerances, each scanned feature falls within the tolerance band defined by the corresponding tolerance band in the reference data, c. locate edge pixels along the perimeter of each feature in the reference data, d. for each edge pixel so located, creating a measurement vector originating from the edge pixel and pointing to the opposite side of the feature, the vector being oriented perpendicular to the perimeter or at a predetermined angle relative to the perimeter, thereby also indicating the edge orientation, and whose vector length is the distance to the edge pixel on the opposite side of the feature, e. storing information representing the vector length, vector orientation, starting edge pixel, and ending edge pixel coordinates of each measurement vector, the stored information defining the measurement position to be used to locate the feature edge when the part is subsequently measured, f. define a linear search range for locating each starting and ending edge pixel of each measurement vector when subsequently measuring the part, i. wherein each linear search range has a search axis oriented at the same angular orientation of its corresponding measurement vector, ii. a search limit in one direction along a search axis extending away from each edge pixel, wherein the limit in one direction extends to or beyond a boundary defined by a tolerance band closest to the edge pixel, and iii. for the same edge pixel, the limit in the other direction along the search axis is equal to or greater than the closest distance between the edge pixel along the measurement axis and its closest tolerance band located on the other side of the feature, g. storing the corresponding search range for each expected starting edge pixel and ending edge pixel position, h. Move each feature in the reference data to all positions within its tolerance band, and for each position of the feature at each previously stored start and end pixel measurement position, search for edge pixels of the feature within the specified search range, and i. When an edge pixel is found on a feature, A. Calculate the length of the measurement vector as the length from the found edge pixel to the edge pixel on the opposite side of the feature, said length being determined at the previously stored measurement angle, and B. storing the calculated length of the measurement vector associated with each measurement position as an allowable result, and C. When the length of the measurement vector exceeds the maximum measurable value, store a representation of the result as an allowable result, ii. When no edge pixel is found on the feature, a representation of the result is stored as an allowable result, i. storing, for each spatially fixed measurement position, the minimum and maximum allowable vector sizes and their allowable representations corresponding to a plurality of positions of the feature within its tolerance band, j. storing and indicating in a measurement parameter list the edge pixel measurement positions of those vectors whose lengths do not change or change only within a small defined minimum critical width tolerance as each feature moves within its corresponding tolerance band, such vectors are called critical width vectors, k. When measuring a feature on a part from a scanned camera image, for each critical width vector, search for edge pixels of the feature within a search range around the specified starting and ending edge pixel measurement locations and calculate a vector length measurement defined as the length between those two found edges, 1. from those vector length measurements, computing statistics as a function of feature type, wherein the statistics are computed over the entire part, or for a section of a part, or for different dies on a part, or sections within different dies on a part, where a given feature type is represented by features of the same size in the reference data, m. The statistic calculated for the feature type is at least one of the minimum, maximum, mean, standard deviation, or other statistic, and n. Reporting of statistical data to characterize a part or for process control.

10. The method according to claim 9, further comprising: include: Use edge-finding filters to compute edge pixel locations with sub-pixel accuracy.

11. The method according to claim 9, additionally include: The scanned camera image of the part is compared to a reference image to look for defects.

12. The method according to claim 9, additionally include: a. Convert CAD data into grayscale images, and b. Convolve or process the grayscale image through a two-dimensional filter to simulate the blur of the imaging optics so that the reference data better matches the scanned camera image.

13. The method of claim 9, wherein the part comprises a conductor and an insulator, the conductor and the insulator being distinguishable from one another in the scanned camera image.

14. The method according to claim 9, additionally include: a. During inspection of a part from a scanned camera image, for a plurality of measurement locations corresponding to feature locations in the reference data, where feature-related measurements are to be calculated, i. searching for edge pixels of the feature in the scanned camera image within a search range around the expected edge pixel locations specified in the reference data, ii. presenting the locations of the edge pixels found associated with each feature to the Feature Shape Processor for analysis, and iii. The feature shape processor then measures one or more feature dimensions and measures the distance of the found edge pixels to a predetermined expected shape, or at least one of the widest dimension, narrowest dimension, or center of the feature, b. calculating from those feature dimensions statistics for features of the same shape and size in the reference data, wherein the statistics are calculated over the entire part, or for a section of the part, or for different dies on the part or sections within different dies on the part, and c. Report the statistical data to characterize the part or for process control.

15. The method of claim 9, wherein the statistical data comprises minimum, maximum, average, and standard deviation feature size, line width, and spacing measurements calculated for an entire part, a section of a part, each die on a part, or a section of each die on a part.

16. The method of claim 10, wherein the edge-finding filter is a Sobel or Pruitt type filter.

17. A computer-implemented method for creating optical measurements of parts and reference data measurement tolerances, Features The method comprises: a. locating the reference data, b. defining a tolerance band that specifies how much one or more features of the part may deviate spatially from the reference data, c. Create measurement vectors at fixed spatial locations around the perimeter of each feature, d. Each measurement vector represents the size of a feature in the reference data from one point on the perimeter to another point on the perimeter at a specified angle, e. A linear search range is defined around each fixed spatial position so that the starting point and the ending point of each measurement vector can be found, f. Each linear search range has a search axis that is oriented at the same angular orientation of the measurement vector to which it corresponds, g. for each feature to be inspected, move the feature to a plurality of positions within the tolerance band of the feature, and h. for each position of the feature and at each previously created spatially fixed measurement position, search for a point along the perimeter of the feature within the defined linear search range, and i. When a point is found, A. measure the vector size at the angle specified by the corresponding measurement vector, and B. storing a vector size or a representation indicating when a measured size exceeds the maximum measurable value as allowable, ii. When no point is found, store a representation of the result as admissible, and i. storing, for each spatially fixed measurement position, the minimum and maximum allowable vector sizes and their allowable representations corresponding to a plurality of positions of the feature within its tolerance band, j. modify each minimum and maximum allowable vector size or its representation by increasing the variance to allow for acceptable process variation, k. when performing optical measurement of the part from a scanned camera image, for each measurement vector, searching for points along the perimeter of the feature within a defined linear search range of the vector, and i. When a point is not located, an indication is generated indicating that the point was not found on the perimeter, ii. When a point is located, the vector size is measured at the angle specified for the vector and an indication is generated indicating when the size exceeds the maximum measurable value 1. When using the optical measurement to detect defects, vector sizes that are smaller than a minimum allowable vector size or a representation thereof or larger than a maximum allowable vector size or a representation thereof are reported as defects.

18. The method of claim 17, wherein edge-finding filters are used to calculate the positions of points along the perimeter of the feature with sub-camera pixel accuracy.

19. The method according to claim 17, additionally include: The scanned camera image of the part is compared to a reference image to look for defects.

20. The method according to claim 17, additionally include: a. Converting computer-aided design (CAD) data into grayscale images, and b. Convolve or process the grayscale image through a two-dimensional filter to simulate the blur of the imaging optics so that the reference data better matches the scanned camera image.

21. The method of claim 17, wherein the part comprises conductors and insulators, the conductors and insulators being distinguishable from one another in the scanned camera image.

22. The method according to claim 17, additionally include: a. During inspection of a part from a scanned camera image, for a plurality of measurement locations corresponding to feature locations in the reference data, where feature-related measurements are to be calculated, i. searching for edge pixels of the feature in the scanned camera image within a search range around the expected edge pixel locations specified in the reference data, ii. presenting the locations of the edge pixels found associated with each feature to the Feature Shape Processor for analysis, and iii. The feature shape processor then measures one or more feature dimensions and measures the distance of the found edge pixels to a predetermined expected shape, or at least one of the widest dimension, narrowest dimension, or center of the feature, and b. Report measured feature dimensions that exceed a predetermined value as defects.

23. The method according to claim 17, additionally include: a. further storing and indicating those measurement vectors, called critical width vectors, whose sizes do not change or change only within a small defined minimum critical dimension tolerance as each feature moves within its corresponding tolerance band in the reference data, and b. When measuring a part from a scanned camera image, for each critical dimension vector, i. defining locations along the perimeter of the feature within the search range of the vector, and ii. at each such perimeter point located, measuring the vector size at the same angle as its corresponding measurement vector in the reference data, c. calculating statistics as a function of feature type from the vector size measurements, wherein the statistics are calculated over the entire part, or for a section of the part, or for different dies on the part or sections within different dies on the part, wherein a given feature type is represented by features of the same size in the reference data, d. Calculate statistics for, but not limited to, minimum, maximum, mean and standard deviation feature sizes for different feature types, and e. Report said statistics for use in characterizing the part or for process control.

24. The method of claim 18, wherein the edge-finding filter is a Sobel or Pruitt type filter.

25. A computer-implemented method for creating optical measurements of parts and reference data measurement tolerances, Features The method comprises: a. Positioning reference data, b. defining tolerance bands that specify how much one or more features may deviate spatially from the reference data, c. Create measurement vectors at fixed spatial locations around the perimeter of each feature, d. Each measurement vector represents the size of a feature in the reference data from one point on the perimeter to another point on the perimeter at a specified angle, e. A linear search range is defined around each fixed spatial position so that the starting point and the ending point of each measurement vector can be found, f. Each linear search range has a search axis that is oriented at the same angular orientation of the measurement vector to which it corresponds, g. for each feature to be inspected, move the feature to a plurality of positions within the tolerance band of the feature, and h. for each position of the feature and at each previously created spatially fixed measurement position, search for a point along the perimeter of the feature within the defined linear search range, and i. When a point is found, A. measure the vector size at the angle specified by the corresponding measurement vector, and B. storing a vector size or a representation indicating when a measured size exceeds the maximum measurable value as allowable, ii. When no point is found, store a representation of the result as admissible, and i. storing, for each spatially fixed measurement position, the minimum and maximum allowable vector sizes and their allowable representations corresponding to a plurality of positions of the feature within its tolerance band, j. further storing and marking those measurement vectors as critical dimension vectors whose dimensions do not change or change only within a defined minimum critical dimension tolerance as each feature moves within its corresponding tolerance band, k. When measuring a part from a scanned camera image, for each critical dimension vector, i. defining locations along the perimeter of the feature within the linear search range defined for the vector, and ii. for each such perimeter point located, measuring the vector size at the same angle as its corresponding measurement vector calculated from the reference data, 1. calculating statistics as a function of feature type based on the critical dimension vector measurements, wherein the statistics are calculated over the entire part, or for a section of the part, or for different dies on the part or sections within different dies on the part, where a given feature type is represented by features of the same size in the reference data, m. The statistic calculated for the feature type is at least one of the minimum, maximum, mean, standard deviation, or other statistic, and n. Reporting of statistical data used to characterize a part or for process control.

26. The method of claim 25, wherein edge-finding filters are used to calculate the positions of points along the perimeter of the feature with sub-camera pixel accuracy.

27. The method according to claim 25, additionally include: The scanned camera image of the part is compared to a reference image to look for defects.

28. The method according to claim 25, additionally include: a. Converting computer-aided design (CAD) data into grayscale images, and b. Convolve or process the grayscale image through a two-dimensional filter to simulate the blur of the imaging optics so that the reference data better matches the scanned camera image.

29. The method of claim 25, wherein the part comprises conductors and insulators, the conductors and insulators being distinguishable from one another in the scanned camera image.

30. The method according to claim 25, additionally include: a. During inspection of a part from a scanned camera image, for a plurality of measurement locations corresponding to feature locations in the reference data, where feature-related measurements are to be calculated, i. searching for edge pixels of the feature in the scanned camera image within a search range around the expected edge pixel locations specified in the reference data, ii. presenting the locations of the edge pixels found associated with each feature to the Feature Shape Processor for analysis, and iii. The feature shape processor then measures one or more feature dimensions, including measuring the distance of the found edge pixels to a predetermined expected shape, or at least one of the widest dimension, narrowest dimension, or center of the feature, b. calculating statistics for features of the same shape and size in the reference data from those feature sizes, wherein the statistics are calculated over the entire part, or for a section of the part, or for different dies on the part or sections within different dies on the part, the statistics being calculated for at least one of the minimum, maximum, mean and standard deviation of the measured feature sizes, and c. Report statistical data for characterizing parts or for process control.

31. The method of claim 26, wherein the edge-finding filter is a Sobel or Pruitt type filter.