CN100520379C - Method for checking flat medium with pattern and its equipment - Google Patents

Abstract

A concurrent low resolution/high resolution parallel scanning system is provided as an improvement in the scanning process of an inspection system for planar objects, such as large flat plates employed in panel displays, whereby lower resolution defect detection efficiently overlaps and parallels higher resolution defect review and classification stages in which defects are automatically defined and resolved. Although the invention is a valid solution for the more general problem of optically inspecting the surface of a flat article for defects, the invention is particularly useful for detecting pattern defects on large glass plates deposited with integrated-circuits for forming LCD flat panel displays.

Description

Method and apparatus for inspecting patterned flat media

Cross References to Related Applications

This application is a continuation of U.S. Patent Application Serial No. 10/439,991, filed May 16, 2003, which claims the benefit of the filing date of U.S. Provisional Application Serial No. 60/423.008, filed November 1, 2002, The entire content of the latter is hereby incorporated by reference.

technical field

This invention relates to the general field of machine observation based inspection techniques, and more particularly to machine observation based detection and classification of defects occurring on large, patterned, flat surfaces. In particular, the present invention addresses the inspection of materials deposited on large substrate glass plates such as liquid crystal display (LCD) panels. Although the invention is applicable in general to the inspection of any patterned flat media, the invention is particularly concerned with the inspection of glass substrates for preformed thin film transistor (TFT) LCD panels.

Background technique

In the manufacture of LCD panels, large, clear sheets of thin glass are used as substrates on which to deposit layers of materials to form circuits that serve as multiple identical display panels. This deposition is usually carried out in stages, where in each stage a specific material, such as metal, indium, or Tin oxide (ITO), silicon, or amorphous silicon. Each processing stage includes various steps such as deposition, masking, etching and stripping.

During each processing stage and within each step within a stage, it is possible to introduce various structurally-affecting production defects that can have electronic and/or visual implications on the final LCD panel product. Such defects include but are not limited to circuit shorts, open circuits, foreign particles, mask issues, feature size issues, over-etch and under-etch. In order for the final LCD panel to function properly, these defects are preferably detected, classified and, if necessary, repaired at the stage of defect generation. The decision to fix is based on the exact classification of the defect, in particular the distinction between "fatal", "repairable" and "process" defects.

The operating resolution of a system for automatic defect detection often has a direct impact on inspection speed and cost of the system. Therefore, only relatively low resolutions are suitable for scanning the entire area of the plate. Unfortunately, at this lower resolution, it has historically not been possible to perform both detection and reliable classification from the same collected image data. In addition, low resolution has an impact on the performance of the detection algorithm, which often generates a considerable number of false alarms that need to be eliminated. Therefore, after the defect detection step, a defect review step is required, in which higher resolution inspection (via camera) is used to obtain defect regions of interest for later verification of candidate defects, and in the Automatic or human-assisted classification is then performed.

In such operations, a low-resolution subsystem is used to detect problem areas of the panel (Defect Detection Subsystem - DDS), while a separate high-resolution camera is used at a later stage to acquire high-resolution images of these problem areas images (Defect Review Subsystem - DRS), so as to achieve more reliable automatic or manual classification. As long as the number of problem areas detected by DDS can be kept within manageable limits, high resolution image acquisition for these single points remains feasible. In addition, this number often has a direct impact on the cycle time of the system (if all defective areas are to be reviewed) or on the system's review performance (if a fixed finite number of defects is to be reviewed).

Automated optical inspection (AOI) equipment has been used for a variety of problems including, but not limited to, printed circuit board (PCB) inspection, silicon very large scale integration (VLSI) circuit die (die) inspection, and LCD panel inspection. Most of the solutions employed are based on spatial domain pattern comparison techniques, which are often combined with sensor-level pixel or sub-pixel precision calibration techniques.

U.S. Patent No. 4,579,455 assigned to Levy et al. discloses a calibration and pattern comparison technique in which a pair of 7×7 windows on test and reference images are considered, and multiple possible 3×7 windows within this window The sum of squared errors over 3 sub-windows is computed. If the minimum error on these combinations exceeds a threshold, it is considered defective. This approach appears to be able to compensate for calibration mismatches down to the sensor pixel level.

The statement relating to the coarse calibration accuracy of the method of US Patent No. 4,805,123 invented by Levy et al. and assigned to Specht et al. describes an improved calibration and comparison technique for detecting defects. In this technique, large windows in test and reference images are used to compute sensor pixel-level correlations between test and reference. The minimum point that produces the sampled relevant surface is found, and a quadratic function is applied to the adjacent surfaces of the minimum point. Using this suitable quadratic function, a subpixel-accurate transformation can be obtained to calibrate the test and reference images. The calibrated images were compared by thresholding the image differences over 2x2 sub-windows on the test and calibrated reference images.

Variations and improvements to these basic techniques have also been proposed, for example, U.S. Patent No. 5,907,628 assigned to Yolles et al., which, among other things, points out the disadvantages of using sampled correlation surfaces to find minima and argues that due to Coarse sampling of the surface makes this point likely not correspond to a true minimum. Therefore, they argue that the subsequent sub-pixel interpolation step does little to improve the detected minima and will produce erroneous calibration results leading to false alarms in the detection. Yolles et al. propose to alleviate these problems by refining the comparison process in terms of improved comparison entities.

As with any of the above methods, the use of a single suitable (relatively low) resolution to scan the surface of the object being inspected will result in a set of possible disadvantages. These possible disadvantages must include both legitimate disadvantages and false alarms, since these methods cannot completely filter out expected changes between test and reference images. This results in alerts for a set of candidate defects and increases the need to verify these to form a true defect map of the inspected object. Additionally, there is a strong need to classify plausible defects into defect classes, thereby aiding in the handling of inspected objects, possibly allowing repair of the object in some applications.

A solution along this direction is proposed in US Patent No. 5,699,447 assigned to Alumot et al., which describes a two-stage scanning scheme. The entire panel is first scanned in one stage at higher speed by a lower resolution non-charge coupled device (CCD) optics stage with a small diameter laser beam in raster scan mode. This is followed by a second stage of scanning using a high-resolution CCD-based optical system. The latter scanning stage extracts higher resolution images of all defect suspect locations detected by the previous scanning stage. Although this solution illustrates the need to extract a higher resolution image of the inspected object for verification, it differs from the present invention in that the first stage of inspection is by raster scanning with a small diameter laser beam, And the two checks are performed in sequential phases. It includes the following disadvantages:

(a) Increased cycle time: the high-resolution imaging phase comes immediately after the main detection phase. The sequential nature of high-resolution imaging has a significant impact on inspection cycle time, as the time required for high-resolution image acquisition, review, and classification is added to the time required for inspection scans.

(b) Idle imaging resources: The high-resolution defect review imager is idle when the inspection imager is active, and the inspection imager is idle when the review imager is active. This results in an inefficient use of system resources within a given time limit.

(c) Non-optimized review process: When classification is required and production environment time constraints prevent imaging of all candidate defect locations, it may leave the user of the system with the difficult task of deciding how many and which candidate defects to use Resolution review imager collection and processing.

What is needed is a fully automated and overlapping high resolution defect review system that operates in conjunction with fast and precise classification techniques to improve the speed and accuracy of inspection and repair.

Contents of the invention

According to the present invention, it provides a concurrent low-resolution/high-resolution parallel scanning system as an improvement to the scanning process of the inspection system, thereby enabling the low-resolution defect detection stage to be effectively integrated with where defects are automatically defined and resolved The high-resolution defect review and classification stages overlap and run in parallel. While the invention is an effective solution to the more general problem of optically inspecting the surface of flat objects for defects, the invention is particularly useful for detecting pattern defects on large glass plates on which integrated circuit LCD flat panels are deposited. Accordingly, the invention will be described with respect to this particular application.

The present invention provides an improved system that is mechanically and electrically capable of concurrently temporally performing low-resolution and high-resolution imaging of an object surface under inspection. This consists of two independent and parallel imaging subsystems with different resolutions, each capable of acquiring images from the surface of the object under inspection in parallel. The low-resolution imaging stage includes the Defect Detection Subsystem (DDS), and the high-resolution imaging stage includes the Defect Review Subsystem (DRS). Furthermore, each of these subsystems may in turn have one or more imaging channels.

DDS covers the entire surface of the object under inspection, typically using multiple identical, relatively low-resolution imaging optics and photoelectric transducers such as CCD devices or complementary metal-oxide-semiconductor (CMOS) photosensitive devices. ). The express purpose of this subsystem is to distinguish candidate defects and optionally pre-classify them according to their location relative to the TFT grid features. This pre-classification can classify candidate defects into: data lines, gate lines, transistors, capacitors, and ITO electrodes. Since the number of defects that can be reviewed in a given time is limited, the results of defect pre-classification can be used to prioritize competing defects for review. Prioritization of candidate defects is accomplished by assigning each candidate defect a review value factor corresponding to the pre-classification results.

The DDS may be mounted on a moving device that moves along the object under inspection, possibly through multiple passes, or alternatively be fixed mounted on a moving surface carrying the object under inspection.

DRS covers a small inspected area with a small number of identical imaging optics and photoelectric transducers with relatively high resolution. The specific purpose of this subsystem is to image at high resolution the candidate defects identified and pre-classified by the DDS subsystem for automatic final defect classification. The DRS imaging channel can be mounted on a moving device that moves along the object under inspection synchronously with the DDS. Each DRS channel is also capable of moving independently through the object under inspection.

By overlapping low resolution defect detection and high resolution defect review and classification functions, the present invention seeks to significantly reduce the overall cycle time for automated optical inspection of large patterned flat objects. This increases the value of (Automated Optical Inspection Instruments) AOI instruments for users who often operate under tight cycle time constraints given the inspection issues applied to TFT LCDs, thus adding AOI systems for in-line 100% TFT -LCD panel inspection opportunity.

Examination is simplified by mechanical movement through the DRS imaging channel, which is dynamically adjusted by a scheduling algorithm. This algorithm seeks to optimize the defect review effort by maximizing the number of high priority defect candidates captured by the DRS while minimizing the distance the subsystem modules move. The input to this algorithm is the spatial distribution of candidate defects detected by DDS within the plane of the glass sheet and the review value factors assigned to candidate defects by DDS. The result of the scheduling algorithm is an optimized motion pattern for each individual DRS module. This sport pattern is then executed by the DRS sport system.

Therefore, the optimal scheduling algorithm minimizes the number of candidate defects that cannot be reviewed by high-resolution imaging of a DRS with a given number of cameras. It also allows defect candidates that are more important for high-resolution imaging to be prioritized. Overall, it either results in minimizing the number of cameras required for a given candidate defect distribution, thereby potentially reducing system cost, or improving system performance with the same number of camera modules.

One particular embodiment of a scheduling algorithm may be based on graph theory, and is specified in this disclosure for demonstration purposes. However, other alternative implementations of the scheduling algorithm are possible, and the invention is not limited to the specific details of the algorithm described.

On-the-fly (OTF) focusing is a typical feature required by the present invention to achieve both low and high resolution overlap by bringing the high resolution imaging system into sharp focus at the candidate defect location concurrent DDS/DRS system. It increases the time efficiency of the overall scheme and enables the DRS to operate non-stop along the y-axis.

The real-time concurrent review and triage that is part of the DRS minimizes the need for human intervention during the inspection process and increases the cycle time efficiency of the system.

The present invention and its components function as an inspection and related processing pipeline which enables all key stages of inspection to be fully active and utilized during scanning of the inspected object.

The present invention will be better understood with reference to the following detailed description taken in conjunction with the accompanying drawings.

Description of drawings

Figure 1 is a top view of a small area of a panel section showing possible defect types;

The perspective views of Figure 2A and Figure 2B respectively show a perspective view of a possible implementation of the present invention corresponding to the relative scanning directions;

Figures 3A and 3B show feasible boundaries;

4A, 4B and 4C illustrate access policies;

5A and 5B illustrate the autofocus strategy.

Detailed ways

Referring to FIG. 1, some possible defects that can be found during LCD panel manufacturing are shown. These defects include metal bumps 110 protruding into the ITO layer 112, ITO bumps 114 protruding into the metal 116, so-called rat bite cracks 118 in the metal 112, open circuits 120 in the metal 116, cracks in the transistor 122 of the pixel. Transistor shorts 124, and foreign particles 126 in any areas. Each type of these defects must be checked, classified and repaired where possible.

According to the present invention, the inspection platform 10 (FIG. 2A or 2B) is equipped with a Defect Review Subsystem (DRS) 12. The DRS 12 includes a plurality of substantially identical higher resolution imaging optics and photoelectric transducers that together form a set of camera modules 14, 16, 18 that are capable of being inspected along with defects on a gantry 20 The scanning motion of the subsystem (DDS) 24 moves together along the length of the inspected object 22 . Each identical camera 14 , 16 , 18 is also capable of independently moving across the scanning motion of the defect detection subsystem 24 .

A suitable mechanical layout of such a system is illustrated in Figures 2A and 2B in two opposing scan directions, where the DSR 12 is placed on its own scan gantry 20, which can be moved along the low-resolution rate scan direction movement. Individual high resolution DRS modules 14, 16, 18 are operated for movement across the main scan direction. For purposes of illustration, twenty DDS24 modules 26-45 and three DRS modules 14, 16, 18 are shown. The actual number of modules may also be determined by the needs of a particular application. The DRS modules 14 , 16 , 18 do not have to be on separate racks 20 , they can also be installed on the same rack, such as the DDS rack 47 . Also, although the use of a linearly driven gantry is diagrammed here, it is not the only approach when considering imaging scans performed on the surface under inspection. There may be other alternative methods of performing this scan, such as a tilt-turn mirror scanner with suitably designed stationary optics, and the algorithms disclosed herein may be employed accordingly.

Operation of the DDS/DRS concurrent system 10 includes causing the lower resolution DDS 24 on the gantry 47 to scan the entire area of the object under inspection, which can be performed in multiple passes, where each pass covers a portion of the total area . During each round, DDS performs basic defect inspection and classification, and generates a list of defects and defect candidates. These candidate defects are then queued for dynamic scheduling imaging with DRS 12 at higher resolutions.

Each such defect candidate is also associated with a DRS value measure to indicate the level of importance of imaging this particular defect candidate with high resolution. The DRS 12 cooperating with the DDS 24 executes an optimal motion pattern to maximize the number and value of candidate defects acquired in the current round. Candidate defects that cannot be obtained in the current round are scheduled in subsequent rounds. The details of this process are described in the following sections.

Small changes in the flatness of the large object under inspection across the scan area combined with the very small depth-of-field of the typical optics of the DRS modules 14, 16, 18 are necessary conditions for the DRS module's dynamic focus scheme. Accordingly, the present invention incorporates a DRS module 14, 16, 18 capable of performing dynamic autofocus while traveling to a candidate defect location. The details of this feature are described in the following sections.

DRS module scheduling

A feature of the present invention is the process of scheduling available high-resolution DRS modules 14, 16, 18 to scheduled defect candidates to optimize the effort for defect review by making the This is achieved by maximizing the number and DRS value while minimizing the distance the modules move. A specific embodiment of such a scheduling algorithm is based on graph theory and will be described for illustrative purposes. The invention is not limited to the specific details of this algorithm, alternative methods providing dynamic scheduling may also be used.

Based on the assumption that the expected distribution of defects is uniform, each DRS module 14, 16, 18 can be considered to be independent and have the minimum amount of interaction required to resolve the x-span partitioning of the individual responsibilities and prevent collisions, To be explained below.

As the DDS 24 scans, it reveals candidate defects to be imaged concurrently by one of the high resolution DRS modules 14,16,18. Due to mechanical acceleration and speed constraints on the system design, a DRS module 14, 16, 18 adhering to these constraints may not be able to capture all revealed candidate defects in the same pass. Full details of a particular scheduling algorithm are not given in this disclosure, but a list of key steps is given here and some key constraints that must be considered are outlined so that a person of ordinary skill in the art can come up with a suitable scheduling algorithm. The scheduling algorithm used to dispatch selected DRS modules 14, 16, 18 to candidate defects takes into account mechanical constraints on module motion, such as maximum acceleration and speed, dynamic focus requirements, number of defects that can be acquired by alternating motion patterns , and the value of the DRS value of each individual defect. A suitable scheduling algorithm also handles the case of multiple rounds of scanning over the scan area, effectively enabling candidate defects that cannot be captured on the current round to be captured in subsequent rounds.

The DRS module scheduling process should consider a fixed or varying look-ahead window in the main direction of the scan (denoted as the y-axis in the x-y plane) to determine the dynamic list candidates subject to the scheduling optimization process of the DRS imaging module defect. In addition, this window can cover the entire glass span in case all untreated defect candidates are considered for this scheduling optimization. For any given distance along the Y-axis, a single DRS module has a field of fit (FOA) of a region within which the module can reach any point within it, assuming the module is initially stationary, And it tends to come to rest again when the movement reaches the end. Figures 3A and 3B illustrate suitable access fields.

As shown in Figures 3A and 3B, each initial position and velocity of a particular DRS module and a preselected look-ahead window on the y-axis result in a particular visited field. This field of access areas determines which defects are reachable from the current location of the DRS module. However, once the module starts to move towards a target location that may be a defect candidate, this area will change and must be recalculated. If the module is not stationary about the x-axis when computing a region, the shape of this region will change.

According to a particular embodiment of the scheduling algorithm described herein, each of the DRS modules of the present invention processes the visible candidate defect map in a selected y-axis window by appropriate access field calculations to perform the following steps:

(a) Construct a forward flow graph with nodes corresponding to candidate defects and current module positions, and arcs corresponding to feasible motions from the current position to and between candidate defects (arc);

(b) Compute a value factor for each arc representing the movement of a module from one candidate defect to another. This value factor includes the distance value of the physical candidate defect to the candidate defect, the penalty of losing other defects (due to the new suitable access field generated in the new position), and the negative cost (negative cost) of acquiring the target defect (or by its benefits of reviewing value decisions);

(c) Find the resulting graph for the minimum cost path from the current position to the end of the considered y-axis window;

(d) Calculate the required motion trajectory for the corresponding DRS module.

4A to 4C illustrate the above process. The motion solution is optimal as long as the set of candidate defects falling within the considered y-axis lookahead window does not change. However, once a new defect, possibly with a different value, appears within the view of this window, the current motion solution may no longer be optimal, and the above process should be repeated to find a new optimal one for the involved DRS modules. path. When computing a new solution, the DRS modules involved may already be in motion towards the defect. Based on the new optimal solution, the module can change its orientation towards different candidate defect locations.

As a feature of the invention, the cost of an arc is determined as an appropriate function of the negative cost (i.e., the benefit) of acquiring a target candidate defect, when an arc is selected and the total distance and energy cost of movement is associated with that arc, the candidate The cost of the defect is lost. The individual factors in the total cost are normalized to compatible dynamic ranges and weighted according to the needs of the specific application to produce the final arc cost.

The resulting graph-theoretic problem of "shortest path finding" is solved for a minimum-cost path from the current module position up to and including the last candidate defect along the y-axis within the module's look-ahead window. After obtaining the last defect candidate in the computational order, the module keeps x static until another new defect candidate appears within the lookahead window, in which case the process repeats. Solutions to graph-theoretic problems arise by using algorithms that are well established in this field. For example, Dijkstra's algorithm for shortest paths or its incremental variant known in the art seems well suited for this task, but other alternative algorithms can also be considered.

Another feature of the invention is the incorporation of boundary conditions and the presence of adjacent modules into well-articulated diagrams by means of additional constraints on suitable access fields. Each of adjacent DRS modules crossing the same x-axis is assigned equal-sized scheduling regions with an appropriate amount of overlap between them. The x-span partitioning into equal-sized regions is based on the assumption of a uniform distribution of defects.

It is possible for a candidate defect to fall within the overlap region between two adjacent DRS modules, in which case the solutions of these two modules will be considered. This situation does not lead to any conflict conditions if only one or no modules include that particular defect in their optimal scheduling trajectories. Where both optimal paths include that particular defect, an algorithm can be chosen to take into account the calculated total optimal outcome costs of both modules, and assign the defect to the one that results in the minimization of the two total optimal costs module. For example, let m _i and m _i+1 be two adjacent modules, C ^* _i and C ^* _i+1 are two optimal outcome costs, then the defect is assigned to module k satisfying the following conditions:

$\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arg</span>}\mathrm{<span\; class="notranslate">\; min</span>}<span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

After finding the optimal scheduling solution for all DRS modules, the corresponding motion data is calculated and sent to the DRS motion control hardware. This relationship holds true for the various stand combinations contemplated by the present invention.

DRS camera module dynamic auto focus

Another feature of the invention is the process of dynamically adjusting the focus of the DRS camera modules and finding the best focus. This is done during the movement of the module towards the target candidate defect location, as explained below.

There are random deviations from flatness throughout the span of the large flat object being inspected. These deviations are assumed to have low spatial frequency (ie occur slowly along the x-y plane). This effect is combined with a very narrow depth of field for the very high microscope magnification required for high resolution imaging. Therefore, the optics must be refocused for each new position being imaged. To maximize defect coverage by the DRS module, this focusing needs to be done on the fly, ie while the module is moving towards a candidate defect location. This is especially required in embodiments where the DRS module is mounted on the same gantry as the DDS, since the latter has constant y-axis motion dictated by low resolution line scan imaging requirements.

In one embodiment, the present invention seeks to achieve optimal focus on a candidate defect location by performing a series of full or partial image acquisitions during movement towards the candidate defect. In any embodiment, the dynamic focus process is initiated and completed at a distance close enough to the target location that the resulting focus is effective at the target location.

This process is carried out by means of the following steps:

(a) starting at a predetermined distance from the target candidate defect and during the movement of the module, acquire a series of image data from the DRS camera module;

(b) sampling a focus quality curve by using the series of image data extracted from the camera module in combination with focus quality measurements computed on those images;

(c) interpolating the samples using a smoothing function to determine maximum focus for adjusting the focusing optics along the focus direction;

(d) directing the focusing optics to a position where the focus quality criterion is maximized to obtain the sharpest focus on the target area.

Figures 5A and 5B illustrate the movement of modules from candidate defect _n2 to _n5 , where _n5 is assumed to still be suitable under additional constraints. Depending on the nature of the required optics, e.g. a line-scan sensor or an area-scan sensor required for the application, the image acquisition step in (a) can be performed either during any diagonal movement of the module (also in the x-axis direction) ( FIG. 5A ), it can also be executed only when the camera becomes x-stationary (FIG. 5B ). In the latter case, the calculation of the appropriate access field takes into account an additional x quiescent period at the end of the motion for focusing purposes.

The focus quality measure used in step (b) may be based on the contrast of the image and the highest frequency content in the image, which is limited by the modulation transfer function (MTF) of the lens system used.

Candidate Defect Classification in DRS Motion

Using the high-resolution images captured by DRS, the system can perform concurrent review and automatic classification of candidate defects found on inspected objects.

DDS generates a stream of candidate defects that are queued and scheduled for imaging by multiple DRS modules. The DRS module is scheduled to perform high-resolution imaging of prominent candidate defects by means of the methods explained in the previous sections. This produces a stream of candidate defects associated with the high resolution image data. These candidate defects undergo a two-stage process, including:

(a) automatic review; and

(b) Automatic classification.

During the automatic review (AR) process, high-resolution candidate images are compared with reference images (or more specifically, data representations of reference images) stored in system memory, which were previously acquired with the same module or with a different module of. This step involves compensation for known variations between test and reference, including corrections for such things as camera sensitivity, sensor pixel sensitivity variations, and spatial misalignment at the sensor pixel level. Due to the high resolution of the DRS module, no sub-pixel calibration is required.

The results of the automated review process either confirm the existence of legitimate defects at alternative locations, or reject defects as "false" defects, ie, artifacts of known limitations of low-resolution DDS. Reasonable defects are forwarded to the automatic classification stage.

During the Auto Classification (AC) stage, the high-resolution defect image is combined with the output of the AR stage to be used to extract relevant features of the defect and a final decision on the defect type is made through the classification process. The individual features that may be of interest depend on the particular application, but should include defect size, defect location, defect shape, and signal level.

In the case where the TFT-LCD panel inspection considered is for a product defect, the main output of the classification system is to decide whether the defect is:

a) process defects,

b) a repairable defect, or

c) Fatal flaws.

Subcategories within these main decisions are considered according to the needs of specific users.

Another feature of the present invention is the ability to implement the above-mentioned stages in parallel (ie, concurrent processing). For example, the AR and AC phases operate concurrently as part of the DRS operation, while the DDS also simultaneously performs a low-resolution scan of the entire surface of the inspected object.

The invention has been explained above with reference to specific embodiments. Other embodiments will be apparent to those of ordinary skill in the art. It is therefore not the intention of the foregoing description to limit the invention, except as indicated by the appended claims.

Claims (14)

1. A method of inspecting patterned flat media comprising: Using relatively low-resolution imaging and localization protocols to detect defects in patterned flat media by imaging equipment; concurrently Imaging is performed using relatively high resolution based imaging and localization protocols to review defects pointed out by the above inspection steps. 2. The method of claim 1, wherein the review step uses dynamically optimized in-motion autofocus imaging. 3. A defect detection method performed concurrently with in-motion review and classification of candidate defects in a patterned flat object under test, said method comprising the following operations: Obtaining an image of an object through a defect detection subsystem; detecting a candidate defect of the object; and assigning a review value to the candidate defect, the defect detection subsystem having a plurality of defect detection subsystem modules, the modules according to a first operate at a relatively low operating resolution; The review and classification operations described therein include: acquiring an image of a smaller area covering the candidate defect by a defect review subsystem; reviewing the smaller area; and classifying the smaller area, the defect review subsystem having a plurality of defect review subsystem modules, The modules operate according to a second relatively higher operating resolution. 4. The method of claim 3, comprising: Utilize the dynamic defect review subsystem module scheduling algorithm to maximize the number of higher-priority candidate defects acquired by the defect review subsystem module, and minimize the distance moved by the defect review subsystem module, so as to Optimized for motion in the concurrent higher resolution review described above. 5. The method according to claim 4, wherein the scheduling algorithm further comprises: reviewing subsystems for each defect and constructing a graph-theoretic forward fit motion graph at each iteration of said motion optimization, said graph having a node corresponding to each candidate defect and its associated module's current position , and have arcs corresponding to suitable motions between said nodes; The resulting graph is obtained by associating each of said arcs with a cost according to an appropriate cost function chosen from among functions representing the cost of missing other defects, the distance required to move, and the review value of the target candidate defect, so said arc represents module movement from a first of said candidate defects to a second of said candidate defects; solving a resulting graph for finding a minimum cost path represented by a sorted order of defect-to-defect transitions from the current position of the defect review subsystem to the end of a window considered along the scan direction; and Motion data is calculated for the defect review subsystem module for use in controlling the motion of the defect review subsystem module. 6. The method of claim 3, comprising: automatically focusing the imaging element starting at a predetermined distance from the target candidate location and during motion of the defect review subsystem module; obtaining a focus quality metric curve using at least samples of the focus quality metric computed on the image; interpolating the samples of the focus quality metric using a smoothing function to determine a point of maximum focus for adjusting the focus optics along the focus direction; and The focusing optics are directed to a position that maximizes the focus quality metric curve to obtain the sharpest focus at the target candidate position. 7. The method according to claim 6, wherein the focusing step comprises: Obtaining a sequence of image data from an imaging element of a defect review subsystem; and wherein said obtaining step comprises A focus quality curve is sampled using a series of image data in combination with said focus quality measurements computed on the images. 8. The method of claim 3, comprising: Generate a list of candidate defects in the defect detection subsystem; Queue and schedule imaging by multiple defect review subsystem modules; scheduling the defect review subsystem module to perform relatively higher resolution imaging of the salient candidate defects to generate a series of candidate defects associated with the relatively higher resolution image data; Defect candidates are subjected to a two-stage process consisting of: Automatic review processing; and Automatic classification processing; The high-resolution candidate image is compared to reference images of known defect states stored in system memory during an automated review process, wherein the comparison process includes compensating for known variations between test and reference, which Including at least one of the following amendments: a) imaging device sensitivity, b) Changes in sensor pixel sensitivity; Compensating for spatial misalignment at the sensor pixel level to produce confirmation of the presence of plausible defects at candidate locations or to reject defects as false defects, including artefacts of known limitations of low-resolution DDS; transmit information about legitimate defects for automatic classification processing; thereafter In the automated classification process, relatively high-resolution defect imaging is used in conjunction with the output of the automated classification process to extract relevant features of the defect; The final conclusion of the defect type is made through the classification process. 9. An apparatus for inspecting patterned flat media comprising: An inspection subsystem for inspecting defect candidates with imaging equipment using relatively low-resolution imaging and localization protocols; A review subsystem concurrently reviews defects indicated by the inspection subsystem using a relatively high resolution imaging and localization protocol. 10. The apparatus of claim 9, wherein the detection subsystem is further capable of assigning a review value value to the candidate defect. 11. A device for defect detection that concurrently performs defect review and classification of phenomena in a measured object while in motion, said device comprising: a defect detection subsystem having a plurality of defect detection subsystem modules for acquiring an image of an object at a first relatively lower operating resolution, detecting a candidate defect, and assigning a review value to the candidate defect; and A defect review subsystem having multiple defect review subsystem modules capable of concurrently acquiring images covering a small area of a candidate defect, reviewing the candidate defect with a relatively high resolution and Defects are classified as defects. 12. The device according to claim 11, wherein the defect detection subsystem is installed on the first movable platform, and the defect review subsystem is installed on the second movable platform. 13. The device according to claim 11, wherein the defect detection subsystem includes a plurality of detection modules fixedly installed on the first movable platform, and the defect review subsystem includes a plurality of detection modules installed along the second Multiple defect review subsystem modules for movable gantry movement. 14. The apparatus of claim 13, wherein movement of a first one of the defect review subsystem modules is constrained by a second position of the defect review subsystem module, and wherein the apparatus is further A controller is included, the controller can perform the following operations: Constructing a forward flow graph having a node corresponding to a candidate defect and the current position of one of the defect review subsystem modules and having links from the current position of the defect review subsystem module to the first selected candidate defect and intermediate the arc corresponding to the feasible motion of the second selected defect candidate; For each arc representing the movement of a module from one candidate defect to another, a cost is associated with the arc as a function of value factors, including the cost of missing other defects, the distance that must be traveled, and the cost of acquired defects. value to get the resulting graph; solving the resulting graph to minimize the cost path from the current location of the defect review subsystem module to the end of the considered window along the length of the inspected object; and Motion data is calculated for the defect review subsystem module for use in controlling the motion of the defect review subsystem module.

Images