DE4026206A1 - Fully automatic two=dimensional measuring method - using microscope detecting orientation markers and verifying mechanical microstructure on substrate - Google Patents

Abstract

The microstructures to be measured are mounted on a table movable in two directions and compared with a reference structure from a CAD system. Three orientation markers are detected and measured and the substrate rotated and translated in the table plane. Substrate tilting is detected and the z-dimension measured via the microscope z-drive stoke. The structure is verified according to CAD demand data taking into account rotation and tilt errors and is divided into several basic geometric shapes which are globally treated to associate the detected elements with the desired structure. ADVANTAGE - Eliminates lengthy adjustment of substrate w.r.t. microstructure.

Description

The invention relates to a method for fully automatic two-dimensional measurement of mechanical microstructures on substrates according to the preamble of the claim.

Quality control is becoming increasingly popular in industry Importance. Constantly increasing demands on the quality of Products force extensive controls during the Fer maintenance. The technical complexity of the articles to be manufactured requires ever more complex and costly production method.

However, checking the end products only prevents that Release a faulty item to the market and not also unnecessary costs of incorrect production processes. Around The exact location, time and type of errors are determined the intermediate products are also increasingly being checked. It is enough only rarely the information that errors are present. Rather, the cause of errors is searched for the product tion process to be able to correct.

The results of a time-consuming and costly visual inspection, those in industrial quality control more often are often subjective and tiring for the examiner dend. In addition, they do not always deliver the desired detail lated information about the dimensional accuracy of the test specimen, since the Extent of measurements the personnel capacity often by far exceeds.

This problem is particularly evident in the microferti supply. With the smallest manufacturing benefit, thousands of Structures integrated. The high cost of production pro Processes force here to an extensive extent in particular exams. The dimensional accuracy of so many structures can be however, no longer determine manually.  

A procedure of the e. G. Art is from "Image Analytical Quality control in microfabrication ", 11th DAGM symposium, Proceedings: "Pattern recognition 1989", IT reports no. 219, Springer Verlag, known.

With this method it is possible to create structures with A CAD system was designed to be two-dimensional measured. However, the assignment of the actual to the target structure is particularly problematic with periodic structures. If too the substrate is not parallel to the microscope stage lies, e.g. B. by dust or particles on the microscope stage, the structures are blurred and cannot more recognized and verified.

The object of the invention is now a method of the beginning to develop the type mentioned, in which any orien in space tated 2-dimensional structural surfaces by dreidimensiona len mechanical microstructures recognized and clearly one Target structure can be assigned.

This task is performed by the defining part of Pa claim resolved.

The method according to the invention is described below using a Example explained in more detail.

It became a system from already available on the market Components put together with which the e. G. method can be carried out.

The system is based on the VME bus. Including the operating Find terminals and the color monitor for image playback all hardware components in a 19 ′ ′ rack with 33 heights units (HE) place.  

As a master computer that also functions as a basic CPU, control processor and bus controller, the Motorola processor 68020 used at 25 MHz. In addition to the mathematical co-pro processor 68881 it has 4 MB of local dynamic RAM. The A 120 MB SCSI disk is allocated directly to the CPU.

The operating system is the real-time system OS-9. Despite the lack of "hard real time conditions" in this System has proven the use of this operating system because it is easy to use, takes up little storage space and is also suitable for software development.

An Ethernet LAN Con is used for communication with the outside world troller available, which uses TCP / IP proto via OS-9 Internet koll masters.

The image is taken in with a video card. The video signal a SW camera is shown here in an image grid of 512 × 512 Pixels digitized with 256 gray levels each. At the same time The card contains an image memory for the recording of two 512 × 512 images. From the digitized image can be a special BV card for image preprocessing in Video norm computes a line image with 1-pixel wide lines will.

A CCD camera with Sony-Sen is attached to the digitizing card sor 2/3 ′ ′ and 756 × 581 pixels connected. The effec tive TV resolution is 560 × 575 pixels.

The multi-processor concept of the slaves is right with transputers alized. Thereby T800 from Inmos with 20 MHz and each 4MB of dynamic RAM used. Through the new to the market 4 MBit chips can be a total of 8 processors of this equipment in a single VME bus slot.  

The transputers are connected in the form of a trinary tree, this means that of the 4 available communications onsleitung (left) one with the higher-level transputer (Father transputer) while connected to the rest Links to three other transputers (sons) getting produced. A link of the root transputer is physika connected to the VME bus. He also leaves the direct Access to its memory from the VME bus too (dual-ported RAM).

A fully automatic light microscope was used as the measuring microscope with computer interfaces and a 6 ′ ′ scanning table det.

The individual process steps are described in more detail below purifies.

Three marking crosses are predefined on the substrate ter position applied that the substrate coordinate system describe. Before starting a substrate measurement, the Substrate coordinates are transformed into table coordinates. This requires a relationship between the microscope stage and the substrate coordinate dinate system. The cross marks will be automatically started and the respective exact position determined in the x, y and z directions. The determination of the z-posi tion takes place through a software-based autofocus that the calculation of the max. Image contrast is based.

The center of gravity coordinates of the marking crosses result in the x and y position. To determine this, for each Cross all image objects to the defined cross geometry tries. The Kreuz, von determines a plausibility check which the center of gravity coordinates are then calculated.

In the rare event that a cross is not the he has waited geometry, it is possible, nevertheless, manually to determine the x-y-z position.  

The transformations take into account both the rotation and also the translation of the substrate with respect to the microscope cal as well as the tilting of the substrate with respect to the Mi croscopic. This gives the possibility of sub strat m. E. in any position (on the microscope stage) measured.

The procedure for object verification is based on a local and global pattern recognition. The usage Both methods enable a structural measurement, the Er accuracy is limited only by the optics used becomes. Positioning inaccuracies - causes e.g. B. by Inexpensive microscope stages - are of no importance.

Local pattern recognition

The pattern recognition takes place in a pixel list which is on the Based on a line calculated from the original grayscale image image with 1-pixel-wide lines.

In picture windows of definable size within the com munifying pixels in the pixel list corners, straight lines, circles Arches, spirals and surfaces verified.

Contrary to the general understanding of a corner here A total of three corner types are distinguished:

• 1. as the intersection of two straight lines,
• 2. as the intersection of a straight line with an arc and
• 3. as the intersection of two arcs.

Corner detection

The parameters of a corner are the center of the corner, the direction of the leg direction and sense of rotation. The corner point we are looking for is the pixel whose neighboring pixels most similar to the leg directions  Have gradient directions in the corresponding direction of rotation. This guarantees the detection of a corner in any case continuous A quality value of the corner is calculated, which is the deviation The actual corner is evaluated by the shape of the target corner.

Line detection

Parameters are the two line end points. All lines between two already verified endpoints are in the Pixel-Li most examined. You go into a straight fit (with least Squares methods). Standard and maximum deviation of all associated pixels are compared to the calculated Gera determined the parameter.

Arc detection

Parameters are circle radius, circle center, start and End point and direction of rotation. The verification is carried out by a Circle fit with standard and maximum deviation as with the Gera detection.

Surfaces

They are verified by the lines surrounding them described. Irregularities can be discovered.

Global pattern recognition

Because the local pattern recognition global geometric aspects can not take into account all verified corner points within an image in a global view checked again. This is especially important if same structures appear in a periodic arrangement in the picture nen. It tries to use least squares methods to Assign corners to the target corners. In an iterative process with a self-adjusting picture window will be wrong Assignments rejected and redetermined. The result is one error-free pattern recognition and a transformation matrix,  that in a rotation and a translation part the place describes the difference between the target and actual image content.

With the method, an individual can still be Carry out a fully automatic measuring program that is not too full constant measurement of all structures and structural parts forces, but the restriction to a definable sub quantity (critical areas). This is achieved through:

• - Breakdown of the objects to be measured into geometric pri mitives (corners, straight lines, arcs, spirals, hyperbolas, Parabolas, etc.),
• - The structuring of pattern recognition commands similar to the geometric primitive describing the picture objects ven, where the hierarchy of commands is related the graph-theoretical proof of freedom from cycles he brings,
• - the use of imaginary geometric primitives,
• - the naming of the critical areas to be measured already with the CAD design,
• - the formulation of image processing commands on the Basis of the CAD original data and the data of the kriti areas,
• - the property of "resolution invariance" that makes it possible turns on and off during the fully automated process to control up to 6 different lenses on the same substrate ern.

The error evaluation can be done both online on the monitor represent, as well as off-line - very detailed - by Ver Execute from target data with the determined actual data ren.

Error evaluation on-line

Errors that are detected by the pattern recognition are "Corners do not exist", "Corner quality value too bad",  "Line not available", "Line interrupted". Mistakes that can be detected by the measurement, "distance is not calculable "," Distance deviates from the nominal size ". The height of the Distance deviation and the degree of roughness of the structure edges are divided into several deviation classes, and who which can be identified by different colors on the monitor makes.

Error evaluation off-line

The actual data determined describes the geometry of the actually existing structures. By comparing the Parameters of the geometry primitives can be very detailed lated deviation of the actual structures from the target structures to calculate. The error information goes over considerably the statement "the structure is faulty", and be writes the deviation of the individual structural part both due to the deviation of the parameters of the respective structural pr mitiven as well as by the deviation of individual contour image Points. This automatically results in the type of error.

A particular advantage of the method is that a time-consuming adjustment of the substrate with the mechanical Microstructures are eliminated. Tilting can be tolerated the. By combining local and global patterns identifier can rely on the use of a high-precision, expensive Mi can be omitted microscopically, without an accuracy penance must be accepted.

Claims (1)

Method for fully automatic two-dimensional measurement of mechanical microstructures on substrates, in which the measurement is carried out through a microscope, in which the mechanical microstructures to be measured are held on a table which can be moved in two directions, and the automatically detected structures with reference structures from one CAD system are compared, characterized in that

• a) by recognizing and measuring three orientation marks on the substrate in front of both rotation and translation of the substrate in the plane of the measuring table surface, and the tilting of the substrate from this plane are determined, the measurement of the z dimension over the stroke of the z -Drive of the microscope is determined, then
• b) the structure is verified element by element on the basis of the CAD target data, the falsifications due to the rotations and the tilting being taken into account and the structures being broken down into a set of predetermined geometric basic shapes, and then
• c) the recognized elements are assigned to the target structure via a global treatment of all recognized geometric basic shapes.