WO1987001806A1

WO1987001806A1 - Method for the automatic optical control of objects such as integrated circuits

Info

Publication number: WO1987001806A1
Application number: PCT/FR1986/000307
Authority: WO
Inventors: Jean-Claude Georgel; Jacques Gerber; Olivier Hignette; Jean-Claude Noack
Original assignee: Bertin & Cie
Priority date: 1985-09-16
Filing date: 1986-09-15
Publication date: 1987-03-26
Also published as: FR2587489A1; JPS63500976A; FR2587489B1; EP0238529A1

Abstract

Disclosed is a method for automatically controlling integrated circuits before their mounting, or similar objects. An image of the object to be studied is recorded by means of a microscope coupled to a TV camera, for example, and the image is compared with a prerecorded reference image, both images being recorded in the form of digital signals; a differential binary image is created by making an anomaly bit signal correspond to the pixels where there is no coincidence, whereafter the parameters (position, height, width, elongation, compactness) are analyzed for each defect comprised of a plurality of adjacent pixels corresponding to an anomaly signal for a possible discard decision. Arrangements are provided to make the system independent of defects which do not disturb the operation of the circuits.

Description

Method for automatic optical inspection of objects such as integrated circuits "

The present invention relates to an apparatus for controlling small integrated electronic circuits, better known under the name of "chips", after their manufacture and before their use.

The integrated electronic circuits each comprise a support made of semiconductor material on which thin layers are deposited and diffused and traced a series of metal tracks leading to connection pads. An insulating layer, called

"passivation layer" covers the tracks. The circuits are produced by mainly photographic techniques which will not be described here, in the form of wafers or slices, for which the technicians use the name of "wafers", of a diameter which is for example approximately 10 cm. , which each contain a large number of identical chips, for example 500, separated from each other by a network of orthogonal lines, carried by the edge and called "cutting paths".

The slices, when they are finished, are glued on a plastic membrane stretched on a rigid frame adapted to be stacked with other similar frames in special baskets. A cutting of the wafer to separate the chips usually takes place while it is held on the plastic membrane, and sawing is done by following the cutting paths. Checks are made during fabrication, but other checks must be made after manufacture and before mounting the chips in the devices that include them. The checks, which can be carried out before or after cutting, usually include an electrical check under tips and a visual inspection under a microscope intended to eliminate the chips having defects liable to harm the subsequent operation of the device in which they are mounted. The most frequently observed faults are:

. scratches on metal tracks; . holes in the passivation layer covering the metal; . ink stains on metal connection pads (ink from the previous marking of a defective chip);

. sawing defects destroying or threatening the integrity of the guard ring;

. lack of metal on a track; . absence of electrical control marks on the connection pads;

. metal bridging between two tracks; . oxide defect under one. metal track;

. bridging of two metal tracks by a dust under the passivation layer.

The optical control also detects the frequent presence of anomalies which are not considered as faults because they are not such as to cause rejection. These are small (non-metallic) dusts and "peppercorns".

We call "peppercorns" surface irregularities of the metal tracks or polysilicon often due to the polycrystalline nature of these layers.

Optical control, exercised by a human operator under a microscope, is tedious and delicate work. In addition, it is not perfectly reliable: according to the operator, and for the same operator according to its degree of fatigue, rejection rates can vary in considerable proportions, up to a ratio of 1 to 10. The current trend is to an increase in the complexity of the circuits, which will make work even more difficult.

There is therefore a need for improvement and automation of the so-called "optical" control.

Automatic optical control devices are already known for controlling the masks used for making chips, and in which an image obtained with the aid of a microscope lens is transformed into a video image which is analyzed and then checked using using suitable software. The problem of checking a mask is comparatively simple, because the comparison of a single image, in black and white, with a reference image makes it possible to detect all the faults. The problem becomes more complex when it comes to the control of a wafer during manufacture, and the complexity becomes maximum for the control of a finished and sawn wafer, because all the defects which we have given the list can be find simultaneously present. Some of these faults are barely visible under lighting conditions, in particular, which allow the detection of other faults and vice versa. As a result, the complication of the apparatus and the time necessary for the control are increasing.

The slice control devices currently in use generally include an optical system involving a human operator for the detection and identification of faults, assisted by automatic means of handling the frames carrying the slices up to the microscope. and operating their alignment under the optical axis thereof. It exists also machines using video image analysis techniques to determine the width of metal tracks.

We have also proposed, cf. article by Mr. George HUANG in the magazine SPIE VOL. 449, 1984, p. 563-571, entitled "A Robotic Alignaient and Inspection System for semi conductor Processing", un. control system operating by direct comparison -: two different optics are directed one towards the object to be controlled the other towards a reference object, and the images are then combined then the combination is sent to a TV camera whose signals are sent to image processing circuits. Such a system requires the use of two separate optics, necessarily spaced from each other by a distance at least equal to the dimension of a wafer carrying the chips to be inspected. Maintaining correct centering with sufficient forecast to allow .combi¬ nation by optical means with the necessary finesse of details is problematic.

We know, in particular from US-A-4481664, a method of optical control of objects such as pla¬ quettes carrying one or more integrated electronic circuits which escapes the above objection. According to this process, we operate by comparing a binarized image of the object, that is to say a black and white image, with an analog reference image, stored in memory beforehand, and we create a binary image. differential by matching a binary anomaly signal to the elementary surfaces of the image (or pixels) where there is no agreement between the recorded signal and the reference signal. The document EP-A-0124113 describes an analogous technique.

Such methods are effective in detecting certain anomalies, such as a notch or protrusion from the edge of a metal strip deposited on an insulator. On the other hand, they can allow other anomalies to escape, such as a flaking of the insulation, or a stain on a metallic part, which do not sufficiently affect the level of reflected or diffused energy to change the value of the binary signal assigned to the corresponding pixel The object of the invention is therefore to provide a method and an automatic control apparatus intended for sorting which, while allowing great safety and a high speed of operation, highlight the defects which do not do not necessarily translate into a very large variation in the level of reflected or diffused energy, and provides numerous information on all the faults noted. The invention provides for this purpose a method of optical control of objects such as wafers carrying one or more integrated electronic circuits, comprising the following steps: a) a digitized image of the object, or of the part of it, is recorded object to be studied, i.e. an image formed by a set of digital signals each corresponding to an elementary surface (or pixel) of the object, defined by its coordinates, the digital value corresponding to the reflected energy and / or scattered for each pixel of the object, b) this digitized image is compared with a reference digitized image stored beforehand by making the difference between the energy levels of the corresponding pixels of said images of the object and of reference, so as to create a differential digital image, c) this image is binarized so as to obtain a binarized image which comprises anomaly signals representing the ele surfaces where there is no agreement between the recorded signal and the signal reference, d) the parameters (position, height, width, elongation, compactness) of each defect constituted by several elementary surfaces having an anomaly signal are determined, and e) a rejection decision is taken when one of these parameters exceeds a limit fixed in advance or, in case of doubt, the image is saved for a subsequent human decision. The essential characteristic of the invention lies in the fact that in step b), the comparison is operated by making the difference between the energy levels of the corresponding pixels - _t which results in a digitized differential image, which is transformed in step c) into the differential binary image provided for in this step.

Thanks to this feature, a considerable number of faults or anomalies do not remain outside the scope of detection, hence increased safety. According to advantageous methods:. the defects constituting "peppercorns", that is to say being found on a metallic or polycrystalline track, due to surface irregularities, are not subjected to the extraction and to the determination of parameters. ¬ of this track and affecting a number of lower elementary surfaces, to a predetermined value;

. several images corresponding to different lighting conditions are recorded, and said images are compared not only to a reference image but also to each other, and in particular in this "case, a visible light image and an image of fluorescence, and the visible defects on the two images which correspond to dust are not subjected to the extraction and determination of parameters;

. before saving the image, we save the position mark image and of orientation being on the object, the difference in position and orientation between these marks and the corresponding marks existing on the reference image is measured, and the object is therefore moved either mechanically , either one of the two images optically or electronically, to bring the recorded image into coincidence with the reference image and preferably the orientation deviations "between the two images are corrected by rotating a prism of Dove around a coaxial axis or light beam. Anyway, it is advantageous to provide that the position and orientation marks are formed by remarkable lines of the object, with high contrast and intersecting substantially at an angle law;

. in a first station, the object is pre-positioned, then it is transferred to a second station for checking, the two stations being provided with locating means placed in a similar manner, and the means of transport being provided with means associated locating means and the means of locating the means of transport ^{* are brought} into coincidence successively with the locating means of the two stations, so that the prepositioning is retained after transport and, advantageously, the at least partially detected faults in the first post;

. if the object bears a mark, for example in ink, which was imposed during an anterior check such as an electrical check under spikes to indicate that it is defective, the anomaly detected corresponding to this mark leads to a rejection decision without extraction and determination of its parameters;

. at least one image obtained with lighting combining a light background and a background is recorded dark in selected proportions;

. at least one image obtained with partially coherent light illumination is recorded; . the reference image previously stored in memory is obtained by recording, under the same conditions as for the image of the object to be examined, the image of an object considered to be faultless by an operator, or an image formed by the juxtaposition of partial images each corresponding to a part of an object considered to be without defect.

The invention will now be described in more detail, using a practical example, given without limitation, and illustrated with the aid of the figures, among which:

Figure 1 is a schematic perspective view of an installation according to the method.

Figure 2 is a diagram of the optical column of the first station. Figure 3 is a diagram of the optical column of the second station.

Figure 4 is a diagram of the main light source of the second station.

Figure 5 is a diagram of the light source for fluorescence.

Figure 6 is a diagram of the command and control station and the electronic image processing.

The installation described in the figures is intended for the automatic control of slices which are glued to a plastic membrane stretched in a frame. The slice, after gluing, may have been sawn in an automatic machine, or still be in one piece. Sawing does not reach the plastic membrane.

The frame is of a type allowing automatic sawing. It consists of a metal ring or made of plastic. Its outer perimeter has four flats, and, on either side of one of them, a V-notch and a straight notch defining a frame of reference specific to the frame. The frames are transported in baskets containing, for example, 10 or 25 superimposed frames.

The installation described in the figures is designed to be placed at the entrance to an assembly line. It receives baskets of wafers having undergone an electrical check between centers, and it delivers baskets of controlled wafers intended for sawing if it has not been done, for removing the marked defective chips, then for assembling the good chips for obtaining various devices.

The installation shown comprises, preferably, groups on the same frame 1:

. a receiving-shipping station 2 which receives the baskets to be checked, the baskets for the sections inspected, the baskets for the sections to be rejected;

. a so-called "macrovision" station 3, which ensures the "prepositioning" of the wafer in centering and orientation, as well as the detection of previously marked chips and certain coarse defects;

. so-called "microinspection" stations 4 which detect and identify faults. Consideration of the respective speeds at the different stations shows that the installation advantageously comprises four to eight microinspection stations;

. a marking station 5 for defective chips;

. a set of handling 6 of the frames; . a command and control station; . electronic circuits, grouped in boxes and associated with memory or program supports.

The receiving-shipping station 2 is, in the embodiment described, permanent static, that is to say that it does not move the baskets or the frames.

It contains means to prohibit or signal the incorrect placement of a basket, or the presence of a basket that does not conform to the intended model. It contains means for keeping the baskets in the proper position.

It is also provided with means for locking a basket, the slices of which are being inspected, and for signaling the corresponding unoccupied levels of the basket.

All these means are within the reach of those skilled in the art, so there is no need to describe them in more detail. ^~ The posted Macrovision receives frames carrying the slices, these frames are placed and immobilized on a support 8 through a clamping block comprising a flat plate and connected to a source of vacuum. This depression presses the membrane against the plate and will therefore immobilize the wafer 9. The flat plate is carried by a turntable, itself carried by a table movable in two directions XY. The flat plate is perpendicular to the optical axis of the observation equipment which will be described later. The setting up of a frame is carried out by taking as reference the two notches provided at its periphery, and which we mentioned above. The imprecision on the position of a wafer, or more precisely of one of the chips which it comprises, after its immobilization on the flat plate, relative to the observation apparatus, 1 t

is evaluated at + 1 mm in translation and + 6 ° in rotation.

The aim of prepositioning is to reduce this inaccuracy to a few tens of micrometers in translation and a few minutes in rotation.

Like this, the cut paths are used as benchmarks, in the case of an unsawed slice, and the saw cuts in the case of an already sawed slice. Another task of the macrovision station is the detection of ink marks made at the previous electrical control. This detection does not require great tracking accuracy, and it makes it possible to decrease the time spent at the microvision station, which is the busiest, by avoiding having to control chips already recognized as defective, and by avoiding a mark detection sequence.

In the same state of mind, we can also plan to detect large defects such as scales coming from sawing and having damaged the guard ring of a chip, or even the absence, on a stud at the macro-vision station. of connection, marks created by the points during the electrical control. The optical equipment comprises an optical column associated with a sensor of the video camera type.

The optical column (cf. FIG. 2) comprises two switchable objectives 10, 11 of respective magnifications x 2 and x 0.3 respectively intended for prepositioning and for the detection of faults or marks, a removable filter 12 for the detection of blemishes. ink, and a lighting device 13 comprising a light source 14, a first group of optical fibers 15 intended to produce bright field lighting by sending a beam through a objective 16 then on a mirror, and a second group of optical films 18 intended to produce lighting with a black background. A mobile screen 19 makes it possible to obtain either a combination of the brightfield-blackfield lighting, or the brightfield alone.

The images received by the video camera 20 are sent to a processing system performing in real time the digitization of the images, their storage in fast memories and the usual arithmetic and logical operations between two images. This system is advantageously achieved to date by a set of electronic cards sold by the company IMAGING TECHNOLOGY and called "AP 512", "FB 512", "ALϋ 512" and "HF 512".

The AP 512 analog processing processor card receives the analog video signal from a standard camera, it digitizes this signal at 512 points per line and quantifies it in 256 gray levels, then this data is transmitted by a specialized video bus to memory cards FB 512 images. These represent the basic image memory, that is to say 512 x 512 pixels coded at 256 gray levels.

The ALϋ 512 calculation unit card performs conventional arithmetic and logical operations on 2 "operating images" of 512 x 512 points each. Each operation is performed at the video rate.

The HF 512 histogram card calculates in real time (TV scan) the gray level histogram and the coordinates of the points of the image having a given gray level.

It is understood that using this equipment, it is possible to carry out the following operations, which represent the pre-positioning operating sequence:. mechanical positioning of the center of rotation of the support frame under the optical axis of the camera;

. iterative search around this area to find in the field of view ,. 2 marking lines (cutting paths or saw cuts) perpenicular;

. mechanical recentering command X, Y, 0 to bring the 2 marking lines on the axes of the field of view: the maximum displacements to be made are of the order of + 1 mm / + 6 °. This refocusing cancels the initial uncertainties in positioning the frame on the clamping mechanism, the wafer on its support and the lithography of the chips on the wafer. . mechanical movement along the X or Y axis of the table towards the edge of the wafer (of the order of 50 mm);

. calculation of the exact position of the intersection of the 2 marking lines displayed in the field of view. No mechanical cropping at this stage;

. mechanical movement along the same axis of the table towards the opposite end of the wafer (of the order of 100 mm);

. calculation of the exact position of the intersection of the 2 saw cuts displayed in the field of view;

. angular mechanical cropping (of the order of 15 ') and possibly in X, Y translation.

In the case of an unsawed slice, the search for the cutting paths is done in a simple manner in a light background. In the case of a sawn edge, it is preferable to determine the saw lines in "mixed" background (therefore mixture of light background and dark background). Magnification objective 2 used for H

this operational phase corresponds to a pixel dimension of the order of 10 micrometers.

For the detection of ink spots corresponding to markings made during previous checks, the lowest magnification, 0.3, is used, corresponding to a larger pixel and a wider field. A 50mm field allows you to cover a 100mm (4 inch) slice in just four passes. We then establish, by the in ormati¬ c means indicated above, a card of the wafer with the position of the marked chips, therefore defective, and which will not have to be checked again. The microinspection station, in its principle, is analogous to the macrovision station, and we will only indicate here the main differences, which result from the much greater finesse of the optical analysis. - The first difference is due to the positioning. The macro-vision station performs initial positioning, both in centering and orientation. The frame is then transferred to the microinspec¬ tion station, taking precautions to keep the position acquired, but slight displacements are inevitable. The precision sought is such that the positioning must be carried out in two stages: first on a central chip of the wafer, with an average optical magnification, to crop all the chips of the wafer; then at high magnification to refocus each chip, in the case of a sawn wafer, or each group of a few chips in the case of an unsawed wafer. It is not possible to operate these positions, especially the last mentioned, in exactly the same way as at the Macrovision station for the reason that the field of examination is substantially the dimensions of a chip, so that it is impossible to use the cutting paths or_ the saw lines, which can be outside the field for the most part. It is therefore necessary to use for the identification of remarkable lines which have been determined beforehand on the chip itself.

This preliminary determination, in the current state of production, is preferably made with human intervention: it involves selecting, in the layout of the chip, well-defined and isolated patterns, comprising two lines intersecting at right angles . The operator can be helped in his work by a call to image processing techniques. For example, on an image of the chip, it can highlight only the vertical lines, or horizontal lines, so as to better see those which are isolated and of good length. ^ It should be noted that, if there is memories, which give images comprising a large number of short lines that cannot be used for locating, these memories are rarely on the edge of the chip, whereas it is preferable for better positioning; that the tracking lines are near the same edges. The identification lines are determined on a reference chip, and it is in principle valid for all chips of the same type.

The procedure is then carried out in the same way as at the macrovision station for determining the centering and orientation deviations with the desired position.

Another difference with the macrovision is that the angular deviations being small, it was possible, instead of mechanically rotating the wafer, to rotate the image on the sensor by an optical device, Dove prism-for example , whose mass is obviously much lower than that of the mechanical assembly supporting the wafer.

On the other hand, the chip-by-chip examination requires a number of displacements in X and Y equal to the number of chips to be inspected, each time with a fine positioning sequence.

The finesse of the optical analysis also results in a very shallow depth of field, which may be less than the tolerances of the planicity of the wafer. It is remedied by an automatic focusing device, well known under the name of "autofocus", and which it is not necessary to describe here. Note, however, that to improve performance (sorting rate), this autofocus is done by moving an optical element and not by moving the slice inspected.

Figures 3 and 4 show the detail of the optical column. It can be seen that as a whole, it has the same structure as that of macrovision, and, for simplicity, the same references have been given to similar parts, although, for example, the magnifications are very different.

A difference also appears in terms of lighting: a coherent imaging mode makes it possible to obtain a much greater contrast of defects and a greater value of the transfer function than in incoherent imaging mode. On the other hand, it gives rise to effects of differentiation of the typical edges (rings, fringes) in agreement with the theory of diffraction, which complicate the image and can make its interpretation difficult. In addition, it tends to accentuate the contrast of the granularity of the silicon, which can be a problem. The tests led to choosing as compromise a lighting having a similar coherence factor 0.5. We can also provide means to vary this factor.

Another important difference appears at the level of illumination. It results from the fact that fluorescent lighting requires a separate, high-power source. This source has been represented at 21, at 22 the optical fiber, and at 23 the "dichro∑que cube" associated therewith (FIG. 5).

The automatic focusing device comprises a laser diode 24, a divergent lens 25 and a two-quadrant diode 26. The signals emitted by the latter control the movement of the focusing optical group 27. The switchable optics 28 allow - obtain different magnifications without changing the focus.

The prism 29, manually switchable, allows to pass to the visual control. It is provided on only one of the micro-inspection stations but can equip them all.

The marks 30 and 31 designate switchable filters, respectively colored and neutral with different densities, to allow control in different colors and rough adaptation of the signal level to the characteristics of the sensor.

Reference 32 designates the Dove prism, the axis of rotation of which coincides with the optical axis of the column.

The optical column that has just been described makes it possible to inspect a chip successively in bright field lighting, in dark field lighting, in mixed bright field / dark field lighting, in fluorescence, and in different colors.

It is also possible, in a more complicated and more expensive variant, to carry out the checks of simultaneously, by providing that after the objective, on the path of the radiation emitted by the object, beam splitters equipped with filters select the beams of different colors and send them to separate receivers (video camera). Brightfield and mixedfield lighting can be created from: sources of different colors. Image processing is done according to the same principles "as above. The marking station is designed to deposit, on electrical order, an ink stain with a diameter of about 0.5 mm on the chips Recognized as defective. The positional accuracy of the spot must itself be of the order of 0.5 mm. The table may be of the same type as that used at the macro-vision station, and the uncertainties arising from the transport from the microinspection station are too small for it to be necessary to make a new positioning by image processing. The table allows displacements in X and Y, under the action of the control signals, to place under the inking device, which is fixed, the defective ^• chips whose position on the wafer was recorded during the ^" macro and microvision steps. The construction of the marking station uses means and techniques known or within the reach of the skilled person. It is therefore not necessary to extend it further. One can also provide for lais¬ ser the fixed edge and to move the inking device. In Figure 1, the marking station is shown associated with the macrovision station. In fact, the table is common to these two positions. Indeed, the times for the passage of a manager at the macrovision station are short compared to the times for the passage at the microinspection station, and even with four microvision stations for a single macro station, it is not occupied full time. As the performances requested at the table of the inking station are not superior to those of the table of the macrovision station, it is advantageous to use the same table for these two stations.

The anchoring device itself is placed a little away from the axis of the macro vision column. It is enough that the amplitude of the horizontal displacements of the common table according to X or Y are sufficient for a frame can be treated in one or the other station.

Of course, if the number of micro-inspection stations were increased, it might be more advantageous to separate the two stations completely.

The functions of the entire handling are as follows:

. extraction of a frame from a basket located at the receiving-shipping station ;. . transfer of this framework to the macro-vision station;

. transfer of the framework from the macro-vision station to a selected micro-inspection station; -

. transfer of the frame to the inking station; . transfer of the frame to the receiving and shipping station and insertion of this frame into the appropriate basket.

In the embodiment described, the receiving-shipping station is static, that is to say it does not move the frames and the baskets, and the macro, microinspection and inking stations do not provide only displacements of very limited amplitude. All handling is operated by a single device, called "robot-server". It would obviously be possible to provide a different structure, with, for example, separate means for extracting and inserting a frame arranged at the receiving-shipping station.

The robot server is provided with an arm which carries gripping members by means of a wrist which can be stiffened by electromagnetic locking. The first of these gripping members is constituted by a clamp intended to penetrate into a basket and grasp the desired frame without hitting the basket itself, which carries superimposed frames with a pitch of 6.35 mm in the commonly used models. The robot-server can apply vertical displacements to the clamp to lift a frame from its support in the basket or table of an inspection post, and horizontal movements of translation in two orthogonal directions and of rotation around d 'a vertical axis, in order to serve the various stations of the installation. If these stations were arranged in line, rotational movements would not be necessary, and if the stations were arranged in a circle, one could be content with rotational movements and radial translation.

The imprecision due to the movements of the robot-server requires a summary positioning of the frames before their tightening by the clamping block of the macrovision station. For this, the clamping block carries two vertical cylindrical pins, and the robot waiter comes to apply the frame against these two pins so that they penetrate one into the V-notch of the frame, the other into the notch. straight. Corresponding joints of the robot server have been relaxed beforehand. The frame is held in abutment on the rods until it is completely immobilized by the holding means on the table. After a prepositioning has been done at the macro-vision station, it should be kept during the transfer to the micro-inspection station. For this, the robot-server is equipped with a suction gripper capable of lifting the frame above a clamping block and supporting a frame during transport. A reference frame X, Y, Z is integral with the vacuum gripper, and associated reference frames X, Y, Z are provided at the macro-vision station and at all the micro-inspection stations. In practice, the repository

X, Y, Z, integral with the suction cup gripper, is formed by the clamp, and the frames of reference X, Y, Z asso¬ ciated by a relief of corresponding shape. The sequence of operations during the transfer is as follows:

. positioning of the server in front of the macrovision station;

. orientation of the empty gripper towards the macrovision station;

. lowering the gripper to the charge level;

. gripper advance; . wrist relaxation; . taking of the local reference frame X, Y, Z by the clamp, which causes the application of the suction cups on the frame;

. grip of the frame by suction cups; . unclamping the frame; . retraction of the gripper, which results in the lifting of the frame and the loss of the local reference frame; . stiffening of the wrist;

. receipt], complete with the chip;

. rise of the gripper;

. moving the robot server to the microinspection station; . deposition of the frame by a sequence analogous to that just described.

We will now describe the real-time image processing system first in terms of hardware, then software.

The equipment consists of 2 parts (see Figure 6):. a general microcomputer; . a specific image part. The microcomputer implements standard cards designed around the INTEL 40 multibus:

. a processor card 41 controlled by the microprocessor 42 of the "80286" type from INTEL associated with its mathematical coprocessor 43 of the "80287" type, also from INTEL;

. a memory card RAM 44 extension of 0.5 megabyte.

The mass memory is made up of 2 disk drives making it possible to write and read: * ". Removable flexible disks 46;

. a fixed hard drive of 200 megabytes 46. The specific image part is made up of 49, 50, 50 bis and 50 ter cards, which are IMAGING TECHNOLOGY, AP512 and FB512, ALU512 and HF512 standards. The 2 image cards are connected to the microcomputer by the MULTIBUS 40 at the back of the basket. The processor controls them via their command and status registers.

The analog video signal delivered by the shooting TV camera 20 is converted into digital data representing 256 gray levels. The image thus scanned in the form of a 512 × 512 dot matrix is stored in random access memory on the FB512 card. This image in memory can be viewed permanently on a TV monitor using the card AP512. In fact, this card converts the digital signal delivered by the FB512 cards. The flow of digital data exchanged between the two transit cards by a very fast specialized bus. The software therefore processes images of

512 x 512 dots coded on 256 gray levels: it is possible to carry out all the current logical and arithmetic operations between any 2 of the image memories. All software can be grouped into 3 families:

. "computer" utilities; . basic image programs; . specific inspection software.

Computer utilities allow the operator to: _

. create its programs; ^

. store data on the hard drive or floppy disk;

. print the results: alphanumeric- or ^" white and black image.

The basic image programs concern:. the acquisition of a current image with a possibility of horizontal and / or vertical zoom of coefficient 1/2 or 2;

. transfers of an image to any one of the image memory planes;

. calculations on the gray levels of an image: cut along a video line, global histogram, spatial signature;

. various artifices at the visualization level;

. certain facilities for graphic plots and protection of pixels in writing. Specific inspection software enables:

. binarize an image;

. extract the cutting paths or saw cuts from the current image;

. isolate ink stains from the electrical test and significant sawing faults;

. generate the remarkable lines of the chip; . automatically crop the current image to its reference;

. detect and extract the anomalies present on the image in bright background and / or black background and (or) colorimetry; . filter the "peppercorns" from the metal areas;

. acquire a direct map of the metal tracks; ^~ "

. judge the severity of the defects. Apart from the various electronic image processing units which have been described above, the command and control means comprise means for intervention by the human operator, that is to say essentially the control keyboard. 48, the TV camera 51 and an eyepiece on at least one optical micro-inspection column as has been said above.

Human operator interventions are routine operations on the one hand: identification of a batch of frames or a frame or section

(this operation can also be computerized by affixing coding on the frames or sections), intervention in the event of operating incidents or breakdowns, order to start and stop the installation , modification of certain elements during a change of characteristics of the sections or frames, mechanical and optical adjustments, which must be very careful due to the necessary precision. Two types of intervention are more specific: for the first, it is the choice of the reference chip. We could, in theory, use as a reference the layout of the chip as it was established by the server which designed it, but this way of doing things creates the obligation to take into account the distortions which result from the reproduction processes. and for forming the circuits of the chip. It is therefore preferable to take as a reference a chip deemed to be without defect, and which will have been chosen from a section submitted for examination. It is the data relating to this chip which will be stored and which will be used to control all the chips of the same type, unless a variation in the manufacturing conditions obliges to create a new reference following a relative change of the chip. If necessary, we can constitute a reference with several chips of which we only take the best part for each.

The second specific intervention by the operator relates to the choice of marking lines in the chosen chip, as explained above. These marking lines are also stored and are used until a variation in the manufacturing conditions also means that other marks have to be sought. As mentioned above, the chosen marking lines can be used on different chips, provided that they are found in the same place. 2S

As mentioned above, among the parameters of each fault which is taken into account when making a rejection decision, is the position of this fault. This point needs to be clarified. Any anomaly 5 which does not interest and does not threaten the metal areas can be eliminated as not constituting a defect. On the contrary, an anomaly on the metal parts or in their immediate environment is presumed to correspond to a defect likely to cause rejection in

10 according to the criteria of a "sorting specification book".

To determine if an anomaly interests or threatens the metallic areas, we start from a ^" binary image of the metallic areas, that is to say from an image in which each pixel can have two states:" white "or" black ". This image can correspond to the image of the masks used to manufacture the chip. However, we prefer to extract it from the image, obtained with" black background "lighting of the reference chip in the device

20 control itself. In this way, all the deformations inherent not only in the manufacturing process are eliminated, but also in the properties of the optics of the apparatus and in the diffraction phenomena due to the lighting mode.

The binary image of the metallic areas is used to create a binary image with "capture areas" by "expanding" the metallic areas. Then, we make a logical comparison between the binary image of the anomalies of each chip and this binary image with

• _ZQ "catching areas". An anomaly which is inside the metal zones with their sensor zones is a potential defect subject to the interpretation process.

Claims

1. Method for optical control of objects such as wafers carrying one or more integrated electronic circuits comprising the following steps: a) a digitized image of the object, or of the part of an object to be studied is recorded, it is that is to say an image formed from a set of digital signals each corresponding to an elementary surface (or pixel) of the object, defined by its coordinates, the digital value corresponding to the energy reflected and/or scattered by each pixel of the object, b) this digitized image is compared to a reference digitized image previously stored in memory, c) a differential binary image is created by matching a binary anomaly signal to the elementary surfaces where it does not there is no agreement between the recorded signal and the reference signal, d) the parameters (position, height, width, elongation, compactness) of each defect consisting of several elementary surfaces presenting an anomaly signal are determined, and e) a rejection decision is made when one of these parameters exceeds a limit fixed in advance or, in case of doubt, the image is recorded for a later decision characterized in that in step b), the comparison is made by making the difference between the energy levels of the corresponding pixels, which abou-. tit to a digitized differential image which is transformed in step c) into the differential binary image provided in this step.

2. Method for controlling integrated circuits according to claim 1, characterized in that the defects constituting "pepper grains" are not subjected to extraction for the determination of the parameters

5, that is to say being on a metallic or polycrystalline track, due to surface irregularities of this track and affecting a number of elementary surfaces less than a predetermined value.

3. Method according to claim 1 or 2, 10 characterized in that several images corresponding to different lighting conditions are recorded, and said images are compared not only to a reference image but also to each other.

4. Method according to one of claims 1 to 3,

^5 characterized in that a visible light image and a fluorescence image are recorded, and the defects visible on the two images and which correspond to dust are not subjected to the extraction and determination of parameters .

20 5. Method according to one of claims 1 to 4, characterized in that before recording the image, we record the image of position and orientation markers located on the object, we measure the gap and position and orientation between these marks and the

-^ corresponding marks existing on the reference image, and we accordingly move either the object mechanically, or one of the two images optically or electronically, to bring the recorded image into coincidence with the reference image.

30 6. Method according to claim 5, characterized in that the orientation differences between the two images are corrected by rotating a Dove prism around a coaxial axis in a light beam.

7. Method according to claim 5 or 6, characterized in that the position and orientation markers tion are constituted by remarkable lines of the object, with strong contrast and intersecting approximately at right angles.

8. Method according to one of claims 1 to 1, characterized in that, in a first station, the object is prepositioned, then it is transferred to a second station for control, the two stations being provided with locating means placed in a similar manner, and the means of transport being provided with associated locating means, and the locating means of the means of transport are brought into coincidence successively with the locating means of the two stations, so that the prepositioning is preserved after transport.

9. Method according to claim 8, characterized in that at least part of the defects are detected in the first station.

10. Method according to one of claims 1 to

9, characterized in that, if the object bears a mark, for example in ink, which was imposed during a previous check such as an electrical check under points to indicate that it is defective, the Anomaly detected corresponding to this mark results in a rejection decision without there being any extraction and determination of its parameters.

11. Method according to one of claims 1 to

10, characterized in that at least one image obtained with lighting combining a light background and a dark background in chosen proportions is recorded.

12. Method according to one of claims 1 to

11, characterized in that at least one image obtained with partially coherent light illumination is recorded.

13. Method according to one of claims 1 to 12, characterized in that the reference image highlighted memory beforehand is obtained by recording, under the same conditions as for the image of the object to be examined, the image of an object considered to be flawless by an operator, or an image formed by the juxtaposition of partial images each corresponding to a part of an object considered to be flawless.

1. Method according to one of claims 1 to 13 _, characterized in that, to take into account the position of each defect, metallic or pαly-crystalline zones of the object, a binary image "with capture zones" by dilation of said metallic or polycrystalline zones, and this image with capture zones is compared with the ^" binary image of the anomalies of the object to be studied, so as to treat the anomalies depending on whether they are found in a metallic or polycrystalline zone, enlarged by incorporating the capture zones, or outside these zones.

15. Method according to claim 14, characterized in that the binary reference image of the metallic or polycrystalline zones is obtained in the manner indicated in claim 13.