DE4003983C1 - Automated monitoring of space=shape data for mfg. semiconductors - compares image signals for defined illumination angle range with master signals, to determine defects - Google Patents

Abstract

The method involves illuminating the components and observing them using a (TV) camera whose signals are evaluated to detect manufacturing defects. The component is first illuminated over a reproducible angle range and a corresponding first set of image signals stored and compared with a further set. Space-shape data of structures of interest are derived from the image signal set differences.

Description

The invention relates to a method for automatic monitoring Chen spatial shape data according to the preamble of claim ches 1 or a device for performing the method according to the preamble of claim 17.

A major problem in the semiconductor industry is that the reliability of components manufactured by the manufacturer must be guaranteed. The reliability of the components depends - provided a correctly manufactured chip - in the first li never depend on the quality of the installation of the chip in the housing. This includes both the state in which the Chip is installed, as well as the location of the chip in the housing as well as the type and quality of the electrical connections between the chip and the housing connection contacts. For this reason the chip surfaces are checked for mechanical ones  Damage or contamination, the location of the chip in the Ge housing, the gluing points between chip and housing as well as the Bond wire connections between the chip and the housing final contacts. This inspection has so far been essentially exclusively by human personnel with the help of micro scope performed. On the one hand, this process is for the per very strenuous and costly for the entrepreneur, on the other hand, with today's high manufacturing ge only random checks of the Components possible.

EP 01 59 354 B1 describes the method and apparatus of the known type. The arrangement shown there However, it is extremely complicated to build and deliver so much wealth of data that needs to be processed that even when using a very fast calculator Samples from a production can be checked.

From DE-OS 24 31 931 it is known for the determination of Spatial shape data of semiconductor devices stored sets of Compare image signals with other sets of image signals chen. Otherwise, according to the teaching of this document Thickness of a measurement object by means of an automatable Measurement checked, but not very extensive usable information about the data mentioned above.

DE 38 06 209 A1 describes a structural defect detection system stem known, for example for an integrated half lead Switching is applicable. With this system one delivers Camera image output signals, which are processed in an image signal device detected and displayed on a monitor who the.

The invention is based, task and method to continue in the direction of the type mentioned at the beginning that that simple to identify the essential data errors in the manufacture of semiconductor components can be derived and checked.  

This object is achieved in the characterizing part of claim 1 or 17 specified features procedurally or vorrich resolved in terms of thinking.

An essential point of the invention is that only those image signals or pixels to which the Corresponding image signals are observed, which for the Review are relevant.

A common cause of errors is incorrect training of the bond wires. In particular, it can be relatively easy happen that bond wires are not bent correctly or too long are trained so that they can later be cast in the building partly protrude beyond the sealing compound. To verify this, you can get the semiconductor devices from a variety of Rich illuminate and with the camera fixed Win sets of image data. Here the lighting takes place from directions that are essentially in one plane with the The bond wires run. The image signals become the positions derived from light reflection points that arise on the bond wires. This is possible because the bond wires have an extremely smooth surface. Out the positions of the light reflection points, the lighting direction and the camera or its optical axis is the Pitch angle of the bond wires at the reflection points calculates. By integrating the pitch angle (via the Length of the bond wires starting at a point of known height) the course or the height of the bond wires in a  Direction calculated that is essentially perpendicular to the upper surface of the chips runs. The calculated height of everyone like this checked component is compared with a target range, so that a component can be discarded when the calculated height falls outside the range.

To save computing capacity and thus the speed of work to increase the speed, it is an advantage if you look at the course of Bond wires in the image plane of the camera knows before that described three-dimensional measurement is made. For this can be made up of two sets of image signals, which are at different Areas of illumination were obtained, those areas as traces of the course of bond wires in an X-Y plane (corresponding to the camera image plane) define that in both Set of image signals due to a shadow cast by the bond wires appear unlit. Of course the Angle between the two lighting angles as large or chosen small so that the shadows overlap.

It is also possible (if necessary additionally) to do this under Seeking semiconductor device coaxial to the optical axis of the Ka to illuminate mera, this optical axis essentially is oriented perpendicular to the surface of the chip. The Ver runs of the bond wires in the X-Y plane then appear darker and can be detected using a known tracking algorithm be communicated. This is because the chip surfaces Illumination with non-polarized light over the entire surface of the light reflect back very strongly directly into the camera while the bond wires most of the light in other rich reflections.

After such an image analysis step, only the that image data as belonging to light reflection points are classified, which are in the area of traces of bondver runs in the X-Y plane and their hel Stand out sufficiently from the environment. Thereby is not just the reduction of computers mentioned at the beginning capacity but also an increase in interference immunity Lich.  

After the reflection points especially in the area of large Radii of curvature appear elongated, it is advantageous if the calculated pitch angle only Priorities are attributed to a precisely defined To be able to win points for the derivation of the altitude course nen.

Since it makes sense to monitor a large number carries out similar semiconductor components, can be in ei a learning step (e.g. using visually checked components) the background image signals to be expected on average and save spatial shape data as learning data. Here it turns the position of the chips in the housing, for example Course of the glue points, their edges around the chips are visible and the surroundings of the bond wires in particular to the chip surface, to an underlying substrate and to the connection contacts of the bond wires. If the The semiconductor devices can be used for average data illuminate in one step from a direction that through a plane runs which is essentially perpendicular to egg ner stands through the bond wires plane, the Illumination direction runs obliquely to the chip surface. With this illumination, the bond wires cast a shadow on the underlying layers (chip, substrate). Out the course of the shadow cast by the bond wires and the learning The course of the bond wires can be in one direction can be expected to be essentially perpendicular to the surface the chips run, i.e. the height of the bond wires. These dates can alternatively or in addition to the Measured values described at the beginning with regard to the bond wire height be used.

If you essentially lower the optical axis of the camera right to the chip surface and lighting in the essentially coaxial with the optical axis of the camera with un makes polarized light, so reflects a flawless Chip essentially uniform over its entire surface away. You can now find flaws on the chip surfaces  by defining that the image sections or Bildsi gnale looks for which corresponds to a lower brightness chen than the surrounding areas. This is because both Contamination particles as well as scratches on the chip surface Scatter the incident light so that the camera the amount of light reflected at these points becomes smaller. If you now save the image data thus obtained and the Lighting in a next step from another win kel, who can be inferred from the difference image what kind of error it is because then the different scattering properties of contamination particles and scratches can be seen. Preferably done the lighting is approximately parallel to the surface of the Chips, so that the errors appearing bright in the difference image pose as impurities, the dark appearing errors Defi as injuries (scratches) on the chip surface nable.

The above-mentioned learning process, in which the spatial form Average data of the semiconductor components to be checked are determined, is suitable for rejecting faulty ones Components, that is, components whose spatial shape data is more as a predetermined amount from the average data give way. It is also, to some extent, a pro regulation production machines due to the error data found Lich. For example, the machine can be corrected be applied, which sticks the chips on the substrate, because for example, an oversized adhesive edge all around the chip an excessive dosage of glue can be concluded. Also from ver run of the bond wires can control data for the bond machine be won. In any case, the data obtained used to separate out incorrectly produced components will.  

A rough check using a learning process can be done be carried out that at least one correctly manufactured Semiconductor component, but better a group of such, under a defined lighting angle or lighting angle area is illuminated and recorded. Read from the image data then reflex areas can be defined, which are correct if ge finished components are to be assigned to the bond wires. Switches then the image data of a semiconductor component to be examined from this sample image data, it lies with a high probability an incorrectly manufactured component. So z. B. would be when the light reflection area of a bond wire is small ner than normal, from excessive curvature of the bond wire going out. So this version of the Procedure for checking characteristic reflection mu which are derived from correctly manufactured components.

Due to the simplified nature of the pattern recognition process a significant increase in speed at the over testing while reducing the necessary rakes and achieve storage capacity. The procedure can also be advantageous with the more precise verification procedure, at wel chem exact statements about the spatial shape data are obtained, combine. So z. B. you can do the more time consuming, accurate Procedure only for every nth component (e.g. every tenth building part) while performing the simpler procedure on everyone Component is carried out. This is a minimum check guaranteed and at the same time the probability high that a systematic error (through the more precise procedure) is recognized, the z. B. by incorrect work a machine slowly creeps in even if the bugs for the simpler procedure within the permissible range ches lie.

If you work with a color camera and different Be lighting directions different colors or different Allocates spectral distributions, so from the color signal additional data can be obtained from the camera. In particular, can in this case, simultaneous lighting from several,  possibly all directions of interest at the same time done. Each reflection point can then namely the color signal of the corresponding image data determined accordingly Illumination directions can be assigned. So here is done no control of the light sources, rather all Information taken from the image data. At this point still noticed that under a "set of image data" not only a variety of gray values or color values can be understood, but each individual pixel still more information, e.g. B. Data on the (statistical) weighting of the individual pixel etc. can be added.

Calibration or, if necessary, calibration can be carried out by performing one in one process step Ball of known size with a reflective surface ver subject to driving. Because every point of the ball is a known one Surface angle or a certain radius of curvature order, calibration or calibration data can be obtained and ge are saved, based on which in the subsequent building le examinations of the obtained values or image data in Ab real values are transferable.  

Single lighting is suitable as lighting sources tion sources, such as LEDs or the like, or others but also fiber optic lighting, where that Light at one end of the light guide (which in different positions are mounted) and the light goes out a single lighting source in the other ends of the Light guide is introduced. To do this controlled (sequentially) aperture devices are suitable for accomplishing this.

It is also possible to provide mirror systems for lighting watch. A cylinder ellip is suitable for this, for example mirror, in one focal axis of which is to be examined Component and directed in its other focal axis  radiant linear light source (cylindrical light source le with coaxial, rotatable slit diaphragm) is arranged. So it is essentially a matter of reproducing bare lighting can be done from different angles.

In a further embodiment of the invention, how indicated at the beginning using the corresponding method, a Lighting device used in which the various which lighting directions are assigned different colors are. This can be done using suitable color filters a light source that is at least that of Color filter passed significant spectral components sends. Such a color filter can, for example, consist of a Slide positive film can be made with a standardized Color spectrum (from blue to red) was photographed. If one around a spherical section-shaped lighting room wants to have the component to be examined, Conducting the bond wire height curve all locations on the ball section-shaped color filters that have the same height (in the optical axis of the camera) have the same color to have.

Special embodiments of the invention result from the Subclaims. In the description below are preferred te embodiments of the invention with reference to images gene explained in more detail. Here show:

Fig. 1 shows an embodiment of the invention in a schematic representation.

FIG. 2 shows an embodiment of a lighting light-guide;

Fig. 3 AC schematic image extracts bonding wire localization;

Fig. 4 is a partial perspective illustration of a chip having bonding wire;

Fig. 5 is a schematic diagram for explaining the bond wire Ver run measurement;

Fig. 6 A, B further illustrations for explaining bond wire courses;

Fig. 7 shows another embodiment of the invention in a schematic representation similar to that of FIG. 1; and

Fig. 8 is a partial perspective view taken along the line VIII-VIII of Fig. 7.

In Fig. 1, an embodiment of a device for monitoring spatial shape data in the manufacture of semiconductor components is shown schematically. This includes a holding tion 22 , to which a plurality of individual light sources 16 a to 16 n is attached. The individual light sources 16 a- 16 n are preferably arranged at equal angular distances from one another and directed to a common center. A holder (not shown) is provided under the holder 22 with the light sources 16 a- 16 n, on which a semiconductor component to be examined can be positioned. The semiconductor device is indicated in the drawing by the schematic representation of a chip 10 , the connection points of which are connected via bonding wires 12 to connection contacts 13 of a housing (not shown).

Above the semiconductor component, a (CCD) camera 14 is held so that the optical axis O of its lens 25 is essentially perpendicular to the surface 11 of the chip 10 .

Behind the lens 25 of the camera 14 , a beam splitter 24 is mounted so that a light source 23 arranged next to the camera 14 can illuminate the semiconductor component coaxially to the optical axis O.

All lighting sources 16 a- 16 n and 23 are in a controlled connection with a processing device 17 which also supplies the image output signals of the camera 14 . With this arrangement, it is possible to illuminate the semiconductor component to be examined successively from different directions in accordance with the light sources controlled by the processing device 17 and to record the image signals generated in the camera 14 for further processing.

Instead of the large number of individual lighting sources 16 a- 16 n, it is possible to fasten light guides with first ends in the holder 22 , in the other ends of which light is emitted. A suitable arrangement is shown (in principle ler representation) in Fig. 2. The ends of the light guides L 1- L 5 shown there are fastened in a holder (not shown) so that they lie on a circle (section). A light source 21 is provided, which emits light in the direction of the ends of the light guides L 1- L 5 . A disk-shaped diaphragm 19 is provided between the light source 21 and the end faces of the light guides and can be rotated by means of a (step) motor 20 . In the aperture there is an opening which is designed such that the light emitted by the light source 21, depending on the position of the aperture 19, can only ever fall onto the end face of one of the light guides L 1- L 5 , depending on the position of the pinhole 19th The circle (section) on which the ends of the light guides are positioned corresponds, of course, to the path of movement of the hole in the aperture 19 .

To explain the description of the method according to the invention made at the outset, the attached FIG . This shows a schematic representation of a semiconductor component, in which a chip 10 is mounted (glued) on a substrate 15 . Terminal contacts on the surface 11 of the chip 10 are connected via bonding wires 12 to terminal contacts 13 , which are connected to contact pins (not shown) protruding outwards (out of the housing). Each of the bond wire 12 is guided as a result of the known bonding method arcuate between the corresponding connection point on the chip 10 and the contact 13 , so that the bond wire 12 runs essentially in a plane A ver, which is substantially perpendicular to the surface 11th of the chip 10 is. The surface 11 runs in an xy plane, the bond wires thus extend in a direction Z upward beyond the surface 11 of the chip 10 .

If one now illuminates such a bonding wire 12 from a certain angle, which is assumed in FIG. 5 with approximately 10 ° to the surface 11 of the chip 10 , then, as shown in FIG. 5, only a small amount becomes due to the curvature of the bonding wire 12 , A portion of the incident light corresponding to a small surface section of the bonding wire 12 is reflected in (the lens of) the camera 14 , the remaining portions of the light are emitted in other directions. Since the roughness of the bond wire upper flat is low, only small stray light components from other bond wire sections get into the camera. Since the angle of incidence is equal to the angle of incidence of the light rays and the relative positions of the respective illuminating light source to the camera 14 are known, the angle can be determined at which the surface section of the bonding wire 12 that reflects the light into the camera 14 extends. If you now let one light source 16 a- 16 n after the other illuminate the bond wires 12 , then different points on the bond wires 12 can be determined one after the other with regard to their angle to the surface 11 of the chip 10 , which means that the angles or the associated locations on the bond wires 12 whose course in the Z direction can be calculated.

This state of affairs is again explained in FIGS . 6A and 6B, FIG. 6A showing a correct bond wire course and FIG. 6B showing an incorrect bond wire course. From these figures it is also apparent that the areas in which the bonding wires reflect light into the camera have different lengths depending on the radius of curvature, which in turn can be used to derive data that can be used for the quality check. In particular, it is possible, using a learning method on correctly manufactured semiconductor components, to obtain sufficiently reliable data about the assessment of the reflex zones at different locations on the bond wires as to whether the bond wires have a correct course ( FIG. 6A) or an incorrect course ( FIG . 6b).

In order to be able to find the reflex points on the bond wires more easily from the image data, it is advantageous if the courses of the bond wires in the xy plane are known. To find out these courses, you can illuminate the bond wires from two different angles, which are preferably arranged symmetrically around the optical axis O of the camera. The two resulting images are shown schematically in FIGS. 3A and 3B. This then results in a brightness pattern 26 for the background, a brightness image 28 for the area in which the respective bonding wire 12 reflects, and a brightness pattern 27 , which corresponds to a shadow that the respective bonding wire casts on the background, for each of the images . By forming a difference between the images of Fig. 3A and Fig. 3B, the image of such 3C results of FIG. Wherein the (dark) shadow area 27, which in the pictures of Fig. 3A and 3B at the same location (in the xy-plane), contrasts with the (lighter) background. The reflection zones (in the xy plane) which are used to determine the height (see FIGS. 5 and 6) must then lie in the region of this “shadow trace” of the bonding wires. Data reduction can thus be achieved to simplify and accelerate information processing. Of course, it is also possible, if the bond wire courses in the xy plane deviate very far from a target course, to reject the associated semiconductor components as defective.

A further preferred embodiment is described in more detail below with reference to FIGS . 7 and 8.

In this embodiment, instead of a multiplicity of light sources 16 a- 16 n, an illumination unit consisting of a plurality of white light sources 31 is provided, the light of which is transmitted through a color filter 32 attached in a holder 22 to the chip 10 to be examined. The color filter 32 is designed so that each lighting direction is assigned a defined color. For example, the color filter 32 in FIGS. 7 and 8 can run through the color spectrum from blue to red from bottom to top. The only thing that matters is that each direction from which the chip 10 is illuminated can be assigned a specific color.

The camera 14 is designed as a color camera, so that a signal (analog or digital) can be obtained from its output signal via a color signal converter 30 , which assigns a color value to each pixel. This color value then supplied to the processing device 17 corresponds to a certain direction of illumination.

The arrangement according to FIG. 7 can also be seen as a section through a hollow spherical lighting device (with camera), in which case the color filter is preferably formed axially symmetrical to the optical axis O of the camera 14 . In this embodiment of the invention, a simplification can be achieved insofar as no separate control of different light sources is required.

Claims (24)

1. A method for automatic monitoring of spatial shape data in the production of semiconductor components, in which the semiconductor components are illuminated by means of an illumination device ( 16 ) and observed via a camera ( 14 ), the image output signals of an image signal processing device ( 17 ) for recognizing Manufacturing errors can be fed, the semiconductor component to be examined being illuminated under at least a first reproducible illumination angle range, at least
a set of image signals corresponding to the illuminated semiconductor is stored which
stored sets of image signals are compared with another stored set of image signals and
from the differences in the sets of image signals spatial shape data of structures of interest are derived, characterized in that the illumination of the semiconductor components takes place from a multitude of directions when the camera is stationary and corresponding sets of image data are obtained therefrom. 2. The method according to claim 1, characterized, that the further stored set of image signals one Set of pattern image signals corresponding to that of a or an averaged variety of correctly made ones Semiconductor components was won. 3. The method according to claim 2, characterized, that the semiconductor component to be examined is then discarded becomes when a set of image signals in position and / or Size of light reflection spots around a predetermined one Amount (spread area) of the set of pattern image signals deviates. 4. The method according to claim 3, characterized in
that to monitor bond wire connections ( 12 ) with circuit contacts ( 13 ) on semiconductor chips ( 10 ) the lighting device from directions that is essentially in one plane (A) with the courses of the bond wire ( 12 ) that the position of on the light reflection points resulting from the bonding wires ( 12 ) are derived from the image signals,
that the positions of the light reflection points, the direction of illumination and the camera are used to calculate the angle of inclination of the bond wires at the reflection points; and
that the course of the bond wires in a direction (Z) is calculated by integrating the pitch angle, which is essentially perpendicular to the surface of the chips. 5. The method according to claim 4, characterized in that from two sets of image signals, which are obtained at various lighting angles, those areas are defined as traces of the courses of bond wires ( 12 ) in a ner xy plane, which are defined in both sets of image signals due to a shadow cast by the bond wires ( 12 ) appear unlit. 6. The method according to claim 4, characterized in that in one step the semiconductor component to be examined is illuminated coaxially to the optical axis (O) of the camera ( 14 ), which is substantially perpendicular to the surface ( 11 ) of the chips ( 10 ) is aligned and the courses of bond wires ( 12 ) in an xy plane are determined by means of an edge tracking algorithm of dark linear image signals. 7. The method according to any one of claims 5 or 6, characterized, that only those image data as belonging to Licht-Refle xion points are classified, which are in the area of the Spu ren of the bond wire courses lie in the x-y plane and differ from the surroundings in terms of their brightness values take off sufficiently. 8. The method according to claim 4, characterized, that to the coherent image signals essentially surfaces of the light Points of reflection are calculated and the calculated pitch angle assigned to these points  the. 9. The method according to any one of the preceding claims, characterized in that when monitoring a plurality of similar brightness ranges in one learning step, the expected background image signals and spatial shape data are stored as learning data, which encompasses the surrounding area of bonding wires ( 12 ) Chip surface ( 11 ), an underlying substrate and connection contacts ( 13 ) for the bond wires ( 12 ) belong. 10. The method according to claim 9, characterized in that the semiconductor components are illuminated in one step from a direction that runs through a plane (B), which is perpendicular to a through the bond wires ( 12 ) extending plane (A), wherein the direction of illumination extends obliquely to the chip surface ( 11 ), and that the course of the bond wires ( 12 ) in a direction (Z) is calculated from the course of a shadow cast by the bond wires ( 12 ) and the learning data, said direction being substantially perpendicular to the surface ( 11 ) of the chips ( 10 ). 11. The method according to claim 1, characterized in
that the illumination of semiconductor chips ( 10 ) takes place essentially perpendicular to the chip surface ( 11 ) coaxially to the optical axis (O) of the camera ( 14 ) with unpolarized light and
that those points on the surface ( 11 ) of a chip ( 10 ) are defined as flaws in a step, the associated image signals of which present a lower brightness re than surrounding areas. 12. The method according to claim 11, characterized in that the lighting takes place in a next step at an acute angle approximately parallel to the surface ( 11 ) of the chip ( 10 ), a difference image from the previously won set of image data and the current image data is generated and the defects that appear bright in the difference image are defined as impurities that appear dark and the defects are defined as injuries to the chip surface ( 11 ). 13. The method according to any one of the preceding claims, characterized in that in a learning process, the average of the image signals corresponding to the relative position of the chips ( 10 ) to Ge housing and / or the courses of Be covered with adhesive range around the edges of the chips ( 10 ) are determined and then an error signal is generated when the currently obtained image signals deviate from the mean values of the image signals by predetermined values. 14. The method according to any one of the preceding claims, characterized, that the spectrum of illuminating light is dependent is set by the angle of illumination, and that from a Color signal of the camera belonging to individual image signals Illumination angle can be derived. 15. The method according to claim 14, characterized, that the lighting at the same time under different loading lighting angles is performed. 16. The method according to any one of the preceding claims, characterized, that in a calibration step a ball of known size subjected to the process and from the image signals Kali Brier values are obtained and saved. 17. Device for carrying out the method according to claims 1 to 16, with an illumination device ( 16 ) for illuminating the semiconductor components ( 10 ), a camera ( 14 ) for image recording and an image signal processing device ( 17 ) for detecting manufacturing defects of the Semiconductor component ( 10 ), characterized in that the lighting device ( 16 ) is designed and arranged to a semiconductor component to be investigated in such a way that it can be illuminated with at least two different reproducible Win angles when the camera is stationary. 18. The apparatus according to claim 17, characterized in that the lighting device ( 16 ) comprises a plurality of individual lighting sources ( 16 a- 16 n). 19. The apparatus according to claim 18, characterized in that all individual lighting sources ( 16 a- 16 n) are essentially the same distance from a point in which chem the semiconductor component to be examined can be positioned. 20. The apparatus according to claim 18 or 19, characterized in that the individual lighting sources ( 16 a- 16 n) comprise light emitting diodes. 21. The apparatus according to claim 18 or 19, characterized in that the individual lighting sources ( 16 a- 16 n) comprise first ends of light guides ( 18 ), in the second ends successively in succession through diaphragm devices ( 19 ) light from a single light source ( 21 ) can be introduced. 22. The apparatus according to claim 17, characterized in that the lighting device ( 16 ) comprises a zylinderringför shaped or spherical segment-shaped light emitting surface ( 29 ) which is designed such that different locations of the light emitting surface are assigned different light spectra, and that the camera ( 14 ) for Generation of image signals is formed, which include color signals. 23. The apparatus according to claim 22, characterized by a device ( 30 ) which converts a chrominance signal from a color television camera into a signal, the different values of which correspond to the different locations of the light-emitting surface ( 29 ) or different illumination angles. 24. Device according to one of claims 22 or 23, characterized in that the lighting device ( 16 ) comprises a broadband light source ( 31 ) and a color filter ( 32 ) with changing colors or Lichtdurchlaßberei Chen over the surface.