JPH04142691A - Automatic photoelectric image count system - Google Patents

Abstract

PURPOSE: To speedily and exactly count the number of dice by providing a personal computer or microprocessor, light source beam projector, scanner, video signal processing and counting device, printer and monitor. CONSTITUTION: This system is provided with a light projector 2 for irradiating objects to be counted with uniform light, photoelectric linearly arranged device 3 for signal processing a two-dimensional video by executing mechanical scanning from one side to the object, circuit for converting the electric video signals of the object to binary digital data, and a personal computer 1 having a suitable peripheral equipment for storing the digital data and the count value of the objects. Thus, a large number of fine dice of semiconductors distributed on one plane can be speedily and exactly counted.

Description

DETAILED DESCRIPTION OF THE INVENTION Background of the Invention After a semiconductor factory completes the fabrication of a wafer (vife, r), the wafer is tilted and cut into dice (d
The wafer is finished into wafers (ice), and then separated to detect inferior products, but it is difficult to determine the number of dice on the final wafer, and a new and accurate count must be performed. Otherwise, it will be difficult to take inventory of the flow of goods within the factory, inventory during market transactions, and acceptance inspection at the time of sale. Currently, figures like the second one are estimated visually at the factory. The precision of artificial counting is unreliable (and extremely time consuming), as the dice no longer form a regular array after being expanded and separated on a viscous plastic film. Diode (LED)
Since such individual elements are small, large in quantity, and low in price, if artificial counting were used, the cost would further increase and profits would decline. Therefore, it is urgent for the five police departments to design an automated counting system to save manpower.

We previously designed a system for this problem and attempted to improve the above drawbacks. The structure was photographed using a television camera (CCTV) and
The video data was written to the memory of the host computer using the ``+7 Access'' method. However, general CCT
Due to the limited resolution of V, its function could only count the number of dice on the child side (+lice). If the number were larger, the resolution would be insufficient and accurate counting would not be possible. The current die manufacturing process has been greatly improved, and the die grain on one wafer is reduced to 1
．． The number of units exceeded 20,000, and the above system became unusable.

After we designed the system, a Japanese company also developed a similar counting system. Its optics and hardware adopted a high resolution pccTv of 800 x 1048, so the captured images were better and could be used to calculate a larger number of dice. However, the high-resolution optical and hardware costs were quite high, and the selling price of the system was more than 100 Taiwan dollars.

SUMMARY OF THE INVENTION The present invention solves the above problems by using a linear CCD scanner and adding a stable brightness light projection structure and circuit with negative feedback to it for scanning. The purpose of counting is achieved quickly and accurately by taking images and processing them using software. The structure of this scanning camera is similar to a general video scanner, so the structure was modified to complete the combination and the cost was very low, but it cost only 100,000 Taiwan dollars. Finished,
It is only a tenth of the above CCTV combination device and has sufficient economical and practical effects.

Example 1. System principle FIG. 1 is a system structure diagram according to the present invention.

The system mainly includes a personal computer or microprocessor 1, a light source light projection 2, a scanner 3, a video signal processing and counting device 4, a printer 5 and a monitor 6. Its structure is as shown in FIG. The present invention further provides one permanent light source (
A light source with stable brightness) includes a feedback control circuit 21 (shown in FIG. 2) and a diffuser plate 22, which control the brightness of light emitted from the light source and serve as the illumination equipment 2 of the scanner. The scanner block includes imaging equipment (not shown in the figure) and image signal processing (not shown in the figure), and the main part of the scanner is a linear charge coupling device (Iiner).
r chxrge coupleB device
(CCD), captures images of the dice that are distributed, and inputs the obtained digital image data to the computer block 1 in the figure using the DMA method. The computer block 1 and the counting processing block 4 are concerned with software counting processing. The monitor block 6 displays the obtained image and also displays the counting result after the counting is completed. The printer block 5 prints the images and counting results input into the computer.

The present invention relates to a system for automatically counting the number of dice distributed over a wide area, and the operating conditions will be explained from the following four aspects.

A. Principle of Operation Figure 1 is an explanatory diagram of the system structure. First, one variable resistor 2 on the permanent light source feedback control circuit shown in FIG.
18 to lock the light source to the optimum image brightness level, the light is further transmitted through one diffuser plate 22 to produce a uniform light source, and projected from above onto the sample 23 awaiting measurement. .

After that, a linear CCD scanner 3 performs one-dimensional mechanical scanning on the underside of the sample to capture a two-dimensional image of the object (g
rabbing). The captured video data is then converted from analog to digital signals and input into the computer's memory for software processing. On the one hand, the images are scanned and on the other hand, the data is input into the computer using DMA method, and the data is transferred to one gray scale level (gre75ctle lew) before further storage.
el) and converts it into a 1-bit signal of second class gray scale. In order to carry out such processing, the speed and memory capacity required for image data processing are saved, which simplifies the software processing work.The software program monitors the image obtained by scanning. This is convenient for checking the quality of the obtained image.The printer can print out the image in its original size and save it as a file for inspection.

After confirming that the video quality meets the requirements, all separated dics are counted using a special law counting method.
e) count the number quickly and accurately and display and store the results; Statistical corrections can then be made to these numbers to determine the number of overlapping dice in a small number of images to achieve a more accurate count.

Optical projection design is originally optimal if parallel light rays can be used, but one size of parallel light source is sufficient.
It is not easy to design a 5×15an square light source, and it is also uneconomical. Therefore, the design of the optical projection of the system is a very important technique from the perspective of the present invention, and only a good and uniform projection will produce a valid image, and the counting results will also be accurate.

As shown in FIG. 1, there is a diffuser piece below the light source 2, which produces a uniform surface light beam and dice 3.
to project onto. This projection method can increase the sensitivity of the object image, reduce background noise, and avoid the occurrence of continuous dice images, and the appearance of non-separation may cause the aliasing phenomenon. This results in a counting error.

The light projection structure of the system utilizes one incandescent light source 2 (lamp R, shown in FIG. 2). The brightness can be completed with a specially designed permanent light source feedback control circuit 21 (as shown in FIG. 2), resulting in a stable light source brightness. As the lamp ages or power supply fluctuations cause a change in brightness, the circuit can automatically adjust through negative feedback to maintain the originally set brightness level.

There are many ways to generate a stable brightness feedback control circuit. FIG. 2 is a permanent light source feedback control circuit designed by the present invention.

This circuit uses one AC power supply.
pplr) 211, 4 diodes D 1D
.

It includes a bridge current transistor 212 consisting of an inverter D3, a single-coupled transistor Q213, a transistor Q3215, a power transistor Q4216, an amplifier 217, and the like.

The operation of this circuit is one UJT.

(uni-junction Tr*n5ijo+) makes one SC
R, Q2214 (Si1icon-eoirolled
-Rectifier) and phase angle (phas
Triggering the e-tngl provides a way to adjust and lock the light. In this circuit, when the power supply is in the positive half cycle, diode D and diode D3 are sequentially biased, and when the SCR conducts, the supply voltage changes from 1-2-3-^- to -4-5 -6-7 and joins Incandescent Light RL2. When the power supply is in a negative half cycle,
Diode D3 and diode D4 are biased in the forward f direction.

If SCR214 is conductive, the power supply is also 7-6-3
-^- is supplied to the load via -4-5-2-1.

Feedback control of the SCR trigger is shown below. Feedback diode D, feedback electric resistance R, and operational amplifier:
'. Current - Voltage (1)
11111 hundredier.

Configure TIA). D, produces a corresponding short circuit current after being exposed to the light of the incandescent lamp RL2, and a voltage VE on the counter electrode E of the transistor Q4216.
is locked, so the current I E = VE on the counter electrode resistor R2
/R2. This drain is a constant thickness current 1 provided by the scratched JFET transistor Q3215.
dss affects the charging time on capacitor CT 219. When charging reaches the peak voltage Vp of UJT, UJ
When T is turned on (ON), one pulse current is generated and flows from the CT terminal to R4.

As a result, the R4 at the BlfjI terminal generates one pulse voltage, which is applied to the gate G of the 5CR 214, causing the 5CR 214 to conduct. The load (brightness) of the incandescent lamp 2 is adjusted by the variable resistor 218, and when the impedance R of the variable resistor 218 is high, V b2bl decreases, and from the relational expression V = ηV b2bl + 0.6V/T It can also be seen that it decreases. Therefore, the charging time of CT becomes shorter, and LIJT213 and SC
All R214 become conductive quickly. The lamp then has a relatively high average load current and brightness.

Conversely, if IR218 is low, charging will be relatively long, conduction will be relatively slow, and therefore a relatively low average load current will be obtained, and the brightness will be relatively low, resulting in an overall For example, the variable resistor VR21B can be used as a brightness adjustment function.

When the brightness of an incandescent lamp fluctuates due to fluctuations in power supply voltage or various external factors such as aging, the feedback photo-induced current IP changes relatively accordingly. If the locked brightness is exceeded, I increases, the current flowing through R also increases accordingly, VE increases relatively, and IE and IC also increase accordingly, so the voltage at VT increases. However, l1ii decreases, the charging time and triggering time of capacitor CT219 are thereby delayed, and the average current, power and brightness of the load are therefore reduced. Conversely, if the lamp becomes dimmer than the set point, it will trigger earlier to increase the load power and brightness. Therefore V
After adjusting R218 and fixing the Re value, the lamp brightness is locked to a suitable level recognized as the optimum image, resulting in one stable luminous intensity feedback control. The DC power supply required for the trigger and feedback control circuit is
One limiting current resistor Rd and one reverse bypass Zener diode (2ener diode) D 5
227, one voltage drop can be reduced to 2 using series and voltage division.
Provide the above control circuit diagram with a 6V power supply.

B. Includes scanner imaging ÷ - twirling element (not shown). When the light source is projected onto the dice 33 from the top to the bottom, the transmitted light is specularly reflected and then passes through the lens to the object 1.
':+ The image is focused by φ on the linear coupling element. The video data passes through an A/D converter and is quickly written and stored in a memory using a digital method. What you need to be careful about here is that if the critical level of the black-white gray scale is too high, the image will be too black and will be confused with the image of the adjacent dice.On the other hand, if the critical level is too low, the dice The number of pixels in the image (dice) was too small to cause a phenomenon that could not be distinguished from noise points, which caused the counting to be inaccurate.Generally speaking, a good light projection results in a good image and a suitable critical value. greatly affects.

One of the advantages of the above-mentioned scanning structure is that it is possible to use the cheap image scanner (image 1 scanner) that is commonly used today, and it can be easily modified, thus significantly reducing the cost of the device. I can do it.

Spatial resolution of CCD 32 in the scanner (Resoluti
on) requires reaching 300D P I,
That is, each inch has 300 points. Since the size of the sample to be measured is approximately 4 inches, the resolution of 1200 x 1200 is better than the resolution of the IQOOX 800 of the CCTV tube used in the conventional system, and a larger number of 33 dice can be counted. In one dimension, the image is taken as one line each time, and the scanning speed for each line is 0.003 seconds, and if the finder is 4 inches x 4 inches, the scanning time will be 3.6 seconds. . This time is not critical to the software counting time (approximately 90 seconds).

Digitized video signals are binary images (bio*
r7 image), the data is just 0
and 1, and one pixel requires only one bit of storage. The image of 1 die (+1 ice) is 1
It is composed of groups of pixels. This group of pixels needs to be adjacent to each other. Therefore, when counting the number of dice (+1iee), it is only necessary to know how many groups of pixels are assembled.

Description of C1 Counting Principle The counting law is based on the one-send-back principle, and each coordinate position on a two-dimensional image plane corresponds to one pixel.

In addition to its peripheral boundary position, the object image can be searched in eight different directions at any other position. If we were to find the current location of A, its coordinates would be (i) (i).

That is, the relationship between A and its eight adjacent points is as shown in FIG. 4, and they are referred to as 0 adjacent points, 1 adjacent points to 7 adjacent points, respectively, in the clockwise direction. Pixel A is referred to as the search center for the other eight adjacent point pixels. In the process of searching and counting, every pixel on the image plane has a chance to become the searching center, which allows us to identify whether its neighboring point is one of the object's pixel set.

Since there are only two types of digital data in this system, 0 and 1, the digital image of an object is an image plane composed of 0 and 1. As illustrated in Figure 5a, a black point on a binary image plane is called a black pixel, and its digital value is 1.
1'. Set another blank black and white value of 10' to a white pixel (vhije
pixel).

The horizontal direction is called the X-axis, and the vertical direction is called the Y-axis. The coordinates of the upper left vertex in FIG. 5 are (0, Q).

The present invention starts the search operation from the position (0,0). In this figure, the largest sets of both black pixels (A, -A7) and (B1 -B4), whose pixels are adjacent to each other, represent two dice as shown in Figure 5b. Other examples (A
SA SA 1), (Bl.

B2) etc. are all connected sets and are not the largest connected set of black pixels, so one dice image cannot be completely displayed yet.

The counting method is stacking (+tacking) and trial and error method (lr
The number of dice is counted by performing a sending back operation using 7 and error).

The data structure of the deposition element is shown below.

5TACK ELEMENT Pixel direction: Recording passes through direction d when moving from position (i, i) to (g, h). The serial numbers of the directions are as shown in FIG. The number in parentheses corresponds to the number in the direction of movement of the search center.

Pixel address second lower search center (1) and vertical position (i)
Record.

Counting the number of dice (+1iee) starts from the upper left corner of this two-dimensional pinary image plane and searches for black pixels along the horizontal scanning line. When searching for the first black pixel, search for it and focus on it. Also, add a mark to it to make it the black pixel that has already been found. Since there are eight adjacent points at each search center, there are eight directions to choose from when searching. Select adjacent points in a clockwise direction. If you select that a certain adjacent point is a black pixel and do not attach a symbol, search for this newly searched black pixel as the center, add a symbol to the black pixel that was the center of the previous search, and then check if it is the previously searched black pixel. Represent a pixel and place it in the stack (+1xck). If the pixel is searched again after that, it will not consider it as a black pixel and will continue to search for the next adjacent point. If there are no black pixels among the eight neighboring points of a certain black pixel, or if the eight neighboring points are marked with black pixels, search back for the previous one search center from the stack, and then search for the next one neighboring point. Search for. By repeatedly searching in this way, when all the configurations of black pixels in which all the largest black pixels are connected and aggregated are found, that is, one complete dice (di
ce) is added to the counter when it is found. Once all pixels of the entire digital image plane have been searched, the counter records the total number of dice in the object plane.

D. Statistical analysis and correction If the image of each dice is digitized, it will affect the size on the binary image plane. If it is too dark, the appearance will be polymerized (alia+ing).
) The dice that cause the phenomenon and separate are connected to each other. If it shines too much, the dice may disappear. The plane size of the digital binary image of the dice also affects the computation time. The larger the size, the longer the processing time, and vice versa, the shorter the processing time, so how to determine the degree of illumination has considerable importance on the function of this device.

The three curves in FIG. 6 are the results of measurements under three different lighting conditions. Curve 3 is the brightest, 2 is next, 1 is even further, and the error is all 0.0.
Although it is less than 5%, the speed is different. As can be seen, locking the flashing feedback to an appropriate level has a significant impact on accuracy and speed.

On the other hand, when a wafer is cut into dice and separated into expanded quantities and then transported as a folding bag, sometimes a very small number of dice are connected to each other, so this also results in a situation where the images are connected to each other. Under this phenomenon, the number of pixels on the dice increases greatly and differs from the general average value. Therefore, it is possible to calculate the average value and standard deviation of the number of pixels, perform statistics, and make some corrections to obtain more accurate results. The attached Table 1 shows the data from the counting test of this system. As can be seen from the table, the errors are extremely small and the results are fully satisfactory.

The above-mentioned apparatus and counting rules for use are very convenient and can quickly and accurately calculate the die size of semiconductors of large quantity and minute size and placed on a flat surface. The device of the invention can also be applied analogously to other similar objects. Such a method is cheaper and more practical than the currently available structure, and is worthy of innovation.

The invention is not limited to the embodiments described and shown, but is intended to cover any variations thereof.

[Brief explanation of drawings]

Fig. 1 is a system structure diagram according to the present invention, Fig. 2 is a photoelectric feedback control circuit diagram designed according to the present invention, Fig. 3 is a structural diagram of the scanner in Fig. 4, and Fig. 4 is an arbitrary pixel in the counting law and its The coordinate relationship diagram of eight adjacent pixels, the first side and Figure 5b are layout diagrams of dice, which are the two largest connected black pixels in the digital video data of this system, and Figure 6 is the light source. FIG. 3 is a diagram showing the relationship between brightness and number of dice versus sensing and counting time. 30... Reflection view, 3rd... Lens set, 32... CCO type equipment 33...
· dice. Fig. Fig. Fig. 5. Jar

Claims (7)

[Claims] (1) An automatic photoelectric video counting system for optically counting a collection of separated or dispersed objects, comprising: a beam projection device that uniformly projects a beam of light onto the object to be counted; a photoelectric linear array device that performs signal processing on a two-dimensional image by performing mechanical scanning on an object from the side; a circuit that converts the electrical signal of the image of the object into binary digital data; An automatic photoelectric video counting system comprising a personal computer with suitable peripherals for storing the counts. (2) The collection of objects to be counted is a semiconductor die that has been cut, unrolled, and separated and affixed to a developable plastic film. automatic photoelectric counting system. (3) a light beam projection device is placed on a diffuser plate and has a light source therewith, the diffuser plate being larger than the area of the object to be counted and in front of the light source generating a uniformly diffused light beam; 2. The automatic photoelectric image counting system according to claim 1, wherein the light source has stable brightness based on a circuit feedback principle and is capable of obtaining a uniform image of the object. (4) A light beam negative feedback circuit is further provided for locking the light beam projection device of the SCR-controlled light source to a fixed level of brightness, and the light beam intensity of the light beam negative feedback circuit is appropriately sensed by a photodiode. A sensing signal is obtained at a position where the sensing signal is fed back to a transimpedance amplifier for current-to-voltage conversion, and then the obtained voltage signal is applied to a voltage-controlled current source to generate a current. A load set between the collector and gate of the transistor changes the above current according to a potential difference, and supplies the current to influence the charging time of a capacitor set at the input terminal of a silicon-controlled trigger device. , thereby causing a substantial change in the trigger face of the silicon controlled trigger device so that the brightness of the light beam is automatically adjusted, and the light source is locked to a preset brightness level by a variable resistor so that the light beam 2. The automatic photoelectric image counting system of claim 1, wherein the system is unaffected by aging or power fluctuations. (5) an optical image forming system in which a photoelectric device for image scanning focuses an image of the object on a linear charge couple device and a mechanical drive device for one-dimensional scanning, and on the charge couple device CCD; a comparator for converting the analog signal output from the charge-coupled device to binary digits, and a synchronization circuit for direct memory access of the digital video data so that the video data is readable by computer software. An automatic photoelectric image counting system as claimed in claim 1. (6) Find the entire group of pixels using the distribution rule, count the pixels that are next to each other as one unit, and then divide and separate the objects into multiple pieces using the above rule. A counting algorithm that accurately determines the total number of dice counted in a group. (7) Furthermore, the correction method for determining the appropriate number of images of large pixel groups of connected objects includes those performed by statistical means, in which some pixel groups are too large and abnormal (i.e., the images formed are 7. The method of counting according to claim 6, including automatically issuing a warning signal in case of irregularity (not always obvious) to alert an operator to check the object to be counted.