JPS6224133A - Automatic binary-coding system - Google Patents

Abstract

PURPOSE:To achieve a better result of the recognition in the appearance inspection of a flaw or the like, by a method wherein the image of the appearance of a resin sealed type semiconductor device to be inspected is divided into multiple parts in terms of the brightness thereof to be memorized and the memorized divided image are further subdivided so that a binary image will be obtained at slice levels determined by the average brightness of the subdivided image segments. CONSTITUTION:The appearance of a resin sealed type semiconductor device to be inspected is photographed with an ITV camera, the image thereof is inputted into an image processor and the input signal is divided in terms of the brightness classified, for example, by 256 graduations to be memorized into a memory section of the image processor. The image memorized is further subdivided and the average brightness of the subdivided image segments are calculated to determine the slice levels and finally, a binary coded image is obtained. Furthermore, an arithmetic processing of the average brightness of the divided image is made on the basis of experimentally determined coefficients to obtain the optimum slice levels and thus, a binary coding of the memory image is implemented based on the resulting levels.

Description

[Detailed Description of the Invention] [Technical Field of the Invention] The present invention relates to an apparatus for automatically inspecting the appearance of resin-sealed semiconductor devices, that is, checking for holes, chips, 4&, leads, bends, lead lengths, and dam cut eccentricity. Mainly related to the applied automatic binarization method.

[Technical background of the invention]

The manufacturing of semiconductor products can be divided into the pre-process of building elements and circuits onto wafers, and the post-process of assembling the elements divided into pellets, but the process is complex and requires a work environment that is extremely dust-free. Currently, the automation of front-end processes lags behind that of back-end processes.

On the other hand, as the production scale of semiconductor products increases, there is a need to establish highly productive manufacturing lines.
Semiconductor wafer size as a ring.

There is a trend toward larger diameters, and in reality, as much as 120 mφ is used, and it is already possible to pull semiconductor single crystals up to 200 nmφ.

As wafers become larger in diameter, it becomes difficult to handle them manually, and automation is inevitably required for wafer transport, setting in equipment, and the like.

Recent semiconductor devices, as exemplified by VLSI, have made remarkable progress in becoming highly integrated and high-performance, and as a result, manufacturing processes have become more complex and diverse, so the impact of the cleanliness of the manufacturing line on yield is Automation is being called for more than ever in order to keep the human body away from the source of dust, and the wave of automation is coming. As a link to this, devices have begun to be developed that automatically perform visual inspections of semiconductor devices, particularly resin-sealed semiconductor devices.

The actual method is to convert the image signal of the resin-sealed semiconductor device to be inspected using an industrial ITV camera into a binary image at a specific slice level, and then conduct an external inspection of the image signal. In this case, the slice level is constant.

On the other hand, it is possible to automatically perform visual inspection using pattern recognition technology, but the error limit for this method is 3 to 3.
Since it is 4%, it is inappropriate for aiming at minute appearance defects in semiconductor devices, that is, an error limit of 0.1%.

The above-mentioned visual inspection device uses visual sensor application equipment that has a built-in dedicated processor and minicomputer necessary for image processing and performs image processing at high speed.

[Problems with background technology]

The above-mentioned resin-sealed semiconductor device to be inspected is imaged with an ITV camera under proper lighting, but it was found that uniform brightness within the surface could not be obtained due to the surface condition of the object. However, this poses a problem in obtaining a binarized image.

When we investigated the distribution state of specular reflection lighting and ring lighting as lighting methods under constant lighting conditions, we found that not only were there variations within the plane, but the brightness level was different for each subject. Furthermore, although the brightness level of ring lighting differs depending on the subject, it was found that the variation within the plane was smaller than the former.

Therefore, if there are portions of different brightness on the inspection surface of the resin-sealed semiconductor device that is the subject, there is a problem in that the fidelity of the image obtained is impaired when binarized at a constant slice level.

[Purpose of the invention]

The present invention provides a new binarization method that eliminates the above-mentioned drawbacks, and is particularly intended to improve recognition results.

[Summary of the invention]

In order to achieve the above object, the present invention images the external appearance of a resin-sealed semiconductor device to be inspected using an ITV camera, inputs the image to an image processing device, and divides this input signal into 256 gradations of brightness. and stored in the memory section of this image device. A method was adopted in which this stored image is divided into a plurality of parts, the average brightness of the divided images is calculated, a slice level is determined, and a binarized image is thereby obtained. Furthermore, a method was adopted in which the average brightness of the divided images is subjected to arithmetic processing using an experimentally determined coefficient to determine the optimal slice level, and the stored image is binarized based on this.

[Embodiments of the invention]

The present invention will be explained in detail with reference to FIGS. 1 to 7.

The image processing device adopted in the method of the present invention has a built-in dedicated processor and minicomputer essential for image processing, and is a TO3P visual sensor application device capable of basic image processing such as binarization processing.
IX-II (trade name, manufactured by Toshiba Corporation) and T-7/20E (manufactured by Toshiba Corporation, product number) were used as the minicomputer. Some of the main functions of the basic commands are listed in Table 1.

Table 1, Figure 1 shows a schematic diagram of an automatic appearance inspection device that employs this method, and the inspection items are shown in Figure 3.

This device has a transport path (3) that transports a resin-sealed semiconductor device (2) to be inspected to a stand (1) equipped with a display panel (6).
are provided, and transfer members (4) (4) are provided in the middle and at the end.
). An extension path (7) extends from the transfer member (4) to the end of the pedestal (1). At the end of the (7) is a stocker (5) consisting of an unloader into which the semiconductor devices (2) separated by the transfer member (4) are delivered. (5) will be installed. Mount (
1) At the other end is a resin-sealed semiconductor device to be inspected (2)...
In addition to arranging the loader (8) that houses the above-mentioned image processing device (9), it is electrically connected to the ITV camera (10) installed at each test station.

The position of the resin-sealed semiconductor device (2) to be inspected transported to each station is controlled by the following means. As shown in FIG. 2, the resin portion that encapsulates this semiconductor device has a hexagonal cross section, long in one direction, and flat on the opposite surface.

The longitudinal position of this resin part is determined by positioning pins (not shown) provided in the conveyance path (3) corresponding to the test station. In addition, the clearance with the resin part is set to 0.1+a to regulate the width direction. Further, by means of the positioning pins, the semiconductor device that is stationary on the transport path corresponding to the test station is fixed by being depressurized by the exhaust hole (7) communicating with the decompression mechanism provided in this transport path, and is prevented from tilting.

A total of five ITV cameras (10) are arranged to take images of the resin-sealed semiconductor devices to be inspected (2) under optimal conditions based on the roles shown in Table 2.

Due to the division of roles shown in Figure 3, the imaging portions of the resin-sealed semiconductor device are the front surface, back surface, and left and right lead surfaces, and the inspection items include cavities, chips, scratches, lead bends, lead length, bends, and dam cuts and eccentricity. It is. This dam cut is an item that indicates the degree of eccentricity caused by removing earth when a so-called tie bar (connecting rod) formed on a lead frame used when mounting a semiconductor element is cut.

Now, even if the exterior of a resin-sealed semiconductor device, that is, the resin part, is illuminated to the optimum level, the brightness often varies locally.

Therefore, as mentioned above, we adopted the method of setting the slice level for each region. Figure 4 shows the state in which the captured image to be automatically binarized is divided into 10 slices, which are determined for each divided region. The levels are shown in Figure 5. This figure shows ITV power under optimal lighting.

The image captured by the camera (10) is processed by the image processing device (9).
) and convert this input signal into 256 gradations (0 to 255).
The brightness is divided into two parts and stored in the memory section. It should be noted that the state in which this stored image has been divided and the slice level determined as described above is illustrated in FIGS. 4 and 5.

In FIG. 4, (11) is the peripheral area of the semiconductor device (2), (12) is the binarization target area including this, and (13) is the state in which the longitudinal direction of this target area is divided into ten equal parts, and (13) is the brightness. The sampling line of 2 in Fig. 4 is shown in (14).
The valued slice level (16) is determined by averaging the brightness of each area or by calculating it with a constant determined experimentally. Further, the obtained brightness distribution for each region is shown in FIG. 5 (15).

Next, specific measures such as scratches and nests will be explained.

First, setting of the inspection area and masking will be explained. In the above-mentioned automatic inspection apparatus, the resin-sealed semiconductor device to be tested is set on a transport path from a loader, transported to each test station, and placed at a standard position at each station by the above-mentioned positioning mechanism. This standard position can naturally be obtained by the image processing device (9), but it is enlarged compared to the actual semiconductor device (2).

As mentioned above, the resin part of the resin-sealed semiconductor device to be inspected (2) has a positioning pin in its longitudinal direction, a side wall part provided in the transport path in width, and an inclination corresponding to each station on the transport path. It is regulated by a pressure reduction mechanism formed in the conveyance path and placed at the standard position.

This resin-sealed semiconductor device to be inspected is imaged by an ITV camera under optimal irradiation conditions and then input to an image processing device (9). The image of this device is a square, and the upper left of the square is predetermined as the origin, and the screen has 512 pixels both vertically and horizontally.
It is made up of pixels. Next, the position of the semiconductor device taken in this image is determined, that is, the inspection area is measured.
In this case, this semiconductor device is assumed to be composed of line segments parallel to each other, and after investigating the density of each pixel in the X and Y directions, Display the sum of E linearly on a separate line. This situation is shown in Figure 6.

AA' and DD' indicate the inspection status of each pixel from each corner.

After determining Xsum and Ysum using this method, the left and right end X coordinates XI2 are determined from the total value with reference to the origin mentioned above.
, XR is obtained and the upper and lower end Y coordinates YL are obtained from the Ysum data in the same way.
, find VD. The area within the constant value from the area surrounded by XL, XR, YU, and VD is set as the area to be inspected. From experience, this constant is 0.045
x 4 pixels (amount of variation). In the image used for this measurement, the resin-sealed semiconductor device to be inspected is illustrated as a dark area, but it is also possible to do the opposite. It should also be noted that when measuring the number of dark areas, if the total number exceeds the standard value, it will be counted.

The resin part of the resin-sealed semiconductor device to be inspected (2) has an ejector pin mark in the center of its flat surface, and a resin layer exists below the pin mark, so the circle of this pin mark is masked. In addition to reducing errors, a semicircular notch is formed at one end edge of the resin layer in the entire thickness direction of the one end edge. Therefore, the entire surface of this notch is XL, Y
As shown in FIG. 5, masks A,
Do B.

Next, defective parts are detected as shown in FIGS. 7(a) and 7(b).

After measuring the inspection area, ie, the position of the semiconductor device, by the peripheral distribution measurement, the area of the cavity and chip S4 is determined at station 1.5, and if the maximum value of each exceeds the reference value, it is determined to be defective.

For station 3, the length of the flaw in the X and Y directions, L4, is calculated, and if it exceeds the standard value, it is determined to be defective.

The reference values are shown in Table 2, which are based on empirical rules.

Next, measurements at stations 2 and 4 will be described.

This measurement differs from stations 1 and 3.5 in that it is an image.

It is very easy to judge because the object to be measured is a metal part, so the slice level is 12.
Fix it to the 0th brightness.

The shaded areas related to dams and eccentricity in Table 2 are the dam parts of the lower leads that have been remade to make them easier to see, so the lower side shows the lead shape. In actual measurement, the eccentricity of the dam E can be determined by determining the peripheral distribution of a rectangular parallelepiped that includes A and B and has C and D as sides in the same manner as shown in FIG. Lead bending and lead length can also be measured using methods similar to those shown in Table 2.

〔Effect of the invention〕

By using this automatic binarization method, it has become possible to suppress variations in the binary image even if there are parts with different brightness on the surface to be inspected.

The method to achieve this is to divide the binarized area, which makes it possible to detect defective parts that could not be found sufficiently, and improves the detection limit accordingly. The effect on mass production is extremely large.

[Brief explanation of drawings]

Figure 1 is a diagram showing an outline of an automatic semiconductor appearance inspection device to which this method is applied, Figure 2 is a cross-sectional view of a resin-sealed semiconductor device to be inspected accommodated in its transport path, and Figure 3 is an inspection diagram of this inspection equipment. Diagrams summarizing the method, Figures 4 to B (a) (b)
The figure is a diagram illustrating an inspection method of an inspection device/to which this method is applied.

Claims (2)

[Claims] (1) Input the image capture screen of the resin-sealed semiconductor device to be inspected into an image processing device, divide the resulting image with a certain number of pixels into multiple parts, and create a binary image at a slice level determined by the average brightness of each part. An automatic binarization method characterized by obtaining. (2) Input the image capture screen of the resin-sealed semiconductor device to be inspected into an image processing device, divide the resulting image with a certain number of pixels into multiple parts, and calculate a constant for the average brightness of each part. An automatic binarization method characterized by obtaining a binary image based on slice levels.