JP2959156B2 - Solder shape inspection method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for inspecting the shape of solder, and more particularly, to a method for inspecting a sample solder registered in a computer by a neural network system .
Then, a solder similar to the solder to be inspected is extracted and
It was to determine the acceptability of the solder shape査subject solder
About the shape inspection method.

[0002]

2. Description of the Related Art IC, LSI, resistor chip, capacitor
The electrodes of electronic components such as chips are soldered to the substrate .
After bonding to the poles , a solder shape inspection is performed. Conventionally, such shape inspection, observe the solder by a camera inspection
The quality of the solder was judged based on the ratio of the brightness of the pixels in the inspection area. However, such conventional means, based on the mere light / dark ratio of the pixels, determines the quality of the solder, so that the inspection accuracy is poor and the shape of the observed solder is accurately determined. It was impossible.

[0003]

[SUMMARY OF THE INVENTION Incidentally, although the solder shape to adhere the electrode portion of the electronic component to an electrode on the substrate is in a different, but still broadly classified into the typical shape of the finite (typically several ten degree in) You can do it. In addition, when solders of various shapes are roughly classified into finite typical shapes,
The solder each roughly classified has been, each having a smooth flat surface and the inclined surface of the unique, yet its surface are those having a specular. Therefore, by irradiating the illumination light from a specific direction to the solder roughly classified according to the shape in this way, and observing the reflected light with a camera provided in a specific direction,
Each of the solder-specific image is obtained.

SUMMARY OF THE INVENTION Accordingly, an object of the present invention is to provide a means for focusing on the optical characteristics of solder as described above, and utilizing this, to easily and easily determine the quality of the shape of the solder. More specifically, a neural network system
To provide means for determining the quality of solder
The purpose is to .

[0005]

SUMMARY OF THE INVENTION The present invention is irradiated with illumination light to the sample and a half field of bonding the electronic component to the substrate, the reflected
Was received by the camera light, this sample solder of the master
Input layer over image, an intermediate layer, neural network having an output layer (New
Input to the input layer of the neural network) . Then, a neural coefficient is obtained in which the difference between the output value of the output layer of the master image of the specific sample solder and the output value of the output layer of the master image of the other sample solder becomes large . Many such operations
Repeat for several sample solders,
A neural coefficient capable of distinguishing the sample solder from each other is registered in the computer as a definite neural coefficient .

[0006] Then solder inspection target observed by the camera, based physician the image to determine nerve factor type, which is and registered in the input layer of the neural network (neural network)
To calculate the output value of the output layer . And in the sample solder
Having an output value that approximates the solder of the output value of the inspected from
Extract the sample solder. Then, the solder to be inspected is
Is determined to be similar in shape to the sampled solder extracted
Is, also this quality of this sample solder shape is known
From the, you can determine the quality of the solder of the shape, which is the object of inspection
You .

[0007]

According to the above configuration, the neural network system
Using a stem to accurately inspect the shape of solder
Can Ru.

[0008]

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a side view of a solder shape inspection apparatus.
You. Electronic components 2 such as IC and LSI are mounted on the substrate 1.
Have been. The lead 3 which is an electrode part of the electronic component 2
Are bonded to the substrate 1 by solder 4. The illumination light is irradiated from above the light source 6 to the solder 4, the reflected light over the C
By receiving the CD camera 5, Ru <br/> to enter the hand image of the solder 4. The CCD camera 5 has a light receiving section 5a and a lens 5b.

As described above, although the shape of the solder for bonding the electrode portion of the electronic component to the substrate is various, it can be roughly classified into a finite typical shape. In addition, the surface of the solder is a metallic mirror surface, and when the solder is roughly classified into typical shapes as described above, each of the roughly classified solders has its own smooth flat surface or inclined surface. . Therefore, when the illumination light is irradiated from a specific direction to the solder roughly classified according to the shape in this way, and the reflected light is received by a camera provided in a specific direction, an image of each solder is obtained. can get. Then, with reference to FIGS.
Some sample solders 4 with shapes (41-4n)
Image will be described.

FIG. 2 shows the light reflection characteristics of the first sample solder 41 having the most typical shape. The sample solder 4 1, a lead 3 and adhere well to the electrode 7 of the substrate 1, the sample solder 4 1 shape is good. Sample solder 4 first surface is curved, the light a incident on the steep inclined surface is reflected to the side, the reflected light C
The light does not enter the CD camera 5 and is observed dark. Also solder
The right end of 41 is a flat surface, and the light b incident on this flat surface is reflected right above and enters the CCD camera 5, so that it is observed brightly. Therefore, the solder 4 1
When observed with the CCD camera 5, a master image I1 having a light-dark distribution shown in FIG. 3 is obtained.

FIG. 4 shows a second sample solder 42 having a typical shape . This sample solder 4 2
A flat, not by bonding a lead 3 to the electrode 7 of the substrate 1, the shape of the sample solder 4 2 Ri poor der,
Such sample solder 42 is poorly welded (poor wed).
ting). This sample solder 42
When light is applied to the entire surface, it is light incident on the center of the surface
The light is reflected immediately above the injury , enters the CCD camera 5, and is
Is observed bright , and therefore the light-dark distribution shown in FIG.
A master image I2 having obtained.

[0013] FIG. 6 have been shown the third sample solder 43
You. The sample solder 43 is a slight excess, the lead 3 is sufficiently adhered to the electrode 7 of the substrate 1, the shape of the sample solder 4 3 is good. This sample solder 4
The top of 3 is flat, and the light c incident thereon is reflected right above and enters the CCD camera 5, so that it is observed brightly. Also, as described with reference to FIG.
The light a incident on the slope is reflected to the side, and the CCD camera 5
Is not observed, so it is observed darkly.
The reflected light b is reflected right above, enters the CCD camera 5, and becomes brighter.
Observed. Therefore, it has the light-dark distribution shown in FIG.
That master image I3 is Ru obtained.

As described above, when the sample solders 4 (41 to 4n) of various shapes are observed by the CCD camera 5, a master image unique to each sample solder 4 (41 to 4n) is obtained.
I1 to In are obtained. As mentioned above, the shape of the solder is
It is roughly divided into several tens of typical shapes.
Although an image having the image is obtained, in this embodiment, FIGS.
The three types of solders 41, 42 and 43 shown in FIG.
As an example, the above description has been given. In this embodiment,
Although described the binary image brightness as an example, the case of other images, such as multi-level or color images, unique image is obtained each sample <br/> solder.

FIG. 8 shows a block diagram of the neural network system computer. Neural network
The over neural network) 10, an input layer 11, intermediate layer 1
2 and an output layer 13. The input layer 11 has a first cell A (A1 to An1). Also, the middle layer 12
Has a second cell B (B1 to Bn2). The output layer 13 has a third cell C (C1 to Cn3). The input layer 11 receives image data of the solder 4 captured by the light receiving section 5 a of the CCD camera 5.

FIG. 9 shows a neural network (neural network).
10 models are shown. I1 to In refer to FIGS.
As described while illuminating, the master image of the sample solder 4 (41 to 4n) observed by the CCD camera 5 is shown in FIG.
You. The light receiving portion 5a of the CCD camera 5 is arranged in a matrix.
That we have a large number of cells D (D1~Dn4). The plurality of cells D are connected to one cell A. The plurality of cells A are connected to one cell B. The plurality of cells B are connected to one cell C. The line connecting the cells is called a neural network ( neural network ). The neural network 10 includes a neural coefficient ki (i = 1, 2, 3,... N) connecting the cell A and the cell B,
Nerve coefficient kj connecting cells C (j = 1, 2, 3,... N)
Exists. The nerve coefficient is calculated as a connection weight (synaptic).
weight) . Next, a method of determining the quality of the solder shape will be described.

First, the sample solders 41-4n
Learning is performed as follows. From many solders,
Sample-shaped solder is used as sample solder 41-4n
To extract. Next, the sample solder 4 is moved by the CCD camera 5.
1 to 4n are observed, and the images I1 to In are input to the first cells A1 to An of the input layer 11 as master images. Then, the first cell A outputs a binary signal of H and L according to the input value from the cell D. Further, the second cell B outputs a binary signal of H and L according to the input value from the first cell A. The third cell C of, depending on the input value from the second cell B, and outputs the multi-level signal down one or more.

Therefore, the conditions of the table shown in FIG.
As shown in (1), first outputs an output value close to 1 (maximum value) with respect to the cell C1 of solder 4 1 master images I1, respect to the other sample solder 4 2 to 4n master images I2~In , 0 nerve coefficient ki 1, kj 1 for outputting an output value close to (minimum), that is, the output value of the particular sample solder 4 1 master images I1, other sample solder 4 2 to 4n master images nerve coefficient k the difference between the output value of I 2 -In can be clearly distinguished I pole Daito Do
Find i 1 , kj 1 . There are many neural coefficients ki1 and kj1 that satisfy the condition (1).

Next, as shown in condition (2) of FIG.
The cell C2 outputs an output value close to 1 with respect to the master image I2 sample solder 4 2, other samples solder 4 1, 4 3 to 4n master images I1, respect to I3～In, close to zero the output Nerve coefficient ki 2 , which outputs a value,
kj2 , that is, the master of the specific sample solder 42
Output value of image I2 and other sample solders 41 and 4
Difference between output values of 3 to 4n master images I1 and I3 to In
Are maximal and are clearly distinguishable neural coefficients Ki2 , Kj
Ask for 2. There are also a large number of neural coefficients ki 2 and kj 2 that satisfy such condition (2). Then conditions (1) and the conditions <br/> matter (2) is satisfied at the same time nervous coefficient ki 1 + 2, kj 1 + 2
The Ru determined (see FIG. 11). Thus two conditions (1), nerve coefficient ki 1 + 2 which satisfy simultaneously the (2), k
There are many j 1 + 2, but the number of conditions increased
The number of neural coefficients that satisfy all conditions gradually decreases
You small.

Similarly, conditions (3) to (3) in FIG.
As shown in (n), the other sample solder 4. 3 to 4 n Nitsu
Repeated the same operation can have, all of the conditions (1)
The final neural coefficient ki ′ = k that satisfies all of (n)
i1 + 2 + ... + n, Ru asked to kj '= kj1 + 2 + ... + n ( Fig. 11 participants
See) . So this nerve coefficient ki ', kj', all of
The sample solders 41 to 4n are registered in the computer as deterministic nerve coefficients that can be distinguished . Such learning is back
Called propagation.

Thus, the sample solders 41 to 4n
If you have learned, solder 4 to be inspected
5 and observe the image using the neural network 1
Type in 0, confirmed nerve coefficient ki, which is registered ', kj' to
The output value of each cell C of the output layer 13 is obtained based on the
Extract sample solder with output value approximating force value
If the shape of the solder 4 to be inspected is
It can be determined that the shape is similar to the shape of the solder . Since the quality (GOOD, NO GOOD) of the shape of the sample solder and its category (incomplete, insufficient, excessive, etc.) are known in advance, the quality of the shape of the solder to be inspected and the category thereof can also be determined. In FIG. 9, GOOD is a good product,
NO GOOD is a defective product. GOOD and NO
The numbers -1, -2, ... -n added to GOOD are:
The category of GOOD and NO GOOD are shown.

[0022]

[Table 1]

As described above, the definite neural coefficients ki ', k
If j ′ is set, the definite nerve coefficients ki ′, k
The shape of the solder to be inspected is inspected based on j ′. This shape inspection is performed as follows.

The solder 4 to be inspected is connected to the CCD shown in FIG.
The image I ′ is observed by the camera 5 and input to the input layer 11 (see FIGS . 8 and 9). And the definite nerve coefficient k
The output values of all the cells C1 to Cn3 are calculated based on i ′ and kj ′. Then, from among the cells C1 to Cn 3, it extracts the cell C that has output the closest output value 1.

[0025] The closest out here, for example, cell C1 1
If the output power value, the solder 4 has a shape similar to the first sample solder 4 1, the shape is determined as good. If also for example cell C2 has output <br/> most output value close to 1, the solder 4 has a shape similar to the second sample solder 4 2, the shape is judged to be defective. If also for example the cell C3 has output closest output value 1, the solder 4 has a shape similar to the third sample solder 4 3, the shape is determined as good.

In this way, the sample solders 41 to 4 n
Has an output value similar to the solder 4 to be inspected
Lever to extract things, the solder 4 is extracted sample half
It is determined to have a field similar shape, also sample half
Since the quality of the shape of the field is known in advance, it can be determined solder 4 shape the quality of the inspection object immediately.

[0027]

According to the method as described above, according to the present invention, the neural network system, it is possible <br/> performing solder shape inspection.

[Brief description of the drawings]

FIG. 1 is a side view of a solder shape inspection apparatus according to the present invention.

Cross-sectional view of a solder according to the present invention; FIG

FIG. 3 is an image diagram of a solder according to the present invention.

Cross-sectional view of a solder according to the present invention; FIG

FIG. 5 is an image diagram of a solder according to the present invention.

[6] cross-sectional view of a solder according to the present invention

FIG. 7 is an image diagram of a solder according to the present invention.

FIG. 8 is a block diagram of a neural network system computer according to the present invention.

FIG. 9 is a diagram of a neural network system according to the present invention.

FIG. 10 is a table showing conditions according to the present invention.

FIG. 11 is an explanatory diagram of a deterministic nerve coefficient according to the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 Substrate 2 Electronic component 3 Electrode part 4 Solder 5 CCD camera 10 Neural network (neural network) 11 Input layer 12 Intermediate layer 13 Output layer

──────────────────────────────────────────────────続 き Continuation of front page (56) References JP-A-63-196980 (JP, A) JP-A-2-298840 (JP, A) JP-A-4-104048 (JP, A) JP-A-4- 222064 (JP, A) JP-A-4-142412 (JP, A) (58) Fields investigated (Int. Cl. ⁶ , DB name) G01N 21/84-21/91 H05K 3/34

Claims (3)

(57) [Claims] 1. A (i) an electronic component by irradiating a sample <br/> irradiating light to Handa to adhere to the substrate, and receives the reflected light to the camera
Steps to get a master image of each sample solder
When, (ii) the steps of each sample solder master image to the input layer of the neural network, (iii) the output value of the output layer of the particular sample solder master image, this other sample solder master images nerve coefficient of the neural network which difference is large in the output value of the output layer of determined
Because, the step of registering the computer the neural coefficient as determined nerve factor, comprising obtaining the <br/> image is observed by the camera solder (iv) inspected, the image of (v) This Input to the input layer, and (iii)
Calculating the output value of the output layer based on the deterministic nerve coefficient registered in the step of (vi) ; and (vi) the step of (v) in the sample solder
In the calculated test output value approximate to the solder of the output values of the target
Extract the sample solder having
Judge as having similar shape to the extracted sample solder
Step a, solder shape inspection method characterized in that it consists of. 2. The method according to claim 1, wherein the image is a binary image. 3. The shape of the solder according to claim 1, wherein in the step (iii), the output value of the specific sample solder is near1 and the output values of the other sample solders are near0. Inspection methods.