JPS62113436A - External appearance inspection method for semiconductor pellet - Google Patents

Abstract

PURPOSE:To carry out a high precision detection with much quantity of information by using a variable density picture by detecting the external appearance image of a semiconductor pellet which is to be inspected as a variable density value. CONSTITUTION:After obtaining a picture S by optically detecting the appearance image of many semiconductor pellets, the picture P of one pellet is cut out. A defective picture E1 which is equal to the absolute value ¦P-R¦ of the difference between a picture P to be inspected and a good reference picture R is made. The mean value of the defective picture E1 is E1A, the value of the brightest point is Emax, the value of the darkest point is Emin, a threshold level SL is calculated by a formula SL=E1A+alpha(Emax-Emin) (wherein, alpha is a constant) and a divalent picture E2 is made with this value. A defective label image EL is obtained by labelling the divalue picture E2 and the external appearance inspection of the semiconductor pellet is carried out by calculating the area, the shape data (width, length, etc.) and the position of a defect by the defective label image EL.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to a method for inspecting the appearance of semiconductor pellets.

[Technical Background of the Invention 1] Due to the development of image processing technology and artificial visual devices, inspection work, etc., which conventionally relied entirely on human vision, is rapidly becoming automated. There are various methods of fine pattern inspection technology that can be applied to such automated inspection work (see, for example, Journal of the Institute of Electrical Engineers of Japan, November 1973 issue, p. 57), and these methods can be applied to printed wiring boards, photomasks, or It has been put to practical use in automatic inspection and automatic assembly of products such as semiconductor pellets.

[Problems in the Background Art] The conventional fine pattern inspection technology described above is a pattern inspection method using binary images, but this method has the following problems, and therefore is not suitable for automatic inspection of semiconductor pellets, etc. was not suitable.

(i) When obtaining an image of an object to be inspected, some type of illumination device is used to illuminate the object, but if there are variations in illuminance, a highly accurate binary image cannot be obtained. In order to obtain a highly accurate binary image, it is necessary to constantly change the threshold level to the optimum value in response to variations in illuminance, but this is difficult in practice. Further, in order to perform stable binarization, it is necessary to uniformly illuminate the object to be inspected, but such illumination is impossible.

(ii) A method using a binary image cannot detect thin dirt adhering to the surface of a semiconductor pellet.

(iii) There is no established method for measuring the shape, position and size of defects, degree of defects (depth of eaves, degree of contamination), etc. after visual inspection.

(iv) Automatic appearance inspection equipment using a binary pattern projection method applying the fine pattern inspection technology has also been put into practical use (for example, Fuji Jiho Vol. 52, No. 8, 197).
9) With this method, visual inspection of multiple defects or complex patterns is not possible, and visual inspection of multiple semiconductor pellets cannot be performed simultaneously.

[Object of the Invention] An object of the present invention is to provide a method for inspecting the appearance of a semiconductor pellet, which solves the problems of the prior art and allows the appearance of a semiconductor pellet to be inspected more accurately and more practically than the conventional method. The goal is to provide the following.

[Summary of the Invention] The method according to the present invention detects the external appearance image of a semiconductor pellet to be inspected as gray scale values, and obtains an image f@! While obtaining P, the absolute value of the difference between the image R of a good semiconductor pellet obtained in advance and the image P is defective image E corresponding to P-R1.
, and then converting this defect image E into a binary image E2, and labeling the binary image E2 to detect defects in appearance of the semiconductor pellet to be inspected. be.

Unlike the conventional method, the method of the present invention uses a grayscale image, so the amount of information is larger than when using a binary image, and highly accurate detection can be performed.

Note that a pattern detection device proposed by the present applicant (see Japanese Patent Application Laid-open No. 137978/1983) can be used to carry out the method of the present invention. That is, when aligning the images P and R, after alignment is performed by a pattern detection method using a correlation coefficient, labeling is performed from an image of a defect equal to the absolute value of the difference between P and R. By doing this, the appearance of the semiconductor pellet is inspected.

[Embodiments of the invention] The present invention will be described in detail below with reference to the drawings.

In the method of the present invention, external images of a large number of semiconductor pellets are optically detected, photoelectrically converted, and converted into digital signals to generate images S as shown in FIGS. 1(a) and 2.
After obtaining the data, an image P for one semiconductor pellet is extracted from this image S by the data extraction circuit 1 shown in FIG. ) and the correlation coefficient calculation circuit 2 extracts the reference image R from the reference image storage circuit 3 in which the reference image R regarding a non-defective semiconductor pellet is stored as shown in FIG.
, the correlation coefficient f between the two is calculated, and this correlation coefficient f is input into the determination circuit 4 to detect the position of the image P within the image S. (Please note that the process up to this point was published in Japanese Unexamined Patent Publication No. 57
The pattern detection device disclosed in Japanese Patent No. 137978 is used. ) Next, a defective image E, whose size is equal to the absolute value of the difference between the image P and the reference image R, I P - Rl, is created as shown in FIG. 1(C). In this case, the average value of the defect image E is EIA, the maximum value is E-(&t is also the value of the bright point),
The minimum value (value of the darkest point) is El, 1. n, and the threshold level SL is calculated using the formula 5L=Eu+α(E n'm −E min ), where α is a constant).
) A binary image E2 is created as shown in FIG.

Then, by labeling (classifying) this binary image E2, a defect classification image EL shown in FIG. 1(e) is obtained,
Using this defect classification image EL, the size, area, shape (width, length, etc.), and position of the defect are calculated and the appearance of the semiconductor pellet is inspected. The labeling is performed, for example, by assigning coordinates P (X, V ) of the number of columns X in the horizontal direction from the upper left edge of the binary image and the number of rows y in the vertical direction from the upper left edge to four pixels A adjacent thereto.

(x-1, y), A2 (x-1, y-1),
At (x, y-1) and A4(X-t1, y
-Classify using the information in 1).

[Effects of the Invention J Navigation The method of the present invention explained in the embodiments can produce the following effects.

(a)iI! Since the light image is used, it is possible to detect dust and dirt that could not be detected accurately using conventional methods, and also to detect minute pattern distortions.

(b) Since the pattern used for pattern position detection and appearance inspection is the same image, memory for the reference pattern can be saved.

(C) Even if the luminous intensity of the illumination light source (illuminance on the semiconductor pellet) drifts to some extent, the method of the present invention automatically calculates the threshold level SL, so that highly accurate images can always be obtained.

(d) Semiconductor pellet positioning device of inspection equipment (
Even if there is poor accuracy or play in the XY table (for example, an XY table), the pattern position can be detected precisely, so there will be no error in determining defective products.

(e) Since a plurality of semiconductor pellets can be inspected, the appearance inspection can be speeded up.

[Brief explanation of drawings]

Figure 1 is a diagram for explaining the principle of the method of the present invention, Figure 2 is a diagram for explaining the principle of the method of the present invention.
The figure is a diagram showing the principle and schematic configuration of a pattern inspection device used to implement a part of the method of the present invention. DESCRIPTION OF SYMBOLS 1... Data extraction circuit, 2... Correlation coefficient calculation circuit, 3... Reference image storage circuit, 4... Judgment circuit.

Claims (1)

[Claims] 1 After optically detecting the appearance image of the semiconductor pellet to be inspected and converting it into an analog electrical signal, further converting it into a digital signal to obtain an image P regarding the semiconductor pellet to be inspected, while detecting a non-defective semiconductor pellet. After creating an image E_1 corresponding to the absolute value of the difference between the image R and the image P, converting this image E_1 into a binary image E_2, and labeling the image E_2, A semiconductor pellet appearance inspection method characterized by detecting the shape, size, position, etc. of defects.