Chip visual positioning method

Abstract

The invention discloses a chip visual positioning method, which comprises the following steps of: 1, shooting salient points on the surface of a chip through a visual system to obtain an original salient point image, and preprocessing the original salient point image to obtain a standard image; step 2, positioning the salient points in the standard image, and determining the positioning center of the chip: firstly, taking a plurality of salient points on the surface of the chip as detection objects, and respectively measuring the distances between the detection objects and the surrounding salient points; and secondly, comparing the measured distance value with a preset standard value, if the difference between the measured distance value and the preset standard value exceeds a preset threshold value, rejecting the salient point from the detection object, and finally, determining the positioning center of the chip by taking the reserved salient point as the detection object. After the technical scheme is adopted, compared with the prior art, the method has the following beneficial effects that individual salient points which do not meet conditions are automatically removed through the design of the standard deviation threshold value of the salient point spacing, so that the whole center of the chip is more accurate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of chip bonding, in particular to a method for visual positioning of a chip. BACKGROUND

[0002] In early chip bonding processes, very thin gold wires are used to connect the pads around the chip to the package pins, which is called "wire bonding". As the functionality of chips becomes more and more powerful, the number of pins increases dramatically (for example, some chips can have up to thousands of pins), and faster signal transmission speed and smaller package size are required, so wire bonding has become unable to meet the requirements.

[0003] Currently, Bump technology is used for chip bonding. In simple terms, Bump is a small metal contact on the surface of a chip, which can be distributed on the entire surface of the chip, greatly increasing the number of connections. At the same time, Bump technology has the advantages of excellent electrical performance and smaller package size due to the extremely short signal path. For example, see Chinese patent application No. CN120033180, which discloses a "micro-bump stacked solder chip assembly and three-dimensional stacked packaging method".

[0004] Since the bonding position must be accurate during chip bonding, it is necessary to position the chip, and currently visual positioning methods are mostly used. For example, two sets of vision systems are used to take pictures of the upper and lower chips and mark the relative positions, and then detect whether the relative positions of the two chips are accurate. During visual positioning, the Bump (bump) on the surface of the chip is usually used as a reference point for positioning analysis. There are currently two main methods: First, use the center of gravity of the template contour as the bump positioning center. The bumps on the surface of the chip are usually arranged in an array, and the contour formed by several bumps can be used as a point reference image, for example, nine bumps in a "nine-square" pattern form a rectangular image, and the center of gravity (or center) of the image contour is used as the positioning center. This positioning method has low positioning accuracy, and may have some bonding deviation during chip bonding.

[0005] The second method uses multiple bumps as reference points and calculates the overall center. This method has high positioning accuracy because multiple bumps are used for positioning. However, this method also has a disadvantage: if individual bumps are shifted, it will affect the overall positioning result. During chip production, transportation and transfer, individual bumps may be shifted. At the same time, due to the extremely small size of the bumps, the shift of individual bumps may have a significant impact on the final overall calculation result.

[0006] In view of this, the present application proposes the following technical solution. SUMMARY

[0007] The technical problem solved by the present application is to overcome the shortcomings of the prior art and provide a method for chip visual positioning.

[0008] To solve the above technical problems, the present application adopts the following technical scheme: a chip visual positioning method, which comprises the following steps: step 1: capturing the surface bumps of the chip through a vision system to obtain an original bump image, and pre-processing the original bump image to obtain a standard image; step 2: positioning the bumps in the standard image to determine the positioning center of the chip, which adopts the following way: first, taking several bumps on the surface of the chip as detection objects, and measuring the distance between each detection object and the surrounding bumps; second, comparing the measured distance value with the preset standard value, if the difference between the two values exceeds the preset threshold, the detection object is removed; finally, taking the remaining bumps as detection objects to determine the positioning center of the chip.

[0009] Further, in the above technical solution, the pre-processing method of the original bump image in step 1 is image background interference pre-processing.

[0010] Further, in the above technical solution, the pre-processing method of the original bump image in step 1 adopts the following way: generating a heat map of the bumps, and then using Gaussian fitting to improve the accuracy of the peak points in the heat map.

[0011] Further, in the above technical solution, the threshold in step 2 is a dynamically adjusted threshold, which is adjusted by manually adjusting the threshold according to historical data and environmental changes.

[0012] Further, in the above technical solution, the threshold adjustment according to environmental changes is a corresponding dynamic adjustment of the threshold according to the environmental temperature drift.

[0013] After adopting the above technical solution, the present application has the following beneficial effects compared with the prior art: by designing the standard deviation threshold of the bump distance, the present application automatically removes individual bumps that do not meet the conditions, making the overall center of the chip more accurate.

[0014] The present application can be used for automatic positioning, screening and quality scoring algorithm of wafer / chip surface bumps based on machine vision, and is suitable for quality detection in semiconductor packaging, microelectronic manufacturing and other scenes. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a schematic diagram of the light standard image of the present application. DETAILED DESCRIPTION

[0016] The present application will be further described below in conjunction with specific embodiments and drawings.

[0017] The application is a chip visual positioning method, which comprises the following steps: Step 1: The surface bump of the chip is photographed by the visual system to obtain the original bump image, and the original bump image is preprocessed to obtain the standard image. In order to facilitate subsequent judgment, the original bump image in step 1 needs to be preprocessed, which can adopt a conventional image processing method, mainly image background interference preprocessing. For example: background suppression based on morphological top-hat transformation, adaptive light compensation. Through preprocessing, the problem of failure under deformation or occlusion in traditional template matching is solved.

[0018] In order to improve the detection accuracy and locate the bumps in the image, while ensuring that the system can run quickly, the application also adopts the following way when preprocessing the original bump image: 1. Improved YOLO network: the input image is output as a bump heat map through an encoder-decoder structure, and the real-time performance is improved through lightweight design such as depthwise separable convolution algorithm. This step is responsible for quickly and roughly finding the possible location area of the "bump", and generating a "heat map".

[0019] 2. Sub-pixel positioning: Gaussian fitting is adopted for the peak point of the heat map to realize ±0.1 pixel level positioning accuracy. This step is responsible for fine processing of the heat map to improve the rough position to a very high accuracy.

[0020] The heat map is a special way of image expression. In the original image, the position of each "bump" will appear as a bright spot (high-light area) in the heat map, and the background is dark. The center of the bright spot (the brightest point) corresponds to the most possible position of the bump. By outputting the heat map, it is convenient to accurately find the center of the bright spot, that is, to determine the center of each bump.

[0021] The depthwise separable convolution (Depthwise Separable Convolution) is to improve the real-time performance and greatly reduce the calculation amount of the model. Since the generated heat map is discrete, the peak point coordinates can only be integers (for example, the 100th row and the 50th column). But the real target position may fall between two pixels. The purpose of this step is to break the pixel limit and obtain more precise coordinates than one pixel.

[0022] The Gaussian fitting of the heat map peak point means that: first, find the rough center position of each bright spot on the heat map. Usually, a simple method such as finding the local maximum value point is used, and the coordinates of this point are integer pixel coordinates. Through Gaussian fitting, the positioning error is reduced from integer pixel level to sub-pixel level. Finally, the positioning accuracy far exceeds the resolution of the original image.

[0023] In addition, in order to avoid over-segmentation of the image, the present application adopts a distance transform + watershed algorithm to avoid over-segmentation by combining the prior size constraint of the convex points.

[0024] Reference Figure 1 As shown, it is a schematic diagram of a standard image. The standard image has convex points A1-A9.

[0025] Step 2: Position the convex points in the standard image to determine the positioning center of the chip, which adopts the following way: First, take the chip surface convex points A1-A9 as the detection object, and measure the distance between each convex point and the surrounding convex points; Second, compare the measured distance value with the preset standard value, if the difference between the two exceeds the preset threshold, the convex point will be excluded from the detection object. For example, if the distance between convex point A5 and convex points A2, A4, A6, A8 exceeds the preset standard value, it is determined that the convex point A5 does not meet the condition, and it is excluded from the detection object.

[0026] Finally, take the remaining convex points as the detection object to determine the positioning center of the chip.

[0027] Still take Figure 1 as an example to illustrate step 2. Take the convex points A1-A9 in the standard image as the detection object, and detect the distance between each convex point and its adjacent convex point. For example, if the difference between the distance of convex point A1 from convex points A2 and A4 and the preset standard value is within the set threshold range, it indicates that the position of the convex point A1 is accurate. Suppose the difference between the distance of convex point A5 from the surrounding convex points and the preset standard value exceeds the set threshold range, it indicates that the position of the convex point A5 is not accurate, and A5 is excluded. Finally, take the remaining detected convex points A1-A4 and A6-A9 as the detection object to determine the positioning center of the chip.

[0028] Further, the threshold in step 2 adopts a dynamic adjustment threshold, which is self-adaptive threshold adjustment according to historical data and environmental changes. Specifically, the self-adaptive threshold adjustment according to environmental changes is corresponding dynamic adjustment of the threshold according to environmental temperature drift. The reason for adopting the dynamic adjustment threshold is that in actual application, based on the fineness of the chip, the spacing between the bumps detected by the same chip will change at different environmental temperatures. For example: the spacing difference between the bumps at an environmental temperature of 15°C and an environmental temperature of 25°C can be enough to cause a large difference in the final positioning result. In order to avoid this situation, the present application analyzes the bump spacing change data caused by environmental changes according to historical data, and correspondingly adjusts the threshold according to the environmental temperature drift. In this way, the error caused by the change of the environmental temperature can be corrected, and the chip positioning accuracy is further improved.

[0029] After adopting the method of the present application, the positioning accuracy of the visual algorithm is improved to ±0.1 μm. The segmentation accuracy of the bonded bumps is >98%. In addition, the present application can reduce the frequency of human operation and the mathematical basis for the operator, and simplify the debugging index.

[0030] Of course, the above only describes specific embodiments of the present application and is not intended to limit the scope of the present application. Any equivalent changes or modifications made to the structure, features and principles described in the present application shall be included in the scope of the present application.

Claims

1. A method for chip visual positioning, characterized in that, The method includes the following steps: Step 1: Use a vision system to capture images of the bumps on the chip surface to obtain the original bump images, and preprocess the original bump images to obtain standard images; Step 2: Locate the bumps in the standard image to determine the chip's positioning center. The method used is as follows: First, using several bumps on the chip surface as the detection objects, the distance between each bump and the surrounding bumps is measured. Secondly, the measured spacing value is compared with the preset standard value. If the difference between the two exceeds the preset threshold, the protrusion is removed from the detection object. Finally, the remaining bumps are used as the detection objects to determine the chip's positioning center.

2. The chip visual positioning method according to claim 1, characterized in that: The preprocessing method for the original convex dot image in step 1 is: image background interference preprocessing.

3. The chip visual positioning method according to claim 1, characterized in that: The preprocessing method for the original convex point image in step 1 is as follows: generate a heat map of the convex points, and then use Gaussian fitting to improve the accuracy of the peak points in the heat map.

4. The chip visual positioning method according to claim 1, characterized in that: The threshold in step 2 is a dynamically adjusted threshold, which is adjusted based on historical data and environmental changes.

5. The chip visual positioning method according to claim 4, characterized in that: The threshold adjustment for environmental changes is a dynamic adjustment of the threshold based on the drift of the ambient temperature.

Images