Substrate surface defect recognition method and detection system

Abstract

This invention discloses a method and detection system for identifying defects on substrate surfaces, belonging to the field of precision manufacturing and automated optical inspection technology. The method includes: establishing an identification benchmark and a defect feature library; dividing the substrate surface into identification points in different substrate regions; collecting steady-state chromaticity values ​​of normal substrates under different wavelengths of detection light at each identification point and dynamic time-series response values ​​during wavelength switching to form a comprehensive identification benchmark including curve range and time-series response range; simultaneously, collecting corresponding features of known defect types to construct a defect feature library; during detection, simultaneously acquiring the steady-state spectral curve and dynamic time-series response values ​​of the substrate under test, first comparing them with the identification benchmark to determine anomalies, and then matching and calculating the similarity between the abnormal features and the defect feature library to achieve accurate identification of defect types. This invention improves the depth and intelligence level of defect detection.

Description

Technical Field

[0001] This invention relates to the fields of precision manufacturing and automated optical inspection technology, and more specifically, to a method and system for identifying defects on the surface of a substrate. Background Technology

[0002] In the electronics industry, the cleanliness and chemical state of the substrate (such as DBC, ceramic substrate) directly affect the reliability of the product. After processes such as soldering and high-temperature treatment, various defects may appear on the substrate surface, such as solder joints turning blue (Sn / Pb oxidation), base plate turning black (Ni oxidation), and flux residue. These defects are usually similar in microscopic and color characteristics, but their causes and hazards are different.

[0003] Currently, mainstream automated optical inspection equipment mainly uses single or composite white light sources to illuminate and acquire two-dimensional images of the substrate surface. Anomalies are detected by comparing the differences in color, brightness, or texture with a standard template. This method has significant limitations: First, it is essentially a "difference detection" method, which can only determine "whether it is different from the standard" but cannot answer "what kind of difference," that is, it cannot identify the specific type of defect. Second, for defects with similar static optical characteristics but different material composition or physical state (such as different metal oxides or different forms of contaminants), it is difficult to effectively distinguish them by static image information alone, which can easily lead to misjudgment or classification failure and cannot provide accurate feedback for process traceability. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and detection system for identifying surface defects on a substrate, so as to solve the problems that the prior art cannot accurately classify surface defects and is insensitive to the microscopic physicochemical state of defects.

[0005] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a method for identifying defects on a substrate surface, comprising the following steps: S1. Establishing an identification benchmark and defect feature library: S11. Obtaining multiple normal substrate samples and performing the following sub-steps: S111. Dividing the substrate surface into multiple identification points according to the substrate distribution, each identification point corresponding to an identification area composed of a single substrate, specifically including dividing the substrate surface according to exposed areas, pad areas, plating areas, component areas, etc., so that the identification area corresponding to each identification point is composed of a single substrate; S112. Irradiating the substrate with at least two different wavelengths of detection light. On the board surface, for each identification point, the steady-state chromaticity value of the surface is collected under the illumination of detection light at each wavelength. When switching between different wavelengths of detection light, the temporal response value is collected, i.e., the rate of change and relaxation time of the surface chromaticity response. The rate of change of the surface chromaticity response refers to the instantaneous speed at which the surface chromaticity value changes with wavelength switching, and the relaxation time refers to the time constant required for the surface chromaticity to reach a steady state when the wavelength is switched. S113: Based on the steady-state chromaticity value of each identification point, a reference curve of its surface chromaticity with respect to the detection light illumination wavelength is generated. The range of the reference curve is determined based on the curve data of multiple normal substrate samples. Based on the temporal response value of each identification point, the temporal response is determined. The reference range, i.e., the range of its surface chromaticity response change rate and relaxation time, together with the range of the reference curve and the time response range, constitutes the identification reference for this identification point. The identification reference is dynamically improved. As the amount of normal substrate sample data acquired increases, the range of the reference curve and the time response range are continuously supplemented and optimized; S12. For specific defect types, acquire various defect samples, and execute steps S111 and S112 to extract the curve features of the surface steady-state chromaticity of various defect samples with respect to the wavelength of the detection light and / or the change rate and relaxation time features of the surface chromaticity response, and store them as defect features in the defect feature library; S2. Detection of the substrate to be tested: S21. Step S111 of the test substrate is to divide the identification points; Step S22 of the test substrate is to execute step S112 to obtain the steady-state curve and timing response value of each identification point, that is, the curve of the steady-state chromaticity of the surface under test with respect to the wavelength of the detection light, as well as the rate of change and relaxation time of the chromaticity response of the surface under test; Step S3 of the surface defect identification: Step S31 of the test substrate is to compare the steady-state curve and timing response value of each identification point with the range and timing response range of the reference curve of the corresponding point in the identification reference; Step S32 of the test substrate is to determine that there is a surface defect at the identification point if the steady-state curve and / or timing response value deviates from the range and timing response range of the reference curve of the corresponding point.S33. For identification points where surface defects are determined to exist, the features of their measured steady-state curves and / or measured time-series response values ​​are matched and similarity calculated in real time with various defect features in the defect feature library. The defect type is identified based on the similarity results. The similarity calculation employs at least one of weighted Euclidean distance, cosine similarity, or machine learning-based classification algorithms.

[0006] In another aspect, the present invention provides a substrate surface defect detection system for implementing the substrate surface defect identification method described above. The system includes an optical detection mechanism, which includes a multi-light source module, an image acquisition module, and a control and processing unit. The multi-light source module is used to generate and switch at least two different wavelengths of detection light. The image acquisition module is used to acquire images of the substrate surface. The control and processing unit is communicatively connected to the optical detection mechanism. The control and processing unit is used to control the multi-light source module to switch sequentially and synchronously trigger the image acquisition module, process the image to divide the identification points and calculate the surface steady-state chromaticity value, the rate of change of surface chromaticity response, and the relaxation time of each point, store and manage the identification benchmark and defect feature library, perform comparative analysis of the steady-state curve and the time-series response value to be tested, and perform matching and similarity calculation of defect features.

[0007] Beneficial Effects: In summary, this invention provides a method and system for identifying defects on substrate surfaces. By introducing a dual-dimensional spectral feature analysis and a dual-layer database comparison mechanism, it not only analyzes the steady-state chromaticity characteristics of the material surface under different wavelengths of detection light, but also captures its dynamic temporal response process when the light changes, namely the rate of change and relaxation time of its surface chromaticity response. It innovatively introduces dynamic temporal response (rate of change and relaxation time) as a key feature. This feature is extremely sensitive to the physical properties of the surface material, such as microstructure, crystal morphology, and adhesion, and can effectively distinguish defects with similar static spectra but different physical states, significantly improving the system's resolution and anti-interference ability.

[0008] Other features and advantages of the present invention will be set forth in the following description. Attached Figure Description

[0009] To more clearly illustrate the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic flowchart of a substrate surface defect identification method according to an embodiment of the present invention; Figure 2This is a schematic diagram of a substrate surface defect detection system according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a reference curve in the CIE Lab color space according to an embodiment of the present invention. Detailed Implementation

[0011] The present invention will be described below with reference to specific embodiments. It should be noted that the embodiments described below are examples of the present invention and are only used to illustrate the present invention, and are not intended to limit the present invention. Other combinations and various modifications within the scope of the present invention can be made without departing from the spirit or scope of the present invention.

[0012] This embodiment provides a method for identifying defects on the substrate surface. Figure 1 This is a flowchart illustrating a method for identifying surface defects on a substrate according to an embodiment of the present invention. Figure 1As shown, it includes the following steps: S1. Establishing an identification benchmark and defect feature library; S11. Obtaining multiple normal substrate samples and performing the following sub-steps: S111. Dividing the substrate surface into multiple identification points according to the substrate distribution on the substrate surface, each identification point corresponds to an identification area composed of a single substrate; S112. Irradiating the substrate surface with at least two different wavelengths of detection light, and for each identification point, collecting its surface steady-state chromaticity value under each wavelength of detection light irradiation, and collecting its time-series response value, i.e., its surface chromaticity response, when switching between different wavelengths of detection light irradiation. The rate of change and relaxation time; S113. Based on the steady-state chromaticity value of each identification point, generate a reference curve of its surface chromaticity with respect to the wavelength of the detection light, and determine the range of the reference curve according to the curve data of multiple normal substrate samples. Based on the time response value of each identification point, determine the time response range, that is, the reference range of its surface chromaticity response rate of change and relaxation time. The range of the reference curve and the time response range together constitute the identification reference of the identification point; S12. For specific defect types, obtain various defect samples, and execute steps S111 and S112 to extract various types of defects. The surface steady-state chromaticity curve characteristics of the defect sample with respect to the wavelength of the detection light and / or the rate of change and relaxation time characteristics of the surface chromaticity response are stored as defect features in the defect feature library; S2, detection of the substrate under test: S21, perform step S111 on the substrate under test to divide the identification points; S22, perform step S112 on the substrate under test to obtain the steady-state curve and timing response value of each identification point, that is, the curve of the steady-state chromaticity of the surface under test with respect to the wavelength of the detection light and the rate of change and relaxation time of the surface chromaticity response; S3, surface defect identification: S31, the surface defect is identified by... The steady-state curve and timing response value of each identification point on the substrate are compared with the range of the reference curve and the timing response range of the corresponding point in the identification reference; S32, if the steady-state curve and / or timing response value deviate from the range of the reference curve and the timing response range of the corresponding point, it is determined that there is a surface defect at the identification point; S33, for the identification points that are determined to have surface defects, the characteristics of the steady-state curve and / or timing response value are matched and similarity calculated in real time with various defect features in the defect feature library, and the defect type is identified based on the similarity result. Specifically, in step S111, the substrate surface is divided into multiple identification points according to the substrate distribution on the substrate surface. Each identification point corresponds to an identification area composed of a single substrate. Specifically, the substrate surface is divided according to the exposed area, pad area, plating area, component area, etc., so that the identification area corresponding to each identification point is composed of a single substrate. Different substrates have different optical properties, such as large differences in reflectivity and absorption spectrum. Mixed substrates can easily lead to the measured colorimetric value being a comprehensive response of multiple materials, resulting in inaccurate characteristic curves and an inability to effectively distinguish between normal and defective materials.

[0013] Specifically, in step S112, at least two different wavelengths of detection light are used to irradiate the substrate surface. Because the detection information of a single wavelength detection light is of a single dimension, it is easy to cause unclear changes and make it difficult to distinguish differences. Therefore, the response of the substrate surface material under different spectral dimensions is obtained by using at least two multi-wavelength detection lights. Different substrates have drastically different sensitivities (i.e., reflection / absorption characteristics) to different wavelengths of light. For example, a certain flux residue may have characteristic absorption in the ultraviolet band, but is almost transparent under visible red light. A metal oxide layer may produce obvious interference color changes in a specific blue-green light band, but does not change much in other bands. If the wavelength of the single wavelength detection light happens to be insensitive to the specific defect that needs to be detected, then even if the defect exists, it is difficult to generate an effective signal difference, thus causing a detection blind zone.

[0014] Specifically, in steps S112 and S22, the rate of change of surface chromaticity response refers to the instantaneous speed at which the surface chromaticity value changes with wavelength switching, and the relaxation time refers to the time constant required for the surface chromaticity to change to a steady state when the wavelength switches. Here, the dynamic temporal response (rate of change and relaxation time) is introduced as a key feature for substrate surface defect detection. The core is that it expands the detection dimension from the traditional static space to the dynamic time domain, thereby revealing the differences in the physical state and microstructure of the material surface. This information is completely impossible to obtain through static spectral analysis (only looking at the steady-state color). The static "detection light wavelength-chromaticity" curve mainly reflects the chemical composition and electronic band structure of the substrate surface material. It represents the material problem on the substrate surface, but static spectral analysis is difficult to describe the form in which this material exists and its structure. The rate of change and relaxation time are directly related to the energy transfer process and excited state lifetime on the substrate surface. For example, a dense, well-crystallized oxide layer and a loose, porous contaminant, even if their chemical compositions are similar, will have different internal carrier mobility and thermal diffusion rates. This will lead to the former responding faster and relaxing more quickly after photoexcitation, while the latter may respond slowly and relax more slowly. That is, by introducing dynamic time-series response, we can not only determine the presence of oxide on the substrate surface, but also further infer whether the oxide layer is dense or not, thus achieving a deeper characterization.

[0015] Specifically, in step S113, the identification benchmark is dynamically improved. As the amount of normal substrate sample data acquired increases, the range of the benchmark curve and the time response range are continuously supplemented and optimized. For example, data that is too large due to error data is eliminated to narrow the range and / or data that is outside the range is included in the range to expand the range because it is normal in subsequent further testing. That is, the identification benchmark can be dynamically improved as qualified product data accumulates, and the defect feature library can also be continuously expanded to include new defect types.

[0016] Specifically, in step S33, the similarity calculation employs at least one of weighted Euclidean distance, cosine similarity, or a machine learning-based classification algorithm.

[0017] Another aspect of this embodiment provides a substrate surface defect detection system. Figure 2 This is a schematic diagram of a substrate surface defect detection system according to an embodiment of the present invention, as shown below. Figure 2 As shown, it includes an optical inspection mechanism, which comprises a multi-light source module, an image acquisition module, and a control and processing unit. The multi-light source module is used to generate and switch at least two different wavelengths of detection light; the image acquisition module is used to acquire images of the substrate surface; the control and processing unit is communicatively connected to the optical inspection mechanism, and is used to control the multi-light source module to switch sequentially and synchronously trigger the image acquisition module, process the image to divide the identification points and calculate the surface steady-state chromaticity value, the rate of change of surface chromaticity response, and relaxation time of each point, store and manage the identification benchmark and defect feature library, perform comparative analysis of the steady-state curve and the timing response value to be tested, and perform matching and similarity calculation of defect features.

[0018] For example, in a specific application scenario, taking the intelligent sorting of surface defects of a power module DBC substrate as an example, this embodiment takes the DBC substrate used in a certain type of IGBT power module as the detection object. After welding, it needs to be treated by a formic acid furnace, and various surface defects often appear. According to this embodiment, the substrate surface defect detection system includes: an optical detection mechanism, including: a multi-light source module: integrating four high-brightness LED light sources of ultraviolet, blue, green and red light, controlled by a driving circuit, which can realize millisecond-level fast sequential switching; Image acquisition module: It adopts a high-resolution area array industrial camera and is equipped with an electric filter wheel. The filter is matched with the wavelength of the light source and the filter is switched synchronously when the light source is switched to ensure the signal-to-noise ratio. Control and Processing Unit: Industrial control computer, implementing the following method for identifying substrate surface defects: S1. Establishing identification benchmarks and defect feature libraries: S11. Acquiring multiple normal DBC substrate samples and executing the following sub-steps: S111. Importing DBC substrate design drawings into the software, identifying "exposed aluminum nitride ceramic areas," "thick copper circuit areas," and "thin copper pad areas" on the DBC substrate according to the substrate distribution on the DBC substrate surface. In each area, generating multiple identification points according to a preset grid size, that is, dividing the DBC substrate surface into multiple identification points, each identification point corresponding to an identification area composed of a single substrate; S112. For each normal DBC substrate sample, controlling the detection light from four light sources—ultraviolet, blue, green, and red—to sequentially switch and illuminate the DBC substrate surface, with the camera simultaneously acquiring images. For each identification point, extracting the chromaticity coordinates of each identification point under the four light sources (ultraviolet, blue, green, and red), for example, CIE... The ab values ​​in the Lab color space, i.e., the surface steady-state chromaticity values ​​under detection light at each wavelength, are used to connect the four chromaticity coordinates to form the steady-state reference curve for that identification point. At the instant of each light source switch, the chromaticity value changes are sampled at a high frame rate to calculate the maximum rate of change Vmax from the steady-state value of the previous wavelength light source to the steady-state value of the current wavelength light source, and the time constant T required for the surface chromaticity to change to steady state when the light source wavelength switches. This is its temporal response value, which is its surface chromaticity response change rate Vmax and relaxation time T; S113, Figure 3 This is a schematic diagram of a reference curve in the CIE Lab color space according to an embodiment of the present invention, as shown below. Figure 3 As shown, for multiple normal DBC substrates located at the same position in the same area, the mean and standard deviation of the identification points at each identification point are calculated based on their steady-state chromaticity values ​​and steady-state reference curves. This yields the range of the reference curve, as shown in the figure. Figure 3The chromaticity coordinates are shown within an area centered on the mean point and with the standard deviation range as the radius. Based on the timing response values ​​of each identification point on multiple normal DBC substrates, the timing response range is determined. Similarly, the mean and standard deviation of Vmax and T can be obtained, thus obtaining the reference range of the rate of change and relaxation time of the surface chromaticity response. The range of the reference curve and the timing response range together constitute the identification reference for this identification point. S12. For specific defect types, such as samples of three known defect types: Type A: bluish, EDS confirmed as Sn / Pb oxide; Type B: black, EDS confirmed as Ni oxide; Type C: dull and lackluster, FTIR confirmed as silicone resin residue; obtain samples of various defects and execute steps S111 and S112 to extract the curve characteristics of the surface steady-state chromaticity of various defect samples with respect to the wavelength of the detection light and / or the rate of change of the surface chromaticity response Vmax. The characteristics of x and relaxation time T are stored as defect features in the defect feature library; S2, detection of the DBC substrate under test: S21, execute step S111 on the DBC substrate under test to divide the identification points; S22, execute step S112 on the DBC substrate under test to obtain the steady-state curve and timing response value of each identification point, that is, the curve of the steady-state chromaticity of the surface under test with respect to the wavelength of the detection light, and the rate of change Vmax and relaxation time T of the chromaticity response of the surface under test; S3, surface defect identification: S31, compare the steady-state curve and timing response value of each identification point on the DBC substrate under test with the range of the reference curve and the timing response range of the corresponding point in the identification reference. For example, for a certain identification point in the "thick copper circuit area", compare the four chromaticity coordinates of its steady-state curve under test with the "curve range" of the corresponding point in the identification reference, that is, determine whether it is located in the area of ​​the thick copper circuit area. Figure 3Within the area shown, centered on the mean point and with the standard deviation range as the radius, the measured Vmax and T values ​​are compared with the time response range. If all data are within the normal range, it is judged as qualified and without surface defects; S32. If the measured steady-state curve and / or the measured time response value deviates from the range of the reference curve and the time response range of the corresponding point, it is determined that the identification point has surface defects. For example, if the chromaticity coordinates of the measured steady-state curve of an identification point in a "thin copper pad area" under blue and green light illumination exceed the above area range, the identification point is judged to be abnormal and has surface defects; S33. For identification points judged to have surface defects, The characteristics of the steady-state curve and / or the timing response value to be tested are matched and similarity calculated in real time with various defect features in the defect feature library. The defect type is identified based on the similarity results. Specifically, the feature vectors of the steady-state curve, Vmax and T at the identification point of the anomaly are compared with the three types of features A, B and C in the defect feature library. The calculation results show that the similarity of the features of the steady-state curve, Vmax and T at the identification point of the anomaly is 0.91 with type A features, 0.65 with type B features, and 0.30 with type C features. That is, the defect type is that the "thin copper pad area" is suspected to have Sn / Pb oxide (matching degree 91%).

[0019] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying defects on a substrate surface, characterized in that, Includes the following steps: S1. Establish an identification benchmark and defect feature library: S11. Obtain multiple normal substrate samples and perform the following sub-steps: S111. Divide the substrate surface into multiple identification points according to the substrate distribution on the substrate surface. Each identification point corresponds to an identification area composed of a single substrate. S112. Irradiate the substrate surface with at least two different wavelengths of detection light. For each identification point, collect its surface steady-state chromaticity value under each wavelength of detection light. When switching between different wavelengths of detection light, collect its time-series response value, i.e., the rate of change and relaxation time of its surface chromaticity response. S113. Based on the steady-state chromaticity value of each identification point, generate a benchmark curve of its surface chromaticity with respect to the wavelength of the detection light. Determine the range of the benchmark curve based on the curve data of multiple normal substrate samples. Based on the time-series response value of each identification point, determine... The timing response range, i.e., the reference range of its surface chromaticity response change rate and relaxation time, together with the range of the reference curve and the timing response range, constitutes the identification reference of the identification point; S12, for a specific defect type, obtain various defect samples, and execute steps S111 and S112 to extract the curve features of the surface steady-state chromaticity of various defect samples with respect to the wavelength of the detection light and / or the change rate and relaxation time features of the surface chromaticity response, and store them as defect features in the defect feature library; S2, detection of the substrate under test: S21, execute step S111 on the substrate under test to divide the identification points; S22, execute step S112 on the substrate under test to obtain the steady-state curve and timing response value of each identification point, i.e., the curve of the steady-state chromaticity of the surface under test with respect to the wavelength of the detection light and the change rate and relaxation time of the surface chromaticity response; S3. Surface Defect Identification: S31. Compare the steady-state curve and timing response value of each identification point on the substrate under test with the range of the reference curve and timing response range of the corresponding point in the identification reference; S32. If the steady-state curve and / or timing response value deviate from the range of the reference curve and timing response range of the corresponding point, it is determined that there is a surface defect at the identification point; S33. For the identification points where surface defects are determined to exist, perform real-time matching and similarity calculation of the characteristics of the steady-state curve and / or timing response value under test with various defect features in the defect feature library, and identify the defect type based on the similarity results.

2. The method according to claim 1, characterized in that, In step S111, the substrate surface is divided according to the substrate distribution on the substrate surface, specifically according to the exposed area, pad area, plating area and component area on the substrate surface, so that the recognition area corresponding to each recognition point is composed of a single substrate.

3. The method according to claim 1, characterized in that, In steps S112 and S22, the rate of change of the surface chromaticity response refers to the instantaneous speed at which the surface chromaticity value changes with wavelength switching, and the relaxation time refers to the time constant required for the surface chromaticity to change to a steady state when the wavelength is switched.

4. The method according to claim 1, characterized in that, In step S113, as the number of normal substrate sample data acquired increases, the curve range and timing response range are continuously supplemented and optimized.

5. The method according to claim 1, characterized in that, In step S33, the similarity calculation employs at least one of weighted Euclidean distance, cosine similarity, or a machine learning-based classification algorithm.

6. A substrate surface defect detection system, used to implement the method as described in any one of claims 1-5, characterized in that, include: An optical inspection mechanism includes: a multi-light source module for generating and switching at least two different wavelengths of inspection light; an image acquisition module for acquiring images of the substrate surface; and a control and processing unit, communicatively connected to the optical inspection mechanism, for: controlling the multi-light source module to switch sequentially and synchronously triggering the image acquisition module; processing the image, dividing it into identification points, and calculating the surface steady-state chromaticity value, the rate of change of surface chromaticity response, and the relaxation time of each point; storing and managing the identification benchmark and defect feature library; performing comparative analysis of the steady-state curve and the time-series response value to be tested, as well as matching and similarity calculation of defect features.

Images