A chip defect detection system and method based on visual recognition

Abstract

The present invention discloses a chip defect detection system and method based on visual recognition, and relates to the technical field of chip defect detection. The present invention comprises: S10: partitioning the surface of an epitaxial wafer, and the inner area of ​​each edge closed line in the grain boundary edge image corresponds to a partition; S20: predicting the defect index of each partition; S30: predicting the resistivity uniformity of each partition; S40: screening the epitaxial wafer with defects. The present invention partitions the surface of the epitaxial wafer to ensure that the infrared light spectrum obtained by the infrared Fourier transform spectrometer can better reflect the defect conditions of each partition, and evaluates the production quality of the epitaxial wafer through multiple physical characteristics of the epitaxial wafer in combination with the resistivity uniformity of each partition, thereby further improving the system's defect detection accuracy for the epitaxial wafer.

Description

Technical Field

[0001] The present invention relates to the technical field of chip defect detection, and in particular to a chip defect detection system and method based on visual recognition. Background Art

[0002] Epitaxy is a type of semiconductor process. In the bipolar process, the bottom layer of the silicon wafer is P-type substrate silicon (sometimes with a buried layer); then a layer of single crystal silicon is grown on the substrate, which is called the epitaxial layer; later, the base region, emitter region, etc. are injected into the epitaxial layer. Finally, a vertical NPN tube structure is basically formed: the epitaxial layer is the collector region, and there are base and emitter regions on the epitaxial layer. The epitaxial wafer is a silicon wafer with an epitaxial layer on the substrate.

[0003] At present, defect detection of epitaxial wafers usually relies on manually placing the epitaxial wafers under a microscope to achieve quality inspection of the epitaxial wafers. The quality inspection standard is usually whether the production specifications of the epitaxial wafers are the same as the standard production specifications. The impact of the uniformity and thickness of the epitaxial wafer resistivity on the production quality of the epitaxial wafers is not taken into account. In addition, the above-mentioned inspection process is not only time-consuming, but also has a relatively low inspection accuracy. Summary of the invention

[0004] The purpose of the present invention is to provide a chip defect detection system and method based on visual recognition to solve the problems raised in the prior art.

[0005] To achieve the above object, the present invention provides the following technical solution: a chip defect detection method based on visual recognition, the method comprising:

[0006] S10: randomly selecting an epitaxial layer from the grown epitaxial layers, placing the selected epitaxial layer on a detection platform, the epitaxial wafer including the epitaxial layer and the substrate, collecting a surface image of the epitaxial wafer by an image acquisition device, graying and edge detecting the collected surface image to obtain a grain boundary edge image, partitioning the surface of the epitaxial wafer based on the grain boundary edge image, and the inner area of ​​each edge closed line in the grain boundary edge image corresponds to a partition;

[0007] S20: taking the central area of ​​each partition as the incident area of ​​the infrared light emitted by the infrared Fourier transform spectrometer, and predicting the defect index of each partition according to the infrared spectrum diagram generated by the infrared Fourier transform spectrometer;

[0008] S30: According to the grain boundary edge image, the image corresponding to each partition is intercepted, and based on the brightness value of each pixel point in each intercepted image, the pixel blocks existing in each intercepted image are analyzed, and the resistivity uniformity of each partition is predicted in combination with the positional relationship between the pixel blocks;

[0009] S40: Screening the defective epitaxial wafers.

[0010] Further, the S20 includes:

[0011] S201: numbering each partition on the surface of the epitaxial wafer, the numbering result is: i=1,2,…,m; m represents the total number of partitions on the surface of the epitaxial wafer, according to the minimum diameter d of the i-th partition i , and the incident angle γ of the infrared light emitted by the infrared Fourier transform spectrometer, the radius R of the central area of ​​the i-th partition i to determine;

[0012] S202: Controlling the infrared Fourier transform spectrometer to emit infrared light to the central area of ​​each partition, marking the extreme values ​​in the infrared spectrum of each partition according to the infrared spectrum of each partition obtained by the infrared Fourier transform spectrometer, and numbering the marked extreme values ​​in the order of appearance time of the extreme values, and the numbering result is: j=1,2,…,n; n represents the total number of extreme values ​​in the infrared spectrum;

[0013] According to the series corresponding to each extreme value and the wavelength at each extreme value, the thickness of the epitaxial wafer in the central area of ​​each partition is predicted. The specific prediction formula is:

[0014] T i =(1 / n)*∑{(Q ij -0.5)*[0.001*h ij / Ö(p 2 -sinγ 2 )]};

[0015] Among them, Q ij represents the level corresponding to the jth marked extreme value in the infrared spectrum of the i-th partition, h j represents the wavelength at the jth marked extreme value in the infrared spectrum of the i-th partition, T i represents the thickness of the epitaxial wafer in the central area of ​​the i-th partition, Ö represents the square root, ∑ represents the summation symbol, the subscript of ∑ is j=1, and the superscript is n;

[0016] S203: According to G i =|fT i | / f predicts the defect index of the i-th partition. Analyzing the defect index of the epitaxial wafer through the defect index of each partition is conducive to improving the detection accuracy of epitaxial wafer defects.

[0017] Furthermore, the central area radius R of the i-th partition is i The specific method for determination is:

[0018] According to the production specifications of the epitaxial wafer, the standard thickness value f of the epitaxial wafer is obtained, and the maximum action distance of infrared light on the epitaxial wafer is calculated according to r=2*f(1+w)*tan[arcsin(sinγ / p)], where w represents the maximum thickness error rate of the epitaxial wafer, and p represents the refractive index of the epitaxial wafer;

[0019] The radius of the central area of ​​the i-th partition is R i =d i -r, the central area refers to R i The semicircular area with a radius of , takes the center of the i-th partition as the origin to construct a plane coordinate system, then the center coordinates of the central area of ​​the i-th partition are (d i / 2-r,0); By controlling the incident infrared light in the central area, the gaps between the grain boundaries can be avoided from affecting the thickness calculation results of the epitaxial wafer, further improving the defect detection effect of the chip.

[0020] Further, the S30 includes:

[0021] S301: Based on the grain boundary edge image, the image corresponding to the i-th partition is intercepted, and the image brightness is calculated according to the image brightness calculation formula L ix =0.3*A ix +0.59*C ix +0.11*B ix Calculate the brightness value of the pixel numbered x in the intercepted image of the i-th partition, where x=1,2,…,X, represents the number corresponding to each pixel in the intercepted image, X represents the total number of pixels, and A ix , C ix , B ix They represent the values ​​of the pixel numbered x in the intercepted image of the i-th partition in the red, green and blue color channels respectively;

[0022] S302: When Ky≤L ix ≤K+y, the pixel numbered x is marked in the intercepted image of the i-th partition, and the marked edge frame in the intercepted image of the i-th partition is determined according to the marking result. The marked edge frame refers to the edge closed line of a pixel block composed of multiple adjacent marked pixel points, wherein K represents the standard brightness value of the epitaxial wafer determined according to the production specifications of the epitaxial wafer, and y represents the error value;

[0023] When Ky>L ix or L ix >K+y, the pixel numbered x in the intercepted image of the i-th partition is not marked;

[0024] S303: Calculate the minimum distance value E between the selected pixel block in the intercepted image of the i-th partition and the pixel block numbered c according to the distance formula between two points.ic Calculate, c=1,2,…,C, represents the number corresponding to each pixel block in the intercepted image except the selected pixel block, and C represents the total number;

[0025] according to The resistivity uniformity of the i-th partition is predicted, where C i represents the total number of pixel blocks in the captured image of the i-th partition except the selected pixel block, U i Represents the resistivity uniformity of the i-th partition. The doping distribution of each partition is analyzed through the brightness value of each pixel in the image corresponding to each partition, and the resistivity uniformity of each partition is analyzed. This process does not require the use of specific instruments, which is convenient for quickly analyzing the production quality of epitaxial wafers, and further improves the system's defect detection efficiency for epitaxial wafer chips.

[0026] Further, the S40: according to The quality index of the selected epitaxial layer is calculated. If T> the set threshold, the selected epitaxial layer is retained. If T< the set threshold, it is considered that the selected epitaxial layer has defects and the selected epitaxial layer is screened out. The production quality of the epitaxial wafer is analyzed based on multiple physical characteristics of the epitaxial wafer to ensure the accuracy of the analysis results.

[0027] A chip defect detection system based on visual recognition, the system comprising a partition processing module, a defect index prediction module, a resistivity uniformity prediction module and a chip screening module;

[0028] The partition processing module is used to grayscale and perform edge detection processing on the epitaxial wafer surface image acquired by the image acquisition device, and obtain a grain boundary edge image, and perform partition processing on the epitaxial wafer surface based on the grain boundary edge image;

[0029] The defect index prediction module is used to predict the defect index of each partition of the epitaxial wafer;

[0030] The resistivity uniformity prediction module is used to predict the resistivity uniformity of each partition of the epitaxial wafer;

[0031] The chip screening module is used to screen defective epitaxial wafers.

[0032] Furthermore, the partition processing module randomly selects an epitaxial layer from the grown epitaxial layers, and places the selected epitaxial layer on the detection platform. The epitaxial wafer includes an epitaxial layer and a substrate. The surface image of the epitaxial wafer is collected by an image acquisition device, and the collected surface image is grayed and edge detected to obtain a grain boundary edge image. Based on the grain boundary edge image, the surface of the epitaxial wafer is partitioned, and the internal area of ​​each edge closed line corresponds to a partition.

[0033] Further, the defect index prediction module includes a central area radius determination unit, an epitaxial wafer thickness prediction unit and a defect index prediction unit;

[0034] The central area radius determination unit determines the central area radius of each partition according to the minimum diameter of each partition, the incident angle of the infrared light emitted by the infrared Fourier transform spectrometer, and the production specifications of the epitaxial wafer;

[0035] The epitaxial wafer thickness prediction unit predicts the epitaxial wafer thickness value in the central area of ​​each partition according to the infrared spectrum of each partition obtained by the infrared Fourier transform spectrometer;

[0036] The defect index prediction unit predicts the defect index of each partition according to the prediction result transmitted by the epitaxial wafer thickness prediction unit.

[0037] Further, the resistivity uniformity prediction module includes a brightness value calculation unit, a marking unit and a resistivity uniformity prediction unit;

[0038] The brightness value calculation unit intercepts the image corresponding to each partition according to the grain boundary edge image, and calculates the brightness value of each pixel point in the intercepted image of each partition by using a brightness calculation formula;

[0039] The marking unit inputs the calculation result transmitted by the brightness value calculation unit into the judgment condition, selects whether to mark the pixel point based on the judgment result, and determines the pixel block formed in the intercepted image of each partition according to the marking result;

[0040] The resistivity uniformity prediction unit predicts the resistivity uniformity of each partition according to the distribution of each pixel block in the intercepted image of each partition.

[0041] Furthermore, the chip screening module calculates the quality index of the selected epitaxial wafer according to the resistivity uniformity of each partition and the defect index of each partition, and selects whether to screen out the selected epitaxial wafer based on the calculation result.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] 1. The present invention performs zoning processing on the surface of the epitaxial wafer to ensure that the infrared light spectrum obtained by the infrared Fourier transform spectrometer can better reflect the defect conditions of each zone. Combined with the resistivity uniformity of each zone, the production quality of the epitaxial wafer is evaluated through multiple physical characteristics of the epitaxial wafer, thereby further improving the system's defect detection accuracy for the epitaxial wafer.

[0044] 2. The present invention analyzes the doping distribution of each partition through the brightness value of each pixel point in the image corresponding to each partition, thereby realizing the analysis of the resistivity uniformity of each partition. This process does not require the aid of specific instruments or manual detection, which facilitates rapid analysis of the production quality of epitaxial wafers and further improves the system's defect detection efficiency for epitaxial wafer chips.

[0045] 3. The present invention controls the incident infrared light within the central area, thereby preventing the gaps between the grain boundaries from affecting the thickness calculation results of the epitaxial wafer, and further improving the defect detection effect on the chip. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A schematic diagram of the workflow of a chip defect detection system and method based on visual recognition according to the present invention;

[0047] Figure 2 A schematic diagram of the working principle and structure of a chip defect detection system and method based on visual recognition according to the present invention;

[0048] Figure 3 It is a schematic diagram of the epitaxial layer structure of a chip defect detection system and method based on visual recognition of the present invention. DETAILED DESCRIPTION

[0049] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making any creative work shall fall within the scope of protection of the present invention.

[0050] Example: Figure 1 , Figure 2 and Figure 3 As shown, the present invention provides a chip defect detection system and method technical solution based on visual recognition, a chip defect detection method based on visual recognition, the method comprising:

[0051] S10: randomly selecting an epitaxial layer from the grown epitaxial layers, placing the selected epitaxial layer on the detection platform, the epitaxial wafer including the epitaxial layer and the substrate, collecting the surface image of the epitaxial wafer by an image acquisition device, the image acquisition device including a camera, graying the collected surface image, edge detection processing to obtain a grain boundary edge image, the specific method of graying the image, edge detection processing to obtain the edge image is the prior art, based on the grain boundary edge image, the grain boundary is the interface between grains with the same structure but different orientations, the epitaxial wafer surface is partitioned, there are multiple edge closure lines in the grain boundary edge image, the edge closure line is the outline of the grain on the surface of the epitaxial wafer, and the inner area of ​​each edge closure line corresponds to a partition;

[0052] S20: taking the central area of ​​each partition as the incident area of ​​the infrared light emitted by the infrared Fourier transform spectrometer. The infrared Fourier transform spectrometer is an infrared spectrometer developed based on the principle of Fourier transform of the infrared light after interference. The defect index of each partition is predicted according to the infrared spectrum generated by the infrared Fourier transform spectrometer.

[0053] The S20 includes:

[0054] S201: numbering each partition on the surface of the epitaxial wafer, the numbering result is: i=1,2,…,m; m represents the total number of partitions on the surface of the epitaxial wafer, according to the minimum diameter d of the i-th partition i , and the incident angle γ of the infrared light emitted by the infrared Fourier transform spectrometer, the radius R of the central area of ​​the i-th partition i The specific determination method is:

[0055] According to the production specifications of the epitaxial wafer, the standard thickness value f of the epitaxial wafer is obtained, and the maximum action distance of infrared light on the epitaxial wafer is calculated according to r=2*f(1+w)*tan[arcsin(sinγ / p)], where w represents the maximum thickness error rate of the epitaxial wafer, p represents the refractive index of the epitaxial wafer, and r represents the maximum action distance of infrared light emitted by the infrared Fourier transform spectrometer on the epitaxial wafer;

[0056] The radius of the central area of ​​the i-th partition is R i =d i -r, the central area refers to R i The semicircular area with a radius of , takes the center of the i-th partition as the origin to construct a plane coordinate system, then the center coordinates of the central area of ​​the i-th partition are (d i / 2-r,0);

[0057] S202: Controlling the infrared Fourier transform spectrometer to emit infrared light to the central area of ​​each partition, the infrared Fourier transform spectrometer acts on each partition separately, marking the extreme values ​​in the infrared spectrum of each partition according to the infrared spectrum of each partition obtained by the infrared Fourier transform spectrometer, and numbering each marked extreme value according to the order of appearance time of the extreme value, and the numbering result is: j=1,2,…,n; n represents the total number of extreme values ​​in the infrared spectrum;

[0058] According to the series corresponding to each extreme value and the wavelength at each extreme value, the thickness of the epitaxial wafer in the central area of ​​each partition is predicted. The specific prediction formula is:

[0059] T i =(1 / n)*∑{(Q ij -0.5)*[0.001*h ij / Ö(p2 -sinγ 2 )]};

[0060] Among them, Q ij represents the level corresponding to the jth marked extreme value in the infrared spectrum of the i-th partition, h j represents the wavelength at the jth marked extreme value in the infrared spectrum of the i-th partition, T i represents the thickness of the epitaxial wafer in the central area of ​​the i-th partition, Ö represents the square root, ∑ represents the summation symbol, the subscript of ∑ is j=1, and the superscript is n;

[0061] S203: According to G i =|fT i | / f predicts the defect index of the i-th partition, G i represents the defect index of the ith partition;

[0062] S30: According to the grain boundary edge image, the image corresponding to each partition is intercepted, and based on the brightness value of each pixel point in each intercepted image, the pixel blocks existing in each intercepted image are analyzed, and the resistivity uniformity of each partition is predicted in combination with the positional relationship between the pixel blocks;

[0063] S30 includes:

[0064] S301: Based on the grain boundary edge image, the image corresponding to the i-th partition is intercepted, and the image brightness is calculated according to the image brightness calculation formula L ix =0.3*A ix +0.59*C ix +0.11*B ix Calculate the brightness value of the pixel numbered x in the intercepted image of the i-th partition, where x=1,2,…,X, represents the number corresponding to each pixel in the intercepted image, X represents the total number of pixels, and A ix , C ix , B ix They represent the values ​​of the pixel numbered x in the intercepted image of the i-th partition in the red, green, and blue color channels, respectively. ix represents the brightness value of the pixel numbered x in the intercepted image of the i-th partition;

[0065] S302: When Ky≤L ix ≤K+y, the pixel numbered x is marked in the intercepted image of the i-th partition, and the marked edge frame in the intercepted image of the i-th partition is determined according to the marking result. The marked edge frame refers to the edge closed line of a pixel block composed of multiple adjacent marked pixel points, wherein K represents the standard brightness value of the epitaxial wafer determined according to the production specifications of the epitaxial wafer, and y represents the error value;

[0066] When Ky>Lix or L ix >K+y, the pixel numbered x in the intercepted image of the i-th partition is not marked;

[0067] S303: construct a plane coordinate system with any point in the intercepted image of the i-th partition as the coordinate origin, determine the coordinates of any pixel point in the intercepted image of the i-th partition based on the constructed plane coordinate system, and calculate the minimum distance value E between the selected pixel block in the intercepted image of the i-th partition and the pixel block numbered c according to the distance formula between two points. ic Calculate, c=1,2,…,C, represents the number corresponding to each pixel block in the intercepted image except the selected pixel block, and C represents the total number;

[0068] according to The resistivity uniformity of the i-th partition is predicted, where C i represents the total number of pixel blocks in the captured image of the i-th partition except the selected pixel block, U i represents the resistivity uniformity of the ith partition;

[0069] S40: screening the defective epitaxial wafers;

[0070] S40: According to The quality index of the selected epitaxial layer is calculated, and T represents the quality index of the selected epitaxial layer. If T>the set threshold, the selected epitaxial layer is retained. If T<the set threshold, it is considered that the selected epitaxial layer has defects and the selected epitaxial layer is screened out.

[0071] A chip defect detection system based on visual recognition, the system includes a partition processing module, a defect index prediction module, a resistivity uniformity prediction module and a chip screening module;

[0072] The partition processing module is used to grayscale and detect the edge of the epitaxial wafer surface image collected by the image acquisition device, and obtain the grain boundary edge image. Based on the grain boundary edge image, the epitaxial wafer surface is partitioned.

[0073] The partition processing module randomly selects an epitaxial layer from the grown epitaxial layer, and places the selected epitaxial layer on the detection platform. The epitaxial wafer includes the epitaxial layer and the substrate. The surface image of the epitaxial wafer is collected by the image acquisition device, and the collected surface image is grayed and edge detected to obtain a grain boundary edge image. Based on the grain boundary edge image, the surface of the epitaxial wafer is partitioned, and the inner area of ​​each edge closed line corresponds to a partition;

[0074] The defect index prediction module is used to predict the defect index of each partition of the epitaxial wafer;

[0075] The defect index prediction module includes a central area radius determination unit, an epitaxial wafer thickness prediction unit and a defect index prediction unit;

[0076] The central area radius determination unit determines the central area radius of each partition according to the minimum diameter of each partition, the incident angle of the infrared light emitted by the infrared Fourier transform spectrometer, and the production specifications of the epitaxial wafer;

[0077] The epitaxial wafer thickness prediction unit predicts the epitaxial wafer thickness value in the central area of ​​each partition according to the infrared spectrum of each partition obtained by the infrared Fourier transform spectrometer;

[0078] The defect index prediction unit predicts the defect index of each partition according to the prediction result transmitted by the epitaxial wafer thickness prediction unit;

[0079] The resistivity uniformity prediction module is used to predict the resistivity uniformity of each partition of the epitaxial wafer;

[0080] The resistivity uniformity prediction module includes a brightness value calculation unit, a marking unit and a resistivity uniformity prediction unit;

[0081] The brightness value calculation unit intercepts the image corresponding to each partition according to the grain boundary edge image, and calculates the brightness value of each pixel point in the intercepted image of each partition by using a brightness calculation formula;

[0082] The marking unit inputs the calculation result transmitted by the brightness value calculation unit into the judgment condition, selects whether to mark the pixel point based on the judgment result, and determines the pixel block formed in the intercepted image of each partition according to the marking result;

[0083] The resistivity uniformity prediction unit predicts the resistivity uniformity of each partition according to the distribution of each pixel block in the intercepted image of each partition;

[0084] The chip screening module is used to screen defective epitaxial wafers;

[0085] The chip screening module calculates the quality index of the selected epitaxial wafer according to the resistivity uniformity of each partition and the defect index of each partition, and chooses whether to screen out the selected epitaxial wafer based on the calculation results.

[0086] Example 1: Assuming that the incident angle γ of the infrared light emitted by the infrared Fourier transform spectrometer is 30°, the refractive index of the epitaxial wafer is p=1.2, the standard thickness value of the epitaxial wafer is f=50 μm, and the maximum thickness error rate of the epitaxial wafer is w=0.05, then the maximum action distance of the infrared light emitted by the infrared Fourier transform spectrometer on the epitaxial wafer is:

[0087] r=2*f(1+w)*tan[arcsin(sinγ / p)]

[0088] =2*50(1+0.05)*tan[arcsin(sin30° / 1.2)]

[0089] =49.35μm;

[0090] Let the minimum diameter of the first partition be d 1 =100 μm;

[0091] Then the radius of the center area of ​​the first partition is R 1 =d 1 -r=100μm-49.35μm=50.65μm;

[0092] The center coordinates of the central area of ​​the i-th partition are (0.65, 0).

[0093] Embodiment 2: According to Figure 3 It can be seen that 1 represents the epitaxial wafer in the epitaxial layer, 2 represents the substrate in the epitaxial layer, 3 represents the image acquisition device, the side of the epitaxial wafer away from the substrate is the surface of the epitaxial wafer, and 4 represents the edge closure line.

[0094] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A chip defect detection method based on visual recognition, characterized in that: The method comprises: S10: randomly selecting an epitaxial layer from the grown epitaxial layers, placing the selected epitaxial layer on a detection platform, the epitaxial wafer including the epitaxial layer and the substrate, collecting a surface image of the epitaxial wafer by an image acquisition device, graying and edge detecting the collected surface image to obtain a grain boundary edge image, partitioning the surface of the epitaxial wafer based on the grain boundary edge image, and the inner area of ​​each edge closed line in the grain boundary edge image corresponds to a partition; S20: taking the central area of ​​each partition as the incident area of ​​the infrared light emitted by the infrared Fourier transform spectrometer, and predicting the defect index of each partition according to the infrared spectrum diagram generated by the infrared Fourier transform spectrometer; The S20 includes: S201: numbering each partition on the surface of the epitaxial wafer, the numbering result is: i=1,2,…,m; m represents the total number of partitions on the surface of the epitaxial wafer, according to the minimum diameter d of the i-th partition i , and the incident angle γ of the infrared light emitted by the infrared Fourier transform spectrometer, the radius R of the central area of ​​the i-th partition i To determine, the specific method is: According to the production specifications of the epitaxial wafer, the standard thickness value f of the epitaxial wafer is obtained, and the maximum action distance of infrared light on the epitaxial wafer is calculated according to r=2*f(1+w)*tan[arcsin(sinγ / p)], where w represents the maximum thickness error rate of the epitaxial wafer, and p represents the refractive index of the epitaxial wafer; The radius of the central area of ​​the i-th partition is R i =d i -r, the central area refers to R i The semicircular area with a radius of , takes the center of the i-th partition as the origin to construct a plane coordinate system, then the center coordinates of the central area of ​​the i-th partition are (d i / 2-r,0); S202: Controlling the infrared Fourier transform spectrometer to emit infrared light to the central area of ​​each partition, marking the extreme values ​​in the infrared spectrum of each partition according to the infrared spectrum of each partition obtained by the infrared Fourier transform spectrometer, and numbering the marked extreme values ​​in the order of appearance time of the extreme values, and the numbering result is: j=1,2,…,n; n represents the total number of extreme values ​​in the infrared spectrum; According to the series corresponding to each extreme value and the wavelength at each extreme value, the thickness of the epitaxial wafer in the central area of ​​each partition is predicted. The specific prediction formula is: T i =(1 / n)*∑{(Q ij −0.5)*[0.001*h ij / Ö(p 2 -synγ 2 )]}; Among them, Q ij represents the level corresponding to the jth marked extreme value in the infrared spectrum of the i-th partition, h j represents the wavelength at the jth marked extreme value in the infrared spectrum of the i-th partition, T i represents the thickness of the epitaxial wafer in the central area of ​​the i-th partition, Ö represents the square root, ∑ represents the summation symbol, the subscript of ∑ is j=1, and the superscript is n; S203: According to G i =|fT i | / f predicts the defect index of the i-th partition; S30: According to the grain boundary edge image, the image corresponding to each partition is intercepted, and based on the brightness value of each pixel point in each intercepted image, the pixel blocks existing in each intercepted image are analyzed, and the resistivity uniformity of each partition is predicted in combination with the positional relationship between the pixel blocks; The S30 includes: S301: Based on the grain boundary edge image, the image corresponding to the i-th partition is intercepted and the image brightness is calculated according to the image brightness calculation formula L ix =0.3*A ix +0.59*C ix +0.11*B ix Calculate the brightness value of the pixel numbered x in the intercepted image of the i-th partition, where x=1,2,…,X, represents the number corresponding to each pixel in the intercepted image, X represents the total number of pixels, and A ix , C ix , B ix They represent the values ​​of the pixel numbered x in the intercepted image of the i-th partition in the red, green and blue color channels respectively; S302: When Ky≤L ix ≤K+y, the pixel numbered x is marked in the intercepted image of the i-th partition, and the marked edge frame in the intercepted image of the i-th partition is determined according to the marking result. The marked edge frame refers to the edge closed line of a pixel block composed of multiple adjacent marked pixel points, wherein K represents the standard brightness value of the epitaxial wafer determined according to the production specifications of the epitaxial wafer, and y represents the error value; When Ky>L ix or L ix >K+y, the pixel numbered x in the intercepted image of the i-th partition is not marked; S303: Calculate the minimum distance value E between the selected pixel block in the intercepted image of the i-th partition and the pixel block numbered c according to the distance formula between two points ic Calculate, c=1,2,…,C, represents the numbers corresponding to the pixel blocks other than the selected pixel block in the intercepted image, and C represents the total number; according to The resistivity uniformity of the i-th partition is predicted, where C i represents the total number of pixel blocks in the captured image of the i-th partition except the selected pixel block, U i represents the resistivity uniformity of the ith partition; S40: screening the defective epitaxial wafers; The S40: The quality index of the selected epitaxial layer is calculated. If T>the set threshold, the selected epitaxial layer is retained. If T<the set threshold, it is considered that the selected epitaxial layer has defects and the selected epitaxial layer is screened out.

2. A chip defect detection system based on visual recognition applied to the chip defect detection method based on visual recognition according to claim 1, characterized in that: The system includes a partition processing module, a defect index prediction module, a resistivity uniformity prediction module and a chip screening module; The partition processing module is used to grayscale and perform edge detection processing on the epitaxial wafer surface image acquired by the image acquisition device, and obtain a grain boundary edge image, and perform partition processing on the epitaxial wafer surface based on the grain boundary edge image; The defect index prediction module is used to predict the defect index of each partition of the epitaxial wafer; The resistivity uniformity prediction module is used to predict the resistivity uniformity of each partition of the epitaxial wafer; The chip screening module is used to screen defective epitaxial wafers.

3. The chip defect detection system based on visual recognition according to claim 2, characterized in that: The partition processing module randomly selects an epitaxial layer from the grown epitaxial layers, and places the selected epitaxial layer on the detection platform. The epitaxial wafer includes an epitaxial layer and a substrate. The surface image of the epitaxial wafer is collected by an image acquisition device, and the collected surface image is grayed and edge detected to obtain a grain boundary edge image. Based on the grain boundary edge image, the surface of the epitaxial wafer is partitioned, and the internal area of ​​each edge closed line corresponds to a partition.

4. The chip defect detection system based on visual recognition according to claim 3, characterized in that: The defect index prediction module includes a central area radius determination unit, an epitaxial wafer thickness prediction unit and a defect index prediction unit; The central area radius determination unit determines the central area radius of each partition according to the minimum diameter of each partition, the incident angle of the infrared light emitted by the infrared Fourier transform spectrometer, and the production specifications of the epitaxial wafer; The epitaxial wafer thickness prediction unit predicts the epitaxial wafer thickness value in the central area of ​​each partition according to the infrared spectrum of each partition obtained by the infrared Fourier transform spectrometer; The defect index prediction unit predicts the defect index of each partition according to the prediction result transmitted by the epitaxial wafer thickness prediction unit.

5. The chip defect detection system based on visual recognition according to claim 4, characterized in that: The resistivity uniformity prediction module includes a brightness value calculation unit, a marking unit and a resistivity uniformity prediction unit; The brightness value calculation unit intercepts the image corresponding to each partition according to the grain boundary edge image, and calculates the brightness value of each pixel point in the intercepted image of each partition by using a brightness calculation formula; The marking unit inputs the calculation result transmitted by the brightness value calculation unit into the judgment condition, selects whether to mark the pixel point based on the judgment result, and determines the pixel block formed in the intercepted image of each partition according to the marking result; The resistivity uniformity prediction unit predicts the resistivity uniformity of each partition according to the distribution of each pixel block in the intercepted image of each partition.

6. The chip defect detection system based on visual recognition according to claim 5, characterized in that: The chip screening module calculates the quality index of the selected epitaxial wafer according to the resistivity uniformity of each partition and the defect index of each partition, and selects whether to screen out the selected epitaxial wafer based on the calculation result.

Images