Chip defect detection method based on three-dimensional point cloud grid reconstruction

Abstract

The invention discloses a chip defect detection method based on three-dimensional point cloud grid reconstruction. The chip defect detection method comprises the following steps: step 1, acquiring a three-dimensional point cloud of a reference chip; step 2, de-noising is carried out; step 3, orderly sampling; step 4, gridding processing: constructing a grid network, and dividing each grid into two triangular patches; step 5, grid optimization: adjusting and optimizing the height coordinate of each triangular patch by minimizing the average distance from all the three-dimensional point clouds in each grid to the corresponding triangular patch so as to obtain an optimized three-dimensional grid model; step 6, collecting a three-dimensional point cloud of a target part of the target chip; and step 7, chip defect detection. On the basis of the three-dimensional grid model, the surface defects (such as deficiency, upwarp and offset) of the chip can be effectively detected, the pin defects (such as excessive tin and insufficient tin) can be quickly positioned, the automation degree and the detection precision are high, and the method is particularly suitable for the manufacturing and quality control process of chips with high precision and high quality requirements.

Description

Technical Field

[0001] The present invention relates to the field of chip defect detection in the semiconductor industry, and in particular to a chip defect detection method based on three-dimensional point cloud grid reconstruction. Background Art

[0002] With the rapid development of the semiconductor industry, the complexity and precision requirements of chip manufacturing are becoming increasingly higher. In the chip production process, defect detection is an important link to ensure product quality and reliability. Traditional defect detection methods are mostly based on two-dimensional image processing technology, and the surface of the chip is inspected by optical microscope or X-ray inspection. Although these methods can effectively detect certain surface defects, they often have difficulty in identifying complex three-dimensional defects due to their reliance on two-dimensional images, especially for the detection of tiny defects, defects with uneven depth, or structures such as pins.

[0003] The limitations of two-dimensional detection technology have prompted the application of three-dimensional vision technology in chip defect detection. By acquiring complete three-dimensional data of the chip surface and internal structure, the three-dimensional detection method can analyze chip defects more intuitively and comprehensively, with high detection precision and accuracy. Traditional three-dimensional detection technology usually relies on methods such as laser scanning and CT scanning. Although these technologies can generate high-quality three-dimensional data, their operation is complex, the cost is high, and the real-time detection capability in the production environment is weak. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a chip defect detection method based on three-dimensional point cloud mesh reconstruction in view of the deficiencies of the above-mentioned prior art. The chip defect detection method based on three-dimensional point cloud mesh reconstruction can realize real-time automatic detection of chip body height defects and pin soldering defects.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A chip defect detection method based on three-dimensional point cloud mesh reconstruction includes the following steps.

[0007] Step 1: Collect the three-dimensional point cloud of the reference chip: For each part in the reference chip that needs to be inspected for three-dimensional defects, collect its corresponding three-dimensional point cloud; one part corresponds to one three-dimensional point cloud set.

[0008] Step 2: De-noising: De-noising is performed on each 3D point cloud set collected in step 1.

[0009] Step 3: Ordered sampling: Orderly sampling is performed on each denoised 3D point cloud set to form a corresponding ordered point cloud set.

[0010] Step 4, grid processing: Take all the ordered sampling points in each ordered point cloud set as grid corner points, build a grid network, and then divide each grid into two triangular faces, so as to form a preliminary three-dimensional grid model; wherein, the three vertices of each triangular face are the three ordered sampling point clouds in the corresponding ordered point cloud set.

[0011] Step 5: Mesh optimization: By minimizing the average distance from all denoised 3D point clouds to the corresponding triangular facets in each square, the height coordinates of each triangular facet are adjusted and optimized to obtain an optimized 3D mesh model.

[0012] Step 6: Collect the three-dimensional point cloud of the target chip and target parts: collect and denoise the three-dimensional point cloud of the target parts in the target chip to be inspected.

[0013] Step 7, chip defect detection: By calling the optimized 3D mesh model corresponding to the target part, the target chip can be subjected to a high degree of defect detection.

[0014] In step 3, the denoised three-dimensional point cloud set P out The method for performing ordered sampling comprises the following steps:

[0015] Step 3-1, obtain the extreme value of point cloud coordinates: traverse the three-dimensional point cloud set P out , the point cloud coordinate extremes in the X and X directions are calculated, which are: the maximum value of the X-direction point cloud coordinate Xmax, the minimum value of the X-direction point cloud coordinate Xmin, the maximum value of the Y-direction point cloud coordinate Ymax, and the minimum value of the Y-direction point cloud coordinate Ymin.

[0016] Step 3-2, set the sampling step: for the three-dimensional point cloud set P out Set the X-direction sampling step size t respectively x and Y-sampling step length t y .

[0017] Step 3-3, divide the sampling area: collect the three-dimensional point cloud P out Divide into m regions along the X direction and n regions along the Y direction, where:

[0018]

[0019] Step 3-4, find the ordered sampling points: In the middle of each sampling area divided in step 3-3, select a point as the found ordered sampling point.

[0020] In step 3-3, the three-dimensional point cloud set P outThe i-th intersection area along the X direction and the j-th intersection area along the Y direction are recorded as the sampling area P(i,j); at this time, the sampling area P(i,j) has variable N(i,j) denoised 3D point clouds, which are: p1, p2, p3, ..., p N(i,j) ; In step 3-4, the ordered sampling points found in the sampling area P(i,j) are recorded as s(i,j), and the X-, Y-, and Z-direction coordinates of s(i,j) are recorded as: s(i,j).x, s(i,j).y, and s(i,j).z, respectively; then the calculation formulas of s(i,j).x and s(i,j).y are:

[0021] s(i,k).x=Xmin+(i-1)*t x

[0022] s(i,j).y=Ymin+(j-1)*t y

[0023] s(i,j).z is determined according to the number of N(i,j), specifically:

[0024] D. When N(i,j)=0, and i=1 or j=1, the specific calculation formula is:

[0025] s(i,j).z=Zstd

[0026] Where Zstd is the height of the reference plane, which is obtained by plane fitting the three-dimensional point cloud of the part collected in step 1;

[0027] E. When N(i,j)=0, i≠1 and j≠1, the specific calculation formula is:

[0028]

[0029] Where s(p,q).z is the Z coordinate of the ordered sampling point corresponding to the pth intersection area along the X direction and the qth intersection area along the Y direction; p is i-1, i and i+1; q is j-1, j and j+1;

[0030] N(p,q) is the number of denoised 3D point clouds in the p-th intersection area along the X direction and the q-th intersection area along the Y direction;

[0031] N is the total number of N(p,q)>0;

[0032] is the sum of the Z coordinates of all ordered sampling points in the region N(p,q)>0;

[0033] F. When N(i,j)≥1, the specific calculation formula is:

[0034]

[0035] In the formula, p1.z, p2.z, p3.z, ..., p N(i,j) .z are three-dimensional point clouds p1, p2, p3, ..., p N(i,j) The Z coordinate of .

[0036] In step 4, the method for constructing a preliminary three-dimensional mesh model includes the following steps:

[0037] Step 4-1, find the corner points of the grid: take the ordered sampling point s(i,j) as the starting corner point of a grid A, and search for the adjacent ordered sampling points to obtain the other three corner points of grid A, namely: s(i+1,j), s(i,j+1), s(i+1,j+1).

[0038] Step 4-2, construct triangular patches: Take s(i,j) in square A as the starting point and construct two triangular patches.

[0039] Step 4-3, construct a preliminary three-dimensional mesh model: In step 3-4, a total of m*n ordered sampling points are obtained. Repeat steps 4-1 to 4-2, traverse each ordered sampling point, and obtain a total of (m-1)*(n-1) squares, and then obtain a preliminary three-dimensional mesh model M containing 2*(m-1)*(n-1) triangles. in .

[0040] In step 4-2, the two triangles of square A are V1(i,j) and V2(i,j), which can be constructed in two ways:

[0041] Construction method 1: V1(i,j) = {s(i,j), s(i+1,j), s(i+1,j+1)}, V2(i,j) = {s(i,j), s(i,j+1), s(i+1,j+1)}.

[0042] Construction method 2: V1(i,j) = {s(i,j), s(i+1,j), s(i,j+1)}, V2(i,j) = {s(i+1,j), s(i,j+1), s(i+1,j+1)}.

[0043] In step 5, the grid optimization method includes the following steps:

[0044] Step 5-1, set the minimum distance threshold Thres of the patch.

[0045] Step 5-1, extract 3D point cloud: The denoised 3D point clouds corresponding to the two triangular patches V1(i,j) and V2(i,j) in grid A are recorded as 3D point cloud sets P1(i,j) and P2(i,j), respectively; where P1(i,j) = {p11, p12, p13…p1n1}, P2(i,j) = {p21, p22, p23…p2n2}; where n1 and n2 are the numbers of 3D point clouds corresponding to the triangular patches V1(i,j) and V2(i,j), respectively.

[0046] Step 5-2, calculate d(i,j): d(i,j) is the average distance from the 3D point cloud in grid A to the corresponding triangle face. The specific calculation formula is:

[0047]

[0048] in:

[0049]

[0050] Where d1(i,j) is the average distance from all 3D point clouds in the 3D point cloud set P1(i,j) to the triangle patch V1(i,j).

[0051] They are the distances from the first, second, third and n1th 3D point clouds in the 3D point cloud set P1(i,j) to the triangle patch V1(i,j).

[0052] d2(i,j) is the average distance from all three-dimensional point clouds in the three-dimensional point cloud set P2(i,j) to the triangle patch V2(i,j).

[0053] They are respectively the distances from the first, second, third and n2th three-dimensional point clouds in the three-dimensional point cloud set P2(i,j) to the triangle patch V2(i,j).

[0054] Step 5-3, calculate the distance D: D is the average value of d(i,j) in all squares. The specific calculation formula is:

[0055]

[0056] Step 5-4, mesh optimization: traverse each square in the preliminary three-dimensional mesh model, calculate its corresponding d(i, j) respectively, and compare it with the minimum distance threshold Thres of the patch, specifically:

[0057] A. When d(i,j)<Thres, the vertical coordinates of the three corner points of the two triangles in the corresponding grid do not need to be adjusted.

[0058] B. When d(i,j)≥Thres, adjust the vertical coordinates z of the three corner points of the two triangles in the corresponding grid and recalculate d(i,j) and D using the least squares method; when D is the smallest, the optimized three-dimensional mesh model M is obtained. out .

[0059] In step 5-4, when adjusting the vertical coordinate z values ​​of the three corner points of the triangle patch, the vertical coordinate z values ​​of each edge ordered sampling point s(1,j), s(m,j), s(i,1) and s(i,n) in the border remain unchanged.

[0060] In step 7, the target chip height type defects include chip body height defects and pin soldering defects; wherein the chip body height defects include chip body height offset defects and chip body warping defects.

[0061] In step 7, the method for detecting the target chip height type defect includes the following steps:

[0062] Step 7-A1, create four detection areas: in the optimized three-dimensional mesh model, with the frame reference edge of the reference chip or the target chip, the four corners of the reference chip or the target chip are offset inward by L and H respectively, to form four detection areas R1, R2, R3 and R4 of the same set size; where L is the set length offset value; H is the set width offset value.

[0063] Step 7-A2, calculate the average height of the reference chip: calculate the average height values ​​of the reference chip in the four detection areas R1, R2, R3 and R4 of the optimized three-dimensional grid model, which are: H model1 , H model2 , H model3 , H model4 .

[0064] Step 7-A3, calculate the height threshold: according to the average height of the reference chip obtained in step 7-A2, set the height threshold; wherein the height threshold includes the minimum height threshold H min and the maximum height threshold H max , the specific expression is:

[0065] Hmin=scale*(H model1 +H model2 +H model3 +H model4 ) / 4

[0066] Hmax=thres+(H model1 1+H model2 +H model3 +H model4 ) / 4

[0067] Where scale is the minimum height threshold proportional coefficient, and scale∈(0,1).

[0068] thres is the maximum height threshold basic setting value.

[0069] Step 7-A4: Set the length direction threshold H len、 and width direction threshold H wid .

[0070] Step 7-A5, calculate the average height of the target chip: calculate the average height values ​​of the target chip in the four detection areas R1, R2, R3 and R4 of the optimized three-dimensional grid model, which are: H1, H2, H3 and H4 respectively.

[0071] Step 7-A6: Target chip height type defect detection, the specific detection expression is:

[0072]

[0073] When H1, H2, H3 and H4 obtained in step 7-A5 satisfy the above five expressions at the same time, the target chip body has no height defects. Otherwise, the target chip body has height defects, specifically:

[0074] A. If the values ​​of H1, H2, H3 and H4 are all less than H min , it means that there is a component missing defect in the target chip itself.

[0075] B. If any value among H1, H2, H3, H4 is less than H min , the rest of the height ranges are in the interval (H min ,H max ), it indicates that the target chip body has a height deviation defect;

[0076] C. If any value among H1, H2, H3, H4 is greater than H max ,

[0077] or Abs(H1-H2)>H wid

[0078] or Abs(H3-H4)>H wid

[0079] or Abs(H1-H3)>H len

[0080] or Abs(H2-H4)>H len

[0081] This indicates that the target chip body has a warping defect.

[0082] In step 7, the pin soldering defect detection method includes the following steps:

[0083] Step 7-B1, creating a pin detection area: In the optimized three-dimensional mesh model, a pin detection area R of the same set size is created for the soldering position of each pin of the reference chip or the target chip.

[0084] Step 7-B2, calculate the extreme value of the average height of the reference chip: the extreme value of the average height of the reference chip includes the minimum extreme value of the average height Hmod min and the maximum mean height extreme value Hmod max ; Among them, Hmod min Hmod is the minimum average height of all pins in the reference chip detection area R; max It is the maximum average height of all pins in the detection area R of the benchmark chip.

[0085] Step 7-B3, set the height threshold H0.

[0086] Step 7-B4, calculate the extreme value of the height of the target chip: calculate the minimum height H of the target chip in all pin detection areas R of the optimized 3D mesh model min and maximum height H max .

[0087] Step 7-B5, Pin soldering defect detection: When H min <Hmod min -H0, the pins have less tin; when H max >Hmod max When +H0, the pins are over-tinned.

[0088] The present invention has the following beneficial effects:

[0089] 1. The present invention is a chip defect detection method based on three-dimensional point cloud reconstruction. This method collects and processes the point cloud data on the chip surface to generate an accurate three-dimensional mesh model, and performs chip height defect analysis on this basis. Compared with traditional two-dimensional methods, three-dimensional point cloud reconstruction technology can provide richer spatial information, so as to more accurately identify defects on the chip surface and pins, such as missing, misaligned, warped, excessive tin in welding, insufficient tin in welding, etc., especially in the detection of small defects, complex shapes and excessive tin in pins, and insufficient tin has significant advantages, and can achieve higher-precision defect positioning and analysis.

[0090] 2. The present invention also has a high level of automation and can perform real-time defect monitoring and quality control during the chip production process, reduce manual inspection errors, and improve production efficiency.

[0091] 3. The present invention is based on 3D point cloud mesh reconstruction technology, which can effectively eliminate the noise at the 3D imaging end, more accurately reconstruct the 3D mesh, and realize accurate 3D mesh reconstruction of the chip. That is, the 3D point cloud mesh reconstruction technology has brought significant technological breakthroughs in the field of chip defect detection, meeting the needs of high precision, high efficiency and automation. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 A flow chart of a chip defect detection method based on three-dimensional point cloud mesh reconstruction of the present invention is shown.

[0093] Figure 2 A schematic diagram showing the formation of a grid of ordered sampling points during the construction of a preliminary three-dimensional grid model of the present invention.

[0094] Figure 3 Two example diagrams of constructing a preliminary three-dimensional grid model according to the present invention are shown; wherein (a) and (b) correspond to two preliminary three-dimensional grid models respectively.

[0095] Figure 4 A schematic diagram of chip body detection is shown; wherein (a) and (b) correspond to two shapes of chip bodies, respectively.

[0096] Figure 5 A schematic diagram of chip pin detection is shown; wherein (a) and (b) correspond to two shapes of pins, respectively. DETAILED DESCRIPTION

[0097] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific preferred embodiments.

[0098] In the description of the present invention, it should be understood that the terms "left side", "right side", "upper part", "lower part" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. "First", "second" and the like do not indicate the importance of the components, and therefore cannot be understood as limiting the present invention. The specific dimensions used in this embodiment are only for illustrating the technical solution, and do not limit the scope of protection of the present invention.

[0099] like Figure 1 As shown, a chip defect detection method based on three-dimensional point cloud mesh reconstruction includes the following steps.

[0100] Step 1: Collect the three-dimensional point cloud of the reference chip: For each part in the reference chip that needs to be inspected for three-dimensional defects, a plurality of sensors are used to scan and collect the corresponding three-dimensional point cloud; one part corresponds to one three-dimensional point cloud set.

[0101] Step 2: De-noising: De-noising is performed on each 3D point cloud set collected in step 1, which specifically includes the following steps.

[0102] Step 2-1: Input the point cloud P in step 1, perform plane fitting, and obtain the reference plane height Z=Zstd.

[0103] Step 2-2: Set the minimum height threshold t1 and the maximum height threshold t2.

[0104] Step 2-3: Extract the point cloud in the interval [Zstd-t1, Zstd+t2] through axial clipping to form a new point cloud P out .

[0105] Step 3: Ordered sampling: Orderly sampling is performed on each denoised 3D point cloud set to form a corresponding ordered point cloud set.

[0106] The above three-dimensional point cloud set P after denoising out The method for orderly sampling comprises the following steps.

[0107] Step 3-1, obtain the extreme value of point cloud coordinates: traverse the three-dimensional point cloud set P out , the point cloud coordinate extremes in the X and Y directions are calculated, which are: the maximum value Xmax of the X-direction point cloud coordinate, the minimum value Xmin of the X-direction point cloud coordinate, the maximum value Ymax of the Y-direction point cloud coordinate, and the minimum value Ymin of the Y-direction point cloud coordinate.

[0108] Step 3-2, set the sampling step: for the three-dimensional point cloud set P out Set the X-direction sampling step size t respectively x and Y-sampling step length t y .

[0109] Step 3-3, divide the sampling area: collect the three-dimensional point cloud P out Divide into m regions along the X direction and n regions along the Y direction, where:

[0110]

[0111] The above three-dimensional point cloud set P out The i-th region along the X direction and the j-th region along the Y direction are recorded as sampling region P(i,j); at this time, the sampling region P(i,j) has variable N(i,j) denoised 3D point clouds, which are: p1, p2, p3, ..., p N(i,j) .

[0112] Step 3-4, find the ordered sampling points: In the middle of each sampling area divided in step 3-3, select a point as the found ordered sampling point.

[0113] The ordered sampling points found in the above sampling area P(i,j) are recorded as s(i,j), and the X-, Y-, and Z-direction coordinates of s(i,j) are recorded as: s(i,j).x, s(i,j).y, and s(i,j).z respectively; then the calculation formulas of s(i,j).x and s(i,j).y are:

[0114] s(i,j).x=Xmin+(i-1)*t x

[0115] s(i,j).y=Ymin+(j-1)*t y

[0116] s(i,j).z is determined according to the number of N(i,j), specifically:

[0117] A. When N(i,j)=0, and i=1 or j=1, the specific calculation formula is:

[0118] s(i,j).z=Zstd

[0119] Where Zstd is the height of the reference plane, which is obtained by plane fitting the three-dimensional point cloud of the part collected in step 1;

[0120] B. When N(i,j)=0, i≠1 and j≠1, the specific calculation formula is:

[0121]

[0122] Where s(p,q).z is the Z coordinate of the ordered sampling point corresponding to the pth intersection area along the X direction and the qth intersection area along the Y direction; p is i-1, i and i+1; q is j-1, j and j+1;

[0123] N(p,q) is the number of denoised 3D point clouds in the p-th intersection area along the X direction and the q-th intersection area along the Y direction;

[0124] N is the total number of N(p,q)>0;

[0125] is the sum of the Z coordinates of all ordered sampling points in the region N(p,q)>0;

[0126] C. When N(i,j)≥1, the specific calculation formula is:

[0127]

[0128] In the formula, p1.z, p2.z, p3.z, ..., p N(i,j) .z are three-dimensional point clouds p1, p2, p3, ..., p N(i,j) The Z coordinate of .

[0129] Step 4, grid processing: Take all the ordered sampling points in each ordered point cloud set as grid corner points, build a grid network, and then divide each grid into two triangular faces, so as to form a preliminary three-dimensional grid model; wherein, the three vertices of each triangular face are the three ordered sampling point clouds in the corresponding ordered point cloud set.

[0130] Step 5: Mesh optimization: By minimizing the average distance from all denoised 3D point clouds to the corresponding triangular facets in each square, the height coordinates of each triangular facet are adjusted and optimized to obtain an optimized 3D mesh model.

[0131] The method for constructing the above preliminary three-dimensional mesh model preferably includes the following steps.

[0132] Step 4-1, find the corner points of the grid: Figure 2 and Figure 3 As shown, the ordered sampling point s(i,j) is taken as the starting corner point of a square A, and the adjacent ordered sampling points are searched to obtain the other three corner points of square A, namely: s(i+1,j), s(i,j+1), and s(i+1,j+1).

[0133] Step 4-2, construct triangular patches: Take s(i,j) in the square A as the starting point and construct two triangular patches. Figure 3 As shown, the two triangular faces of square A are V1(i,j) and V2(i,j), which can be constructed in two ways:

[0134] Construction method 1: V1(i,j) = {s(i,j), s(i+1,j), s(i+1,j+1)}, V2(i,j) = {s(i,j), s(i,j+1), s(i+1,j+1)}.

[0135] Construction method 2: V1(i,j) = {s(i,j), s(i+1,j), s(i,j+1)}, V2(i,j) = {s(i+1,j), s(i,j+1), s(i+1,j+1)}.

[0136] Step 4-3, construct a preliminary three-dimensional mesh model: In step 3-4, a total of m*n ordered sampling points are obtained. Repeat steps 4-1 to 4-2, traverse each ordered sampling point, and obtain a total of (m-1)*(n-1) squares, and then obtain a preliminary three-dimensional mesh model M containing 2*(m-1)*(n-1) triangles. in .

[0137] The above-mentioned grid optimization method preferably includes the following steps.

[0138] Step 5-1, set the minimum distance threshold Thres of the patch.

[0139] Step 5-1, extract 3D point cloud: The denoised 3D point clouds corresponding to the two triangular patches V1(i,j) and V2(i,j) in grid A are recorded as 3D point cloud sets P1(i,j) and P2(i,j), respectively; where P1(i,j) = {p11, p12, p13…p1n1}, P2(i,j) = {p21, p22, p23…p2n2}; where n1 and n2 are the numbers of 3D point clouds corresponding to the triangular patches V1(i,j) and V2(i,j), respectively.

[0140] Step 5-2, calculate d(i,j): d(i,j) is the average distance from the 3D point cloud in grid A to the corresponding triangle face. The specific calculation formula is:

[0141]

[0142] in:

[0143]

[0144] Where d1(i,j) is the average distance from all 3D point clouds in the 3D point cloud set P1(i,j) to the triangle patch V1(i,j).

[0145] They are the distances from the first, second, third and n1th 3D point clouds in the 3D point cloud set P1(i,j) to the triangle patch V1(i,j).

[0146] d2(i,j) is the average distance from all three-dimensional point clouds in the three-dimensional point cloud set P2(i,j) to the triangle patch V2(i,j).

[0147] They are respectively the distances from the first, second, third and n2th three-dimensional point clouds in the three-dimensional point cloud set P2(i,j) to the triangle patch V2(i,j).

[0148] Step 5-3, calculate the distance D: D is the average value of d(i,j) in all squares. The specific calculation formula is:

[0149]

[0150] Step 5-4, mesh optimization: traverse each square in the preliminary three-dimensional mesh model, calculate its corresponding d(i, j) respectively, and compare it with the minimum distance threshold Thres of the patch, specifically:

[0151] A. When d(i,j)<Thres, the vertical coordinates of the three corner points of the two triangles in the corresponding grid do not need to be adjusted.

[0152] B. When d(i,j)≥Thres, adjust the vertical coordinates z of the three corner points of the two triangles in the corresponding grid and recalculate d(i,j) and D using the least squares method; when D is the smallest, the optimized three-dimensional mesh model M is obtained. out .

[0153] When the vertical coordinate z values ​​of the three corner points of the triangle patch are adjusted, the vertical coordinate z value of each edge ordered sampling point s(1,j), s(m,j), s(i,1) and s(i,n) in the border remains unchanged.

[0154] The present invention adjusts the ordinate z value of each ordered sampling point in the preliminary three-dimensional grid model and adjusts the model triangular facets through the least squares method, which can effectively suppress cusp noise and solve the noise fluctuation caused by highlighting of components.

[0155] Step 6: Collect the three-dimensional point cloud of the target chip and target parts: collect and denoise the three-dimensional point cloud of the target parts in the target chip to be inspected.

[0156] Step 7, chip defect detection: By calling the optimized 3D mesh model corresponding to the target part, the target chip can be subjected to a high degree of defect detection.

[0157] The above-mentioned target chip height type defects include chip body height defects and pin soldering defects; wherein, chip body height defects include chip body height offset defects and chip body warping defects.

[0158] The above-mentioned method for detecting the target chip height type defect comprises the following steps.

[0159] Step 7-A1, create four detection areas: In the optimized three-dimensional mesh model, with the frame reference edge of the reference chip or the target chip, the four corners of the reference chip or the target chip are offset inward by L and H respectively, forming the following Figure 4 The four detection areas R1, R2, R3 and R4 with the same set size are shown as follows. Figure 4 As shown; wherein, L is the set length offset value; H is the set width offset value.

[0160] Step 7-A2, calculate the average height of the reference chip: calculate the average height values ​​of the reference chip in the four detection areas R1, R2, R3 and R4 of the optimized three-dimensional grid model, which are: H model1 , H model2 , H model3 , H model4 .

[0161] Step 7-A3, calculate the height threshold: according to the average height of the reference chip obtained in step 7-A2, set the height threshold; wherein the height threshold includes the minimum height threshold H min and the maximum height threshold H max , the specific expression is:

[0162] Hmin=scale*(H model1 +H model2 +H model3 +H model4 ) / 4

[0163] Hmax=thres+(H model1 1+H model2 +H model3 +H model4 ) / 4

[0164] Where scale is the minimum height threshold proportional coefficient, and scale∈(0,1).

[0165] Thres is a basic setting value of a maximum height threshold. In this embodiment, the height unit is preferably set to micrometers, so thres>50 micrometers.

[0166] Step 7-A4: Set the length direction threshold H len、 and width direction threshold H wid .

[0167] Step 7-A5, calculate the average height of the target chip: calculate the average height values ​​of the target chip in the four detection areas R1, R2, R3 and R4 of the optimized three-dimensional grid model, which are: H1, H2, H3 and H4 respectively.

[0168] Step 7-A6: Target chip height type defect detection, the specific detection expression is:

[0169]

[0170] When H1, H2, H3 and H4 obtained in step 7-A5 satisfy the above five expressions at the same time, the target chip body has no height defects. Otherwise, the target chip body has height defects, specifically:

[0171] A. If the values ​​of H1, H2, H3 and H4 are all less than H min , it means that there is a component missing defect in the target chip itself.

[0172] B. If any value among H1, H2, H3, H4 is less than H min , the rest of the height ranges are in the interval (H min ,H max), it indicates that the target chip body has a height deviation defect;

[0173] C. If any value among H1, H2, H3, H4 is greater than H max ,

[0174] or Abs(H1-H2)>H wid

[0175] or Abs(H3-H4)>H wid

[0176] or Abs(H1-H3)>H len

[0177] or Abs(H2-H4)>H len

[0178] This indicates that the target chip body has a warping defect.

[0179] The above-mentioned pin soldering defect detection method preferably includes the following steps.

[0180] Step 7-B1, create a pin detection area: In the optimized three-dimensional mesh model, for each pin soldering location of the reference chip or the target chip, a pin detection area R with the same set size is created, as follows: Figure 5 shown.

[0181] Step 7-B2, calculate the extreme value of the average height of the reference chip: the extreme value of the average height of the reference chip includes the minimum extreme value of the average height Hmod min and the maximum mean height extreme value Hmod max ; Among them, Hmod min Hmod is the minimum average height of all pins in the reference chip detection area R; max It is the maximum average height of all pins in the detection area R of the benchmark chip.

[0182] Step 7-B3, set the height threshold H0.

[0183] Step 7-B4, calculate the extreme value of the height of the target chip: calculate the minimum height H of the target chip in all pin detection areas R of the optimized 3D mesh model min and maximum height H max .

[0184] Step 7-B5, Pin soldering defect detection: When H min <Hmod min -H0, the pins have less tin; when H max >Hmod max When +H0, the pins are over-tinned.

[0185] The preferred embodiments of the present invention are described in detail above; however, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A chip defect detection method based on three-dimensional point cloud mesh reconstruction, characterized in that: The steps include: Step 1, collecting the three-dimensional point cloud of the reference chip: for each part in the reference chip that needs to be tested for three-dimensional defects, collect the corresponding three-dimensional point cloud; wherein one part corresponds to one three-dimensional point cloud set; Step 2, denoising: denoising is performed on each 3D point cloud set collected in step 1; Step 3, ordered sampling: Orderly sampling is performed on each 3D point cloud set after denoising to form a corresponding ordered point cloud set; Step 4: Gridding: Take all ordered sampling points in each ordered point cloud set as grid corner points, build a grid network, and then divide each grid into two triangular facets to form a preliminary three-dimensional grid model; the three vertices of each triangular facet are the three ordered sampling point clouds in the corresponding ordered point cloud set; Step 5, mesh optimization: by minimizing the average distance from all denoised 3D point clouds to the corresponding triangular facets in each square, the height coordinates of each triangular facet are adjusted and optimized, thereby obtaining an optimized 3D mesh model; Step 6, collecting the three-dimensional point cloud of the target chip and target parts: collecting and denoising the three-dimensional point cloud of the target parts in the target chip to be inspected; Step 7, chip defect detection: By calling the optimized 3D mesh model corresponding to the target part, the target chip can be subjected to a high degree of defect detection.

2. The chip defect detection method based on three-dimensional point cloud mesh reconstruction according to claim 1 is characterized in that: In step 3, the denoised three-dimensional point cloud set P out The method for performing ordered sampling comprises the following steps: Step 3-1, obtain the extreme value of point cloud coordinates: traverse the three-dimensional point cloud set P out , calculate the extreme values ​​of the point cloud coordinates in the X and Y directions, which are: the maximum value Xmax of the X-direction point cloud coordinates, the minimum value Xmin of the X-direction point cloud coordinates, the maximum value Ymax of the Y-direction point cloud coordinates, and the minimum value Ymin of the Y-direction point cloud coordinates; Step 3-2, set the sampling step: for the three-dimensional point cloud set P out Set the X-direction sampling step size t respectively x and Y-sampling step length t y ; Step 3-3, divide the sampling area: divide the three-dimensional point cloud set P out Divide into m regions along the X direction and n regions along the Y direction, where: Step 3-4, find the ordered sampling points: In the middle of each sampling area divided in step 3-3, select a point as the found ordered sampling point.

3. The chip defect detection method based on three-dimensional point cloud mesh reconstruction according to claim 2 is characterized in that: In step 3-3, the three-dimensional point cloud set P out The i-th intersection area along the X direction and the j-th intersection area along the Y direction are recorded as the sampling area P(i,j); at this time, the sampling area P(i,j) has variable N(i,j) denoised 3D point clouds, which are: p1, p2, p3, ..., p N(i,j) ; In step 3-4, the ordered sampling points found in the sampling area P(i,j) are recorded as s(i,j), and the X-, Y-, and Z-direction coordinates of s(i,j) are recorded as: s(i,j).x, s(i,j).y, and s(i,j).z, respectively; then the calculation formulas of s(i,j).x and s(i,j).y are: s(i,j).x=Xmin+(i-1)*t x s(i,j).y=Ymin+(j-1)*t y s(i,j).z is determined according to the number of N(i,j), specifically: A. When N(i,j)=0, and i=1 or j=1, the specific calculation formula is: s(i,j).z=Zstd Where Zstd is the height of the reference plane, which is obtained by plane fitting the three-dimensional point cloud of the part collected in step 1; B. When N(i,j)=0, i≠1 and j≠1, the specific calculation formula is: Where s(p,q).z is the Z coordinate of the ordered sampling point corresponding to the pth intersection area in the X direction and the qth intersection area in the Y direction; p is i-1, i and i+1; q is j-1, j and j+1; N(p,q) is the number of denoised 3D point clouds in the p-th intersection along the X direction and the q-th intersection along the Y direction; N is the total number of N(p,q)>0; It is the sum of the Z coordinates of all ordered sampling points in the region N(p,q)>0; C. When N(i,j)≥1, the specific calculation formula is: In the formula, p1.z, p2.z, p3.z, ..., p N(i,j) .z are three-dimensional point clouds p1, p2, p3, ..., p N(i,j) The Z coordinate of .

4. The chip defect detection method based on three-dimensional point cloud mesh reconstruction according to claim 3 is characterized in that: In step 4, the method for constructing a preliminary three-dimensional mesh model includes the following steps: Step 4-1, find the corner points of the grid: take the ordered sampling point s(i,j) as the starting corner point of a grid A, and search for the adjacent ordered sampling points to obtain the other three corner points of the grid A, namely: s(i+1,j), s(i,j+1), s(i+1,j+1); Step 4-2, construct triangular facets: take s(i,j) in grid A as the starting point, and construct two triangular facets; Step 4-3, construct a preliminary three-dimensional mesh model: In step 3-4, a total of m*n ordered sampling points are obtained. Repeat steps 4-1 to 4-2, traverse each ordered sampling point, and obtain a total of (m-1)*(n-1) squares, and then obtain a preliminary three-dimensional mesh model M containing 2*(m-1)*(n-1) triangles. in .

5. The chip defect detection method based on three-dimensional point cloud mesh reconstruction according to claim 4 is characterized in that: In step 4-2, the two triangles of square A are V1(i,j) and V2(i,j), which can be constructed in two ways: Construction method 1: V1(i,j) = {s(i,j), s(i+1,j), s(i+1,j+1)}, V2(i,j) = {s(i,j), s(i,j+1), s(i+1,j+1)}; Construction method 2: V1(i,j) = {s(i,j), s(i+1,j), s(i,j+1)}, V2(i,j) = {s(i+1,j), s(i,j+1), s(i+1,j+1)}.

6. The chip defect detection method based on three-dimensional point cloud mesh reconstruction according to claim 5 is characterized in that: In step 5, the grid optimization method includes the following steps: Step 5-1, set the minimum distance threshold Thres of the patch; Step 5-1, extract 3D point cloud: The denoised 3D point clouds corresponding to the two triangular facets V1(i,j) and V2(i,j) in the square A are recorded as 3D point cloud sets P1(i,j) and P2(i,j), respectively; where P1(i,j) = {p11, p12, p13…p1n1}, P2(i,j) = {p21, p22, p23…p2n2}; where n1 and n2 are the numbers of 3D point clouds corresponding to the triangular facets V1(i,j) and V2(i,j), respectively; Step 5-2, calculate d(i,j): d(i,j) is the average distance from the 3D point cloud in grid A to the corresponding triangle face. The specific calculation formula is: in: Where d1(i,j) is the average distance from all 3D point clouds to the triangle patch V1(i,j) in the 3D point cloud set P1(i,j); are the distances from the first, second, third and n1th three-dimensional point cloud to the triangle patch V1(i,j) in the three-dimensional point cloud set P1(i,j); d2(i,j) is the average distance from all three-dimensional point clouds to the triangle patch V2(i,j) in the three-dimensional point cloud set P2(i,j); are the distances from the first, second, third and n2th three-dimensional point clouds to the triangle patch V2(i,j) in the three-dimensional point cloud set P2(i,j); Step 5-3, calculate the distance D: D is the average value of d(i,j) in all squares. The specific calculation formula is: Step 5-4, mesh optimization: traverse each square in the preliminary three-dimensional mesh model, calculate its corresponding d(i, j) respectively, and compare it with the minimum distance threshold Thres of the patch, specifically: A. When d(i,j)<Thres, the vertical coordinates of the three corner points of the two triangles in the corresponding grid do not need to be adjusted; B. When d(i,j)≥Thres, adjust the vertical coordinates z of the three corner points of the two triangles in the corresponding grid and recalculate d(i,j) and D using the least squares method; when D is the smallest, the optimized three-dimensional mesh model M is obtained. out .

7. The chip defect detection method based on three-dimensional point cloud mesh reconstruction according to claim 6 is characterized in that: In step 5-4, when adjusting the vertical coordinate z values ​​of the three corner points of the triangle patch, the vertical coordinate z values ​​of each edge ordered sampling point s(1,j), s(m,j), s(i,1) and s(i,n) located in the border remain unchanged.

8. The chip defect detection method based on three-dimensional point cloud mesh reconstruction according to claim 1 is characterized in that: In step 7, the target chip height type defects include chip body height defects and pin soldering defects; wherein the chip body height defects include chip body height offset defects, component missing defects and chip body warping defects.

9. The chip defect detection method based on three-dimensional point cloud mesh reconstruction according to claim 8, characterized in that: In step 7, the method for detecting the target chip height type defect includes the following steps: Step 7-A1, create four detection areas: in the optimized three-dimensional mesh model, with the frame reference edge of the reference chip or the target chip, the four corners of the reference chip or the target chip are offset inward by L and H respectively, to form four detection areas R1, R2, R3 and R4 of the same set size; wherein L is the set length offset value; H is the set width offset value; Step 7-A2, calculate the average height of the reference chip: calculate the average height values ​​of the reference chip in the four detection areas R1, R2, R3 and R4 of the optimized three-dimensional grid model, which are: H model1 , H model2 , H model3 , H model4 ; Step 7-A3, calculate the height threshold: according to the average height of the reference chip obtained in step 7-A2, set the height threshold; wherein the height threshold includes the minimum height threshold H min and the maximum height threshold H max , the specific expression is: Hmin=scale*(H model1 +H model2 +H model3 +H model4 ) / 4 Hmax=thres+(H model1 1+H model2 +H model3 +H model4 ) / 4 Where scale is the minimum height threshold proportional coefficient, and scale∈(0,1); thres is the maximum height threshold basic setting value; Step 7-A4: Set the length direction threshold H len、 and width direction threshold H wid ; Step 7-A5, calculating the average height of the target chip: calculating the average height values ​​of the target chip in the four detection areas R1, R2, R3 and R4 of the optimized three-dimensional grid model, which are: H1, H2, H3 and H4; Step 7-A6: Target chip height type defect detection, the specific detection expression is: When H1, H2, H3 and H4 obtained in step 7-A5 satisfy the above five expressions at the same time, the target chip body has no height defects. Otherwise, the target chip body has height defects, specifically: A. If the values ​​of H1, H2, H3 and H4 are all less than H min , it means that there is a component missing defect in the target chip itself; B. If any value among H1, H2, H3, H4 is less than H min , the rest of the height ranges are in the interval (H min ,H max ), it indicates that the target chip body has a height deviation defect; C. If any value among H1, H2, H3, H4 is greater than H max , or Abs(H1-H2)>H wid or Abs(H3-H4)>H wid or Abs(H1-H3)>H len or Abs(H2-H4)>H len This indicates that the target chip body has a warping defect.

10. The chip defect detection method based on three-dimensional point cloud mesh reconstruction according to claim 1, characterized in that: In step 7, the pin soldering defect detection method includes the following steps: Step 7-B1, creating a pin detection area: in the optimized three-dimensional mesh model, a pin detection area R with the same set size is created for each soldering position of the pin of the reference chip or the target chip; Step 7-B2, calculate the extreme value of the average height of the reference chip: the extreme value of the average height of the reference chip includes the minimum extreme value of the average height Hmod min and the maximum mean height extreme value Hmod max Among them, Hmod min Hmod is the minimum average height of all pins in the reference chip detection area R; max is the maximum average height of all pins in the detection area R of the reference chip; Step 7-B3, setting the height threshold H0; Step 7-B4, calculate the extreme value of the height of the target chip: calculate the minimum height H of the target chip in all pin detection areas R of the optimized 3D mesh model min and maximum height H max ; Step 7-B5, Pin soldering defect detection: When H min <Hmod min -H0, the pins have less tin; when H max >Hmod max When +H0, the pins are over-tinned.

Images