A semiconductor workpiece cutting path planning algorithm based on image processing

The image processing algorithm is used to identify single crystal and hybrid crystal regions on the semiconductor chip, and the cutting path is planned, which solves the resource waste and safety hazards caused by manual cutting in the prior art, and achieves efficient and accurate cutting of semiconductor workpieces.

CN116797570BActive Publication Date: 2024-12-06TAIYUAN UNIVERSITY OF SCIENCE AND TECHNOLOGY
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202310751582.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-06-25
Publication Date
2024-12-06
Estimated Expiration
2043-06-25

Smart Images

  • Figure CN116797570B_ABST
    Figure CN116797570B_ABST
Patent Text Reader

Abstract

A semiconductor workpiece cutting path planning algorithm based on image processing belongs to the field of image computing technology and solves the technical problem of manually removing the impurity crystal area on the wafer. It includes the following steps: 1. Obtaining the semantic boundary of the image; 2. Finding the largest inscribed rectangle MRect in the single crystal image X1; 3. Segmenting the contour area in the single crystal area and finding the corresponding cutting line; 4. Determining the final cutting line. The invention can plan the cutting line to realize the automatic cutting of the wafer and can better retain the required single crystal area.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of image computing, and in particular relates to a semiconductor workpiece cutting path planning algorithm based on image processing. Background Art

[0002] When growing epitaxial layers of semiconductor materials, clean single crystal slices are required. However, after the semiconductor material is cut from a crystal rod into wafers, there will be multiple different crystal phase regions on each wafer, so it is necessary to remove the impure crystal region on the wafer and retain the required single crystal portion.

[0003] In the prior art, the semiconductor wafer cutting process is as follows: after the semiconductor material is cut from the crystal rod into wafers, the boundary between the single crystal and the impurity crystal is manually identified by naked eyes, and then manually cut with a knife to obtain a clean single crystal slice. Most semiconductor materials contain valuable precious metals, and the existing manual processing method will waste a lot of human resources; moreover, when cutting the wafer, the wafer is often cut incorrectly or damaged due to manual instability, resulting in material waste; in addition, heavy metal materials are easily inhaled into the body during manual operation, which will cause harm to the operator. Summary of the invention

[0004] The main purpose of the present invention is to overcome the deficiencies in the prior art and solve the technical problem of manually removing the impurity crystal area on the wafer. The present invention provides a semiconductor workpiece cutting path planning algorithm based on image processing.

[0005] The design concept of the present invention is:

[0006] When cutting wide bandgap semiconductor materials, it is necessary to retain the required single crystal part on a wafer with multiple crystal phases, so the cutting path must be planned to ensure that the usable part on the wafer can be maximized. The present invention uses an image processing algorithm to plan the cutting path according to the position of the single crystal on the wafer to achieve the purpose of automatic cutting. The algorithm proposed in the present invention first finds the largest inscribed rectangle in the single crystal outline, and then plans the path based on the intersection of the inscribed rectangle and the single crystal outline. Specifically, the following steps are included: obtain the workpiece image and preprocess it → extract the single crystal outline → find the largest inscribed rectangle → plan the cutting path.

[0007] The present invention is implemented by the following technical solution: a semiconductor workpiece cutting path planning algorithm based on image processing, comprising the following steps:

[0008] S1. Get the semantic boundary of the image:

[0009] S1-1. Obtain a semantic image X through a deep learning model, and distinguish different crystal phase regions in the wafer according to the semantic image X, wherein the crystal phase region to be retained is a single crystal region, and the rest is a mixed crystal region; the semantic image X is set as a grayscale image, and the grayscale value of the single crystal region is different from the grayscale value of the mixed crystal region;

[0010] S1-2, adjusting the color of the impure crystal region in the semantic image X obtained in step S1-1 to be the same as the background color, to obtain a single crystal image X1;

[0011] S2. Find the largest inscribed rectangle MRect in the single crystal image X1:

[0012] S2-1, find all the contours in the single crystal image X1, take the contour with the largest area as the processing contour of the required product, record it as contour contour, then store all the contour points on contour contour in order, and make the position difference between two adjacent contour points no greater than 1;

[0013] S2-2, randomly select a contour point on the contour and record it as start(x,y), loop the points in the contour with start(x,y) as the starting point, record the current point on the contour as cur(x,y), and use the starting point start(x,y) and the current point cur(x,y) as the diagonal points of the rectangle to construct a rectangular contour;

[0014] S2-3, judging whether the rectangular contours constructed in step S2-2 are all located inside the contour contour, and if so, recording the area of ​​the rectangular contour and the corresponding diagonal point coordinates in this state;

[0015] S2-4, repeating steps S2-2 to S2-3, traversing all points on the contour contour, obtaining the inscribed rectangular contour maxRect with the largest area in the contour contour, and recording the coordinates and area of ​​the inscribed rectangular contour maxRect with the largest area;

[0016] S2-5. Calculate the spatial moment m of the contour according to formula (1) ji :

[0017] m ji =∑ x,y ((x, y)*x j *y i ); (1)

[0018] In formula (1), x and y are the coordinate values ​​of the contour, i and j are the image orders;

[0019] S2-6. When the image order i = 0 or 1, j = 0 or 1, the m obtained by formula (2) 00 、m 10 、m 01 Calculate the centroid center(x, y) of the contour contour:

[0020]

[0021] S2-7, taking the centroid center(x, y) of the contour contour determined in step S2-6 as the center, rotating the single crystal image X1, and obtaining a single crystal image X1 every 5° rotation. i , i is an integer greater than 1, a total rotation of 360°, in each single crystal image X i Repeat steps S2-2 to S2-6 to obtain each single crystal image X i All the contours in contour;

[0022] S2-8. Compare each single crystal image X i The area of ​​the largest inscribed rectangle contour maxRec among all contours contour, and the largest inscribed rectangle MRect in the single crystal region is determined;

[0023] S3. Split the contour area within the single crystal area and find the corresponding cutting line:

[0024] S3-1, extending the single sides of the four sides of the inscribed rectangle MRect outward respectively, and the two extension lines at each vertex of the inscribed rectangle MRect divide the single crystal area outside the inscribed rectangle MRect into four areas Con1, Con2, Con3, and Con4;

[0025] S3-2, select the vertex or extension line of the inscribed rectangle MRect in the Con1 area and the intersection point with the outline of the single crystal area as P1(x, y), and form a straight line with any point P2(x, y) in the Con1 area, loop all the points in the Con1 area, calculate the area of ​​the impure crystal area removed by the straight line formed by P2(x, y) and P1(x, y), and sort them according to the size of the removed area of ​​the impure crystal area, and select the group with the largest removed area of ​​the impure crystal area as the candidate cutting line L1; if the removed impure crystal areas are the same, select the group with the largest area of ​​the single crystal area as the candidate cutting line L1;

[0026] Select another vertex or extension line of the inscribed rectangle MRect in Con1 and mark the intersection point with the outline of the single crystal region as P3(x, y), and form a straight line with any point P2(x, y) in the Con1 region. Loop through all points in the Con1 region, calculate the area of ​​the impurity crystal region removed by the straight line formed by P2(x, y) and P3(x, y), and determine the candidate cutting line L2 in the same way as step S3-2;

[0027] Similarly, the candidate cutting lines L3, L4, L5, and L6 corresponding to the Con2 area, the Con3 area, and the Con4 area are obtained respectively;

[0028] S4. Determine the final cutting line:

[0029] Traverse the six candidate cutting lines L1, L2, L3, L4, L5 and L6 determined in step S3, and select the path that can form a convex polygonal single crystal area with the largest area, the largest area of ​​the removed miscellaneous crystal area, and the smallest number of straight line cuts as the final cutting line path. The path optimization method is:

[0030] Assume that the area of ​​the single crystal pixel formed by segmentation is S x , A(·) represents the area of ​​the single crystal region removed, and Z(·) represents the area of ​​the heterocrystalline region removed:

[0031]

[0032]

[0033] Where p represents the number of currently selected cutting lines; l k represents the kth cutting line selected; m represents the total number of all candidate cutting lines, that is, the maximum number; represents the number of permutations and combinations of selecting p cutting lines from all m cutting lines; r e Indicates the path planned under the current e-th permutation and combination; shape e represents the shape of the wafer cut by the e-th path; Φ represents the value range of e; r opt1 Represents the currently selected optimal path, which consists of the p selected cutting lines.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The present invention provides a semiconductor workpiece cutting path planning algorithm based on image processing. The algorithm realizes the automation of the selection path planning. The planned path can realize the selection of the wafer, and the actual processing requirements are met while the number of cutting knives is small. The single crystal obtained by the cutting result does not contain impurities, and while satisfying the maximum retention of the single crystal part, the integrity of the impurities can be guaranteed, reducing the waste of materials and improving the processing efficiency of the wafer selection. Compared with the existing manual dicing, it avoids the damage to the wafer caused by manual dicing and the damage to the human body caused by long-term contact with the wafer during the processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of the present invention;

[0037] Figure 2 A crystal graph of a semantic image X obtained by a deep learning model in a specific implementation manner;

[0038] Figure 3 for Figure 2 Crystal map of the single crystal region obtained after preprocessing (the color of the impurity crystal region is adjusted to the same color as the background);

[0039] Figure 4 for Figure 3 Crystallographic map of the single crystal region obtained after rotation;

[0040] Figure 5 It is a schematic diagram of the contour area within the single crystal area;

[0041] Figure 6 is a schematic diagram of candidate cutting lines;

[0042] Figure 7 It is a cutting path diagram in a specific implementation manner. DETAILED DESCRIPTION

[0043] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0044] like Figure 1 The semiconductor workpiece cutting path planning algorithm shown in the figure based on image processing comprises the following steps:

[0045] S1. Get the semantic boundary of the image:

[0046] S1-1. Obtain a semantic image X through a deep learning model, and distinguish different crystal phase regions in the wafer according to the semantic image X, wherein the crystal phase region to be retained is a single crystal region, and the rest is a mixed crystal region; the semantic image X is set as a grayscale image, and the grayscale value of the single crystal region is different from the grayscale value of the mixed crystal region. In this specific implementation manner, the grayscale value of the single crystal region is 50, and the grayscale value of the mixed crystal region is 100;

[0047] S1-2, adjusting the color of the impurity crystal region in the semantic image X obtained in step S1-1 to be the same as the background color, to obtain a single crystal image X1 (eg Figure 3 shown);

[0048] S2. Find the largest inscribed rectangle MRect in the single crystal image X1:

[0049] S2-1, find all the contours in the single crystal image X1, take the contour with the largest area as the processing contour of the required product, record it as contour contour, then store all the contour points on contour contour in order, and make the position difference between two adjacent contour points no greater than 1;

[0050] S2-2, randomly select a contour point on the contour and record it as start(x,y), loop the points in the contour with start(x,y) as the starting point, record the current point on the contour as cur(x,y), and use the starting point start(x,y) and the current point cur(x,y) as the diagonal points of the rectangle to construct a rectangular contour;

[0051] S2-3, judging whether the rectangular contours constructed in step S2-2 are all located inside the contour contour, and if so, recording the area of ​​the rectangular contour and the corresponding diagonal point coordinates in this state;

[0052] S2-4, repeating steps S2-2 to S2-3, traversing all points on the contour contour, obtaining the inscribed rectangular contour maxRect with the largest area in the contour contour, and recording the coordinates and area of ​​the inscribed rectangular contour maxRect with the largest area;

[0053] S2-5. Calculate the spatial moment m of the contour according to formula (1) ji :

[0054] m ji =∑ x,y ((x, y)*x j *y i ); (1)

[0055] In formula (1), x and y are the coordinate values ​​of the contour, i and j are the image orders;

[0056] S2-6. When the image order i = 0 or 1, j = 0 or 1, the m obtained by formula (2) 00 、m 10 、m 01 Calculate the centroid center(x, y) of the contour contour:

[0057]

[0058] S2-7, taking the centroid center(x, y) of the contour contour determined in step S2-6 as the center, rotating the single crystal image X1, and obtaining a single crystal image X1 every 5° rotation. i (like Figure 4 As shown), i is an integer greater than 1, and the total rotation is 360°. i Repeat steps S2-2 to S2-6 to obtain each single crystal image X i All the contours in contour;

[0059] S2-8. Compare each single crystal image X i The area of ​​the largest inscribed rectangle contour maxRec among all contours contour, and the largest inscribed rectangle MRect in the single crystal region is determined;

[0060] S3. Split the contour area within the single crystal area and find the corresponding cutting line:

[0061] S3-1, extend the four sides of the inscribed rectangle MRect outward respectively, and the two extension lines at each vertex of the inscribed rectangle MRect divide the single crystal area outside the inscribed rectangle MRect into four areas: Con1, Con2, Con3, and Con4 (such as Figure 5 shown);

[0062] S3-2, select the vertex or extension line of the inscribed rectangle MRect in the Con1 area and the intersection point with the outline of the single crystal area as P1(x, y), and form a straight line with any point P2(x, y) in the Con1 area, loop all the points in the Con1 area, calculate the area of ​​the impure crystal area removed by the straight line formed by P2(x, y) and P1(x, y), and sort them according to the size of the removed area of ​​the impure crystal area, and select the group with the largest removed area of ​​the impure crystal area as the candidate cutting line L1; if the removed impure crystal areas are the same, select the group with the largest area of ​​the single crystal area as the candidate cutting line L1;

[0063] In Con1, select another vertex or extension line of the inscribed rectangle MRect and the intersection point with the outline of the single crystal area as P3(x, y), and form a straight line with any point P2(x, y) in the Con1 area, loop all the points in the Con1 area, calculate the area of ​​the impure crystal area removed by the straight line formed by P2(x, y) and P3(x, y), and determine the candidate cutting line L2 in the same way as step S3-2, that is: sort according to the size of the removed area of ​​the impure crystal area, and select the group with the largest removed area of ​​the impure crystal area as the candidate cutting line L2; if the removed impure crystal area is the same, select the group with the largest area of ​​the single crystal area as the candidate cutting line L2;

[0064] Similarly, the candidate cutting lines L3, L4, L5, and L6 corresponding to the Con2 area, the Con3 area, and the Con4 area are obtained respectively (such as Figure 6 shown);

[0065] S4. Determine the final cutting line:

[0066] To ensure that the number of cutting lines is as small as possible, we need to optimize the path. Traverse the six candidate cutting lines L1, L2, L3, L4, L5 and L6 determined in step S3, and select the path that can form a convex polygonal single crystal area with the largest area, the largest area of ​​the removed miscellaneous crystal area, and the smallest number of straight line cuts as the final cutting line path (such as Figure 7 As shown), the path optimization method is:

[0067] Assume that the area of ​​the single crystal pixel formed by segmentation is S x , A(·) represents the area of ​​the single crystal region removed, and Z(·) represents the area of ​​the heterocrystalline region removed:

[0068]

[0069]

[0070] Where p represents the number of currently selected cutting lines; l k represents the kth cutting line selected; m represents the total number of all candidate cutting lines, that is, the maximum number; represents the number of permutations and combinations of selecting p cutting lines from all m cutting lines; r e Indicates the path planned under the current e-th permutation and combination; shape e represents the shape of the wafer cut by the e-th path; Φ represents the value range of e; r opt1 Represents the currently selected optimal path, which consists of the p selected cutting lines.

[0071] In this specific implementation method, the cutting line planning result is as follows Figure 7As shown, the single crystal is divided into regions by finding the largest inscribed rectangle within the single crystal region outline, and then cutting lines are planned in each region. There will be no redundant cutting lines, and sufficient and non-redundant cutting lines can be planned.

[0072] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A semiconductor workpiece cutting path planning algorithm based on image processing, characterized in that: The following steps are involved: S1. Get the semantic boundary of the image: S1-1. Obtain a semantic image X through a deep learning model, and distinguish different crystal phase regions in the wafer according to the semantic image X, wherein the crystal phase region to be retained is a single crystal region, and the rest is a mixed crystal region; the semantic image X is set as a grayscale image, and the grayscale value of the single crystal region is different from the grayscale value of the mixed crystal region; S1-2, adjusting the color of the impure crystal region in the semantic image X obtained in step S1-1 to be the same as the background color, to obtain a single crystal image X1; S2. Find the largest inscribed rectangle MRect in the single crystal image X1: S2-1, find all the contours in the single crystal image X1, take the contour with the largest area as the processing contour of the required product, record it as contour contour, then store all the contour points on contour contour in order, and make the position difference between two adjacent contour points no greater than 1; S2-2, randomly select a contour point on the contour and record it as start(x,y), loop the points in the contour with start(x,y) as the starting point, record the current point on the contour as cur(x,y), and use the starting point start(x,y) and the current point cur(x,y) as the diagonal points of the rectangle to construct a rectangular contour; S2-3, judging whether the rectangular contours constructed in step S2-2 are all located inside the contour contour, and if so, recording the area of ​​the rectangular contour and the corresponding diagonal point coordinates in this state; S2-4, repeating steps S2-2 to S2-3, traversing all points on the contour contour, obtaining the inscribed rectangular contour maxRect with the largest area in the contour contour, and recording the coordinates and area of ​​the inscribed rectangular contour maxRect with the largest area; S2-5. Calculate the spatial moment m of the contour according to formula (1) ji : m ji =∑ x,y (x,y)*x j *y i ); (1) In formula (1), x and y are the coordinate values ​​of the contour, i and j are the image orders; S2-6. When the image order i = 0 or 1, j = 0 or 1, the m obtained by formula (2) 00 、m 10 、m 01 Calculate the centroid center(x, y) of the contour contour: S2-7, taking the centroid center(x, y) of the contour contour determined in step S2-6 as the center, rotating the single crystal image X1, and obtaining a single crystal image X1 every 5° rotation. i , i is an integer greater than 1, a total rotation of 360°, in each single crystal image X i Repeat steps S2-2 to S2-6 to obtain each single crystal image X i All the contours in contour; S2-8. Compare each single crystal image X i The area of ​​the largest inscribed rectangle contour maxRect among all the contours contour, and the largest inscribed rectangle MRect in the single crystal region is determined; S3. Split the contour area within the single crystal area and find the corresponding cutting line: S3-1, extending the single sides of the four sides of the inscribed rectangle MRect outward respectively, and the two extension lines at each vertex of the inscribed rectangle MRect divide the single crystal area outside the inscribed rectangle MRect into four areas Con1, Con2, Con3, and Con4; S3-2, select the vertex or extension line of the inscribed rectangle MRect in the Con1 area and the intersection point with the outline of the single crystal area as P1(x, y), and form a straight line with any point P2(x, y) in the Con1 area, loop all the points in the Con1 area, calculate the area of ​​the impure crystal area removed by the straight line formed by P2(x, y) and P1(x, y), and sort them according to the size of the removed area of ​​the impure crystal area, and select the group with the largest removed area of ​​the impure crystal area as the candidate cutting line L1; if the removed impure crystal areas are the same, select the group with the largest area of ​​the single crystal area as the candidate cutting line L1; Select another vertex or extension line of the inscribed rectangle MRect in Con1 and mark the intersection point with the outline of the single crystal region as P3(x, y), and form a straight line with any point P2(x, y) in the Con1 region. Loop through all points in the Con1 region, calculate the area of ​​the impurity crystal region removed by the straight line formed by P2(x, y) and P3(x, y), and determine the candidate cutting line L2 in the same way as step S3-2; Similarly, the candidate cutting lines L3, L4, L5, and L6 corresponding to the Con2 area, the Con3 area, and the Con4 area are obtained respectively; S4. Determine the final cutting line: Traverse the six candidate cutting lines L1, L2, L3, L4, L5 and L6 determined in step S3, and select the path that can form a convex polygonal single crystal area with the largest area, the largest area of ​​the removed miscellaneous crystal area, and the smallest number of straight line cuts as the final cutting line path. The path optimization method is: Assume that the area of ​​the single crystal pixel formed by segmentation is S x , A(·) represents the area of ​​the single crystal region removed, and Z(·) represents the area of ​​the heterocrystalline region removed: Where p represents the number of currently selected cutting lines; l k represents the kth cutting line selected; m represents the total number of all candidate cutting lines, that is, the maximum number; represents the number of permutations and combinations of selecting p cutting lines from all m cutting lines; r e Indicates the path planned under the current e-th permutation and combination; shape e represents the shape of the wafer cut by the e-th path; Φ represents the value range of e; r opt1 Represents the currently selected optimal path, which consists of the p selected cutting lines.

Citation Information

Patent Citations

  • Pattern size measuring device, and pattern area measuring method

    CN101523155A

  • Quick pose detection method based on geometric features of inside and outside contours of workpiece

    CN106846404A