Detection Method for Dense Gold Wires on Chip Surface

Abstract

The present invention relates to the technical field of chip detection and image processing, and discloses a method for detecting dense gold wires on the surface of a chip, comprising the following steps: 1. Read in the collected image of the dense gold wires on the surface of the chip, the detection area and the number of gold wires; 2. Divide the gold wires within the detection area into several segments, and fit each segment of the gold wire into a straight line segment; 3. Select a number of radial cross-sections at equal intervals in the area of each straight line segment, and use Gaussian curve fitting to obtain the position of the center point of each cross-section; 4. If the number of center points of the cross-sections is equal to the number of gold wires, assign the center points of the cross-sections to the point sequences corresponding to the straight line segments. If they are not equal, use the center points as spare points; 5. Perform linear fitting on all the point sequences on the same straight line segment; 6. Repeat steps 3 to 5 in other segmented areas of the detection area; 7. Connect all the fitted straight line segments into complete gold wires. The detection performance of the present invention is excellent, with strong robustness and good compatibility.

Description

Technical Field

[0001] The present invention relates to the technical field of chip detection and image processing, and particularly to a method for detecting dense gold wires on the surface of a chip. Background Art

[0002] The chip manufacturing process is developing towards a higher density direction, which poses higher requirements for the optical inspection imaging detection technology on the chip surface. Especially in the detection of dense lines, it faces great challenges. Usually, algorithms such as edge detection, region growing detection, and curvelet transform detection are used in online target detection. However, these algorithms will be significantly interfered when the online targets gradually become dense, thus affecting the detection performance and resulting in serious misdetection. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems, and provide a method for detecting dense gold wires on the surface of a chip. Based on the characteristics of the cross-sectional curve of the optical response image of the gold wire, the least square fitting of the Gaussian function is used to find the position points of the gold wire at the cross-section, and the position points of the corresponding gold wire are allocated using the quantity relationship of the gold wires. The detection performance is excellent, the robustness is strong, and the compatibility is good.

[0004] The technical solution adopted by the present invention is as follows:

[0005] A method for detecting dense gold wires on the surface of a chip, characterized by including the following steps:

[0006] Step 1: Read in the acquired image of the dense gold wires on the chip surface, the detection area, and the number of gold wires.

[0007] Step 2: Divide the gold wires in the detection area into several segments, and fit each segment of the gold wire into a straight line segment.

[0008] Step 3: Select several radial cross-sections at equal intervals in each straight line segment area, and use Gaussian curve fitting to obtain the center point position of each cross-section.

[0009] Step 4: If the number of center points of the cross-section is equal to the number of gold wires, then allocate the center points of the cross-section to the point sequence of the corresponding straight line segment. If not, then use the center points as spare points.

[0010] Step 5: Perform linear fitting on all the point sequences on the same straight line segment.

[0011] Step 6: Repeat Steps 3 to 5 in other segmented areas of the detection area.

[0012] Step 7: Connect all the fitted straight line segments into a complete gold wire.

[0013] Further, after Step 4, it further includes:

[0014] Step 4-1: Add the spare point to the point sequence of the nearest straight line segment.

[0015] Furthermore, in the said Step 3, the process of using Gaussian curve fitting based on the peak-valley data of the cross-section is as follows:

[0016] Describe a set of data with a Gaussian function

[0017]

[0018] σ is the half-peak width of the Gaussian curve, y max and x c are the peak height and peak position respectively, y i is the cross-section gray value, x i is the cross-section pixel value,

[0019] Take the natural logarithm of both sides of the above formula

[0020]

[0021] Denote

[0022] Y i = ln y i

[0023]

[0024]

[0025]

[0026] Then there is the following quadratic fitting function

[0027]

[0028] Use all the data and write it in matrix form

[0029]

[0030] That is

[0031] Y N×1 = X N×3 · A 3×1

[0032] Based on the least squares fitting principle, it can be obtained

[0033]

[0034] Then

[0035]

[0036] Furthermore, the straight line fitting algorithm in the said Step 5 is as follows:

[0037] Define the straight-line equation

[0038] y = ax + b

[0039] Write it in matrix form

[0040]

[0041] That is

[0042] Y N×1 = X N×2 ·A 2×1

[0043] Based on the least squares fitting principle, it can be obtained that

[0044]

[0045] Furthermore, a point that belongs to the same point on two or more gold wires is defined as "crossing"; a point where there is no center point on the cross-section of the gold wire is defined as "missing"; a point that does not belong to any gold wire is defined as "interference". After supplementing the point set, the point sequence on the same gold wire is linearly fitted again. When fitting, points with too large a distance are excluded, and their positions are recorded and defined as "interference".

[0046] Furthermore, after all the gold wire segments are fitted, they are connected into a complete gold wire, and all detection anomalies are merged.

[0047] The beneficial effects of the present invention are as follows:

[0048] (1) Based on the characteristics of the cross-sectional curve of the optical response image of the gold wire, the present invention uses the least squares fitting of the Gaussian function to find the position points of the gold wire at the cross-section, realizing the fitting of the gold wire;

[0049] (2) Allocate the position points of the corresponding gold wires according to the number relationship of the gold wires;

[0050] (3) By applying the linear fitting algorithm twice, after the first linear fitting, the spare points at the cross-section are added to the gold wire point sequence; the second linear fitting will exclude the abnormal points;

[0051] (4) It can detect defects such as broken wires, interference, and crossing, and can be compatible with the situation where one gold wire is split into two wires due to different imaging conditions. Description of the Drawings

[0052] Attached Figure 1 is the flow chart of the detection method of the present invention;

[0053] Attached Figure 2 is a schematic diagram of dividing the gold wire into two approximately straight lines;

[0054] Attached Figure 3It is a schematic diagram of the coordinate transformation of the cross-section of the gold wire;

[0055] Appendix Figure 4 It is a schematic diagram of the cross-section transformation of multiple gold wires within a region;

[0056] Appendix Figure 5 It is a schematic diagram of the cross-section transformation when a gold wire is split into two from the middle;

[0057] Appendix Figure 6 It is a schematic diagram of multiple cross-section transformations in the case where the number of nodes is twice the number of gold wires;

[0058] Appendix Figure 7 It is a schematic diagram of calculating the distance from a spare point to each gold wire within a region;

[0059] Appendix Figure 8 It is a schematic diagram of the center points of the gold wires extracted from all cross-sections of a small region containing 2 gold wires in the example;

[0060] Appendix Figure 9 It is a schematic diagram of the result after linear fitting for each gold wire; [[ID= 32]]

[0061] Appendix Figure 10 It is a schematic diagram of the result after linear fitting after processing the candidate points for each gold wire. Specific implementation manner

[0062] The following will make a detailed description of the specific implementation manner of the method for detecting dense gold wires on the chip surface of the present invention with reference to the accompanying drawings.

[0063] Refer to Appendix Figure 1 , the method for detecting dense gold wires on the chip surface of the present invention has the following process:

[0064] First, read in the image to be processed and the region where gold wire detection is required, and input the number of gold wires within the region; then run the region segmentation program and select the equal division method for segmentation; for each small region, obtain cross-sections at equal intervals along the gold wire direction and extract the center points of each gold wire in this cross-section.

[0065] Based on the characteristics of the cross-section curve of the optical response image of the gold wire, the present invention uses the least squares fitting of the Gaussian function to find the position points of the gold wire at the cross-section, and uses the gold wire quantity relationship to allocate the corresponding position points of the gold wire. The present invention uses the linear fitting algorithm twice. After the first linear fitting, the spare points at the cross-section are added to the gold wire point sequence; the second linear fitting will eliminate the abnormal points. The present invention can detect defects such as broken wires, interference, and crossovers, and can be compatible with the situation where one gold wire is split into two due to different imaging conditions.

[0066] The algorithm steps described in the present invention are as follows: (1) Read in information such as images, dense gold wire detection area information, and the number of gold wires; (2) Segment the gold wire detection area along the direction of the gold wire so that each gold wire within each segment is approximately a straight line; (3) Select several radial cross-sections at equal intervals in each straight line segment area for analysis. The cross-section curve is approximately a Gaussian distribution curve, and the central position of the wire is obtained by fitting the Gaussian curve; (4) The central points that meet the gold wire quantity condition on the cross-section curve are assigned to the corresponding gold wire point sequence, and the central points that do not meet the condition are used as spare points; (5) Perform linear fitting on all point sequences on the same gold wire; (6) According to the principle of the distance from a point to a straight line, add the spare points that have not participated in the gold wire allocation. If the same point belongs to two or more gold wires, it is defined as "crossing". If there is no central point of the gold wire at a certain cross-section, it is defined as "missing". Points that do not belong to any gold wire are defined as "interference"; (7) After supplementing the point set, perform linear fitting on the point sequence on the same gold wire again. When fitting, eliminate the points with too large distances and record their positions, which are defined as "interference"; (8) Repeat the process of (3) to (7) for other segmented areas of the current detection area; (9) Connect all gold wire segments into a complete gold wire and merge all detection anomalies, that is: crossing, missing, interference, etc.

[0067] See the appendix Figure 2 , where, after segmenting the gold wire area in step (2), each segment is approximately a straight line segment. In the figure, a segment of the gold wire is divided into two approximate straight lines, 1 and 2.

[0068] See the appendix Figure 3 , in step (3), several cross-sections can be selected at equal intervals in each straight line segment area for analysis. The cross-section curve is approximately a Gaussian distribution curve, and the central position of the wire can be obtained by fitting the Gaussian curve.

[0069] According to Figure 3 the curve in, describe a set of data with a Gaussian function

[0070]

[0071] σ is the half-peak width of the Gaussian curve, y max and x c are the peak height and peak position respectively.

[0072] Take the natural logarithm of both sides of the above formula

[0073]

[0074] Denote

[0075] Y i = ln y i

[0076]

[0077]

[0078]

[0079] There is the following quadratic fitting function

[0080]

[0081] Using all the data and writing it in matrix form

[0082]

[0083] That is

[0084] Y N×1 = X N×3 · A 3×1

[0085] Based on the least squares fitting principle, we can obtain

[0086]

[0087] Then

[0088]

[0089] See Appendix Figure 4 , multiple gold wires are allowed in a region, which are manifested as multiple peaks (or valleys) on the profile curve. The central position of each peak (or valley) is obtained by fitting with a Gaussian curve. When the number of central points corresponds to the number of gold wires, the central points are sequentially assigned to each gold wire. For the case where there is a missing gold wire at the profile in the region, the number of peak (or valley) central position points calculated is inconsistent with the number of gold wires. The central coordinate point sequence is saved for later use. After the gold wires are fitted, the belonging of each central point is determined by calculating the distance from the point to the line.

[0090] The x c1 , x c2 , x c3 coordinates obtained by curve fitting calculation need to be translated and rotated to obtain the actual central coordinates

[0091]

[0092] See Appendix Figure 5 , for the case where a gold wire is split into two in the middle due to imaging or other reasons, two central position points are also obtained according to the algorithm in step (3), and the average coordinate of the two point coordinates is calculated, which is approximately the central point of the actual gold wire.

[0093]

[0094] See Appendix Figure 6, in step (4), for the case where the number of points obtained for the cross-sectional curve is equal to the number of gold wires, each gold wire is assigned one point and saved to the corresponding point set; for the case where the number of points is twice the number of gold wires, each gold wire is assigned two neighboring points, and the average value of the two points is saved to the corresponding point set; the cross-sectional points in other cases are saved for later use. Attached Figure 6 The three different cross-sectional cases in

[0095] , and the point sets for the three gold wires.

[0096] The straight-line equation

[0097] y = ax + b

[0098] Written in matrix form

[0099]

[0100] That is

[0101] Y N×1 = X N×2 · A 2×1

[0102] Based on the principle of least squares fitting, we can obtain

[0103]

[0104] See attached Figure 7 , in step (6), calculate the distance from the spare points to each gold wire within the region. Based on the principle of the closest distance and meeting the distance range requirements, each spare point is assigned to the gold wire point set. For the case where two points belong to the same straight line, calculate their average value as the gold wire point; for the case where one point belongs to two straight lines, record its cross-sectional position, and output the cross-wire result at this position in the final result; for the case where there is no belonging point for the straight line, record its cross-sectional position, and output the missing gold wire segment result at this position in the final result.

[0105] Through the distance criterion, among the above-mentioned spare cross-sectional points, (x p1 , y p1 ), (x p4 , y p4 ) belong to the first and third gold wires respectively, (x p2 , y p2 ), (x p3 , y p3 ) belong to the second gold wire, and recalculate their average value point as the point on the second gold wire.

[0106] In step (7), perform straight line fitting on the gold wire point sequence of the supplementary point set again, and the algorithm is the same as that in step (5). After the fitting is completed, eliminate the points that cause large errors and record their cross-sectional positions. For a small number of error points, they can be judged as interference, and for a large number of error points, they are judged as curved lines. Output the interference detection or curved line result at this position in the final result.

[0107] In step (9), connect the gold wire segments detected in each segment end to end in sequence to form a complete gold wire, and merge all detection abnormalities, that is: crossing, missing, interference, etc.

[0108] See the appendix Figure 8 , through the above method, an example is illustrated. In the test in the appendix Figure 8 , in a small area containing 2 gold wires, the center points of the gold wires extracted at all cross-sections are sorted at their respective cross-sections, and screened based on the principle of 1 times the number of gold wires and the principle of 2 times the number of gold wires. For each gold wire that meets the 1-fold principle, assign one point, and for each gold wire that meets the 2-fold principle, assign 2 points, and calculate the average value and save it to the corresponding gold wire point sequence. Then perform straight line fitting on each gold wire, and the final result is as Figure 9 shown.

[0109] For the candidate center points of the remaining cross-sections, calculate their distances to each gold wire respectively. If the distance meets the threshold requirement and is the smallest, then assign this point to the corresponding gold wire point sequence. If there is no suitable point to assign to a certain gold wire at the cross-section, output the defect "missing"; if the same point meets the requirements of more than two gold wires, output the defects "overlap" or "crossing". After the candidate points are processed, perform straight line fitting on each gold wire again, and eliminate the abnormal points that cause large errors, and output the defect category "interference". The gold wire detection result is as Figure 10 shown.

[0110] Finally, connect the gold wire segments of all small areas and merge all detection results, that is, complete the gold wire detection of the area to be detected in the current image.

[0111] The above is only the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the principle of the present invention, several improvements and retouches can still be made, and these improvements and retouches should also be regarded as the protection scope of the present invention.

Claims

1. A method for detecting dense gold wires on the surface of a chip, characterized in that: The steps are as follows: Step 1: Read in the collected image of the dense gold wires on the chip surface, the detection area, and the number of gold wires; Step 2: Divide the gold wires within the detection area into several segments, and fit each segment of the gold wire into a straight line segment; Step 3: Select a number of radial cross-sections at equal intervals in each straight line segment area, and use Gaussian curve fitting to obtain the center point position of each cross-section; Step 4: If the number of center points of the cross-sections is equal to the number of gold wires, then assign the center points of the cross-sections to the point sequence corresponding to the straight line segment. If not, then use the center points as spare points; Step 5: Perform linear fitting on all the point sequences on the same straight line segment; Step 6: According to the principle of the distance from a point to a straight line, add the spare points that did not participate in the gold wire assignment: If the same point belongs to two or more gold wires, it is defined as "crossing"; If there is no center point of the gold wire at a certain cross-section, it is defined as "missing"; If a point does not belong to any gold wire, it is defined as "interference"; Step 7: After supplementing the point set, perform linear fitting on the point sequences on the same gold wire again. When fitting, exclude the points with too large distances and record their positions, which are defined as "interference"; Step 8: Repeat Steps 3 to 7 for other segmented areas in the current detection area; Step 9: Connect all the gold wire segments into a complete gold wire, and merge all the detection anomalies, namely: crossing, missing, and interference.

2. The chip surface dense gold wire detection method according to claim 1, characterized in that: After the above Step 4, it further includes: Step 4-1: Add the spare points to the point sequence of the nearest straight line segment.

3. The method for detecting dense gold wires on the chip surface according to claim 2, wherein: In the above Step 3, according to the peak-valley data of the cross-section, the process of using Gaussian curve fitting is as follows: Describe a set of data with a Gaussian function σ is the full width at half maximum of the Gaussian curve, y max and x c are the peak height and peak position respectively, y i is the cross-sectional gray value, x i is the cross-sectional pixel value, Take the natural logarithm of both sides of the above formula Denote Y i = ln y i Then there is the following quadratic fitting function Use all the data and write it in matrix form That is Y N×1 = X N×3 · A 3×1 Based on the least squares fitting principle, it can be obtained that Then 4. The method for detecting dense gold wires on the chip surface according to claim 3, wherein: The linear fitting algorithm in the above Step 5 is as follows: Define the straight line equation y = ax + b Write it in matrix form That is Y N×1 = X N×2 · A 2×1 Based on the least squares fitting principle, it can be obtained that

Images