Detection area formation method, grain surface detection method, and grain surface hotspot determination method

Abstract

A detection area formation method, a grain surface detection method, and a grain surface hotspot determination method. The detection area formation method comprises: selecting characterization areas on an integrated circuit layout and forming a detection area from the characterization areas, wherein the characterization areas are used for characterizing local areas having a similar structure, and the size of the local areas is within a range of 0.5 micrometers to 50 micrometers. The grain surface detection method uses the detection area formed by the detection area formation method to perform morphology detection on the surface of a grain. The grain surface hotspot determination method determines, on the basis of a result of the morphology detection, whether a hotspot exists. Compared with the prior art, the present invention has a higher detection compression ratio, which is more conducive to popularizing application of a high-resolution detection means in a production line.

Description

Detection area forming method, die surface detection method, and die surface hotspot judgment method TECHNICAL FIELD

[0001] The present application relates to the field of detecting the surface of a die after CMP processing, and in particular to a detection area forming method, a die surface detection method, and a die surface hotspot judgment method. BACKGROUND

[0002] Chemical mechanical polishing (CMP), also known as chemical mechanical planarization, is a key process for achieving global uniform planarization of a wafer in a semiconductor manufacturing process, and is also the most frequently used process in the manufacturing process (about 25% of all processes). As shown in FIG. 1, the purpose of the CMP process is to achieve planarization of the wafer surface. In the drawings of the present patent, the dark portions are semiconductor substrates, and the light portions are functional portions of metal material. In the present patent, the functional portion refers to other portions in addition to the semiconductor substrate. The most typical functional portion is of metal material, and can also include an insulating portion. The metal portion is often used to realize the function of the circuit, such as connection and contact. The insulating portion is used for insulation and isolation, and is often formed by filling a trench of semiconductor material with insulating material.

[0003] Whether a CMP process has defects is mainly determined by whether a die on the wafer has a hotspot. The hotspot will have a continuous impact on subsequent processes of semiconductor processing, and ultimately lead to a decrease in yield. Therefore, determining whether a die has a hotspot after CMP processing is very important for improving yield. The generation of the hotspot can be caused by the design of the integrated circuit layout, or by the material used in the CMP process or the related parameters of the CMP process. Therefore, determining whether a die has a hotspot after CMP processing is also very important for improving the design of the layout and adjusting the CMP process. In reality, there are mainly two manifestations of hotspots after the CMP process. One manifestation is erosion of the functional portion, as shown in FIG. 2. The erosion occurs in the functional portion of metal material, and generally appears along the width of the metal line. Therefore, the detection of erosion needs to detect the topography (especially the height) at the nanometer level in the size range of nanometers or tens of nanometers. The other manifestation is dishing of the surface area, as shown in FIG. 3. The dishing occurs in the surface area of micrometers or tens of micrometers, and is strongly related to the line width of the functional portion in the surface area and the density of the functional portion. Similarly, nanometer-level detection of the topography is also needed.

[0004] In order to determine whether there is a hot spot on the die on the wafer, it is necessary to detect the surface topography of the die (also known as roughness detection) after each CMP process. The resolution of such detection should be nanometer level, and in order to achieve line efficiency, such detection should be completed within 4 hours.

[0005] In the prior art, a white light interferometer is the default option for detecting the surface topography of the die. Blunt et al. used a white light interferometer to measure the surface roughness of a semiconductor polished substrate and an epitaxial layer, and Zhang et al. also applied it to measure the step height of a sputtered functional layer. Studies have shown that although white light interferometric measurement basically meets the requirements in terms of longitudinal resolution (0.1 nanometer) and measurement time (about 8 hours for each die), its lateral resolution is micrometer level or slightly less than 1 micrometer, so the detection accuracy cannot meet the requirements. Moreover, when the thickness of the die to be measured is too small (less than 50 nanometers), the die itself will appear translucent, at which time the measurement result will be affected due to light scattering. To this end, a layer of material needs to be applied to the surface of the die to enable white light interferometric measurement, but at the same time the die will be damaged.

[0006] Atomic force microscopy (AFM) can achieve a spatial resolution of 1 nm in the X-axis and Y-axis directions, and has a higher resolution in the Z-axis direction (height direction), which can provide a truly three-dimensional atomic-level surface image. At the same time, AFM does not require any special treatment of the sample, and the sample has a small chance of being damaged during preparation, so it can be used for non-destructive testing. As one of the powerful nanoscale resolution surface topography detection tools, AFM completely meets the resolution (accuracy) requirements of semiconductor die surface topography detection, and will not cause damage to the die during measurement. Xu et al. used AFM to non-destructively measure the density of etch pits generated after CMP of a GaN wafer, and Shi et al. also first realized in-situ observation of AFM of local CMP behavior on sapphire. However, the imaging speed of AFM is slow. For example, at a lateral resolution of 10 nm, it takes about 5 minutes to image an area of 100 square micrometers. At this time, it takes several weeks to several months to measure a typical 33x27 mm 2 die. Therefore, it is difficult to popularize the use of AFM and other detection methods with high detection resolution for die surface topography detection on the production line. SUMMARY

[0007] The present application aims to overcome the above-mentioned defects or problems in the background art, and provides a detection area forming method, a die surface detection method, a die surface hotspot judgment method, a computer program, a computing device and a detection device, which have a higher detection compression ratio compared with the prior art, and are more conducive to popularizing the detection means with high resolution on the production line.

[0008] To achieve the above-mentioned purpose, the following technical solutions are adopted:

[0009] The first technical solution relates to a detection area forming method for determining a detection area for surface detection of a die after CMP processing, which selects a representative area on an integrated circuit layout and forms a detection area from each representative area; the representative area is used to represent a local area with similar structure, and the size of the local area in the X-axis and Y-axis is in the range of 0.5 microns to 50 microns.

[0010] The second technical solution is based on the first technical solution, wherein the size of the local area in the X-axis and Y-axis is in the range of 10 microns to 30 microns.

[0011] The third technical solution is based on the second technical solution, wherein the size of the local area in the X-axis and Y-axis is 20 microns.

[0012] The fourth technical solution is based on the first technical solution, which includes the following steps: region division is performed on a to-be-detected part on an integrated circuit layout to form a plurality of local areas, and all local areas cover the to-be-detected part; the to-be-detected part covers all processing area groups, and each processing area group includes processing areas with similar structures; all local areas are classified based on structural similarity to obtain local area groups; at least one local area is selected from each local area group as a representative area, and the set of all representative areas forms a detection area.

[0013] The fifth technical solution is based on the fourth technical solution, wherein adjacent local areas overlap with each other.

[0014] The sixth technical solution is based on the fifth technical solution, wherein the local areas are obtained by sliding a virtual window on the to-be-detected part.

[0015] The seventh technical solution is based on the fourth technical solution, wherein the to-be-detected part is determined by the following steps: edge diffusion is performed on each processing area in the integrated circuit layout to form a corresponding diffusion area; all diffusion areas are classified based on structural similarity to obtain diffusion area groups; at least one diffusion area is selected from each diffusion area group as a to-be-detected area, and the set of all to-be-detected areas forms the to-be-detected part.

[0016] The eighth technical solution is based on the seventh technical solution, wherein the structural similarity is rotational symmetry and / or mirror symmetry.

[0017] The ninth technical solution is based on the seventh technical solution, wherein the to-be-detected part is the shortest in the first path in each alternative to-be-detected part; the alternative to-be-detected part is a set of alternative to-be-detected regions, the alternative to-be-detected part covers all diffusion region groups and each diffusion region group has and only has a preset number of alternative to-be-detected regions; and the first path is the shortest one-way path in the alternative to-be-detected part passing through the center points of all alternative to-be-detected regions.

[0018] The tenth technical solution is based on any one of the fourth to ninth technical solutions, wherein the detection region is the shortest in the second path in each alternative detection region; the alternative detection region is a set of alternative characterization regions, the alternative detection region covers all local region groups and each local region group has and only has a preset number of alternative characterization regions; and the second path is the shortest one-way path in the alternative detection region passing through the center points of all alternative characterization regions.

[0019] The eleventh technical solution relates to a grain surface detection method for detecting a grain surface after CMP processing, which is based on the detection region formation method according to any one of the first to tenth technical solutions to form a detection region and perform topographic detection on the grain surface.

[0020] The twelfth technical solution is based on the eleventh technical solution, wherein the topographic detection on the grain surface is performed by using an AFM.

[0021] The thirteenth technical solution is based on the twelfth technical solution, wherein the sizes of the X and Y axes of the local region are less than or equal to the sizes of the X and Y axes of the maximum imaging range of the AFM.

[0022] The fourteenth technical solution relates to a grain surface hotspot judgment method for judging whether a grain surface after CMP processing has a hotspot, which performs topographic detection on the grain surface by using the grain surface detection method according to any one of the tenth to thirteenth technical solutions, and judges that the grain surface has a hotspot if there is erosion of the functional part and / or depression of the characterization region in the detection result.

[0023] Compared with the prior art, the above-mentioned solutions have the following beneficial effects:

[0024] In the paper Hotspot Prevention Using CMP Model in Design Implementation Flow by Norma Rodriguez et al., it is pointed out that the planarization of CMP depends on the density and width of the metal part of the local region on the grain. At different densities and widths, different degrees of dish-shaped depression and erosion may occur.

[0025] In the paper "Hotspot detection and design recommendation using silicon calibrated CMP model" by Colin Hui et al., it is pointed out that recess and erosion occur in the CMP process, recess or erosion is formed in the lower layer, which leads to copper residue or copper accumulation in the upper layer. Recess and erosion are strongly dependent on the density and line width of the design pattern. Wide functions are more prone to recess than narrow functions.

[0026] From the above paper, it can be seen that the functional part structure of the local area on the surface of the wafer (reflected as a pattern structure on the integrated circuit layout) is a key factor affecting the hotspot distribution of the local area after the CMP process is completed. The structure of the functional part covers two important indicators of design pattern density and line width. Of course, the local area without a functional part should also be considered as a functional part structure. Local areas with different structures have different possibilities of forming hotspots during the CMP process.

[0027] The first technical solution is based on the law revealed in the above paper, and the local area is used to represent the local area with similar structure, so that among a large number of local areas with similar structure, only a small number of representative areas need to be detected, which can represent the possibility of hotspots in these local areas, thereby greatly improving the detection compression ratio, providing a basis for shortening the detection time of the wafer, and more conducive to popularizing the detection method with high resolution on the production line.

[0028] In the first technical solution, the size of the local area is comparable to the size of the surface area where recess may occur, so it can cover the detection of both recess and erosion hotspots.

[0029] In the first technical solution, the integrated circuit layout refers to the design layout of the integrated circuit on the wafer. In some processes, multiple CMP processes need to be performed, and if each CMP process has a different integrated circuit topology, the integrated circuit layout refers to the design layout corresponding to this CMP process. If it is a planar semiconductor integrated circuit, the integrated circuit layout is the final design layout of the integrated circuit. In the first technical solution, the integrated circuit layout is used to select the detection area, which has the advantages of being more convenient and shorter detection time compared to dividing the local area based on the macro image formed by low resolution in the detection process.

[0030] The second and third technical solutions are the optimization of the size of the local area.

[0031] The fourth technical solution provides a specific way to determine the characterization region and then form the detection region. In the fourth technical solution, the measurement part covers all the processing region groups, and all the local regions cover the measurement part, which can ensure that the set of local regions used for similarity classification can cover all the processing regions of the structure type.

[0032] In the fifth technical solution, the adjacent local regions overlap with each other, which can traverse all the functional part structures in the measurement part range, further reduces the influence of factors (such as the starting position) related to the division of the local region on the detection result, and is beneficial to the characterization region to better characterize the possibility of hot spot occurrence.

[0033] The sixth technical solution is a specific implementation of the fifth technical solution.

[0034] In the seventh technical solution, the measurement part is based on the diffusion region, which not only enables the measurement part to include all types of processing regions, but also can contain regions adjacent to the processing regions, and can more comprehensively express the structures of various local regions. Classify the diffusion regions based on structural similarity, and select at least one diffusion region in each diffusion region group as a measurement region to form a measurement part, which can reduce the area of the measurement part, reduce the number of local regions used for classification by using the repetition of the macrostructure of the grain surface, while ensuring that the functional part structures with different structural similarities are included in the measurement part, avoiding missed detection in the detection process.

[0035] In the eighth technical solution, the structural similarity is rotational symmetry and / or mirror symmetry, so it can cover all different functional part structures.

[0036] In the ninth technical solution, the measurement part has the shortest first path among the various alternative measurement parts, which can provide a basis for shortening the final detection path as much as possible under the premise of traversing all diffusion region groups, further shortening the detection time.

[0037] In the tenth technical solution, the detection region has the shortest second path among the various alternative detection regions, which can shorten the final detection path as much as possible under the premise of traversing all local region groups, further shortening the detection time.

[0038] The eleventh technical solution performs topographic detection on the grain surface based on the detection region formed by the above detection region forming method, which can use high compression ratio to shorten the detection time of the grain, and is more conducive to popularizing the detection method with high resolution on the production line.

[0039] The twelfth technical solution is a preferred implementation of the tenth technical solution, which uses AFM to perform surface topographic detection on the grain, which not only ensures the resolution compared to using a white light interferometer, but also performs non-destructive detection on the grain, no longer needs to smear materials on the grain surface, and avoids damaging the detected grain.

[0040] In the thirteenth technical solution, the size of the local region is less than or equal to the size of the maximum imaging range of the AFM, so that the AFM only needs to move the probe when detecting the characterization region, without moving other parts, and without causing errors in the detection field due to the movement of other parts. The correspondence between the detected characterization region in the actual detection process and the characterization region on the integrated circuit layout can be ensured.

[0041] In the fourteenth technical solution, the grain surface is detected by using the above grain surface detection method, and it is judged whether there is a hot spot on the grain surface based on the detection result. Compared with the prior art, it can more quickly judge whether a hot spot exists, and is more conducive to popularizing the detection means with high resolution on the production line. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments, the following briefly introduces the drawings needed to be used:

[0043] FIG. 1 is a schematic view of a partial longitudinal section of a wafer before and after a CMP process;

[0044] FIG. 2 is a schematic view of a partial longitudinal section of a wafer containing an erosion hot spot;

[0045] FIG. 3 is a schematic view of a partial longitudinal section of a wafer containing a recessed hot spot;

[0046] FIG. 4 shows an integrated circuit layout in embodiment one;

[0047] FIG. 5 shows a diffusion region in embodiment one;

[0048] FIG. 6 shows a diffusion region group in embodiment one;

[0049] FIG. 7 shows a tentative measurement part in embodiment one;

[0050] FIG. 8 shows a local region in embodiment one;

[0051] FIG. 9 shows a local region group in embodiment one;

[0052] FIG. 10 shows a detection path in embodiment one;

[0053] FIG. 11 shows a detection region and a detection path in embodiment one;

[0054] FIG. 12 shows an integrated circuit layout in embodiment two;

[0055] FIG. 13 shows a typical local region of a local region group in embodiment two;

[0056] FIG. 14 shows a detection path in embodiment two. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments will be clearly and completely described below with reference to the drawings. Embodiment One

[0058] Referring to FIG. 4, FIG. 4 shows the integrated circuit layout in embodiment one. In the embodiments, the integrated circuit layout refers to the design layout of the integrated circuit corresponding to the die. In some processes, multiple CMP processes need to be performed, and if each CMP process has a different integrated circuit topology, the integrated circuit layout refers to the design layout corresponding to the CMP process. If it is a planar semiconductor integrated circuit, the integrated circuit layout is the final design layout of the integrated circuit. Using the integrated circuit layout to select the detection area has the advantages of being more convenient and shorter detection time compared to dividing the local area based on the macro image formed by low resolution in the detection process.

[0059] As shown in FIG. 4, the size of the integrated circuit layout in the X-axis direction is 12.5 microns, and the size of the integrated circuit layout in the Y-axis direction is 8.5 microns in the embodiments. In the embodiments, there are six processing areas on the integrated circuit layout, which are A1 to A6. Each processing area only includes a functional part. That is, in the embodiments, the processing area does not include a non-functional part. The functional part in the embodiments is a metal part, specifically a copper layer. The sizes of A1, A3, A4, and A6 are the same, the size of the X-axis direction is 2 microns, and the size of the Y-axis direction is 3 microns; the sizes of A2 and A5 are the same, the size of the X-axis direction is 6 microns, and the size of the Y-axis direction is 3 microns. The distance between two adjacent processing areas along the X-axis direction is 0.25 microns, and the distance between two adjacent processing areas along the Y-axis direction is 0.5 microns. The distance from the processing area to the adjacent edge of the integrated circuit layout is 1 micron along the X-axis direction and also 1 micron along the Y-axis direction.

[0060] In the embodiments, based on the integrated circuit layout, the detection area E and the detection path F are formed by the following steps.

[0061] Step 1: respectively edge-diffuse each processing area in the integrated circuit layout to form a diffusion area, and the part to be measured is based on the diffusion area, so that the part to be measured can include all types of processing areas, can contain areas adjacent to the processing areas, and can more comprehensively express the structure of various local areas. For details, refer to FIG. 4 and FIG. 5. FIG. 4 and FIG. 5 show the diffusion area in the embodiments. As shown in FIG. 4, the red box in FIG. 4 is the diffusion area formed after the edge of the processing area A1 is diffused. In the embodiments, each edge of the processing area is diffused outward by 1 micron. The specific way of setting the diffusion distance will be described in detail later. As shown in FIG. 5, the processing areas A1 to A6 are diffused after the edge to form corresponding diffusion areas B1 to B6;

[0062] Step 2: classify all diffusion regions based on rotational symmetry and / or mirror symmetry to obtain diffusion region groups; in this embodiment, the classification manner of the diffusion region groups is structural similarity, specifically rotational symmetry and mirror symmetry. In other embodiments, it can be rotational symmetry or mirror symmetry, and in some cases where the accuracy requirement of the result is not high, it can also be classified based on Hamming distance and other structural similarity indicators. Classifying the diffusion regions based on rotational symmetry and / or mirror symmetry, and selecting at least one diffusion region in each diffusion region group as a quasi-measurement region to form a quasi-measurement part together, can reduce the area of the quasi-measurement part, reduce the number of local regions used for classification by using the repetitiveness of the macrostructure of the grain surface, and at the same time can ensure that all functional part structures are included in the quasi-measurement part C, avoiding missing detection in the detection process. For details, see FIG. 6. FIG. 6 shows the diffusion region groups in this embodiment. As shown in FIG. 6, the number of diffusion region groups in this embodiment is two, which are BC1 and BC2. Among them, the diffusion region group BC1 is the set of diffusion regions B1, B3, B4 and B6, wherein B1 and B3 are mirror symmetric along the X-axis direction, B1 and B4 are mirror symmetric along the Y-axis direction, and B1 and B6 are 180-degree rotationally symmetric with each other. The diffusion region group BC2 is the set of diffusion regions B2 and B5, wherein B2 and B5 are mirror symmetric along the Y-axis direction. Any one of the diffusion regions in the diffusion region group BC1 is not mirror symmetric or rotationally symmetric or mirror symmetric and rotationally symmetric with any one of the diffusion regions in the diffusion region group BC2;

[0063] Step 3: selecting at least one diffusion region in each diffusion region class group to form a quasi-measuring region, and forming a quasi-measuring part C by all quasi-measuring regions; in this embodiment, one diffusion region in each diffusion region class group is selected to form a quasi-measuring region. In this embodiment, the quasi-measuring part C has the shortest first path in each candidate quasi-measuring part. The candidate quasi-measuring part is a set of candidate quasi-measuring regions, the candidate quasi-measuring part covers all diffusion region class groups and each diffusion region class group has only a preset number of candidate quasi-measuring regions; the first path is the shortest one-way path in the candidate quasi-measuring part that passes through the center points of all candidate quasi-measuring regions. The quasi-measuring part C has the shortest first path in each candidate quasi-measuring part, which can provide a basis for shortening the final detection path F as much as possible under the premise of traversing all diffusion region class groups, thereby shortening the detection time. In this embodiment, the preset number is 1. Specifically, this embodiment includes 8 candidate quasi-measuring parts, each candidate quasi-measuring part is a set of two candidate quasi-measuring regions, and the two candidate quasi-measuring regions come from two diffusion region class groups respectively. The 8 candidate quasi-measuring parts are the set of B1 and B2, the set of B1 and B5, the set of B3 and B2, the set of B3 and B5, the set of B4 and B2, the set of B4 and B5, the set of B6 and B2, and the set of B6 and B5. In this embodiment, the first path is the path between the center points of the two candidate quasi-measuring regions in the candidate quasi-measuring part. When the number of diffusion region class groups exceeds 3, the first path is the shortest one-way path in the candidate quasi-measuring part that passes through the center points of all candidate quasi-measuring regions. According to the above rule, the first path of the candidate quasi-measuring part containing B1 and B2 is the shortest. Of course, the first paths of the candidate quasi-measuring parts containing B3 and B2, B4 and B5, and B5 and B6 are also the shortest. The above four candidate quasi-measuring parts can be selected as the quasi-measuring part C. Referring to FIG. 7, FIG. 7 shows the quasi-measuring part C in this embodiment. As shown in FIG. 7, in this embodiment, the quasi-measuring part C is a set of quasi-measuring regions B1 and B2. It should be noted that in other embodiments, the quasi-measuring part C can also be a part of the integrated circuit layout that covers all processing region class groups. The quasi-measuring part can also be the entire integrated circuit layout. Of course, if the entire integrated circuit design layout is selected as the quasi-measuring part C, the calculation time of forming the detection region E and the detection path F will be longer, which is not efficient;

[0064] Step 4: divide the to-be-tested part C into several local regions, all of which cover the to-be-tested part C; all of the local regions cover the to-be-tested part C, which can ensure that the set of local regions used for similarity classification can cover all the processing regions of the structure types. In this embodiment, the adjacent local regions overlap with each other, and in other embodiments, the adjacent local regions can also be adjacent to each other. The adjacent local regions overlap with each other, which can traverse all the functional part structures in the to-be-tested part range, further reduce the influence of factors (such as the starting position) related to the division of the local regions on the detection result, and facilitate the characterization region to better characterize the possibility of hotspot occurrence.

[0065] Referring to FIGS. 7 and 8, FIGS. 7 and 8 show the local regions in this embodiment. As shown in FIG. 7, in this embodiment, the size of the local region along the X-axis direction is 2 microns, and the size along the Y-axis direction is also 2 microns. Specifically, in this embodiment, the local region is obtained by sliding the virtual window of the red frame shown in FIG. 7 in the to-be-tested part C, and the size of the virtual window is the same as that of the local region. When sliding the virtual window, the step value is 1 micron each time. In order to be more efficient in detection, the size of the local region can be set to be between 0.5-50 microns, more preferably between 10 microns and 30 microns, and more preferably 20 microns. It should be noted that the size of the local region can or can not be related to the setting of the aforementioned diffusion distance. In order to improve efficiency, the diffusion distance can generally be set to be less than or equal to the size of the local region, and more preferably, the diffusion distance can be set to be the step value when sliding the window, i.e., the size of the overlapping part of the adjacent local regions. As shown in FIG. 8, according to the above-mentioned method of dividing the local regions, the number of local regions in this embodiment is 40, which are D1 to D40 respectively;

[0066] Step 5: Classify all local regions based on the similarity of functional part structure in the local region to obtain local region groups. The similarity of functional part structure should at least include the similarity of the proportion of functional part in the area of the local region, the similarity of the line width of the functional part. The structure of the functional part refers to the shape of the functional sub-domain if the functional part only contains one functional sub-domain, the shape of each functional sub-domain and the positional relationship between the functional sub-domains if the functional part contains more than two functional sub-domains, and a separate local region group if the local region does not contain a functional part. In the classification based on the similarity of the functional part structure, one classification method that may result in a large number of local region groups is to classify based on rotational symmetry and / or mirror symmetry, i.e. only local regions with rotational symmetry and / or mirror symmetry can be classified into a local region group. The embodiment adopts the classification method of classifying based on rotational symmetry and / or mirror symmetry. On this basis, local regions with rotational symmetry and / or mirror symmetry after a slight translation can also be classified into the same local region group. Specifically, refer to FIG. 9, which shows eight local region groups formed based on the above classification method in the embodiment. Among them, the local region group DC1 contains the local region D1; the local region group DC2 contains the local regions D2, D5 to D9, D11 and D12; the local region group DC3 contains the local regions D3, D4 and D10; the local region group DC4 contains the local regions D12, D15 to D19, D22, D25 to D29; the local region group DC5 contains the local regions D13, D14, D20, D23, D24 and D30; the local region group DC6 contains the local regions D33, D34 and D40; the local region group DC7 contains the local regions D32 and D35 to D39; and the local region group DC8 contains the local region D31.

[0067] Step 6: At least one local region is selected from each local region class group to form a representative region, and a set of all representative regions forms a detection region E; in this embodiment, the detection region E has the second shortest path among the alternative detection regions; the alternative detection region is a set of alternative representative regions, the alternative detection region covers all local region class groups and each local region class group has and only has a preset number of alternative representative regions; the second path is the shortest one-way path in the alternative detection region that passes through the center points of all alternative representative regions. The detection region has the second shortest path among the alternative detection regions, which can shorten the final detection path as much as possible on the premise of traversing all local region class groups, and further shorten the detection time. In this embodiment, the preset number is 1. Referring to FIG. 10 and FIG. 11, FIG. 10 and FIG. 11 show the detection region E and the detection path F. As shown in FIG. 10 and FIG. 11, in this embodiment, the detection region E includes 8 representative regions, which are D1, D2, D3, D12, D23, D33, D32 and D31 respectively. The 8 representative regions cover all 8 local region class groups and each local region class group has and only has one representative region. In this embodiment, the number of alternative representative regions is very large, and it can be intuitively seen from FIG. 10 that the second path (red path in FIG. 10) of the detection region E is the shortest. As shown in FIG. 11, the detection path F is obtained based on the second path of the detection region E, specifically, in this embodiment, the detection path F can be D1 sequentially through D2, D3, D12, D23, D33, D32 to D31.

[0068] In this embodiment, the above-mentioned detection region forming method can be realized by executing a computer program. In the computer program, the algorithm for generating the diffusion region class group from each diffusion region and the algorithm for generating the local region class group from each local region can be a combination of one or more of support vector machine (SVM), K-nearest neighbor algorithm (KNN), random forest, convolutional neural network (CNN), transfer learning algorithm and reinforcement learning algorithm, and this embodiment selects support vector machine (SVM). The algorithm for generating the pseudo-measurement part based on each diffusion region class group and the algorithm for generating the detection region E based on each local region class group can be a combination of one or more of genetic algorithm, Dijkstra algorithm, grid algorithm, particle swarm optimization (PSO) algorithm and ant colony optimization (ACO) algorithm. Specifically, this embodiment selects genetic algorithm.

[0069] The embodiment also discloses a detection device, which comprises the detection apparatus, the computing apparatus and the control apparatus. The detection apparatus is an atomic force microscope (AFM) for detecting the topography of the wafer after CMP processing and outputting the topography detection result. The AFM is used to detect the surface topography of the wafer, which can ensure the resolution and perform non-destructive detection of the wafer compared with the white light interferometer, so that the wafer under test is not damaged by smearing materials on the wafer surface. In the embodiment, the X-axis direction size of the maximum imaging range of the AFM is 2 microns, and the Y-axis direction size is 2 microns. Therefore, in the embodiment, the size of the local region is equal to the size of the maximum imaging range of the AFM, and in other embodiments, the size of the local region can be set to be less than the size of the maximum imaging range of the AFM. The size of the local region is less than or equal to the size of the maximum imaging range of the AFM, so that the AFM only needs to move the probe when detecting the characterization region, without moving other parts, and will not cause errors in the detection field due to the movement of other parts. The correspondence between the characterization region detected in the actual detection process and the characterization region on the integrated circuit layout can be ensured. The computing apparatus comprises an input unit, an output unit, a storage unit and a computing unit. The input unit is used to obtain the integrated circuit layout, the output unit is used to output the detection region and the detection path, the storage unit is used to store the computer program, and the computing unit is used to call and execute the computer program to generate the detection region and the detection path based on the integrated circuit layout. The control apparatus is used to obtain the detection region and the detection path from the computing apparatus to control the detection apparatus to detect the topography of the wafer in the detection region according to the detection path. In the embodiment, the AFM detects the topography in the selected local region by moving the probe, and the movement of the probe can adopt a raster scanning mode or a line scanning mode. The raster scanning mode is adopted in the embodiment. In the embodiment, the movement of the AFM between the characterization regions is mainly realized by the displacement of other parts except the probe, such as the movement of the cantilever or the support arm or other mechanisms equipped with the probe along the X-axis and the Y-axis.

[0070] The embodiment also discloses a crystal grain surface hot spot judgment method for judging whether the crystal grain surface after CMP processing has a hot spot. Specifically, the crystal grain surface hot spot judgment method judges whether there is erosion of the functional part and / or depression of the characterization area based on the detection result output by the detection device. Optionally, if the height of the functional part in each characterization area is lower than the height of the semiconductor substrate and the height difference exceeds a first threshold value, it is judged that there is erosion of the functional part. The first threshold value can be set to 10 nanometers, or can be specifically set according to process requirements. If the overall height of the characterization area with the functional part is lower than the overall height of the characterization area without the functional part or with less functional part, and the height difference exceeds a second threshold value, it is judged that there is depression of the characterization area. The second threshold value can be set to 20-50 nanometers, or can be specifically set according to process requirements. As long as any one of erosion or depression exists, it is judged that the crystal grain surface has a hot spot. The crystal grain surface is detected by using the above-mentioned crystal grain surface detection method, and whether the crystal grain surface has a hot spot is judged based on the detection result. Compared with the prior art, whether the hot spot exists can be judged more quickly, and it is more conducive to popularizing the detection method with high resolution on the production line.

[0071] In the embodiment, the total area of the crystal grain is 106.25 square microns, the detection area E includes 8 characterization areas, and the area of each characterization area is 4 square microns, so the area of the detection area is 32 square microns, and the compression ratio is 3.32. Of course, the embodiment is only for illustrating the method of the patent, and in actual application, due to the high repeatability of the processing area and the high similarity of the local area, the actual compression ratio is much larger than that of the embodiment. Embodiment two

[0072] Referring to FIG. 12, FIG. 12 shows a schematic diagram of the integrated circuit layout of the crystal grain in Embodiment Two. As shown in FIG. 12, the integrated circuit layout in Embodiment Two contains two types of processing regions, which are arranged along the X-axis and repeated as a repeating unit. The entire integrated circuit layout has 48 repeating units along the X-axis direction and 96 repeating units along the Y-axis direction. In one repeating unit, the processing region on the left has a size of 10 microns along the X-axis and a size of 10 microns along the Y-axis, and has both functional parts and non-functional parts. The functional parts are lighter in color, which are copper layers, and the non-functional parts are darker in color, which are semiconductor substrates. The functional parts include eight square rings nested with each other, each ring has a line width of 0.3 microns, and each ring is spaced apart from each other by 0.3 microns. The processing region on the right has a size of 9.9 microns along the X-axis and a size of 10 microns along the Y-axis, and also has both functional parts and non-functional parts. The functional parts are lighter in color, which are copper layers, and the non-functional parts are darker in color, which are semiconductor substrates. The functional parts only have one, which is sandwiched by several square non-semiconductor substrate sub-domains. The semiconductor substrate sub-domains are square, each has a size of 0.3 microns along the X-axis and a size of 0.3 microns along the Y-axis, and each is spaced apart from each other by 0.3 microns along the X-axis and by 0.3 microns along the Y-axis. In each repeating unit, the spacing between the left and right processing regions is 0.3 microns. Along the X-axis, the spacing between two repeating units is 0.3 microns. Along the Y-axis, the spacing between two repeating units is 0.3 microns. Based on the above dimensions, in Embodiment Two, the area of each repeating unit is 20120 square microns, there are 4608 repeating units in total, and the area of the entire integrated circuit layout is approximately 92712960 square microns.

[0073] The detection regions and the detection path of Embodiment Two can be obtained in the same way as Embodiment One. Different from Embodiment One, in Embodiment Two, the diffusion distance is 5 microns, the size of the local region along the X-axis is 10 microns, the size of the local region along the Y-axis is 10 microns, and the step value is 5 microns. In this embodiment, the method for obtaining local region classes based on structural similarity is to classify local regions with rotational symmetry and / or mirror symmetry after a slight translation into the same local region class. Referring to FIG. 13, FIG. 13 shows the characterization regions of each local region class in this embodiment. As shown in FIG. 13, there are 11 local region classes in this embodiment, which are DC01 to DC11, and the characterization regions in each local region class are shown in FIG. 13. FIG. 14 shows the detection path in this embodiment. As shown in FIG. 14, the detection path in this embodiment is indicated by the red line. The corresponding detection regions are also determined by the set of characterization regions.

[0074] The computing device for outputting the detection area and the detection path in the second embodiment is basically the same as that in the first embodiment, and the detection device for detecting the topography of the wafer after CMP processing is also basically the same as that in the first embodiment, except that the maximum imaging range of the AFM is 10 microns x 10 microns.

[0075] The wafer surface detection method and the wafer surface hotspot judgment method in the second embodiment are the same as those in the first embodiment.

[0076] In the second embodiment, the detection area includes 11 representative areas, each representative area has an area of 100 microns, the detection area has an area of 1100 square microns, and the detection compression ratio is 84284.5. Therefore, by using the detection area forming method of the present patent, a very large detection compression ratio can be achieved, which is conducive to greatly shortening the detection time and facilitating the popularization and application of high-resolution detection means on the production line.

[0077] As can be seen from the above embodiments, the representative areas are used to represent local areas with similar structures, so that among a large number of local areas with similar structures, only a small number of representative areas need to be detected, which can represent the possibility of hotspots in these local areas, thereby greatly improving the detection compression ratio, shortening the detection time of the wafer, and more conducive to the popularization and application of high-resolution detection means on the production line.

[0078] The above embodiments use integrated circuit layout to select the detection area, which has the advantages of being more convenient and having shorter detection time compared to forming a macro image at a low resolution during the detection process and then dividing local areas based on the macro image.

Claims

1. A method of forming a detection area for determining a detection area when performing surface inspection of a die after CMP processing, characterized by, Selecting a representation region on an integrated circuit layout and forming a detection region from the representation region; the representation region is used to represent a local region with similar structure, the size of the local region in X-axis and Y-axis is in the range of 0.5-50 microns.

2. The method for forming a detection area according to claim 1, wherein The size of the local region in X-axis and Y-axis is in the range of 10-30 microns.

3. The method for forming a detection area according to claim 2, characterized by, The size of the local region in X-axis and Y-axis is 20 microns.

4. The method for forming a detection area according to claim 1, characterized by, It comprises the following steps: Dividing a measurement part on an integrated circuit layout into local regions, the measurement part covers all processing region groups, each processing region group comprises processing regions with similar structure, and all local regions cover the measurement part; Classifying all local regions based on structural similarity to obtain local region groups; Selecting at least one local region from each local region group as a representation region, and the collection of all representation regions forms a detection region.

5. The method for forming a detection area according to claim 4, wherein Adjacent local regions overlap with each other.

6. The method for forming a detection area according to claim 5, wherein The local regions are obtained by sliding a virtual window on the measurement part.

7. The method for forming a detection area according to claim 4, wherein The measurement part is determined by the following steps: Edge diffusion is performed on each processing region in the integrated circuit layout to form a corresponding diffusion region; Classifying all diffusion regions based on structural similarity to obtain diffusion region groups; Selecting at least one diffusion region from each diffusion region group as a measurement region, and the collection of all measurement regions forms a measurement part.

8. The method for forming a detection area according to claim 7, characterized by, The structural similarity is rotational symmetry and / or mirror symmetry.

9. The method for forming a detection area according to claim 7, wherein The measurement part has the shortest first path among various alternative measurement parts; the alternative measurement part is a collection of alternative measurement regions, the alternative measurement part covers all diffusion region groups and each diffusion region group has only a predetermined number of alternative measurement regions; the first path is the shortest one-way path in the alternative measurement part that passes through the center points of all alternative measurement regions.

10. The detection region forming method according to any one of claims 4 to 9, characterized by, The detection region has the shortest second path among various alternative detection regions; the alternative detection region is a collection of alternative representation regions, the alternative detection region covers all local region groups and each local region group has only a predetermined number of alternative representation regions; the second path is the shortest one-way path in the alternative detection region that passes through the center points of all alternative representation regions.

11. A method for detecting a surface of a crystal grain after CMP processing, characterized by, Based on the detection region formed by the detection region formation method of any one of claims 1-10, the surface of the die is topographically detected.

12. The crystal grain surface inspection method according to Claim 11, wherein The surface of the die is topographically detected by AFM.

13. The crystal grain surface inspection method according to Claim 12, characterized by, The size of the local region in X-axis and Y-axis is less than or equal to the size of the maximum imaging range of AFM in X-axis and Y-axis.

14. A method for judging the presence of a hot spot on the surface of a crystal grain after CMP processing, characterized by, The surface of the die is topographically detected by the die surface detection method of any one of claims 10-13, and if there is erosion of the functional part and / or depression of the representation region in the detection result, it is determined that there is a hot spot on the surface of the die.

Images