A clustering-based triple patterning lithography layout decomposition method

Through the triple pattern lithography layout decomposition method based on clustering, the problem of difficult to eliminate conflicting edge relationships between patterns in triple pattern lithography technology is solved, and faster and more efficient layout decomposition is achieved, reducing the decomposition overhead and solution time. It is applicable In the manufacturing of ultra-large-scale integrated circuits.

CN114937048BActive Publication Date: 2024-11-05SOUTHEAST UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210610203.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-05-31
Publication Date
2024-11-05
Estimated Expiration
2042-05-31

AI Technical Summary

Technical Problem

When the existing triple pattern lithography technology is made of layouts below 10 nm, the conflicting edge relationship between patterns is difficult to completely eliminate, resulting in lithography distortion and circuit performance problems. The solution time of existing algorithms when decomposing layouts in ultra-large scale integrated circuits Too long, the decomposition is more expensive.

Method used

Using the triple pattern lithography layout decomposition method based on clustering, the combination of pattern segmentation, conflict map construction, stitching edge insertion, conflict map simplification and SDP algorithm, the faster layout decomposition decomposition is achieved, and the inter-cluster color adjustment is used to reduce the decomposition overhead . Specific steps include: pattern segmentation, conflict graph construction, stitch edge insertion, simplification into clustering, vertex color matching in the SDP algorithm, and optimization of decomposition quality through inter-cluster color matching.

Benefits of technology

It significantly accelerates the solution speed of layout decomposition, reduces the decomposition overhead, improves the decomposition quality, can effectively handle layout decomposition tasks in ultra-large-scale integrated circuits, and reduces the adjustment workload of engineers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114937048B_ABST
    Figure CN114937048B_ABST
Patent Text Reader

Abstract

The present invention discloses a clustering-based triple pattern lithography layout decomposition method, including: polygons are divided into rectangular storage, and a conflict graph is constructed according to the distance between polygons; a stitching edge is added to the conflict graph by calculating the candidate stitching edge position and selecting a suitable stitching edge; a plurality of conflict graph simplification algorithms such as independent component calculation, hiding vertices with vertex degrees less than 3, bridge edge disassembly, clustering, etc. are used to reduce the scale of the layout decomposition problem; each sub-conflict graph is solved by a semidefinite programming algorithm relaxed from integer linear programming; the vertices between clusters are color-adjusted to reduce the use of stitching edges, and finally all sub-conflict graphs are merged to obtain the final layout decomposition result. The present invention makes the scale of the sub-conflict graph smaller by clustering, thereby accelerating the algorithm solution speed. At the same time, the present invention also achieves a higher algorithm solution quality through a new inter-cluster color adjustment algorithm suitable for triple pattern lithography.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the field of semiconductor lithography manufacturing, and in particular to a clustering-based triple pattern lithography layout decomposition method. Background Art

[0002] With the continuous development of integrated circuits, the number of transistors that can be accommodated on each chip has been growing rapidly at the speed of Moore's Law. Today, it has rapidly grown from the initial few transistors to more than 57 billion transistors. While the chip area remains basically unchanged, the feature size of the standard unit on each chip is also shrinking. Today, the feature size of the most advanced technology node has reached less than 10nm. However, the 193nm light source used for layout lithography manufacturing has insufficient lithography resolution, and direct use in the manufacture of advanced process standard units will cause serious lithography distortion. We call this close relationship between patterns that causes lithography distortion a conflict edge relationship. Figure 1 The lithography distortion under different processes has an impact. In order to continue Moore's Law, triple patterning lithography has become a practical means for the industry to manufacture layouts below 10nm.

[0003] Triple pattern lithography technology refers to the technology of decomposing the pattern in the original layer of layout with higher pattern density into three different mask plates, and then realizing the manufacturing of denser layout through three exposures and etchings. Figure 2 The decomposition results of double patterning lithography (DPL) and triple patterning lithography (TPL) on the same layout. It can be found that for Figure 2 In the layout like the one on the left, patterns b and c still conflict after the double patterning lithography layout is decomposed. In triple patterning lithography, the three patterns are assigned to three masks, and the conflicts are completely eliminated. Therefore, triple patterning lithography is easier to eliminate conflicting edges between patterns in layout manufacturing of advanced processes below 10nm. At the same time, in order to reduce the number of conflicting edges, stitching edges are introduced to divide the original pattern into multiple patterns. Each separated pattern can be assigned a different color, but when the colors are different, there will be a stitching edge, and when the colors are the same, it means that the stitching edge is not used. If two patterns with conflicting edge relationships are in the same mask layer, the circuit structure represented by the two patterns in the layout will have circuit function problems such as short circuit and open circuit, which will have a greater impact on the circuit; if two patterns with stitching edge relationships are assigned to different masks, the circuit structure represented by the two patterns will have circuit performance problems such as resistance and capacitance fluctuations, which will have a smaller impact on the circuit.

[0004] In triple patterning lithography, the most important thing is how to quickly decompose the layout so that the pattern in the original layout can achieve the minimum decomposition cost after being divided into three layers of mask plates. The decomposition cost is represented by the weighted number of unsolvable conflicting edges and used stitching edges in the layout. The smaller the decomposition cost, the higher the decomposition quality, and the less work the layout engineer has to do to adjust the layout in the later stage. As another important criterion for judging the layout decomposition algorithm, the faster the decomposition speed, the better the decomposition algorithm.

[0005] The decomposition problem of triple patterning lithography is usually converted into a three-coloring problem in graph theory for solution. Different coloring of the vertices in the graph means that the corresponding patterns are assigned to different masks. In 2008, ABKahng et al. abstracted the conflicts between the layout polygons into a conflict graph, and then completed the layout decomposition by processing the conflict graph. This abstract method has become the most commonly used representation method for most people to deal with the decomposition of multi-patterning lithography layouts. For the color matching problem of triple patterning lithography layout decomposition, some existing algorithms such as integer linear programming (ILP) and its relaxed semidefinite programming (SDP), graph theory algorithms and heuristic algorithms are relatively mature. However, if these algorithms are directly used for the layout decomposition of today's VLSI layout, the algorithm solution time will be very long. In order to speed up the solution of the algorithm, there are now conflict graph simplification methods for TPL layout decomposition, such as independent component calculation, hidden vertices with a degree less than 3, and bridge edge disassembly. However, these methods are not enough to completely simplify the conflict graph. After being simplified by the above three methods, the vertex scale of some complex graphs is still large. Therefore, the present invention proposes a more thorough conflict graph simplification method to further accelerate the decomposition of the map. While accelerating the speed of the algorithm solution, the map decomposition algorithm proposed by the present invention also has a method for adjusting the vertex color matching in each simplified sub-conflict graph, so as to achieve acceleration while maintaining or even reducing the overhead of map decomposition. Summary of the invention

[0006] In view of this, the purpose of the present invention is to provide a clustering-based triple-pattern lithography layout decomposition method. Compared with some existing layout decomposition methods, this method can achieve a faster solution speed while reducing the decomposition overhead of layout decomposition, which is an improvement on the existing layout decomposition methods.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] A clustering-based triple pattern lithography layout decomposition method, the method comprising:

[0009] Step S1, performing pattern segmentation on a layout, segmenting the polygons in the layout into multiple rectangles, and then obtaining the distance between the polygons based on the rectangles, and then constructing a conflict map of the layout based on the distance and the minimum rule spacing value of triple pattern lithography;

[0010] Step S2: inserting stitching edges for the conflict graph obtained in step S1, including: first calculating a candidate stitching edge insertion position by projection, and then selecting a stitching edge in the conflict graph according to the candidate stitching edge insertion position;

[0011] Step S3, performing a layout decomposition operation on the conflict graph into which the stitching edge is inserted in step S2, which includes: simplifying the conflict graph by a conflict graph simplification algorithm, simplifying it into multiple clusters to reduce the scale of the problem, wherein the conflict graph simplification algorithm includes: an algorithm for independent component calculation, an algorithm for hiding vertices with a vertex degree less than 3, an algorithm for bridge edge disassembly, and an algorithm for clustering;

[0012] Step S4, executing the SDP algorithm for color matching on the conflict graph simplified in step S3, and, when matching colors, adjusting the colors of two vertices with stitching edge relationships between clusters to improve the decomposition quality; wherein, when matching colors, firstly executing the SDP algorithm for the first cluster to obtain a corresponding result, and when matching colors for the second cluster and subsequent clusters, performing inter-cluster color matching adjustment, the inter-cluster color matching adjustment includes: for the second cluster, judging whether there is a stitching edge relationship between each vertex of the cluster and the vertex of the first cluster, if yes, giving an initial suggested color matching to the vertex in the cluster, the initial suggested color matching is the same as the vertex color matching in the first cluster, if no, executing the SDP algorithm for the two clusters to obtain a corresponding result, and the same method is used for the third cluster and the i-th cluster;

[0013] Step S5: for the sub-clusters after the color matching operation in step S4, the internal optimal solution of each sub-conflict graph simplified by the three conflict graph simplification algorithms except clustering is obtained; these sub-conflict graphs are merged with the optimal solutions to obtain a complete layout decomposition result.

[0014] Furthermore, in step S1, the conflict graph is constructed according to the following method, including:

[0015] Step S101, dividing the layout pattern into rectangles along the horizontal or vertical direction;

[0016] Step S102: For the distance between two rectangles, if there is a partial overlap between the two rectangles in the X or Y axis direction, the distance is the parallel distance or the perpendicular distance between the two rectangles; if there is no overlap between the two rectangles in the X or Y axis direction, the distance is the distance between the two closest vertices.

[0017] Step S103, comparing the polygon distances obtained in step S101 and step S102 with the minimum regular spacing value of triple pattern lithography to determine whether conflicting edges are formed between the polygons;

[0018] Step S104, using vertices to represent patterns in the layout and solid lines to represent conflict edge relationships between patterns, thereby abstractly publishing a conflict graph corresponding to the layout.

[0019] Furthermore, in the step S2, for the polygons in the layout, the rectangles therein are projected to obtain a projection sequence on each rectangle, and then the candidate stitching edge insertion position is obtained according to the projection sequence, which specifically includes:

[0020] Step S2011, select a polygon, determine the polygons around it that are in conflict with the polygon; then divide the rectangles of these polygons into regions according to the minimum spacing rule value of triple pattern lithography, and find the region on each rectangle of the selected polygon that is in conflict with the rectangles of the surrounding polygons;

[0021] Step S2012: At this time, the currently selected rectangle is divided into multiple segments by these projections; each segment is marked with the number of projections of other adjacent rectangles on the currently selected rectangle, and the sequence composed of the number of projections on these segments is the projection sequence; there is also a rule that the endpoints must be zero, that is, the projection sequence must start and end with 0;

[0022] Step S2013: select all possible candidate stitching edge insertion positions according to the obtained projection sequence, wherein the selection is performed according to the following rules, which include:

[0023] Rule 1: All regions with projection sequence 0 are candidate stitching edge locations;

[0024] Rule 2: Among the candidate seam edge positions in Rule 1, if the projection sequence starts with "01010", these candidate seam edge positions are redundant;

[0025] Rule 3: If the projection sequence of the currently selected rectangle contains a subsequence "xyz", where x, y, z>0, and x>y, z>y, then there is a missing candidate seam edge in the segment with y projections, where a subsequence refers to a projection sequence containing at least 3 non-zero projections; missing refers to the seam edge position that cannot be found by the algorithm that only inserts seam edges in areas with zero projections.

[0026] Further, for the candidate stitching edges obtained in step S2013, the candidate stitching edges that will cause other design rule check violations are removed, and the remaining candidate stitching edges are used as stitching edges in the conflict graph, wherein the candidate stitching edge positions that meet the following conditions need to be removed, including:

[0027] Condition 1: After inserting the stitching edge at this position, a candidate stitching edge position of a rectangle smaller than the minimum lithography pattern size requirement fs_min will be generated;

[0028] Condition 2: After the stitching edge is inserted at this position, when the segment rectangle is divided into two rectangles, the cross-coverage length of each rectangle does not meet the candidate stitching edge position of the minimum cross-coverage length m_0 required by the process;

[0029] Condition 3: The candidate stitching edge position is at the corner of the rectangle;

[0030] Condition 4: After inserting the seam edge at this position, a candidate seam edge position of a small rectangle that has no conflicting edge relationship with other polygons will be generated.

[0031] Furthermore, in step S3, the independent component calculation algorithm includes: clustering vertices with edge relationships through depth-first traversal of the vertices in the conflict graph to divide the vertices in the original conflict graph into a number of independent components.

[0032] Furthermore, in step S3, the algorithm for hiding vertices with vertex degrees less than 3 includes:

[0033] First, all vertices are set to visible, and a flag is set to indicate whether any vertices are hidden in this iteration;

[0034] Next, perform iteration. In the iteration, after selecting a vertex v, first check its vertex status. If the vertex is hidden, skip it and check the next vertex; otherwise, set the degree of the vertex to 0.

[0035] Then, for the selected vertex v, traverse all its adjacent vertices u. If the vertex u has been hidden, skip the adjacent vertex and check the next adjacent vertex. Otherwise, get the type of the edge between vertex u and vertex v. If it is a stitching edge, add 1 to the stitching edge degree and continue the loop; if it is a conflicting edge, add 1 to the conflicting edge degree of vertex v.

[0036] After traversing all adjacent vertices of vertex v, determine whether the conflict edge degree of vertex v is less than 3 and the stitching edge degree is 0. If the condition is met, it means that the current vertex v can be hidden, and its state is set to hidden, and the flag hide_flag is set to true, and continue to the next iteration;

[0037] Finally, if after traversing all the vertices in the graph, no new vertices with conflicting edge degrees less than 3 and vertices with stitching edge degrees of 0 are found, or all vertices have been hidden, the flag hide_flag will remain false, thus exiting the while loop and simplification is completed.

[0038] Furthermore, in step S3, the bridge edge disassembly algorithm includes:

[0039] First, after selecting an initial vertex, the access order of the vertex and the minimum serial number value of the reachable vertex are both its access order;

[0040] Next, push the vertex into the stack and traverse each adjacent vertex v of vertex u. If v has not been visited, continue depth-first exploration. When this vertex and its deeper vertices have been explored, start from the deepest vertex and trace back upwards in sequence to update the minimum reachable vertex number of each shallow vertex. If the minimum reachable vertex of the adjacent vertex v is larger than the access sequence number of the current vertex u, the edge (u, v) forms a bridge edge. If v has been visited and v is still in the stack, it means that the access sequence of vertex v is smaller than that of vertex u, and vertex u has found a vertex with a smaller access sequence that it can reach. At this time, it is necessary to update the vertex with the minimum access sequence that vertices v and u can reach.

[0041] Then, after searching all the edges of the current vertex u, if it is found that the access order of node u is the same as the minimum sequence value of the vertex that can be reached by the vertex, it means that the vertex is the root vertex of a biconnected subgraph; at this time, the elements on the top of the stack are continuously popped until the popped element is the same as the root vertex, and all the vertices of a sub-conflict graph are obtained;

[0042] Finally, the remaining vertices in the original conflict graph form another doubly connected subgraph. Using the bridge edge as the boundary, a large conflict graph can be divided into two small sub-conflict graphs to reduce the number of vertices in the sub-conflict graphs.

[0043] Furthermore, in step S3, the cluster algorithm specifically includes:

[0044] By performing depth-first exploration only on conflicting edges, the vertices in an independent component are subdivided into multiple small clusters, and there are only stitching edge relationships but no conflicting edge relationships between clusters.

[0045] Furthermore, the step S5 comprises:

[0046] For each cluster formed by clustering, the merging process is carried out together with the color matching process of each cluster; the merging process and the color matching process form such a cycle: for each cluster, first determine whether each vertex in the cluster has an inter-cluster stitching edge relationship with the vertices in other clusters; if so, and the vertices in the other cluster have obtained the optimal color matching through the SDP color matching algorithm, then an initial recommended color matching is given to the vertices in the cluster connected by this inter-cluster stitching edge; then the SDP color matching algorithm is solved in the cluster; if there is no inter-cluster stitching edge, or although there is an inter-cluster stitching edge, but the color matching has not been performed in the other cluster, the SDP algorithm is directly solved for the cluster; this process is repeated until all clusters have obtained color matching and the algorithm ends; the merging process of the optimal color matching results of each cluster is realized in this cycle;

[0047] For the sub-conflict graphs formed by bridge edge decomposition, after obtaining the solution with the best decomposition quality in each sub-conflict graph, since the two sub-conflict graphs are solved relatively independently and connected by a bridge edge, it is necessary to adjust the color scheme in one of the sub-conflict graphs according to the colors of the two vertices connected by the bridge edge; if the colors of the two vertices connected by the bridge edge are the same, then the colors of all vertices in the second sub-conflict graph connected by the bridge edge are offset by 1 to merge the sub-conflict graphs obtained by bridge edge decomposition.

[0048] Furthermore, the step S5 further includes:

[0049] For those hidden vertices, when matching their colors, we only need to use the colors of the adjacent vertices of the hidden vertices to get the color matching;

[0050] Since there is no edge relationship between the independent components, they are relatively independent when solved. Therefore, it is only necessary to directly merge the color matching results in each independent component to obtain the layout decomposition result of the entire layout.

[0051] The beneficial effects of the present invention are:

[0052] 1. The present invention simplifies the conflict graph more thoroughly, so that the scale of each sub-conflict graph is minimized, and the scale of the SDP algorithm to solve each sub-conflict graph becomes smaller, so the solving speed is faster than that of the existing algorithm;

[0053] 2. The present invention adopts a new inter-cluster color matching adjustment algorithm suitable for TPL after clustering, so that the quality of layout decomposition is not lost due to clustering, and appropriate inter-cluster color matching adjustment can also make the quality of layout decomposition higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A schematic diagram of the effects of lithography distortion under different processes described in the background technology;

[0055] Figure 2 It is a schematic diagram of the decomposition of the DPL and TPL layout described in the background technology;

[0056] Figure 3 A schematic flow chart of a clustering-based triple pattern lithography layout decomposition method provided by the present invention;

[0057] Figure 4 is a schematic diagram of the layout segmentation process of an example layout; wherein the Figure 4 (a) is the polygon before segmentation. Figure 4 (b) is a schematic diagram after polygon segmentation;

[0058] Figure 5 Based on Figure 4 (a) Constructed conflict graph;

[0059] Figure 6 Schematic diagram of the calculation of the candidate stitch edge insertion position;

[0060] Figure 7 A schematic diagram of a stitching edge position selected as a stitching edge of a conflict graph;

[0061] Figure 8 is an example of a candidate seam edge position where Figure 8 (a) indicates a violation of the fs_min constraint, Figure 8 (b) indicates a violation of the m_0 constraint, Figure 8 (c) indicates a violation of the non-rectangular corner constraint. Figure 8 (d) represents the generation of a small rectangle without other conflicting edges;

[0062] Fig. 9 Schematic diagram calculated for individual components;

[0063] Fig.10 It is a schematic diagram of hidden vertices with degree less than 3;

[0064] Fig.11 This is a simple schematic diagram of the disassembly of the bridge side, where: Fig.11 (a) is a schematic diagram before the bridge side dismantling. Fig.11 (b) is a schematic diagram after disassembly;

[0065] Fig.12 This is a detailed diagram of the dismantling process of the bridge side, where: Fig.12 (a) indicates a depth-first traversal to vertex 2, and labels the Order and Low values ​​of vertices 0, 1, 3, and 2; Fig.12 (b) indicates updating the Low values ​​of vertices 2 and 3 during the backtracking process; Fig.12(c) indicates a depth-first traversal of the other branch of vertex 3, and labels the Order and Low of vertices 5, 4, and 6; Fig.12 (d) indicates that after the depth-first exploration is completed, the Low values ​​of vertices 5, 4, and 6 are updated; Fig.12 (e) indicates finding the bridge edge in the graph based on the Low value of each vertex; Fig.12 (f) indicates updating the Low values ​​of vertices 1 and 0;

[0066] Fig.13 is a schematic diagram of clustering;

[0067] Fig.14 Schematic diagram of color matching for patterns within a cluster, where Fig.14 (a) is the schematic diagram before color matching. Fig.14 (b) is the schematic diagram after color matching;

[0068] Fig.15 This is a schematic diagram of inter-cluster color matching adjustment, where: Fig.15 (a) Schematic diagram before inter-cluster color matching adjustment. Fig.15 (b) is the schematic diagram after the inter-cluster color matching is adjusted;

[0069] Fig.16 This is a schematic diagram of the bridge edge disassembly sub-graph merging, where: Fig.16 (a) Two sub-conflict graphs that have not been merged. Fig.16 (b) The conflict graph after the direct merge method. Fig.16 (c) After the method of this embodiment is adopted, there is no conflict graph with conflicts;

[0070] Fig.17 is a schematic diagram of hidden vertex recovery, where Fig.17 (a) is a schematic diagram before hidden vertex recovery. Fig.17 (b) is a schematic diagram after hidden vertex recovery. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0072] Example 1

[0073] See also Figure 3-Figure 17 This embodiment provides a clustering-based triple pattern lithography layout decomposition method, the process of which is as follows: Figure 3 As shown, the method specifically includes:

[0074] Step S1, performing pattern segmentation on a GDS-II layout, segmenting the polygons in the layout into multiple rectangles, and then obtaining the distance between the polygons based on the rectangles, and then constructing a conflict map of the layout based on the distance and the minimum rule spacing value of triple pattern lithography.

[0075] Specifically, in this embodiment, the conflict graph is constructed by the following method, including:

[0076] Step S101, divide the layout pattern into small rectangles in the horizontal or vertical direction as required and store them. This is because the layout patterns in the input layout file have various shapes, which makes it inconvenient to calculate the distance between two patterns to determine whether the two patterns form a conflicting edge, and to perform projection operations on the layout patterns. After dividing into small rectangles, the distance between polygons can be determined by the minimum value of the distance between the rectangles, and the distance calculation between rectangles is much simpler than the distance calculation between polygons.

[0077] Step S102: For the distance between two rectangles, if there is a partial overlap between the two rectangles in the X or Y axis direction, the distance is the parallel distance or the perpendicular distance between the two rectangles; if there is no overlap between the two rectangles in the X or Y axis direction, the distance is the distance between the two closest vertices.

[0078] Step S103, comparing the polygon distances obtained in step S101 and step S102 with the minimum regular spacing value of triple pattern lithography to determine whether conflicting edges are formed between the polygons;

[0079] Step S104: Use vertices to represent patterns in the layout, and use solid lines to represent conflict edge relationships between patterns, so as to abstractly generate a conflict graph corresponding to the layout.

[0080] More specifically, in this embodiment, if Figure 4 and Figure 5 As shown, the method of step S101 is adopted to Figure 4 The polygon in (a) is cut into several rectangles along the horizontal direction, such as Figure 4(b) As shown. Next, according to step S102, the distance between the constituent rectangles of the two layout polygons is calculated, and the distance between polygon A and polygon B is obtained as the distance between the upper right vertex of the constituent rectangle A1 and the lower left vertex of the constituent rectangle B2, and the distance between polygon A and polygon C is the vertical distance between the constituent rectangle A2 and the constituent rectangle C1. The distances between other polygons are similar. Then, according to steps S103 and S104, the distance between the polygons is judged to be the size of the minimum regular spacing of the triple pattern lithography to determine whether there is a conflicting edge relationship between the polygons. Use vertices to represent polygons, straight lines to represent conflicting edges, and construct a conflict graph as shown in FIG. Figure 5 shown.

[0081] Step S2: inserting stitching edges for the conflict graph obtained in step S1, including: calculating candidate stitching edge insertion positions by projection, and selecting some suitable stitching edges as stitching edges in the conflict graph according to design rules;

[0082] Specifically, in this embodiment, the stitching edge of the conflict graph is obtained by the following method, including:

[0083] Step S201: for polygons in the layout, project rectangles therein to obtain a projection sequence on each rectangle, and then obtain a candidate stitching edge insertion position according to the projection sequence;

[0084] More specifically, in this embodiment, the projection sequence is obtained by the following method, including:

[0085] Step S2011, select a polygon, determine the polygons around it that are in conflict with the polygon; then divide the rectangles of these polygons into regions according to the minimum spacing rule value of triple pattern lithography, and find the region on each rectangle of the selected polygon that is in conflict with the rectangles of the surrounding polygons;

[0086] Step S2012: At this time, the currently selected rectangle is divided into multiple segments by these projections; each segment is marked with the number of projections of other adjacent rectangles on the currently selected rectangle, and the sequence composed of the number of projections on these segments is the projection sequence. For convenience, there is also a rule that the endpoints must be zero, that is, the projection sequence must start and end with 0;

[0087] Step S2013: According to the obtained projection sequence, all possible candidate stitching edge insertion positions are obtained according to the following rules:

[0088] Rule 1: All regions with projection sequence 0 are candidate stitching edge locations;

[0089] Rule 2: Among the candidate stitching edge positions in Rule 1, if the projection sequence starts with "01010", these candidate stitching edge positions are redundant. This is because, for the first 0, if a stitching edge is inserted at this position of 0, it will not have any impact on the conflict relationship of the rectangles and there will be one more stitching edge, so it is not inserted. As for the candidate stitching edge positions corresponding to the second and third 0s, the currently selected rectangle and its two adjacent rectangles have a total of only three patterns. In TPL, no matter how rectangles b and c are colored, rectangle a can always get a legal color scheme, and there is no need to introduce an additional stitching edge. Secondly, since the projection sequence on a selected rectangle may be symmetrical, if the tail end of the projection sequence on the selected rectangle is also "01010", these candidate stitching edge positions at the tail end are also redundant;

[0090] Rule 3: If the projection sequence of the currently selected rectangle contains a subsequence "xyz" where x, y, z>0 and x>y, z>y, then there is a missing candidate seam edge in the segment with y projections. A subsequence is a projection sequence with at least 3 non-zero projections. A missing seam edge is a position that cannot be found by the algorithm that inserts seams only in the region with zero projections.

[0091] Step S202: For the candidate stitching edges obtained in step S2013, remove the candidate stitching edges that will cause other design rule check violations, and the remaining candidate stitching edges are used as stitching edges in the conflict graph to participate in the calculation of subsequent algorithms.

[0092] More specifically, candidate seam edge positions that meet the following conditions need to be removed:

[0093] Condition 1: After inserting the stitching edge at this position, a candidate stitching edge position of a rectangle smaller than the minimum lithography pattern size requirement fs_min will be generated;

[0094] Condition 2: After the stitching edge is inserted at this position, when the segment rectangle is divided into two rectangles, the cross-coverage length of each rectangle does not meet the candidate stitching edge position of the minimum cross-coverage length m_0 required by the process;

[0095] Condition 3: The candidate stitching edge position is at the corner of the rectangle;

[0096] Condition 4: After inserting the stitching edge at this position, a candidate stitching edge position of a small rectangle with no conflicting edge relationship with other polygons will be generated;

[0097] More specifically, in this embodiment, according to step S201, the polygons with conflicting edge relationships are projected with the minimum regular spacing of triple pattern lithography as the radius to obtain a projection sequence on each component rectangle of each polygon. Figure 6In the projection sequence of polygons b, c, d, e, and f on the rectangle of polygon a is 01212101010, where the 0 at the end is the default 0 added according to the rule that endpoints must be zero. Figure 6 In , each polygon is exactly a rectangle. Then, according to the rule in step S2013, Figure 6 The stitching edge at the projection sequence 01010 in is marked as a redundant stitching edge, and a missing stitching edge is also found in sequence 212. The candidate stitching edge positions obtained after step S201 are as follows: Figure 6 Next, according to the rule in step S202, the generated candidate stitching edges are selected to obtain the final stitching edges added to the conflict graph as shown in Figure 7 As shown. Figure 6 and Figure 7 In addition to the rules shown in the figure, the remaining stitching edge selection rules are shown in the figure below. Figure 8 shown.

[0098] Step S3, performing a layout decomposition operation on the conflict graph into which the stitching edge is inserted in step S2, which includes: simplifying the conflict graph by a conflict graph simplification algorithm, simplifying it into multiple clusters to reduce the scale of the problem, wherein the conflict graph simplification algorithm includes: an algorithm for independent component calculation, an algorithm for hiding vertices with a vertex degree less than 3, an algorithm for bridge edge disassembly, and an algorithm for clustering;

[0099] Specifically, in this embodiment, in step S3, the conflict graph simplification algorithm specifically includes:

[0100] 1. An algorithm for calculating independent components, which includes: clustering vertices with edge relationships by depth-first traversal of the vertices in the conflict graph, thereby dividing the hundreds of millions of vertices in the original conflict graph into several independent components with a smaller number of vertices.

[0101] 2. An algorithm for hiding vertices with a degree less than 3. This simplification method is a process of iteratively hiding the original conflict graph, which includes: first, set all vertex states to visible (not hidden), and set a flag to indicate whether a vertex is hidden in this iteration. Then start the iteration. In the iteration, after selecting a vertex v, first check its vertex state. If the vertex has been hidden, skip it directly and check the next vertex. Otherwise, set the degree of the vertex to 0. Then, for the selected vertex v, traverse all its adjacent vertices u (u and v may be connected by conflicting edges or stitching edges). If the vertex u has been hidden, skip the adjacent vertex and check the next adjacent vertex. Otherwise, get the type of edge between vertex u and vertex v. If it is a stitching edge, add 1 to the stitching edge degree and continue the cycle. If it is a conflicting edge, add 1 to the conflicting edge degree of the vertex v. After traversing all adjacent vertices of vertex v, determine whether the conflict edge degree of vertex v is less than 3 and the stitching edge degree is 0. If the condition is met, it means that the current vertex v can be hidden, and its state is set to hidden, and the flag hide_flag is set to true, and continue to the next iteration. If all vertices in the graph are traversed and no new vertices with conflict edge degrees less than 3 and stitching edge degrees of 0 are found, or all vertices have been hidden, the flag hide_flag will remain false, thus exiting the while loop and simplification ends.

[0102] 3. Bridge edge disassembly algorithm. This simplification method marks the access order of each vertex and the minimum serial number value of the vertex that can be reached by the vertex by setting two arrays Order and Low. It includes: first, after selecting an initial vertex, mark the access order of the vertex and the minimum serial number value of the vertex that can be reached as its access order. Then push the vertex into the stack. Then traverse each adjacent vertex v of vertex u. If v has not been visited, continue depth-first exploration. When this vertex and its deeper vertices have been explored, start from the deepest vertex and trace back upward in sequence to update the minimum serial number value of the reachable vertex of each shallow vertex. If the minimum reachable vertex of the adjacent vertex v is larger than the access serial number of the current vertex u, the edge (u, v) constitutes a bridge edge. If v has been visited and v is still in the stack, it means that the access order of vertex v is smaller than that of vertex u, and vertex u has found a vertex with a smaller access order that it can reach. At this time, it is necessary to update the vertex with the minimum access order that vertex v and u can reach. After searching all the edges of the current vertex u, if it is found that the access order of node u is the same as the minimum serial number value of the vertex that can be reached by the vertex, it means that the vertex is the root vertex of a doubly connected subgraph. At this time, the elements on the top of the stack are continuously popped until the popped element is the same as the root vertex, and all the vertices of a sub-conflict graph are obtained. Similarly, the remaining vertices in the original conflict graph form another doubly connected subgraph. Using the bridge edge as the boundary, a larger conflict graph can be divided into two smaller sub-conflict graphs, reducing the number of vertices in the sub-conflict graphs.

[0103] 4. Clustering algorithm, including: by depth-first exploration of conflicting edges only, and temporarily ignoring stitching edges, the vertices in an independent component are subdivided into multiple small clusters. There are only stitching edge relationships but no conflicting edge relationships between clusters. There are a large number of conflicting edges and a few stitching edge relationships within a cluster. This is because although the vertices within the cluster are obtained by conflicting edge exploration, it cannot be ruled out that some vertices are directly connected by stitching edges and indirectly connected by conflicting edges.

[0104] More specifically, in this embodiment, according to the independent component calculation algorithm, the original conflict consisting of 24 vertices Fig. 9 , after independent component calculation, we get two independent components 1 and 2 consisting of 10 and 14 vertices respectively.

[0105] According to the algorithm for hiding vertices with degree less than 3, Fig.10 In the first iteration, vertices 4 and 5 are hidden, and in the second iteration, the remaining vertices 3, 2, and 1 are hidden. Fig.10 For example, after simplification of vertices with hidden vertex degrees less than 3, the status of all vertices becomes hidden.

[0106] According to the algorithm of bridge side disassembly, similar Fig.11 The conflict graph in (a) is transformed into two smaller sub-conflict graphs after being decomposed by the bridge edge, such as Fig.11 (b) is shown. The detailed process of bridge side dismantling is shown in Fig.12 shown. Fig.12 In (a), we first explore vertices 0, 1, 3, and 2 in depth-first order. After exploring vertex 2, we find that the vertex with the minimum access sequence that can be reached is vertex 0, so we update the minimum access sequence value that can be reached to vertex 0, which is access sequence 0. Similarly, we update the Low value of vertex 3. Fig.12 In (c), a depth-first exploration is performed on another branch toward vertex 3, and the access order and initial Low value are set for each vertex. After the exploration of vertex 6 is completed, it is found that the vertex with the minimum access order that it can reach is vertex 5, so the minimum access sequence value that it can reach is updated to the access order 4 of vertex 5. Similarly, the Low value of vertex 4 is updated. After the update is completed, Fig.12 In (e), the bridge edge can be found according to the Low value of each vertex. Finally, the Low value of the remaining vertex 1 that has not been updated is also updated. The bridge edge search is completed, and the two sub-conflict graphs obtained by disassembly are self-evident.

[0107] According to the clustering algorithm, the original independent components consisting of 14 points, such as Fig.13 As shown, it is divided into 4 clusters with fewer vertices. The number of vertices in each cluster is less than 6, and the number of vertices is reduced by more than half.

[0108] Step S4, executing the SDP algorithm for color matching on the conflict graph after simplification in step S3, and, when matching colors, adjusting the colors of two vertices with stitching edge relationships between clusters to improve the decomposition quality; wherein, when matching colors, firstly executing the SDP algorithm for the first cluster to obtain a corresponding result, and when matching colors for the second cluster and subsequent clusters, performing inter-cluster color matching adjustment, the inter-cluster color matching adjustment includes: for the second cluster, determining whether there is a stitching edge relationship between each vertex of the cluster and the vertex of the first cluster, if yes, giving an initial suggested color matching to the vertices in the cluster, the initial suggested color matching is the same as the vertex color matching in the first cluster, if no, executing the SDP algorithm for the two clusters to obtain a corresponding result, and the same method is used for the third cluster and the i-th cluster.

[0109] Specifically, in this embodiment, in step S4, the SDP algorithm is solved for a cluster by the following method, including:

[0110] Three unit vectors (1, 0), (-1 / 2, √3 / 2) and (-1 / 2, -√3 / 2) are used to represent the three colors; the inner product between two unit vectors is expressed as:

[0111] Then, for two vertices i and j, the following properties apply:

[0112]

[0113] In formula (1) and formula (2), v ik , v jk represents the color of the i-th or j-th vertex, that is, one of the three vectors. i1 Represents the value of the first component of the color vector of vertex i, v i2 Represents the value of the second component of the color vector of vertex i.

[0114] Therefore, the decomposition cost expression of the layout decomposition can be written as follows (3):

[0115]

[0116]

[0117] In formula (3), e ij represents an edge between vertex i and vertex j (which may be a conflict edge or a stitching edge). α represents the ratio of the cost brought by a stitching edge to the cost brought by a conflict edge. In the present invention, α is taken as 0.1.

[0118] By relaxing the requirement for the discrete value of vector vi in ​​equation 3, we can obtain the formula shown in equation 4:

[0119]

[0120]

[0121]

[0122] Any feasible solution in formula (3) can be By Set to Construct a feasible solution of formula (4), that is, each It can be understood as an approximate solution to the solution of (3) after the vector dimension is expanded.

[0123] Thus, for a solution Z of formula (4) R , and a solution OPT corresponding to formula (3) has Z R<=OPT. That is, the solution of formula (4) is an approximate solution of the solution of formula (3). Because the constant will not affect the solution of the minimum value of formula (4), the constant term in formula (4) is removed to obtain the following formula (5):

[0124]

[0125]

[0126]

[0127] Finally, convert formula (5) into the standard semidefinite programming form, as shown in formula (6):

[0128] min A·X (6)

[0129]

[0130]

[0131] X≥0 (6d)

[0132] In formula (6), A represents the cost coefficient between each vertex, and the cost coefficient of each position in the matrix is ​​shown in formula (7); α is 0.1, which represents the ratio of the influence of the stitching edge to the influence of the conflicting edge; X represents whether the colors of the vertices obtained at the end are the same or not. If they are the same, the inner product of the colors corresponding to the two vertices is 1, and X ij =1. If different, the minimum value of the inner product is -1 / 2, X ij >=-1 / 2. A·X represents the inner product between matrices A and X, that is, ∑ i ∑ j A ij X ij Constraint (10d) indicates that X is positive semidefinite, because X ij and X ji Both represent the same edge in the conflict graph.

[0133]

[0134] The obtained X matrix is ​​rounded to make the elements of X become 1 and -0.5. 1 means that the two vertices at the corresponding positions in the X matrix have the same color, while -0.5 means that the two vertices at the corresponding positions in the X matrix have different colors. At this point, the color similarity and difference relationship between the vertices is obtained, and then the color of the other vertices is obtained by assuming the color of a vertex and based on the color similarity and difference relationship between the vertices.

[0135] Specifically, in this embodiment, the inter-cluster color matching adjustment is performed by the following method:

[0136] First, perform the SDP color matching solution in a sub-cluster, and then record the solution results. Before solving the next cluster, it will first determine whether each vertex in the second cluster has a stitching edge relationship with the vertices in other clusters. If so, it will further determine whether the vertices in the external cluster have been colored by the SDP algorithm. If the vertices in the external cluster already have a color, the color of the corresponding vertex in the current second cluster will give an initial suggested color, which has the same color as the vertex in the external cluster. The reason for the suggested color is that in the current second cluster, the edge relationships between vertices are mainly conflicting edges, and the cost of conflicting edges is much greater than the cost of stitching edges between clusters. Therefore, if the best color matching result of the intra-cluster color matching algorithm needs to adjust the color of the vertices that have a stitching edge relationship with the vertices in the external cluster, the color will still be adjusted.

[0137] More specifically, in this embodiment, Fig.14 The cluster shown in (a) is color matched, and the color matching result is as follows Fig.14 (b) is shown. Fig.14 In the conflict graph shown in (a), there are 4 stitching edges generated by step S2. After the intra-cluster SDP algorithm is used to solve and color each vertex, the optimal coloring solution makes the colors of the two vertices connected by some stitching edges consistent, and only one stitching edge is actually used, as shown in Fig.14 As shown in (b).

[0138] More specifically, in this embodiment, Fig.15 In the algorithm, after performing intra-cluster pattern color matching in cluster 1, the algorithm will perform inter-cluster color matching before performing intra-cluster pattern color matching in cluster 2. For the vertices in cluster 2, the inter-cluster color matching adjustment algorithm scans each vertex in the cluster to check whether it has a stitching edge relationship with the vertex in cluster 1 that has been colored. If so, the color of the vertex is set to the color of the corresponding vertex in cluster 1, such as Fig.15 (b) shows this. This color matching is called a suggested color matching. The reason why it is a suggested color matching is that the edge relationships between vertices in the current second cluster are mainly conflicting edges, and the cost of conflicting edges is much greater than the cost of inter-cluster stitching edges. Therefore, if the best color matching result of the intra-cluster color matching algorithm needs to adjust the color of the vertices that have stitching edge relationships with vertices in the external cluster, the color matching will still be adjusted. This inter-cluster color matching algorithm can reduce the use of inter-cluster stitching edges as much as possible, thereby reducing the total decomposition cost of the algorithm.

[0139] Step S5: for the sub-clusters after the color matching operation in step S4, the internal optimal solution of each sub-conflict graph simplified by the three conflict graph simplification algorithms except clustering is obtained. The optimal solutions of these sub-conflict graphs are merged to obtain a complete layout decomposition result.

[0140] Specifically, in this embodiment, the merging operation is performed by the following method, including:

[0141] 1. For each cluster formed by clustering, the merging process is actually carried out together with the color matching process of each cluster. The merging process and the color matching process form such a cycle: for each cluster, first determine whether each vertex in the cluster has an inter-cluster stitching edge relationship with the vertices in other clusters. If so, and the vertices in the other cluster have obtained the optimal color matching through the SDP color matching algorithm, then an initial recommended color matching is given to the vertices in the cluster connected by this inter-cluster stitching edge. Then start to solve the SDP color matching algorithm in the cluster. If there is no inter-cluster stitching edge, or although there is an inter-cluster stitching edge, but the color matching has not been performed in the other cluster, the SDP algorithm is directly solved for the cluster. This process continues until all clusters have obtained color matching and the algorithm ends. The merging process of the optimal color matching results of each cluster is realized in this cycle.

[0142] 2. For the sub-conflict graphs formed by bridge edge decomposition, after obtaining the best decomposition solution in each sub-conflict graph, since the two sub-conflict graphs are solved relatively independently and connected by a bridge edge (conflict edge), it is necessary to adjust the color scheme in one of the sub-conflict graphs according to the colors of the two vertices connected by the bridge edge. If the colors of the two vertices connected by the bridge edge are the same, the colors of all vertices in the second sub-conflict graph connected by the bridge edge will be offset by 1 to merge the sub-conflict graphs obtained by bridge edge decomposition;

[0143] 3. For those hidden vertices, when matching their colors, you only need to use the colors of the adjacent vertices of the hidden vertices to get the appropriate color matching;

[0144] 4. Since there is no edge relationship between the independent components, they are relatively independent when solved. Therefore, you only need to directly merge the color matching results in each independent component to get the layout decomposition result of the entire layout.

[0145] More specifically, in this embodiment, for the sub-conflict graphs formed by bridge edge disassembly, the sub-conflict graphs obtained by bridge edge disassembly are first merged. Fig.16 In (a), each sub-conflict graph obtains its optimal color scheme, such as Fig.16 As shown in (a), if the two sub-conflict graphs are directly merged, the following may occur Fig.16 To solve this problem, after obtaining the color schemes of two vertices connected by a bridge edge, if their color schemes are the same, all vertex color values ​​in the second sub-conflict graph are offset by 1 to resolve the conflict between the sub-conflict graphs. Fig.16As shown in (c), the vertex colors in the sub-conflict graph below are offset by 1, so that there is no additional overhead when the bridge edge disassembly sub-graph is merged.

[0146] For the hidden vertices, Fig.17 Vertex 1 in (a) is hidden in the conflict graph simplification step of hiding vertices with degrees less than 3 because it has only two conflicting edges connecting it to other vertices, and the conflicting edge degrees are less than 3. After the other vertices with more complex conflicting edge relationships are colored, the hidden vertex pops up the hidden list again and obtains the correct color of the vertex based on the colors of its two adjacent vertices, as shown in Fig.17 (b) as shown.

[0147] The method of this embodiment is written in C++ language and runs on an Intel(R) Core(TM) i54210M quad-core eight-thread processor, using the Ubuntu 14.04 operating system. The experimental results of the algorithm on the metal layer 1 of the ISCAS-85&89 benchmark test circuit are compared with the experimental results in the literature [1], as shown in Tables 1 and 2. Among them, #pattern represents the number of layout patterns in the layout file, #cn is the number of conflicting edges (conflict edge number) that still exist after the layout decomposition is completed, and #sn is the number of stitching edges (stitch edge number) used after the layout decomposition is completed. Cost represents the final layout decomposition cost of each layout, and its calculation formula is similar to formula (3a). Time represents the time used for layout decomposition of each layout under the corresponding algorithm, in seconds.

[0148] Among them, the above-mentioned literature [1] is: [1] B. Yu, K. Yuan, B. Zhang, D. Ding, and D. Z. Pan, “Layout decomposition for triple patterning lithography,” Proc. ICCAD, 2011.

[0149] Table 4-1 Decomposition results of each map using the algorithm in reference [1] and this paper

[0150]

[0151] Table 4-2 Decomposition results of each map using the algorithm in reference [1] and this paper

[0152]

[0153] The matters not described in detail in the present invention are all known technologies to those skilled in the art.

[0154] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A clustering-based triple pattern lithography layout decomposition method, characterized in that: The method comprises: Step S1, performing pattern segmentation on a layout, segmenting the polygons in the layout into multiple rectangles, and then obtaining the distance between the polygons based on the rectangles, and then constructing a conflict map of the layout based on the distance and the minimum rule spacing value of triple pattern lithography; Step S2: inserting stitching edges for the conflict graph obtained in step S1, including: first calculating a candidate stitching edge insertion position by projection, and then selecting a stitching edge in the conflict graph according to the candidate stitching edge insertion position; Step S3, performing a layout decomposition operation on the conflict graph into which the stitching edge is inserted in step S2, which includes: simplifying the conflict graph by a conflict graph simplification algorithm, simplifying it into multiple clusters to reduce the scale of the problem, wherein the conflict graph simplification algorithm includes: an algorithm for independent component calculation, an algorithm for hiding vertices with a vertex degree less than 3, an algorithm for bridge edge disassembly, and an algorithm for clustering; Step S4, executing the SDP algorithm for color matching on the conflict graph simplified in step S3, and, when matching colors, adjusting the colors of two vertices with stitching edge relationships between clusters to improve the decomposition quality; wherein, when matching colors, firstly executing the SDP algorithm for the first cluster to obtain a corresponding result, and when matching colors for the second cluster and subsequent clusters, performing inter-cluster color matching adjustment, the inter-cluster color matching adjustment includes: for the second cluster, judging whether there is a stitching edge relationship between each vertex of the cluster and the vertex of the first cluster, if yes, giving an initial suggested color matching to the vertex in the cluster, the initial suggested color matching is the same as the vertex color matching in the first cluster, if no, executing the SDP algorithm for the two clusters to obtain a corresponding result, and the same method is used for the third cluster and the i-th cluster; Step S5: for the sub-clusters after the color matching operation in step S4, the internal optimal solution of each sub-conflict graph simplified by the three conflict graph simplification algorithms except clustering is obtained; these sub-conflict graphs are merged with the optimal solutions to obtain a complete layout decomposition result.

2. The clustering-based triple patterning lithography layout decomposition method according to claim 1, characterized in that: In step S1, a conflict graph is constructed according to the following method, including: Step S101, dividing the layout pattern into rectangles along the horizontal or vertical direction; Step S102: For the distance between two rectangles, if there is a partial overlap between the two rectangles in the X or Y axis direction, the distance is the parallel distance or the perpendicular distance between the two rectangles; if there is no overlap between the two rectangles in the X or Y axis direction, the distance is the distance between the two closest vertices. Step S103, comparing the polygon distances obtained in step S101 and step S102 with the minimum regular spacing value of triple pattern lithography to determine whether conflicting edges are formed between the polygons; Step S104, using vertices to represent patterns in the layout and solid lines to represent conflict edge relationships between patterns, thereby abstractly publishing a conflict graph corresponding to the layout.

3. The clustering-based triple patterning lithography layout decomposition method according to claim 2, characterized in that: In the step S2, for the polygons in the layout, the rectangles therein are projected to obtain a projection sequence on each rectangle, and then the candidate stitching edge insertion position is obtained according to the projection sequence, which specifically includes: Step S2011, select a polygon, determine the polygons around it that are in conflict with the polygon; then divide the rectangles of these polygons into regions according to the minimum spacing rule value of triple pattern lithography, and find the region on each rectangle of the selected polygon that is in conflict with the rectangles of the surrounding polygons; Step S2012: At this time, the currently selected rectangle is divided into multiple segments by these projections; each segment is marked with the number of projections of other adjacent rectangles on the currently selected rectangle, and the sequence composed of the number of projections on these segments is the projection sequence; there is also a rule that the endpoints must be zero, that is, the projection sequence must start and end with 0; Step S2013: select all possible candidate stitching edge insertion positions according to the obtained projection sequence, wherein the selection is performed according to the following rules, which include: Rule 1: All regions with projection sequence 0 are candidate stitching edge locations; Rule 2: Among the candidate seam edge positions in Rule 1, if the projection sequence starts with "01010", these candidate seam edge positions are redundant; Rule 3: If the projection sequence of the currently selected rectangle contains a subsequence "xyz", where x, y, z>0, and x>y, z>y, then there is a missing candidate seam edge in the segment with y projections, where a subsequence refers to a projection sequence containing at least 3 non-zero projections; missing refers to the seam edge position that cannot be found by the algorithm that only inserts seam edges in areas with zero projections.

4. The clustering-based triple patterning lithography layout decomposition method according to claim 3, characterized in that: For the candidate stitching edges obtained in step S2013, the candidate stitching edges that will cause other design rule check violations are removed, and the remaining candidate stitching edges are used as stitching edges in the conflict graph, wherein the candidate stitching edge positions that meet the following conditions need to be removed, including: Condition 1: After inserting the stitching edge at this position, a candidate stitching edge position of a rectangle smaller than the minimum lithography pattern size requirement fs_min will be generated; Condition 2: After the stitching edge is inserted at this position, when the segment rectangle is divided into two rectangles, the cross-coverage length of each rectangle does not meet the candidate stitching edge position of the minimum cross-coverage length m_0 required by the process; Condition 3: The candidate stitching edge position is at the corner of the rectangle; Condition 4: After inserting the seam edge at this position, a candidate seam edge position of a small rectangle that has no conflicting edge relationship with other polygons will be generated.

5. The clustering-based triple patterning lithography layout decomposition method according to claim 1, characterized in that: In step S3, the independent component calculation algorithm includes: performing a depth-first traversal on the vertices in the conflict graph to cluster the vertices with edge relationships, so as to divide the vertices in the original conflict graph into a number of independent components.

6. The clustering-based triple patterning lithography layout decomposition method according to claim 1, characterized in that: In step S3, the algorithm for hiding vertices with vertex degrees less than 3 includes: First, all vertices are set to visible, and a flag is set to indicate whether any vertices are hidden in this iteration; Next, perform iteration. In the iteration, after selecting a vertex v, first check its vertex status. If the vertex is hidden, skip it and check the next vertex; otherwise, set the degree of the vertex to 0. Then, for the selected vertex v, traverse all its adjacent vertices u. If the vertex u has been hidden, skip the adjacent vertex and check the next adjacent vertex. Otherwise, get the type of the edge between vertex u and vertex v. If it is a stitching edge, add 1 to the stitching edge degree and continue the loop; if it is a conflicting edge, add 1 to the conflicting edge degree of vertex v. After traversing all adjacent vertices of vertex v, determine whether the conflict edge degree of vertex v is less than 3 and the stitching edge degree is 0. If the condition is met, it means that the current vertex v can be hidden, and its state is set to hidden, and the flag hide_flag is set to true, and continue to the next iteration; Finally, if after traversing all the vertices in the graph, no new vertices with conflicting edge degrees less than 3 and vertices with stitching edge degrees of 0 are found, or all vertices have been hidden, the flag hide_flag will remain false, thus exiting the while loop and simplification is completed.

7. The clustering-based triple patterning lithography layout decomposition method according to claim 1, characterized in that: In step S3, the bridge edge disassembly algorithm includes: First, after selecting an initial vertex, the access order of the vertex and the minimum serial number value of the reachable vertex are both its access order; Next, push the vertex into the stack and traverse each adjacent vertex v of vertex u. If v has not been visited, continue depth-first exploration. When this vertex and its deeper vertices have been explored, start from the deepest vertex and trace back upwards in sequence to update the minimum reachable vertex number of each shallow vertex. If the minimum reachable vertex of the adjacent vertex v is larger than the access sequence number of the current vertex u, the edge (u, v) forms a bridge edge. If v has been visited and v is still in the stack, it means that the access sequence of vertex v is smaller than that of vertex u, and vertex u has found a vertex with a smaller access sequence that it can reach. At this time, it is necessary to update the vertex with the minimum access sequence that vertices v and u can reach. Then, after searching all the edges of the current vertex u, if it is found that the access order of node u is the same as the minimum sequence value of the vertex that can be reached by the vertex, it means that the vertex is the root vertex of a biconnected subgraph; at this time, the elements on the top of the stack are continuously popped until the popped element is the same as the root vertex, and all the vertices of a sub-conflict graph are obtained; Finally, the remaining vertices in the original conflict graph form another doubly connected subgraph. Using the bridge edge as the boundary, a large conflict graph can be divided into two small sub-conflict graphs to reduce the number of vertices in the sub-conflict graphs.

8. The clustering-based triple patterning lithography layout decomposition method according to claim 1, characterized in that: In step S3, the cluster algorithm specifically includes: By performing depth-first exploration only on conflicting edges, the vertices in an independent component are subdivided into multiple small clusters, and there are only stitching edge relationships but no conflicting edge relationships between clusters.

9. The clustering-based triple patterning lithography layout decomposition method according to claim 1, characterized in that: The step S5 comprises: For each cluster formed by clustering, the merging process is carried out together with the color matching process of each cluster; the merging process and the color matching process form such a cycle: for each cluster, first determine whether each vertex in the cluster has an inter-cluster stitching edge relationship with the vertices in other clusters; if so, and the vertices in the other cluster have obtained the optimal color matching through the SDP color matching algorithm, then an initial recommended color matching is given to the vertices in the cluster connected by this inter-cluster stitching edge; then the SDP color matching algorithm is solved in the cluster; if there is no inter-cluster stitching edge, or although there is an inter-cluster stitching edge, but the color matching has not been performed in the other cluster, the SDP algorithm is directly solved for the cluster; this process is repeated until all clusters have obtained color matching and the algorithm ends; the merging process of the optimal color matching results of each cluster is realized in this cycle; For the sub-conflict graphs formed by bridge edge decomposition, after obtaining the solution with the best decomposition quality in each sub-conflict graph, since the two sub-conflict graphs are solved relatively independently and connected by a bridge edge, it is necessary to adjust the color scheme in one of the sub-conflict graphs according to the colors of the two vertices connected by the bridge edge; if the colors of the two vertices connected by the bridge edge are the same, then the colors of all vertices in the second sub-conflict graph connected by the bridge edge are offset by 1 to merge the sub-conflict graphs obtained by bridge edge decomposition.

10. The clustering-based triple patterning lithography layout decomposition method according to claim 9, characterized in that: The step S5 further comprises: For those hidden vertices, when matching their colors, we only need to use the colors of the adjacent vertices of the hidden vertices to get the color matching; Since there is no edge relationship between the independent components, they are relatively independent when solved. Therefore, it is only necessary to directly merge the color matching results in each independent component to obtain the layout decomposition result of the entire layout.