Neural network based mask synthesis for integrated circuits.
A neural network-based approach for mask synthesis in integrated circuits addresses the inefficiencies of existing OPC methods by directly processing circuit design polygons, improving runtime efficiency and accuracy in generating mask designs.
Patent Information
- Application Number
- JP2022513227
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2020-08-26
- Filing Date
- 2020-08-28
- Publication Date
- 2024-12-09
- Estimated Expiration
- 2040-08-28
AI Technical Summary
The computational expense and inefficiency of existing optical proximity correction (OPC) methods in mask synthesis for integrated circuits, particularly due to the mismatch between lithography systems and shrinking feature sizes, lead to unwanted discrepancies in pattern transfer and increased computational resources.
A neural network-based approach that processes circuit design polygons directly, determining offset distances to generate mask design polygons without requiring pixel-based conversions, thereby improving runtime efficiency and accuracy.
This method reduces computational resources and enhances the quality of mask synthesis results by directly processing polygon edges, facilitating faster and more accurate mask generation for integrated circuits.
Smart Images

Figure 0007596364000001 
Figure 0007596364000002 
Figure 0007596364000003
Abstract
Description
[Technical field]
[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims the benefit of U.S. Patent Application No. 62 / 893,214, filed Aug. 29, 2019, the contents of which are incorporated herein by reference.
[0002] The present disclosure relates generally to the field of electronic design automation, and more particularly to mask synthesis for integrated circuits based on machine learning based models such as neural networks. [Background technology]
[0003] In semiconductor processing, photoresist masks are used to define patterns for various layers of semiconductor devices. Photoresist masks are formed by depositing a layer of light-sensitive material (i.e., photoresist) over a semiconductor structure, exposing the photoresist layer to a radiation source through a patterned photolithography mask ("photomask"), and then removing the unexposed (or exposed) portions of the photoresist layer. The material exposed through the photomask (e.g., metal) can be etched, thereby transferring the pattern on the photomask to the semiconductor device. Interference and other process effects can result in unwanted deviations between the desired pattern and the actual pattern transferred to the photoresist material. These unwanted deviations can significantly affect the performance of the resulting semiconductor device. To minimize these deviations, various compensation schemes have been implemented, including correcting the pattern imparted on the photomask using optical proximity correction (OPC) techniques. The various processes involved in mask synthesis OPC are computationally expensive, which makes the design process time-consuming and expensive. Summary of the Invention
[0004] Disclosed are arrangements (e.g., systems, methods, non-transitory computer-readable storage media having instructions executable by a processor) for generating a mask design for a circuit design. The system receives a circuit design including a circuit design polygon. The system accesses a neural network configured to receive as input features describing the circuit design within a neighborhood of an input location and output an offset distance from an edge of the circuit design polygon. The system uses the neural network to determine an offset distance for a location associated with the circuit design polygon. For example, the system may select a location along an edge of the circuit design polygon and generate an offset distance from the edge of the polygon. The system generates a mask design polygon based on the offset distance. The mask design polygon may represent a feature corresponding to the circuit design polygon, e.g., a main feature or an assist feature. The system generates a mask design for the circuit design using the generated mask design polygon.
[0005] Features corresponding to a location of the circuit design may include the width of a circuit design polygon near the location, the curvature of a circuit design polygon near the location, the depth of the gap between a polygon and a neighboring polygon, etc. The system selects sample points near the location and uses values that describe the circuit design at the sample points. For example, the system may sample distances to design edges, or perform a lithography simulation of the circuit design to generate a resist image and use features of the resist image at various sample points, such as the maximum, minimum, or slope of the resist image.
[0006] The neural network is trained using a labeled dataset based on a pre-generated mask design.
[0007] The features and advantages described in this summary and the detailed description that follows are not all-inclusive, and many additional features and advantages will become apparent to those of ordinary skill in the art upon review of the drawings, specification, and claims of this application.
[0008] The present disclosure will be more fully understood from the detailed description set forth below and from the accompanying figures of embodiments of the present disclosure. These figures are used to provide an understanding and understanding of the embodiments of the present disclosure, and are not intended to limit the scope of the present disclosure to these specific embodiments. Furthermore, these figures are not necessarily drawn to scale. [Brief description of the drawings]
[0009] [Figure 1] FIG. 2 illustrates an overall process of mask synthesis for a circuit design, according to an embodiment. [Diagram 2] FIG. 1 illustrates an example mask design with offsets from the circuit design for mask features according to an embodiment. [Diagram 3] FIG. 1 illustrates an example of goal optimization in mask synthesis, according to an embodiment. [Figure 4] FIG. 1 is a block diagram of a system architecture of a mask synthesis system according to an embodiment. [Diagram 5] 1 is a flowchart of a process for performing mask synthesis of a circuit design according to an embodiment. [Figure 6] FIG. 1 illustrates feature extraction using sample points in the neighborhood of a location according to one embodiment. [Figure 7] FIG. 2 illustrates feature extraction based on level set transformation of a layout according to an embodiment. [Figure 8] FIG. 2 illustrates feature extraction based on a resist image of a layout according to an embodiment. [Figure 9] FIG. 1 illustrates feature extraction based on the medial axis of a layout, according to an embodiment. [Figure 10]FIG. 13 illustrates the output of a neural network for a set of target locations, according to one embodiment. [Figure 11] FIG. 13 illustrates generation of assist features based on offset distance according to one embodiment. [Figure 12] FIG. 13 illustrates generation of mask design polygons based on offset distance according to an embodiment. [Figure 13] FIG. 13 illustrates target optimization based on offset distance, according to an embodiment. [Figure 14] FIG. 2 illustrates an architecture of a neural network used in mask design, according to an embodiment. [Figure 15] FIG. 2 illustrates an architecture of a neural network with multiple outputs used for mask design, according to one embodiment. [Figure 16] FIG. 2 illustrates an architecture of a neural network that provides input to another neural network used in mask design, according to an embodiment. [Figure 17] 1 is a flowchart of various processes used during the design and manufacture of integrated circuits according to some embodiments. [Figure 18] FIG. 1 is an abstract diagram of an example computer system in which embodiments may operate. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0010] The electronic design automation (EDA) process includes a tape-out stage that generates data that will be used to fabricate lithography masks. During mask data synthesis, the "tape-out" data is used to fabricate lithography masks, which are then used to fabricate the finished integrated circuits. In very large scale integrated circuit (VLSI) design flows, mask optimization is a key challenge due to the mismatch between lithography systems and ever-shrinking feature sizes. Optical proximity correction (OPC) is a resolution enhancement technique (RET) used to improve the printability of masks. However, due to shrinking chip dimensions and increasing circuit complexity, the mask optimization process is consuming more and more computing resources.
[0011] Some techniques for mask design utilize convolutional neural networks that represent the mask and circuit designs as pixelated images. The mask synthesis optical proximity correction (OPC) problem is not formulated in terms of pixelated images. The mask synthesis OPC problem is formulated in terms of incremental edge offsets from the design layout for primary features and larger edge offsets for auxiliary feature (AF) placements. This requires pixel-based convolutional neural network mask synthesis to require a computationally intensive step of converting data between grids and polygons, which results in poor run-time performance and poor quality results.
[0012] In contrast, the disclosed neural network according to various embodiments does not require conversion between pixelated representations. The disclosed neural network receives as input features describing the circuit design in a neighborhood of a location and outputs offset distance values. The offset distance values are used to generate mask design polygons that represent the features of the mask design. The disclosed model is a natural fit into the edge-based OPC framework, making it easy and efficient to integrate with this framework in run-time execution.
[0013] Certain pixel-based solutions run into another problem because the placement of the pixel grid relative to the edges of the design is inconsistent when the design pitch is not aligned with the pixel spacing. This occurs because the design pitch is smaller than the dimensions of one pixel. As a result, conversion of data to and from a pixel representation causes inaccuracies, which results in poor quality results. In contrast, the disclosed embodiments process polygon edges, which do not require conversion to and from a pixel-based representation, and therefore do not encounter this problem.
[0014] Furthermore, the techniques disclosed herein use circuit design polygons instead of pixel-based solutions, which changes the problem from a two-dimensional grid to a one-dimensional set of edges. This improves the run-time of the process of generating masks for circuit designs by efficiently obtaining either a final mask solution or an intermediate mask solution. Furthermore, these techniques improve the quality of results for a given amount of computing resources used by replacing computationally expensive techniques, such as iterative target optimization or inverse lithography (ILT) techniques, with less computationally expensive techniques.
[0015] Overall System Environment FIG. 1 illustrates an overall process of mask design for a circuit design, according to one embodiment. A mask design system 100 receives a circuit design 110 as input and generates a mask design. The mask design system 100 uses a machine learning model to determine offsets from edges of geometric shapes of the circuit design to generate corresponding geometric shapes of the mask design. The mask design system generates features for the mask design based on the offsets output by the machine learning model. Details of the mask design system 100 are shown in and described in conjunction with FIG. 2.
[0016] 2 illustrates an example mask design with offsets from a circuit design with respect to features of the mask, according to one embodiment. Polygon 210 represents the geometric shape of the circuit design. Polygon 220 represents a main feature of the mask generated by mask design system 100 from the circuit design. The edge of polygon 220 representing the main feature is determined based on an offset 240 from a corresponding edge of polygon 210 of the circuit design. The position of a secondary feature represented by polygon 230 is determined using an offset 250 from an edge of polygon 210 representing the circuit design.
[0017] A technique used in mask synthesis called target optimization allows the circuit design to be modified. This technique is used for overly constrained cases that may be difficult to manufacture as drawn, but may be easier to manufacture by introducing perturbations. Similar to OPC mask optimization, target optimization is formulated as a search for edge offsets from the circuit design and depends on the local geometric environment of each target edge.
[0018] 3 illustrates an example of goal optimization in mask synthesis, according to one embodiment. Polygons 310 represent the geometric shapes of a circuit design. These polygons are modified according to the goal optimization to obtain polygons 320. For example, polygon 310a is modified to obtain polygon 320a, polygon 310b is modified to obtain polygon 320b, and polygon 310c is modified to obtain polygon 320c. The modification of polygon 310 to obtain the modified polygon is determined using offsets 330.
[0019] Thus, feature specifications, including primary and secondary features, as well as modifications to the circuit design for the target optimization are determined using offsets from edges of polygons in the circuit design. Mask design system 100 uses machine learning based models to determine the offsets used for these techniques.
[0020] The machine learning based approach disclosed herein allows additional OPC corrections to the results provided by the machine learning model, which allows the output of the machine learning model to be further improved for generating a final mask solution. Thus, the machine learning based mask generation acts as a shortcut to bring the mask closer to the final solution and make the OPC easier to complete the corrections. This accelerates the entire design process. Some embodiments perform target optimization by applying the machine learning based model to determine offsets for adjusting the circuit design polygons, and then perform OPC based on this new target using standard OPC methods to correct the mask. The resulting mask solution has improved runtime efficiency and higher accuracy compared to conventional techniques.
[0021] Mask Synthesis System 4 illustrates a block diagram of a system architecture of a mask synthesis system, according to one embodiment. Mask design system 100 includes a neural network 410, a training component 420, a feature extraction component 430, a mask synthesis component 440, a circuit design database 450, and a mask design database 460. Other embodiments may include more or fewer components than those shown herein. Functionality illustrated herein as being performed by particular modules may be performed by modules other than those illustrated herein.
[0022] Circuit design database 460 stores the circuit design that is processed by mask design system 100 to generate a mask. In one embodiment, the representation of the circuit design that is processed by mask design system 100 is based on a Graphic Design System (GDS) format, such as the GDSII format or the Open Artwork System Interchange Standard (OASIS) format. Other embodiments may represent the circuit design using other formats. The generated mask is stored in mask design database 460.
[0023] The neural network 410 is configured to receive as input features describing the circuit design. The features describe the circuit design in the vicinity of a location associated with a particular circuit design polygon. For example, the location may represent a point on an edge of the polygon, a point inside the polygon, or a point outside the polygon. The neural network 410 generates one or more offset distances from the circuit design polygon. The offset distances may be distances along a normal to the edge of the circuit design polygon. The offset distances may be used to generate main features of the mask design, auxiliary features for the mask design, or a target circuit design obtained by modifying the circuit design for the target optimization. The offset distances may be positive or negative for both the main features and auxiliary features, which corresponds to external and internal auxiliary feature polygon generation. An offset for the target setting may also be output. In an embodiment, the neural network 410 is based on a multi-layer perceptron. Although the embodiments described herein are based on neural networks, the techniques disclosed herein are also compatible with other types of machine learning based models.
[0024] The training component 420 trains the neural network using a labeled training data set generated from pre-generated mask designs for the circuit design, e.g., mask designs generated using conventional design tools. The training component 420 applies the neural network to the training data set and adjusts the weights of the neural network based on a loss function that measures the difference between the predicted output of the neural network and the labeled data. The training component 420 applies backpropagation to adjust the weights of the neural network based on the loss function, e.g., using steepest descent.
[0025] The feature extraction component 430 extracts various features of the circuit design to provide as input to the neural network 410. The feature extraction component 430 extracts features based on geometric measurements of polygons in the circuit design, such as width, space, curvature, etc., near a location. Features may be determined using functions of local geometry, such as level set field values, Voronoi diagrams, convolutions of rasterized design polygons, and other transformations. The feature extraction component 430 also extracts features based on lithography simulation-based measurements, such as aerial image values or resist image values, near a location, such as functions of values based on lithography simulation-based measurements, such as slope, minimum, maximum, etc. Although the embodiments disclosed herein describe features related to polygons, the disclosed techniques are applicable to other types of geometric shapes, such as curved shapes.
[0026] The mask synthesis component 440 invokes various components of the mask design system 100 to generate a mask for a circuit design, in this case, the mask synthesis component 440 loads the circuit design from the circuit design database 450, selects portions of the circuit design, e.g., polygons, from the circuit design, selects locations associated with each polygon, extracts features using the feature extraction component, applies the neural network 410 to the extracted features to generate offset distances, and generates a mask based on the offset distances.
[0027] In some embodiments, mask design system 100 is implemented as a distributed system having multiple computer processors that process different portions of the circuit design in parallel. For example, mask synthesis component 230 may process different portions of the circuit design in parallel and generate masks for the different portions of the circuit design in parallel.
[0028] The processes performed by the components of mask design system 100 are described in detail below.
[0029] Overall process of mask synthesis FIG. 5 illustrates a flowchart of a process for performing mask synthesis of a circuit design, according to one embodiment. The steps illustrated in the flowchart of FIG. 5 may be performed in an order different than that shown in the flowchart. For example, certain steps may be performed in parallel with other steps. Furthermore, various steps illustrated in the flowchart may be performed using parallel or distributed systems.
[0030] Mask design system 100 receives 510 a circuit design representing the physical layout of the circuit. Mask design system 100 accesses neural network 410. Mask design system 100 repeats the following steps for each of a set of circuit design polygons obtained from the circuit design: For each circuit design polygon, mask design system 100 identifies one or more locations associated with the circuit design polygon. The locations may represent points on an edge of the polygon, or points inside or outside the polygon.
[0031] Mask design system 100 repeats steps 520, 530, 540, and 550 for each location. Mask design system 100 selects 520 a location associated with a circuit design polygon for processing. Mask design system 100 determines 530 a set of features describing the circuit design within a neighborhood of the selected location. A neighborhood may be defined as a portion of the circuit design that is within a threshold distance of the location. Mask design system 100 provides 540 the set of features as input to neural network 410. Mask design system 100 executes 550 neural network 410 to determine an offset distance from an edge of the circuit design polygon.
[0032] The mask design system 100 generates 560 a mask design polygon that corresponds to the circuit design polygon based on the offset distance. The mask design polygon may represent a main feature or a secondary feature of the mask design. For example, the mask design system 100 may move an edge of the circuit design polygon by the offset distance to obtain an edge of the main feature. Alternatively, the mask design system may generate a mask design polygon that represents a secondary feature that is at an offset distance from an edge of the circuit design polygon. In one embodiment, the mask design system 100 modifies the circuit design polygon using the offset distance to obtain a new target polygon that is used to generate the mask.
[0033] The mask design system 100 uses the generated mask design polygons to generate a mask design for the circuit design 570. For example, the mask design system 100 combines the various resulting mask design polygons to generate an overall mask design.
[0034] Various steps of the process shown in Figure 5 may be performed using multiple processors, for example using a distributed architecture. For example, different portions of a circuit design may be processed in parallel to generate corresponding portions of a mask design. In this case, a large circuit design may be divided across multiple processors to speed up the execution of the process.
[0035] Feature Extraction The feature extraction component 430 extracts various features for each location X and provides them as input to the neural network 410. The location X is assumed to be associated with an edge E of a circuit design polygon C. For example, edge E may be the edge on which the location X is located, or the edge closest to the location X if the location X is inside or outside C. The feature extraction component extracts various features for each point P in the neighborhood of the location X, i Sample the set of
[0036] 6 illustrates feature extraction using sample points in the neighborhood of a location, according to one embodiment. The mask design system 100 extracts a point P in the neighborhood of a location X. i The neighborhood may represent areas of the circuit design that are within a threshold distance of location X. Mask design system 100 orders these points such that the same ordering is used for the sample points during training of the neural network and at inference time, i.e., at run-time when the neural network is run on a new circuit design.
[0037] According to one embodiment, the mask design system 100 moves a distance D1 along a normal to the edge E of the circuit design polygon C. Starting with a point that is distance D1 from the edge E, the mask design system 100 moves in a circle (either in a clockwise or counterclockwise direction) to sample points at the same distance. FIG. 6 shows the mask design system 100 moving in a counterclockwise direction to sample points P1, P2, P3, ..., and so on. Once the mask design system 100 has completed one revolution in the circle at distance D1, the mask design system 100 moves to a longer distance D2 normal to the edge E. The mask design system 100 repeats the counterclockwise rotation at distance D2, and completes another revolution in the circle at distance D2 to sample points at the same distance. The mask design system repeats this process until a threshold distance that defines a neighborhood is reached. This is an example mechanism for consistently ordering points in the neighborhood of location X, and the mask design system 100 may use other mechanisms for ordering the sample points.
[0038] The feature extraction component 430 extracts points P or sample points P that may represent location X. i The feature extraction component 430 may move in the direction of the normal to edge E such that the normal is oriented toward the inside of the polygon to determine the other edge of polygon C. The feature extraction component 430 uses the distance that it extends as the width of the polygon. The feature extraction component 430 may move in a direction different from the normal to edge E to determine the width at an angle, for example, if point P is at or near a corner of the polygon where two edges meet. (traverse) possible.
[0039] The feature extraction component 430 moves along the normal of the outward facing edge E of polygon C to find the closest neighboring polygon. (traverse) By doing this, we can extract features that represent the spaces between the polygons. (traverse)The distance represents the width of the space around the polygon. The feature extraction component 430 may move along a direction different from the normal to reach a point near the intersection of two edges of the circuit design polygon C, for example. (traverse) possible.
[0040] The feature extraction component 430 may extract features that represent the curvature of the polygon in a neighborhood of point P. The feature extraction component 430 may fit a circle through a set of edges of the polygon C and use the inverse of the radius of this circle as a measure of curvature.
[0041] 7 illustrates feature extraction based on a level set transformation of a layout, according to one embodiment. As shown in FIG. 6, the feature extraction component 430 extracts points P i , where t is the distance from the pixel to the edge of the circuit design polygon. In one embodiment, the level set is generated using a distance function, where the value of each pixel represents the shortest distance from that pixel to a point on the edge of the circuit design polygon. The feature extraction component 430 uses the level set value at each sample point as a feature. Thus, the feature value corresponding to a sample point represents the distance from that point to the nearest edge of the circuit design polygon. FIG. 7 illustrates a threshold 710 such that sample points with level set values below the threshold are inside the circuit design polygon, sample points with level set values above the threshold are outside the circuit design polygon, and sample points with values equal to the threshold are on the edge of the circuit design polygon. Other embodiments may use features based on any technique to determine the distance from a sample point to an edge of a circuit design shape, for example using a geometric search.
[0042] 8 illustrates feature extraction based on a resist image of a layout, according to one embodiment. The feature extraction component 430 may use values of the resist image at various sample points as features. In one embodiment, the feature extraction component 430 uses measurements based on lithography simulation as features. The feature extraction component 430 uses an initial mask, e.g., a circuit design polygon, as an initial approximation of the mask. The initial mask may include the circuit design polygon along with some auxiliary features that are determined based on the circuit design polygon.
[0043] The feature extraction component 430 performs a lithography simulation using the initial mask to generate aerial or resist image values. The feature extraction component 430 uses aerial or resist image values near a location X. For example, the feature extraction component 430 may generate sample points near a location X as shown in FIG. 6 and determine image values at the sample points as features. The feature extraction component 430 may use image tilt at the sample points as features. The feature extraction component 430 may use other attributes of the image, such as a maximum or minimum value of the image near a point as a feature.
[0044] In one embodiment, the feature extraction component 430 rasterizes a portion of the circuit design to generate a pixelated representation and convolves the pixelated image, for example, with a Gaussian kernel. The feature extraction component 430 uses the results of the convolution as features.
[0045] FIG. 9 illustrates feature extraction based on a central axis 910 of a layout, according to one embodiment. A central axis represents a point in a circuit design where at least two closest points are on the boundary of one or more polygons of the circuit design, e.g., both points are on the same polygon, as shown in portion 910a of the central axis, or one point is on one polygon and a second point is on another polygon, as shown in portion 910b of the central axis. Thus, each point on the central axis is equidistant from at least two points that are each on the polygon that is closest to that point. The feature extraction component 430 determines a central axis for the circuit design and uses the distance between the central axis and the closest edge of the circuit design polygon as a feature. The feature extraction component 430 extracts features that represent the distance from the determined central axis at various sample points. For example, the feature extraction component 430 can convert the central axis into a pixelated representation, e.g., by computing the distance to the central axis at every pixel in the region. The feature extraction component 430 uses the distance value at the sample point from the pixelated representation of the central axis as a feature.
[0046] Mask generation based on offset distance FIG. 10 illustrates the output of a neural network for a set of target locations, according to one embodiment. As shown in FIG. 10, the output of the neural network 410 may be parameterized for a set of points, e.g., x0, x1, and x2 as shown in FIG. 10. The mask synthesis component 440 extends a vertical ray 1020 in a direction normal to the edge 1010 of the circuit design polygon. The mask synthesis component 440 identifies points 1040 at offset distances d1 and d2 that are output by the neural network 410 along the ray 1020. The mask synthesis component 440 generates primary and secondary features based on these points 1040, and also uses these points for the goal optimization.
[0047] FIG. 11 illustrates generation of an auxiliary feature based on an offset distance, according to one embodiment. The mask compositing component 440 may connect corresponding points 1040 to define edges of the primary feature or the auxiliary feature. For example, the mask compositing component 440 may generate edges by connecting points at the same distance d1 from the edge 1010 to generate an edge for the primary feature. The mask compositing component 440 may combine the edge from connecting points at distance d1 with an edge formed by connecting points at distance d2 to generate an auxiliary feature. The edges formed as shown in FIG. 11 are provided as input to a tool for post-processing polygons.
[0048] 12 illustrates generation of mask design polygons based on offset distances, according to one embodiment. For traditional OPC output where offsets are stored for each sub-segment of a circuit design edge, mask synthesis component 440 provides the offset distance d output by neural network 410 to an OPC tool that generates mask polygon 1220 from circuit design polygon 1210 using offset distance d1. The OPC tool may use offset distance d2 to generate auxiliary feature polygon 1230 from circuit design polygon 1210. The OPC tool may obtain mask design polygons 1220 and 1230 by combining edges obtained by moving segments of edges of circuit design polygon 1210 by the offset distances.
[0049] 13 illustrates a target optimization based on offset distance, according to one embodiment. For a target optimization in which offsets are also stored for each sub-segment of a design edge, the mask synthesis component 440 provides the offset distance d output by the neural network 410 to an OPC tool that creates a target polygon 1320 by modifying an input circuit design polygon 1310. The OPC tool can obtain the target circuit design polygon 1320 by combining edges obtained by moving segments of the edges of the circuit design polygon 1310 by the offset distance.
[0050] Neural Network Architecture FIG. 14 illustrates the architecture of a neural network used for mask design, according to one embodiment. As shown in FIG. 14, the neural network 410 is a multi-layer connected neural network, e.g., a multi-layer perceptron. The neural network includes an input layer 1410, one or more hidden layers 1420, and an output layer 1430. This embodiment uses a different neural network for each geometry type of mask design to be generated. For example, one neural network may output offset distances used to generate primary features, another neural network outputs offset distances used to generate secondary features, and another neural network outputs offset distances used for goal optimization.
[0051] 15 shows an architecture of a neural network with multiple outputs 1510 used for mask design, according to one embodiment. In this embodiment, the same neural network generates multiple outputs 1510, one for each geometry type of the mask design being generated. Thus, the neural network may generate one output representing the offset distance used to generate the primary features, another output representing the offset distance used to generate the secondary features, and another output representing the offset distance used for goal optimization.
[0052] 16 illustrates an architecture of a neural network that provides input to another neural network used in mask design, according to one embodiment. For example, neural network 1610 outputs offset distances that are used for goal optimization. The results of the goal optimization are provided as input to neural network 1620, which further generates offset distances that are used to generate primary and secondary features.
[0053] Electronic Design Automation Process FIG. 17 illustrates a set of example processes 1700 for translating and verifying design data and instructions representing an integrated circuit for use during the design, verification, and manufacture of such products. Each of these processes may be structured and implemented as multiple modules or operations. The term "EDA" refers to the term "Electronic Design Automation." These processes begin with the creation of a product idea 1710 using information provided by a designer, which is translated to produce a product using a set of EDA processes 1712. Once the design is finalized, the design is taped out 1734, at which time artwork (e.g., geometric patterns) for the integrated circuit is sent to a manufacturing facility for the manufacture of a mask set, which is then used to manufacture the integrated circuit. After tape-out, a semiconductor die is manufactured 1736, and packaging and assembly processes 1738 are performed to produce a completed integrated circuit 1740.
[0054] Specifications of circuits or electronic structures can range from low-level transistor material layouts to high-level description languages. To design circuits and systems, high-level abstractions using hardware description languages ("HDLs") such as VHDL, Verilog, SystemVerilog, SystemC, MyHDL, or OpenVera can be used. HDL descriptions can be converted to logic-level register transfer level ("RTL") descriptions, gate-level descriptions, layout-level descriptions, or mask-level descriptions. Each lower level of abstraction adds more useful details to the design description, for example, more details about the module that contains the description. Lower levels of abstraction can be computer-generated, derived from design libraries, or produced by another design automation process. An example of a specification description language at a lower level of abstraction for describing more detailed descriptions is SPICE, which is used for detailed descriptions of circuits with many analog components. The description at each level of abstraction can be used by the corresponding tool (e.g., formal verification tool) at that layer. The design process may use the sequence depicted in Figure 17. The process described may be implemented by an EDA product (or tool).
[0055] In system design 1714, the functionality of the integrated circuit to be manufactured is described. The design may be optimized for desired characteristics such as, for example, power consumption, processing power, area (physical area and / or lines of code), and reduced cost. Partitioning of the design into different types of modules or components can occur at this stage.
[0056] In logic design and functional verification 1716, modules or components in a circuit are specified in one or more description languages and the functional accuracy of the specification is checked. For example, components of a circuit may be verified to produce outputs that conform to the requirements of the specification of the circuit or system being designed. Functional verification may use simulators and other programs such as testbench generators, static HDL checkers, and formal verifiers. In some embodiments, special systems of components, called "emulators" or "prototyping systems," are used to speed up functional verification.
[0057] In synthesis and design for test 1718, the HDL code is converted into a netlist. In some embodiments, the netlist may be a graph structure, where edges of the graph structure represent components of the circuit, and where nodes of the graph structure represent the manner in which the components are interconnected. Both the HDL code and the netlist are hierarchical products that an EDA product can use to verify, during manufacturing, that an integrated circuit will perform according to a specified design. The netlist may be optimized for a target semiconductor manufacturing technology. Furthermore, the completed integrated circuit may be tested to verify that it meets the requirements of the specification.
[0058] In netlist verification 1720, the netlist is checked for compliance with timing constraints and for correspondence with the HDL code. In design planning 1722, an overall floorplan for the integrated circuit is constructed and analyzed for timing and top-level routing.
[0059] In the layout or physical implementation 1724, physical placement (positioning of circuit components such as transistors or capacitors) and routing (connection of circuit components by multiple conductors) may be performed to select cells from a library to realize a particular logic function. As used herein, the term "cell" refers to a set of transistors, other components, and interconnects that provide a Boolean logic function (e.g., AND, OR, NOT, XOR) or a memory function (e.g., a flip-flop or latch). As used herein, a circuit "block" may refer to two or more cells. Both cells and circuit blocks may be referred to as modules or components, and they may be realized both as physical structures and as simulations. Parameters such as size are specified for the selected cells (based on "standard cells") and made accessible in a database for use by the EDA product.
[0060] In analysis and extraction 1726, the circuit functionality is verified at the layout level, which allows refinement of the layout design. During physical verification 1728, the layout design is checked to ensure that manufacturing constraints, such as DRC constraints, electrical constraints, lithography constraints, etc., are correct and that the circuit functionality meets the HDL design specifications. In resolution enhancement 1730, the geometry of the layout is transformed to improve how the circuit design will be manufactured.
[0061] At tape-out, data is produced that is used to fabricate a lithography mask (possibly after applying lithography refinements). During mask data preparation 1732, the "tape-out" data is used to create a lithography mask, which is then used to create the finished integrated circuit.
[0062] A storage subsystem of a computer system (e.g., mask design system 100 of FIG. 1) may be used to store programs and data structures used by some or all of the EDA products described herein, as well as products used for developing cells for the libraries and for physical and logical design using the libraries.
[0063] Embodiments can be used to process circuit designs at various stages of an electronic design automation process to store the circuit layout, such as layout or physical implementation 1724, analysis and extraction 1726, mask data preparation 1732, etc.
[0064] Computer Architecture 18 illustrates an example machine of a computer system 1800 on which a set of instructions may be executed to cause the machine to perform any one or more of the methodologies discussed herein. In alternative implementations, the machine may be connected (e.g., networked) with other machines in a LAN, an intranet, an extranet, and / or the Internet. The machine may operate within the processing capacity of a server or a client machine in a client-server network environment, as a peer machine in a peer-to-peer (or distributed) network environment, or as a server or client machine in a cloud computing infrastructure or environment.
[0065] The machine may be a personal computer (PC), tablet PC, set-top box (STB), personal digital assistant (PDA), cell phone, web appliance, server, network router, switch or bridge, or any machine capable of executing a set of instructions (sequential or otherwise) that specify actions to be taken by the machine. Additionally, although a single machine is depicted, the term "machine" shall also be construed to include any collection of machines that individually or coupled together execute a set (or sets) of instructions to perform any one or more of the methodologies discussed herein.
[0066] The example computer system 1800 includes a processing device 1802, a main memory 1804 (e.g., read only memory (ROM), flash memory, dynamic random access memory (DRAM), such as synchronous DRAM (SDRAM), static memory 1806 (e.g., flash memory, static random access memory (SRAM), etc.), and a data storage device 1818, which communicate with each other via a bus 1830.
[0067] The processing device 1802 represents one or more processors, such as a microprocessor, a central processing unit, or the like. More specifically, the processing device may be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, or a processor implementing other instruction sets, or a processor implementing a combination of instruction sets. The processing device 1802 may also be one or more special-purpose processing devices, such as an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), a network processor, or the like. The processing device 1802 may be configured to execute instructions 1826 to perform the operations and steps described herein.
[0068] The computer system 1800 may further include a network interface device 1808 for communicating over a network 1820. The computer system 1800 may also include a video display unit 1810 (e.g., a liquid crystal display (LCD) or a cathode ray tube (CRT)), an alphanumeric input device 1812 (e.g., a keyboard), a cursor control device 1814 (e.g., a mouse), a graphics processing unit 1822, a signal generating device 1816 (e.g., a speaker), a graphics processing unit 1822, a video processing unit 1828, and an audio processing unit 1832.
[0069] The data storage device 1818 may include a machine-readable storage medium 1824 (also known as a non-transitory computer-readable medium) on which is stored one or more sets of instructions 1826 or software embodying any one or more of the methodologies or functions described herein. The instructions 1826 may also reside, in whole or at least in part, in the main memory 1804 and / or in the processing device 1802 during execution thereof by the computer system 1800, with the main memory 1804 and the processing device 1802 also constituting machine-readable storage media.
[0070] In some implementations, the instructions 1826 include instructions for implementing functionality corresponding to the present disclosure. Although the machine-readable storage medium 1824 is shown as being a single medium in certain exemplary implementations, the term "machine-readable storage medium" should be interpreted to include a single medium or multiple media (e.g., a centralized or distributed database and / or associated caches and servers) that store one or more sets of instructions. The term "machine-readable storage medium" should also be interpreted to include any medium capable of storing or encoding a set of instructions for execution by a machine and causing the machine and processing device 1802 to perform any one or more of the methodologies of the present disclosure. The term "machine-readable medium" should therefore be interpreted to include, but is not limited to, solid-state memory, optical media, and magnetic media.
[0071] Some portions of the preceding detailed descriptions are presented in terms of algorithms and symbolic representations of operations on data bits within a computer memory. These algorithmic descriptions and representations are the means used by those skilled in the data processing arts to most effectively convey the substance of their work to others skilled in the art. An algorithm may be a sequence of operations leading to a desired result. The operations require physical manipulations of physical quantities. Such quantities may take the form of electrical or magnetic signals capable of being stored, combined, compared, and otherwise manipulated. Such signals may be referred to as bits, values, elements, symbols, characters, terms, numbers, or the like.
[0072] It should be borne in mind that these and similar terms are to be associated with the appropriate physical quantities, but are merely convenient labels applied to these quantities.Unless expressly stated otherwise, as will be apparent from this disclosure, throughout the description, it will be appreciated that certain terms refer to actions and processes of a computer system or similar electronic computing device that manipulate and transform data and memory represented as physical (electronic) quantities in the computer system's registers and memory into other data that are similarly represented as physical quantities in the computer system's memory or registers or other such information storage devices.
[0073] The present disclosure also relates to an apparatus for performing the operations herein. This apparatus may be specially constructed for the intended purpose or may include a computer selectively activated or reconfigured by a computer program stored in the computer. Such computer programs may be stored in a computer-readable storage medium, such as any type of disk, including but not limited to floppy disks, optical disks, CD-ROMs, and magneto-optical disks, read-only memory (ROM), random access memory (RAM), EPROM, EEPROM, magnetic or optical cards, or any type of medium suitable for storing electronic instructions, any of which may be coupled to a bus of a computer system.
[0074] The algorithms and displays presented herein are not inherently related to any particular computer or other apparatus. Various other systems may be used with programs in accordance with the teachings herein, or it may prove convenient to construct more specialized apparatus for carrying out the methods. In addition, this disclosure is not directed to any particular programming language in its description. It will be appreciated that a variety of program languages may be used to implement the teachings of the disclosure described herein.
[0075] The present disclosure may be provided as a computer program product or software, which may include a machine-readable medium having stored thereon instructions that may be used to program a computer system (or other electronic device) to perform a process according to the present disclosure. A machine-readable medium includes any mechanism for storing information in a form readable by a machine (e.g., a computer). For example, a machine-readable (e.g., computer-readable) medium includes a machine (e.g., computer) readable storage medium, such as, for example, read-only memory ("ROM"), random access memory ("RAM"), magnetic disk storage media, optical storage media, flash memory devices, and the like.
[0076] In the above disclosure, implementations of the disclosure have been described with reference to specific exemplary implementations thereof. However, it will be apparent that various modifications thereto can be made without departing from the broader spirit and scope of the implementations of the disclosure as set forth in the following claims. When the disclosure refers to some singular tense elements, the figures may depict more than one element, and like elements are labeled with like numerals. The disclosure and drawings should therefore be considered in an illustrative sense and not in a limiting sense.
Claims
1. 1. A method for performing machine learning based mask synthesis for circuit design, comprising: receiving, by a processing device, the circuit design representing a physical layout of a circuit, the circuit design including a plurality of circuit design polygons; by the processing device, for each of the plurality of circuit design polygons: selecting a location associated with said circuit design polygon; determining a set of features comprising feature values sampled at a plurality of sample points within a threshold distance of said location, said feature values being ordered by moving said sample points using an ordering method used during training of the machine learning based model; providing the set of features as input to the machine learning based model; executing the machine learning based model to determine an offset distance from an edge of the circuit design polygon; and generating a mask-design polygon corresponding to the circuit-design polygon based on the offset distance; generating, by the processing device, a mask design for the circuit design using the generated mask design polygons.
2. The machine learning based model is receiving as input features describing a circuit design within a threshold distance of an input location associated with a particular circuit design polygon; a neural network configured to output the offset distance from the edge of the particular circuit design polygon; The method of claim 1 , wherein the neural network is run with the set of features provided as input to the neural network.
3. The method of claim 2 , wherein the neural network is a multi-layer perceptron.
4. The method of claim 1, wherein the set of features further includes a feature representing a width of the circuit design polygon in a vicinity of the location.
5. The method of claim 1, wherein the set of features further includes a feature representing a curvature of the circuit design polygon in a vicinity of the location.
6. The method of claim 1 , wherein the feature value sampled at each sample point represents a distance of the sample point from an edge of the circuit design feature.
7. The method of claim 1 , wherein the feature value sampled at each sample point represents a distance of the sample point from a central axis of the circuit design.
8. The method of claim 7, further comprising: performing a lithography simulation of the circuit design to generate a resist image; 2. The method of claim 1, comprising: using a feature value of the resist image at each sample point, the feature value of the resist image being one of a maximum, a minimum, or a slope.
9. Generating the mask design polygon corresponding to the circuit design polygon includes: identifying a plurality of points, each point being at an offset distance from a location of the circuit design polygon; and determining a mask design polygon representing an assist feature connecting the plurality of points.
10. The method of claim 1 , wherein the offset distance is a distance determined along a normal from an edge of the circuit design polygon.
11. The method of claim 1 , wherein the location is a point on an edge of the circuit design polygon.
12. The method of claim 1 , wherein the offset distance represents a distance between an edge of a polygon of the mask design and a corresponding edge of a corresponding polygon of the circuit design.
13. 2. The method of claim 1, wherein the mask design polygon that corresponds to a circuit design polygon represents a main feature of the mask design, and one or more edges of the mask design polygon are at an offset distance from a corresponding edge of the circuit design polygon.
14. Generating the mask design polygon corresponding to the circuit design polygon includes:
2. The method of claim 1, further comprising performing a target optimization by modifying the circuit design polygon according to the offset distance to obtain a target circuit design polygon that is further used to generate the mask design polygon.
15. 1. A non-transitory storage medium storing instructions that, when executed by a processor, cause the processor to perform operations for performing mask synthesis for machine learning based circuit design, the operations comprising: receiving a circuit design representing a physical layout of a circuit, the circuit design including circuit design polygons; For each of a plurality of circuit design polygons, selecting a location associated with said circuit design polygon; determining a set of features comprising feature values sampled at a plurality of sample points within a threshold distance of said location, said feature values being ordered by moving said sample points using an ordering method used during training of the machine learning based model; providing the set of features as input to the machine learning based model; executing the machine learning based model to determine an offset distance from an edge of the circuit design polygon; and generating a mask-design polygon corresponding to the circuit-design polygon based on the offset distance; and generating a mask design for the circuit design using the generated mask design polygons.
16. A processor; 11. A non-transitory storage medium storing instructions that, when executed by the processor, cause the processor to perform operations for performing mask synthesis for machine learning based circuit design, the operations comprising: receiving the circuit design representing a physical layout of a circuit, the circuit design including a plurality of circuit design polygons; For each of the plurality of circuit design polygons, selecting a location associated with said circuit design polygon; determining a set of features comprising feature values sampled at a plurality of sample points within a threshold distance of said location, said feature values being ordered by moving said sample points using an ordering method used during training of the machine learning based model; providing the set of features as input to the machine learning based model; executing the machine learning based model to determine an offset distance from an edge of the circuit design polygon; and generating a mask-design polygon corresponding to the circuit-design polygon based on the offset distance; and generating a mask design for the circuit design using the generated mask design polygons; A system comprising:
Citation Information
Patent Citations
Mask design device, mask design method, program, and method of manufacturing semiconductor device
JP2003315973A
Mask pattern manufacturing method using optical proximity effect correction method based on yield
JP2005092212A
Method for creating pattern data
JP2009294440A
Pattern predicting method, program, and apparatus
JP2010044101A
Feature-quantity extracting method, test pattern selecting method, resist model creating method, and designed-circuit-pattern verifying method
JP2010156866A
