Static voltage drop prediction method and system
Through the coordination of machine learning models and mapping tables, static voltage drops are predicted and optimized in electronic circuit design in advance, solving the problems of SIR analysis time spent in the existing technology and frequent trial and error in the wiring stage, achieving more efficient SIR prediction and Layout optimization.
Patent Information
- Application Number
- CN202310522609.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-10
- Publication Date
- 2024-11-12
AI Technical Summary
The prior art static voltage drop (SIR) analysis in electronic circuit design takes a long time and lacks early protection, resulting in frequent trial and error in the wiring stage, making it difficult to find the best layout, increasing the risk of SIR divergence.
The machine learning model is used to predict the static voltage drop results based on the plane layout data, and through mapping tables and compensation values, the static voltage drop is predicted and optimized in the placement stage in advance, reducing SIR violations, and automatically adjusting bump assignments to reduce SIR risks.
It significantly shortens the SIR query time, improves the accuracy of prediction results, reduces the number of trial and error times, improves layout optimization efficiency, and reduces SIR risks.
Smart Images

Figure CN118940713A_ABST
Abstract
Description
Technical Field
[0001] One embodiment of the present disclosure is directed to a static voltage drop prediction method and system, and more particularly to a static voltage drop prediction method and system for predicting the static voltage drop before a placement stage. Background Art
[0002] In the electronic circuit design process, one or more electronic design automation (EDA) tools may be utilized to design, optimize, and verify a semiconductor device design, such as a circuit design in a semiconductor chip.
[0003] During routing, conductive lines or interconnects may be formed to connect the various circuit components of the placement layout. After routing, a physical verification process may be performed on the semiconductor device, and then a voltage drop analysis may be performed. The voltage drop analysis may be referred to as an IREM analysis. During an IREM analysis, the semiconductor device is analyzed to determine if static voltage drops (SIR drops) exist that exceed or otherwise violate design rules. The SIR drops of a semiconductor device are at least partially due to the actual wiring or interconnections between the various circuit components or nodes in the device, and therefore the SIR drop analysis is typically performed after the device is routed.
[0004] In some approaches, the SIR signature flow is time-consuming and lacks early protection, which consumes time. In addition, the trial-and-error method cannot find the best bump position within limited attempts and floorplan layout corrections after inducing long fixed iterations during the routing stage, thus creating the risk of SIR divergence. Summary of the invention
[0005] One embodiment of the present disclosure provides a static voltage drop prediction method, comprising the following steps: receiving planar layout data of a planar layout of a semiconductor device by a static voltage drop prediction circuit; generating a first static voltage drop result based on the planar layout data by a machine learning model; generating a first similarity value based on a comparison of the planar layout data with a plurality of training data; generating a second static voltage drop result based on the first static voltage drop result in a mapping table and a first compensation value corresponding to the first similarity value; and generating bump assignment data to update the planar layout data based on a comparison between the second static voltage drop result and a plurality of predetermined static voltage drop values.
[0006] One embodiment of the present disclosure provides a static voltage drop prediction system, including a static voltage drop prediction circuit having a machine learning circuit, the machine learning circuit being trained based on a plurality of training data associated with a plurality of electronic device designs to predict a static voltage drop result, and being used to perform the following operations: receiving plane layout data of a plane layout; generating a static voltage drop result and a first similarity result; adjusting the static voltage drop result based on the first similarity result; and predicting an optimized static voltage drop result based on the static voltage drop result and a plurality of bump assignments before placing a plurality of units in the plane layout.
[0007] One embodiment of the present disclosure provides a static voltage drop prediction method, comprising the following steps: training a machine learning model in a machine learning circuit based on multiple training data and multiple test data associated with multiple electronic device designs; generating a mapping table based on the training data and the test data, the mapping table including multiple compensation values and multiple similarity values; and predicting a static voltage drop result by the machine learning model based on the floor plan data and the mapping table. BRIEF DESCRIPTION OF THE DRAWINGS
[0008] An aspect of an embodiment of the present disclosure is described in the accompanying Figure 1 The following detailed description is best understood when read together. Please note that, in accordance with standard industry practice, various features are not drawn to scale. In fact, the dimensions of the various features may be arbitrarily increased or reduced for clarity of discussion.
[0009] Figure 1 a block diagram of an illustrated electronic device design system according to some embodiments;
[0010] Figure 2 is a block diagram illustrating a static voltage drop (SIR) prediction system according to some embodiments;
[0011] Figure 3A is a schematic diagram illustrating a portion of a floorplan equipped with bump assignments in which one or more SIR violations are predicted or otherwise determined to exist, according to some embodiments;
[0012] Figure 3B is a schematic diagram illustrating a portion of a floorplan equipped with bump assignments in which one or more SIR violations are predicted or otherwise determined to exist, according to some embodiments;
[0013] Figure 4 is a flow chart illustrating a method of SIR prediction according to one or more embodiments;
[0014] Figure 5 A diagram according to one or more embodiments corresponds to Figure 4A flowchart of the method;
[0015] Figure 6 is a block diagram of an electronic design automation (EDA) system for designing an integrated circuit layout according to some embodiments of the present disclosure;
[0016] Figure 7 is a block diagram of an IC manufacturing system according to some embodiments and an IC manufacturing process associated with the IC manufacturing system.
[0017]
Explanation of symbols
[0018] 10: Electronic device design system
[0019] 20: Electronic Design Platform
[0020] 21: Floor plan tools
[0021] 22: Data Extraction Tools
[0022] 24: Placement tools
[0023] 26: Clock Tree Synthesis (CTS) Tool
[0024] 28: Wiring Tools
[0025] 29: Verification Tools
[0026] 40: Static Voltage Drop (SIR) Prediction Platform
[0027] 42: Static Voltage Drop (SIR) Prediction Tool
[0028] 44: Bump Assignment Adjustment Tool
[0029] 100: Static Voltage Drop (SIR) Prediction System
[0030] 102: Floor plan database
[0031] 104: Static Voltage Drop (SIR) Database
[0032] 106: Bump assignment database
[0033] 110: Machine Learning Circuits
[0034] 112: Static Voltage Drop (SIR) Results Database
[0035] 114: Mapping file database
[0036] 120: Static voltage drop (SIR) prediction circuit
[0037] 130: Bump assignment adjustment circuit
[0038] 302: Bump
[0039] 304: Static Voltage Drop (SIR) violation
[0040] 400: Static Voltage Drop (SIR) Prediction Method
[0041] 401: Operation
[0042] 402: Combination
[0043] 403: Combination
[0044] 404: Operation
[0045] 405: Operation
[0046] 406: Operation
[0047] 407: Operation
[0048] 408: Operation
[0049] 409: Mapping table
[0050] 410: Model training branch
[0051] 420: Static Voltage Drop (SIR) prediction branch
[0052] 421: Floor plan data
[0053] 422: Operation
[0054] 423: Predicted Static Voltage Drop (SIR) Results
[0055] 424: Operation
[0056] 425: Operation
[0057] 426: Operation
[0058] 427: Operation
[0059] 428: Compensation value
[0060] 429: Operation
[0061] 430: Operation
[0062] 431: Operation
[0063] 432: Operation
[0064] 433: Operation
[0065] 434: Operation
[0066] 435: Operation
[0067] 436: Operation
[0068] 437: Operation
[0069] 438: Operation
[0070] 439: Operation
[0071] 440: Operation
[0072] 441: Operation
[0073] 442: Operation
[0074] 443: Operation
[0075] 444: Operation
[0076] 445: Operation
[0077] 450: Bump Optimization Branch
[0078] 600:Electronic Design Automation System
[0079] 602: Processor
[0080] 604: Non-transitory storage media
[0081] 606: Computer program code (instructions)
[0082] 608: Bus
[0083] 610:I / O interface
[0084] 612: Network interface
[0085] 614: Network
[0086] 616: Manufacturing Tools
[0087] 620: Integrated Circuit (IC) Layout
[0088] 622: Design Specifications
[0089] 700: Integrated Circuit (IC) Manufacturing System
[0090] 720: Design Studio
[0091] 722: Integrated Circuit (IC) Design Layout
[0092] 730: Screen Room
[0093] 732: Data Preparation
[0094] 744:Mask Manufacturing
[0095] 745: veil
[0096] 750: Integrated Circuit (IC) Manufacturer / Fab / Fab
[0097] 752: Chip Manufacturing
[0098] 753:Semiconductor wafer
[0099] 760: Integrated Circuit (IC) Device DETAILED DESCRIPTION
[0100] The following disclosure provides many different embodiments or examples for implementing the different features of the provided subject matter. The following describes specific examples of components and configurations to simplify an embodiment of the present disclosure. Of course, these components and configurations are only examples and are not intended to be restrictive. For example, in the following description, the formation of a first feature above or on a second feature may include an embodiment in which the first and second features are directly contacted, and may also include an embodiment in which an additional feature may be formed between the first and second features so that the first and second features may not be directly contacted. In addition, an embodiment of the present disclosure may repeat reference numbers and / or letters in various examples. This repetition is for simple and clear purposes, and does not itself indicate the relationship between the various embodiments and / or configurations discussed.
[0101] The terms used in this specification generally have their ordinary meanings in the art and in the specific context in which each term is used. The use of examples in this specification, including examples of any term discussed herein, is illustrative only and in no way limits the scope and meaning of an embodiment of the present disclosure or any exemplified term. Likewise, an embodiment of the present disclosure is not limited to the various embodiments given in this specification.
[0102] Although the terms "first," "second," etc., may be used herein to describe various components, such components should not be limited by such terms. Such terms are used to distinguish one component from another component. For example, a first component may be referred to as a second component, and similarly a second component may be referred to as a first component without departing from the scope of the embodiments. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.
[0103] As used herein, the terms "comprising," "including," "having," "containing," "involving," and the like should be construed as open ended, ie, meaning including, but not limited to, including.
[0104] As used herein, "approximately," "about," "roughly," or "substantially" shall generally refer to any approximation or range of a given value, wherein the value or range varies according to the various technologies to which the value is related, and the range of values shall be interpreted in the broadest sense as understood by a skilled artisan familiar with an embodiment of the present disclosure to cover all such modifications and similar structures. In some embodiments, a value or range shall generally mean within 20%, preferably within 10%, and more preferably within 5% of a given value or range. The quantities given herein are approximate, thereby meaning that the terms "approximately," "about," "roughly," or "substantially" may be inferred without explicit statement, or mean other approximate values.
[0105] Embodiments provided herein include static voltage drop (SIR) prediction systems and methods for predicting SIR violations in a floorplan layout before placement is performed on the floorplan layout. In some embodiments, machine learning techniques are used to generate and / or modify a machine learning model, and a SIR prediction circuit can predict whether one or more SIR violations will exist in a particular floorplan layout by comparing floorplan data extracted from the floorplan layout with the machine learning model. In some embodiments, in response to the predicted SIR results generated by the machine learning model, bump assignment optimization is performed to find the best bump assignment that reduces the SIR results. In some embodiments, both SIR prediction and bump assignment optimization are performed automatically, and semiconductor devices are manufactured based on updated floorplan data with low SIR risk.
[0106] Figure 1 is a block diagram of an illustrated electronic device design system 10 according to one or more embodiments of the present disclosure. The electronic device design system 10 is operable to generate a floor plan layout and optimize the floor plan layout after floor plan synthesis has been performed on an electrical device (which may be referred to herein as floor plan design) before placing the floor plan layout. The electronic device design system 10 includes an electronic design platform 20 and a static voltage drop (SIR) prediction platform 40. In some embodiments, the electronic design platform 20 and / or the SIR prediction platform 40 may be implemented in hardware, firmware, software, or any combination thereof. For example, in some embodiments, the electronic design platform 20 and / or the SIR prediction platform 40 may be at least partially implemented as instructions stored on a computer-readable storage medium, which may be provided by the computer-readable storage medium illustrated in FIG. 1 and discussed in the following paragraphs. Figure 6 One or more computer processors or processing circuits in an electronic design automation (EDA) system 600 read and execute.
[0107] In some embodiments, the electronic design platform 20 and the SIR prediction platform 40 may be included in or otherwise implemented by the same device, such as the same computing system or device. In other embodiments, the electronic design platform 20 and the SIR prediction platform 40 may be included in or otherwise implemented by separate devices, such as separate and remotely located computing systems or devices.
[0108] The electronic design platform 20 includes electronic device design tools that can be used, for example, to design high-level programming descriptions of analog and / or digital circuits for electronic devices. In some embodiments, the high-level programming description content can be implemented using: a high-level programming language such as C, C++, LabVIEW, MATLAB; a general-purpose system design or modeling language such as SysML, SMDL and / or SSDL; or any other suitable high-level programming language. In some embodiments, the electronic design platform 20 may include various additional features and functionality, including, for example, one or more tools suitable for simulating, analyzing and / or verifying high-level programming descriptions of circuits for electronic devices.
[0109] In some embodiments, the electronic design platform 20 includes a floorplanning (e.g., also referred to as synthesis) tool 21, a data extraction tool 22, a placement tool 24, a clock tree synthesis (CTS) tool 26, and a routing tool 28, each of which may be at least partially implemented as a software tool accessible to and executable by one or more computing devices, processors, or the like.
[0110] The floorplan tool 21 places the blocks or macros in the chip area or core area of the electronic device design according to the following about the floorplan layout: predetermined floorplan specifications (e.g., size and orientation of the blocks or macros), predetermined power consumption specifications (e.g., the sum of the estimated power of the cells in the block or macro), or / and predetermined bump plane specifications (e.g., bump assignments, including domains, center orientations, types of bumps of contact structures, which are used to transmit power supply voltages to the cells). In some embodiments, the floorplan tool 21 also translates one or more characteristics, parameters of characteristics of the floorplan layout, or attributes of the electronic device into one or more logical operations, one or more arithmetic operations, one or more control operations, or the like, which can then be translated into high-level programming description content based on analog circuits and / or digital circuits.
[0111] The data extraction tool 22 is used to extract floorplan data characterized by parameters of the floorplan layout generated by the floorplan tool 21. For example, the grouped features include properties associated with cells (e.g., standard logic cells), such as relative orientations of cells in a block or macro (e.g., center, border, corner), power-related features (e.g., total power consumption of cells located in a region with a certain radius about the center where the SIR violation occurs), power grid-related features (e.g., density of metal tracks), physical properties of cells to bumps (e.g., bump pitch), manufacturing process parameters (e.g., N5, N7, ... technology nodes), combinations thereof, and / or any other suitable parameters for estimating SIR violation information.
[0112] Placement tool 24 generates cells that correspond to or otherwise implement one or more logic operations, one or more arithmetic operations, one or more control operations, or the like generated by floorplanning tool 21. Cells may include geometric shapes that correspond to various features of semiconductor devices including, for example, diffusion layers, polysilicon layers, metal layers, and / or interconnects between layers.
[0113] In some embodiments, the geometry of some of the analog circuits and / or digital circuits may be defined based on standard cells from a predefined library of standard cells associated with a technology library. A standard cell represents one or more semiconductor devices and their interconnect structures that are configured and arranged to provide a logic function such as AND, OR, XOR, XNOR, or NOT; or a storage unit such as a flip-flop or latch. The predetermined library of standard cells may be defined based on geometries corresponding to diffusion layers, polysilicon layers, metal layers, and / or interconnects between layers. Thereafter, the placement tool 24 assigns locations to the geometries on a printed circuit board (PCB) and / or a semiconductor substrate.
[0114] The CTS tool 26 performs clock tree synthesis (CTS) on a design generated, for example, by the placement tool 24. Clock tree synthesis generally refers to the process of synthesizing a clock tree to achieve zero or minimum slope and insert delays, and includes inserting clock tree cells that correspond to clock operations of an electronic device or otherwise implement clock operations. The clock cells may include geometric shapes that correspond to circuits or logic devices that implement clock features of a semiconductor device, including, for example, buffers, inverters, or the like. In some embodiments, each of the clock tree cells includes one or more buffers or inverters electrically positioned along a clock path of the electronic device design. Additionally, in some embodiments, one or more of the clock cells may include a clock gating cell, such as an integrated clock gating cell (ICG). Clock gating is a common technique for reducing clock power by shutting off a clock to a module or circuit element by utilizing a clock enable signal, and such clock gating may be implemented using an integrated clock gating cell. The integrated clock gating cell may include one or more logic circuit components, such as an OR gate, an AND gate, or a latch.
[0115] For example, after clock tree synthesis has been performed on the layout by CTS tool 26, routing tool 28 generates physical connections between cells or geometries in the layout provided by placement tool 24. In some embodiments, routing tool 28 assigns interconnects between geometries using: a textual or image-based netlist describing analog circuits, digital circuits, technology libraries, semiconductor foundries used to manufacture electronic devices; and / or semiconductor technology nodes used to manufacture electronic devices.
[0116] The electronic design platform 20 may include a variety of additional tools, including, for example, a verification tool 29. The verification tool 29 performs various verifications or checks on the electronic device layout, such as after placement, CTS, and routing. For example, in some embodiments, the verification tool 29 may analyze the electronic device layout and may provide static timing analysis (STA), SIR analysis also known as IREM analysis, clock domain cross verification (CDC check), formal verification also known as model checking, equivalence checking, or any other suitable analysis and / or verification. In some embodiments, the verification tool 29 may perform alternating current (AC) analysis such as linear small signal frequency domain analysis and / or direct current (DC) analysis such as nonlinear static point calculation or a calculated series of nonlinear operating points while sweeping voltages, currents, and / or parameters to perform STA, IREM analysis, or the like.
[0117] The verification tool 29 verifies whether the electronic device design, including the layout of cells or geometries provided by the placement tool 24, the placement and functionality of clock tree cells provided by the CTS tool 26, and the interconnections between cells or geometries provided by the routing tool 28, meets one or more specifications, rules, or the like associated with the electronic device design. The verification tool 29 may perform physical verification, in which the verification tool 29 verifies whether the electronic device design is physically manufacturable and that the resulting wafer will meet the design specifications and will not have physical defects that prevent the wafer from functioning as designed.
[0118] The verification tool 29 may perform a static voltage drop (SIR) analysis on the electronic device design. The SIR analysis may be performed, for example, as part of an IREM analysis performed by the verification tool 29. During the SIR analysis, the verification tool 29 analyzes the electronic device design, which may be a physical semiconductor device, to determine whether a SIR drop exists that exceeds or otherwise violates a design rule or parameter specified for the electronic device design. SIR generally refers to a reduction in the VDD voltage level caused by the resistance of metal conductors in a power distribution grid that distributes power or voltage to various components within the electronic device design. The verification tool 29 may utilize any conventional or other techniques to perform the SIR analysis, including, for example, applying electrical test vectors to the electronic device design and measuring or monitoring the voltage drop throughout the electronic device design. In some embodiments, the verification tool 29 may perform the SIR analysis via simulation, for example, using software tools that simulate the application of voltage to the electronic device design and that measure or monitor the resulting SIR drop throughout the electronic device design.
[0119] In some embodiments, the verification tool 29 may generate a SIR map that indicates the value of the static voltage drop throughout the electronic device design. In some embodiments, the verification tool 29 may generate a SIR map that indicates the location of SIR violations in the electronic device design (e.g., SIR values (results) that exceed a predetermined threshold value). Figures 2 to 5 As will be discussed in further detail, as the SIR map generated by the self-verification tool 29 can be derived, the known values (results) and locations of SIR violations for various different electronic device designs can be used as training data to train a machine learning model to predict or determine SIR results and / or SIR violations in a new floor plan before placement has been performed.
[0120] In some embodiments, after placement by placement tool 24, such as may be provided from floorplanning tool 21 and data extraction tool 22, SIR prediction platform 40 is used to predict or determine the presence of SIR violations in a particular electronic circuit floorplan based on features of the floorplan layout. As will be discussed in further detail herein, SIR prediction platform 40 may predict or determine the presence of SIR violations by implementing one or more machine-continuation methods, such as where training data (such as data indicating the presence and / or location of SIR violations in an electronic device design after routing has been performed) and corresponding SIR results are used to train a machine learning model to predict or determine the presence of SIR violations based on similarities or deviations between a new electronic circuit placement layout (e.g., before placement is performed) and the training data. In some embodiments, SIR prediction platform 40 may predict or determine a complete SIR map including SIR values (results) throughout the floorplan layout, and SIR prediction platform 40 may further predict or determine the location of the SIR violation based on the SIR map.
[0121] The SIR prediction platform 40 may include a plurality of electronic device analysis and / or design tools that may be implemented at least in part as software tools that, when executed by one or more computing devices, processors, or the like, may be used to analyze one or more electronic device layouts, including floorplan data for a floorplan layout of an electronic device or circuit, which floorplan data may be received, for example, from the electronic design platform 20 (e.g., from the data extraction tool 22). Additionally, in some embodiments, once the layout is routed, for example, by the routing tool 28, the SIR prediction platform 40 may be used to adjust or otherwise provide information to the electronic design platform 20 indicating one or more adjustments to be made to the floorplan layout in order to avoid or otherwise reduce the presence of a SIR violation in another placement and CTS layout.
[0122] In some embodiments, the SIR prediction platform 40 includes a SIR prediction tool 42 and a bump assignment adjustment tool 44, each of which may be implemented at least in part as a software tool that is accessible to and executable by one or more computing devices, processors, or the like. In some embodiments, the SIR prediction tool 42 and / or the bump assignment adjustment tool 44 may be implemented as a circuit that is operable to perform any of the functions described herein with respect to the SIR prediction tool 42 and / or the bump assignment adjustment tool 44. In some embodiments, the electronic design platform 20 and the SIR prediction platform 40 may be integrated and may be implemented in the same platform. For example, each of the various tools described herein with respect to the electronic design platform 20 and the SIR prediction platform 40 may be accessed or otherwise implemented at least in part by the same device such as a computer device.
[0123] In some embodiments, the SIR prediction platform 40 receives the floorplan layout and floorplan data from the electronic design platform 20 after features of the floorplan are determined by the floorplan tool 21 and the corresponding floorplan data is extracted by the data extraction tool but before placement of cells in the placement layout is performed, for example, by the placement tool 24. The SIR prediction platform 40 may implement a machine learning method to predict or determine SIR results in the floorplan layout, and update the floorplan data of the floorplan layout by providing an indication of one or more recommended adjustments to the bump assignments in order to optimize and improve the overall SIR behavior of the floorplan and even further the placement and CTS layout.
[0124] For example, as will be described in further detail below, in some embodiments, the SIR prediction tool 42 may include Figure 2 The machine learning circuit 110 may be trained to predict or determine the presence of a SIR violation based on the training data. The bump assignment adjustment tool 44 may also receive floorplan data (the aforementioned features of the floorplan layout associated with the predicted or determined SIR result from the SIR prediction tool 42) and the corresponding initial bump assignments. The bump assignment adjustment tool 44 may adjust or recommend adjustments to the bump assignments of the floorplan layout in order to optimize or improve the floorplan based on the predicted or determined SIR result.
[0125] Figure 2 FIG. 1 is a block diagram of a SIR prediction system 100 according to an embodiment of the present disclosure. The SIR prediction system 100 may be combined with the Figure 1 4. The SIR prediction system 100 may be used by the SIR prediction platform 40 in the electronic design platform 20, or may include one or more of the features and functionalities of the SIR prediction platform 40. In some embodiments, the SIR prediction system 100 is included in the SIR prediction platform 40 or implemented by the SIR prediction platform 40. For example, in some embodiments, the SIR prediction system 100 may be implemented or otherwise accessed by the SIR prediction tool 42 and the bump assignment adjustment tool 44 to predict or determine SIR results in a floorplan layout received, for example, from the floorplan tool 21 and data extraction tool of the electronic design platform 20 and adjust the bump assignments of the floorplan layout.
[0126] like Figure 2As shown in FIG. 1 , the SIR prediction system 100 includes a SIR prediction circuit 120 and a bump assignment adjustment circuit 130. The SIR prediction circuit 120 and the bump assignment adjustment circuit 130 may include or otherwise be executed by a computer processor, which is used to perform various functions and operations described herein. For example, the SIR prediction circuit 120 and the bump assignment adjustment circuit 130 may be executed by a computer processor, which is selectively activated or reconfigured by a stored computer program; or may be a specifically constructed computing platform for implementing the features and operations described herein.
[0127] In some embodiments, the SIR prediction circuit 120 includes a memory storing instructions for performing one or more of the features or operations described herein, and the SIR prediction circuit 120 may be operable to execute instructions stored, for example, in the memory to perform the functions of the SIR prediction circuit 120 described herein.
[0128] The SIR prediction circuit 120 may be communicatively coupled to the floor plan database 102 and the SIR database 104. The SIR prediction circuit 120 may access the floor plan layout and corresponding floor plan data from the floor plan database 102. The term floor plan layout is used herein to mean a layout after the floor plan placement has been performed, for example, the floor plan layout includes blocks and macros generated by the floor plan tool 21. The floor plan layout and floor plan data stored in the floor plan database 102 may be provided, for example, from the floor plan tool 21 and the data extraction tool 22 of the electronic design platform 20. The floor plan database 102 may be stored in one or more computer readable memories. In some embodiments, the floor plan database 102 stores training data and test data to train a machine learning model in the machine learning circuit 110, and the SIR database 104 stores SIR results (values) corresponding to the training data and test SIR results (values) corresponding to the test data used to train the machine learning model.
[0129] The SIR prediction circuit 120 receives and analyzes the floor plan data retrieved from the floor plan database 102 to generate SIR results and similarity results based on the comparison of the floor plan data with the training data by the machine learning model trained by the training data. In some embodiments, the SIR prediction circuit 120 can separately detect each of the multiple zones of the floor plan layout. The detected zones of the floor plan layout can have any size and / or shape. For example, the floor plan layout can be divided into zones based on a power grid, and each unit or unit of the power grid can have a size corresponding to the size of each zone of the floor plan layout.
[0130] In some embodiments, the SIR prediction circuit 120 may generate a SIR violation map based on the SIR results using one or more artificial intelligence or machine learning techniques to indicate the location of SIR violations in the floorplan layout. Figure 3A , there is a bump 302 and a SIR violation 304 in the floorplan 30 indicating a SIR violation map of the floorplan layout. In some embodiments, it may be implemented at least in part by the machine learning circuit 110. Some or all of the determinations described herein made by the SIR prediction circuit 120 may be performed automatically by the SIR prediction circuit 120, for example, in response to receiving the floorplan layout from the floorplan database 102 and the SIR database 104.
[0131] "Artificial intelligence" is used broadly herein to describe any computationally intelligent systems and methods that can learn knowledge (e.g., based on training data), and use such learned instructions to adapt methods to solve one or more problems, for example by making inferences based on received inputs, such as floorplan layouts. Machine learning generally refers to a subfield or category of artificial intelligence, and is used broadly herein to describe any algorithm, mathematical model, statistical model, or the like, implemented in one or more computer systems or circuits, such as processing circuits, and which builds one or more models based on sample data (or training data) in order to make predictions or decisions.
[0132] The SIR prediction circuit 120 and / or the machine learning circuit 110 may use, for example, neural networks, deep learning, convolutional neural networks, Bayesian program learning, support vector machines, and pattern recognition techniques to solve problems such as predicting or determining the presence and location of SIR violations in a placement layout. In addition, the SIR prediction circuit 120 and / or the machine learning circuit 110 may implement any one or combination of the following computational algorithms and / or techniques: classification, regression, supervised learning, unsupervised learning, feature learning, clustering, decision trees, or the like.
[0133] As an example, an artificial neural network may be utilized by the SIR prediction circuit 120 and / or the machine learning circuit 110 to develop, train and / or update one or more machine learning models that may be used to generate SIR results to predict or determine the presence and location of SIR violations in a floorplan layout. An example artificial neural network may include a plurality of interconnected "neurons" that exchange information between each other. The connections have numerical weights that may be tuned based on experience, and thus the neural network adapts to the input and is capable of learning. The "neurons" may be included in a plurality of separate layers, such as an input layer, a hidden layer, and an output layer, that are connected to each other. The neural network may be trained by providing training data (e.g., past data indicating the presence and location of SIR violations in an electronic device design after routing has been performed) to the input layer. Through training, the neural network may generate and / or modify hidden layers representing weighted connections that map the training data provided at the input layer to known output information at the output layer (e.g., classifying the input electronic device design as including SIR violations and their locations after routing has been performed). The relationships among neurons in the input layer, hidden layer, and output layer formed through the training process and which may include weighted connection relationships may be stored in the machine learning circuit 110 as one or more machine learning models or otherwise accessible to the machine learning circuit 110.
[0134] Once the neural network has been sufficiently trained, the neural network may be provided with test data at an input layer. Using the SIR violation knowledge (e.g., as stored in a machine learning model, and which may include, for example, weighted connection information between neurons of the neural network), the neural network may determine the received planar data of the floorplan layout at an output layer. For example, the neural network may generate SIR results (including locations and / or SIR values) to predict or determine the presence and location of SIR violations in the floorplan layout.
[0135] Using one or more computationally intelligent and / or machine learning techniques, the SIR prediction circuit 120 may learn (e.g., by developing and / or updating a machine learning algorithm or model based on training data) SIR results, and in some embodiments, the SIR prediction circuit 120 may make some predictions or decisions based at least in part on knowledge, inferences, or the like developed or otherwise learned via training the machine learning circuit 110.
[0136] The machine learning circuitry 110 may be implemented in one or more processors having access to instructions, which may be stored in any computer-readable storage medium, that may be executed by the machine learning circuitry 110 to perform any of the operations or functions described herein.
[0137] In some embodiments, machine learning circuit 110 is communicatively coupled to floorplan database 102 and SIR database 104, and is trained by training data, test data, SIR results (values) corresponding to the training data, and SIR results (values) corresponding to the test data stored in floorplan database 102 and SIR database 104. In some embodiments, the aforementioned data and values correspond to an actual electronic device design, for example, after wiring has been performed. Figure 2 As shown in FIG. 1 , in some embodiments, the predicted SIR results are stored in a SIR result database 112 that is communicatively coupled to the SIR prediction circuit 120 .
[0138] According to some embodiments, other training data may include manually entered inputs, such as one or more variables or adjustable parameters, coefficient values, labels, classifiers, or the like, to adjust or otherwise manage a machine learning model developed in the machine learning circuit 110 and / or stored in the floorplan database 102 via a training procedure. The training may be based on a wide variety of learning algorithms or models, including, for example, support vector machines, linear regression, logistic regression, naive Bayes, linear discriminant analysis, decision trees, k-nearest neighbors, neural networks, or the like. Examples of training the machine learning circuit 110 based on a neural network are provided previously herein.
[0139] In addition to training the machine learning circuit 110 to predict the SIR results corresponding to the floor plan data, the SIR prediction circuit 120 further classifies the training data of the adversarial model to generate a similarity value based on the classified training data and test data, and generates a compensation value based on the comparison between the predicted SIR value generated by the machine learning circuit 110 and the test SIR value corresponding to the test data. For example, since the known difference in the bump spacing between the training data and the test data dominates the determination of the similarity value, other features in the training data are classified from the training data, and the value of the bump spacing in the training data is sent to the adversarial model. The compensation value is generated by calculating the difference between the predicted SIR value and the actual test SIR value. In some embodiments, the similarity value is inversely proportional to the compensation value. For example, when the similarity value A is relatively high, indicating that the difference between the training data and the test data is small, the corresponding compensation value is relatively small, because the prediction of the machine learning model provides a relatively high accuracy.
[0140] In some embodiments, the SIR prediction circuit 120 is communicatively coupled to a mapping file database 114 , which is configured to store compensation values and similarity values in a mapping table.
[0141] After the machine learning model in the machine learning circuit 110 is trained, the training machine learning circuit 110 can generate a SIR result and a first similarity result based on the received floor plan. The SIR prediction circuit 120 further searches the mapping table for a first compensation value corresponding to the first similarity result, and further adjusts the SIR result by adding the first compensation value to the SIR result.
[0142] When the SIR result is greater than the SIR threshold, before placing the cell in the floorplan layout, the SIR prediction circuit 120 further compares the SIR result with the SIR threshold and predicts an optimized SIR result based on the SIR result and the bump assignments stored in the bump assignment database 106. FIG. 3A to FIG. 3B As shown in the embodiment of Figure 3A The maximum SIR value of the SIR results in (eg, 3.56%) is predicted to be greater than a threshold value (eg, 3.5%).
[0143] Therefore, the bump assignment optimization is applied in response to the SIR prediction circuit 120 predicting the optimized SIR result. Figure 1 , the bump assignment adjustment tool 44 can adjust the analyzed floor plan layout based on the predicted SIR results. In some embodiments, the bump assignment adjustment tool 44 is Figure 4 The bump assignment adjustment circuit 130 is implemented by the bump assignment adjustment circuit 130. Specifically, the bump assignment adjustment circuit 130 receives the floor plan data and the predicted SIR result. In addition, the bump assignment adjustment circuit 130 adjusts parameters (e.g., bump spacing between bumps in the bump assignment, the orientation of the bumps) to generate a number of bump assignments in the bump assignment database 106.
[0144] The SIR prediction circuit 120 retrieves bump assignments from the bump assignment database 106, and the machine learning circuit 110 further generates bump SIR results each corresponding to one of the floorplan data and the bump assignments. The SIR prediction circuit 120 determines the lowest value of the bump SIR results as the optimized SIR result, and further updates the bump floorplan data by the bump assignment corresponding to the lowest SIR result. For example, comparing FIG. 3A to FIG. 3B In the embodiment, the maximum SIR value of the SIR result is reduced by moving the bump 302 toward the positive y direction. Figure 3B The mid-bump assignment mid-bump pitch is reduced from 3.56% to about 3.16%, for example. In addition, the number of SIR violations is reduced by about 65.32%.
[0145] Figure 4 FIG. 4 is a flow chart illustrating a method 400 of SIR prediction according to one or more embodiments. Figure 4, the method includes two branches, a model training branch 410 and a SIR prediction branch 420. In the model training branch 410, training data and corresponding SIR results (e.g., SIR values) in known electronic device designs may be used to train a machine learning model, such as a machine learning model implemented by the machine learning circuit 110 and which may be at least partially stored in, for example, the floorplan database 102 and the SIR database 104. In the SIR prediction branch 420, new floorplan data for the floorplan (e.g., as generated by the floorplan tool 21 and the data extraction tool 22) is analyzed (e.g., by the SIR prediction circuit 120) to generate predicted SIR results in the floorplan layout before placement is performed on the floorplan layout. Additionally, bump optimization will be initiated in response to the SIR results generated in the SIR prediction branch 420.
[0146] In operation 401 in model training branch 410 , prior to a placement phase utilizing placement tool 24 , training data is extracted from a training floorplan layout stored, for example, in floorplan database 102 , and test data is extracted from a test floorplan layout.
[0147] The combination 402 of training data and corresponding training SIR results and the combination 403 of test data and corresponding test SIR results are used for operation 404 of training a machine learning model in the machine learning circuit 110 .
[0148] In operation 405, the features of the SIR prediction model are classified by importance. Alternatively, the training data is classified to obtain selected key features as input for building an adversarial detection model.
[0149] At operation 406, similarity values are generated by the adversarial detection model implemented in the machine learning circuit 110 and trained based on the classified training data and test data. Specifically, in some embodiments, binary classification is performed to calculate similarity values between the training data and the test data. As the difference between the training data and the test data becomes smaller, the corresponding similarity value becomes larger.
[0150] At operation 407, a compensation value is generated based on a difference between a predicted SIR value generated by the machine learning model and a test SIR value corresponding to the test data.
[0151] At operation 408, a mapping table 409, which may be stored in the mapping file database 114, is generated to include compensation values and similarity values. In some embodiments, an exemplary mapping table is given as follows:
[0152]
[0153] Table I: Mapping table of similarity values and compensation values
[0154] In the SIR prediction branch 420, detected floorplan data 421 extracted from the floorplan is received by the machine learning model.
[0155] At operation 422 , a predicted SIR result 423 is generated by the machine learning model based on the floor plan data 421 .
[0156] At operation 424, the training data is compared to the floorplan data by the SIR prediction circuit 120, generating a similarity value.
[0157] At operation 425, the SIR prediction circuit 120 determines whether the similarity value generated in operation 424 is within the boundaries of the mapping table 409. For example, when the similarity value is outside the range (e.g., <0.5349 or >0.5962 as shown in Table I for reference), operation 426 is performed to retrain the machine learning model with the floor plan data. Alternatively, when the similarity value is excluded from the mapping table 409, the machine learning model is retrained. In contrast, when the similarity value is within the range of the boundaries of the mapping table, operation 427 is performed. In some embodiments, the highest and lowest values of the similarity value are referred to as threshold values of the boundaries of the mapping table 409.
[0158] At operation 427, the SIR prediction circuit 120 searches the corresponding compensation value 428 in the mapping table 409 according to the similarity value. For example, the similarity value between the test data and the training data is 0.559 and is within the range of 0.5581 to 0.561. The corresponding compensation value in the mapping table I is 0.0743.
[0159] At operation 429, a SIR result is generated by compensating the predicted SIR result 423 with a compensation value 428. The SIR result is equal to the sum of the predicted SIR result and the compensation value. In some embodiments, one or more SIR violations in the same floorplan layout are compensated by the same compensation value.
[0160] At operation 430, the SIR prediction circuit 120 further determines whether the SIR result generated in operation 429 satisfies the design specification. When the SIR result satisfies the design specification (e.g., including the upper limit of the allowable voltage drop of the cell), the SIR result indicates that the floorplan layout with the floorplan data is verified for other operations in the electronic design platform 20. When the SIR result does not satisfy the design specification, the method 400 further includes a bump optimization branch 450.
[0161] Figure 5 A diagram according to one or more embodiments corresponds to Figure 44 is a flow chart of the method 400 of FIG. 4. In the bump optimization branch 450, at operation 431, the bump assignment adjustment circuit 130 determines whether the SIR result satisfies the predefined SIR optimization threshold set in operation 434. When the SIR result does not meet the predefined SIR optimization threshold, operation 432 is performed to find the best bump assignment. In some embodiments, the bump optimization branch 450 of the method 400 is called a heuristic method to roughly find the global optimal combination of bump assignments based on the SIR results.
[0162] Additionally, operations 433 to 435 are performed. At operation 433, bump assignments are prepared. In some embodiments, the amount of bump assignments determines the extent of bump assignment optimization. Alternatively, the more bump assignments there are, the more options there are for selecting the best bump assignments for floorplan layout.
[0163] At operation 434 , the SIR optimization threshold is set as one of the stop conditions for bump optimization.
[0164] At operation 435 , one of the bump assignments corresponding to the SIR result obtained at operation 429 is predicted to be an initial bump assignment having a poor bump SIR result, and the SIR result OLD is used for the following operations.
[0165] After operation 435, operation 436 is performed to generate a new bump assignment based on one of the bump assignments in operation 433. The bump assignment specifies fixed bumps and moving bumps in a region with respect to the initial bump assignment.
[0166] At operation 437, the machine learning model generates a bump SIR result, which is referred to as the SIR result corresponding to the bump assignment from operation 436 and the floorplan data. NEW .
[0167] At operation 438, the SIR results are compared. NEW With SIR results OLD To obtain the updated bump assignment. When the SIR result NEW Less than SIR result OLD When , operation 439 is performed to update the best bump assignment by the bump assignment in operation 436 and replace the best bump SIR result by the bump SIR result in operation 437 .
[0168] In comparison, when the SIR result NEW Not less than SIR result OLD When , operation 440 is performed to update the best bump assignment and replace the best bump SIR result based on the Metropolis rule defined as follows:
[0169]
[0170] where n corresponds to a random value between 0 and 1, T corresponds to the current temperature, and k is the Boltzmann constant.
[0171] When the Metropolis rule is satisfied, the best bump assignment is updated by the bump assignment in operation 436 and the best bump SIR result is replaced by the SIR result in operation 437 .
[0172] Operations 436 to 440 are repeated until the maximum iteration of the optimization reaches a predetermined number at operation 441, and operation 442 is performed. At operation 442, it is determined whether the current temperature T is equal to the stop temperature T_end. When the current temperature T is equal to the stop temperature T_end, operation 443 is performed to output the updated optimal bump assignment by the bump assignment adjustment circuit 130 to further update the floorplan data stored in the floorplan database 102 at operation 445.
[0173] In contrast, when the current temperature T is indeed equal to the stop temperature T_end, operation 444 is performed to multiply the current temperature T by the number Δ and further suspend the number of iterations of performing operation 436 .
[0174] In some embodiments, the method 400 further includes an operation of comparing the updated best bump SIR result with the SIR optimization threshold value set in operation 434 to output an updated best bump assignment of the bump assignment data. For example, when the updated best bump SIR result is less than or equal to the SIR optimization threshold value, the bump assignment adjustment circuit 130 outputs the updated best bump assignment to update the floorplan data.
[0175] See together Figures 4 to 5 At operation 445, the floorplan data 421 is updated and operations 422 to 430 are performed to generate and verify that the SIR result is the sum of the new predicted SIR result and the compensation value associated with the updated floorplan data and the training data.
[0176] The method 400 further includes operations for fabricating a semiconductor device including an electronic device based on the updated floorplan layout design.
[0177] Figures 1 to 5 The configurations are given for illustrative purposes. Various implementations are within the contemplated scope of an embodiment of the present disclosure. For example, in some embodiments, the data extraction tool 22 and the SIR prediction platform 40 are used to collaborate to predict SIR results after one or more processes performed by the placement tool 24, the CTS tool 26, and the routing tool 28.
[0178] View now Figure 6 . Figure 6 FIG. 6 is a block diagram of an electronic design automation (EDA) system 600 for designing an integrated circuit layout according to some embodiments of the present disclosure. The EDA system 600 is used to implement Figures 4 to 5 One or more operations of the method 400 disclosed in Figures 1 to 3B Explanation: In some embodiments, EDA system 600 includes an APR system.
[0179] In some embodiments, EDA system 600 is a general purpose computing device that includes a processor 602 and a non-transitory storage medium 604. Storage medium 604 is encoded with, i.e., stores, among other things, computer program code (instructions) 606, i.e., a set of executable instructions. Execution of computer program code 606 by processor 602 represents (at least in part) an EDA tool that implements, for example, a portion or all of method 400.
[0180] The processor 602 is electrically coupled to the storage medium 604 via a bus 608. The processor 602 is also electrically coupled to an I / O interface 610 and a manufacturing tool 616 by the bus 608. A network interface 612 is also electrically coupled to the processor 602 via the bus 608. The network interface 612 is connected to a network 614, so that the processor 602 and the (computer-readable) storage medium 604 can be connected to external components via the network 614. The processor 602 is used to execute computer program code 606 encoded in the computer-readable storage medium 604 so that the EDA system 600 can be used to perform part or all of the processes and / or methods mentioned. In one or more embodiments, the processor 602 is a central processing unit (CPU), a multiprocessor, a distributed processing system, an application specific integrated circuit (ASIC), and / or a suitable processing unit.
[0181] In one or more embodiments, the storage medium 604 is an electronic, magnetic, optical, electromagnetic, infrared, and / or semiconductor system (or device or apparatus). For example, the storage medium 604 includes semiconductor or solid-state memory, magnetic tape, removable computer disk, random access memory (RAM), read-only memory (ROM), rigid disk, and / or optical disk. In one or more embodiments using optical disks, the storage medium 604 includes compact disk-read only memory (CD-ROM), compact disk-read / write (CD-R / W), and / or digital video disc (DVD).
[0182] In one or more embodiments, the storage medium 604 stores computer program code 606 for enabling the EDA system 600 (where such execution represents (at least a portion of) an EDA tool) to perform some or all of the procedures and / or methods described. In one or more embodiments, the storage medium 604 also stores information that facilitates the performance of some or all of the procedures and / or methods described. In one or more embodiments, the storage medium 604 stores IC layout diagrams 620 of standard cells, including such standard cells as disclosed herein, for example, including in connection with FIG. 3A to FIG. 3B The floor plan is arranged with 30 units.
[0183] The EDA system 600 includes an I / O interface 610. The I / O interface 610 is coupled to external circuits. In one or more embodiments, the I / O interface 610 includes a keyboard, a keypad, a mouse, a trackball, a trackpad, a touch screen, and / or arrow keys for conveying information and commands to the processor 602.
[0184] The EDA system 600 also includes a network interface 612 coupled to the processor 602. The network interface 612 allows the EDA system 600 to communicate with a network 614 to which one or more other computer systems are connected. The network interface 612 includes a wireless network interface such as Bluetooth, WIFI, WIMAX, GPRS, or WCDMA; or a wired network interface such as Ethernet, USB, or IEEE-1064. In one or more embodiments, part or all of the procedures and / or methods described are implemented in two or more EDA systems 600.
[0185] The EDA system 600 also includes a manufacturing tool 616 coupled to the processor 602. Based on the design file, for example, the processing by the processor 602 FIG. 3A to FIG. 3BThe planar arrangement 30, the manufacturing tool 616 is used to manufacture integrated circuits.
[0186] The EDA system 600 is used to receive information via the I / O interface 610. The information received via the I / O interface 610 includes one or more of the following: instructions, data, design rules, standard cell libraries, and / or other parameters for processing by the processor 602. The information is transmitted to the processor 602 via the bus 608. The EDA system 600 is used to receive information related to the UI via the I / O interface 610. The information is stored in the computer storage medium 604 as the design specification 622.
[0187] In some embodiments, a part or all of the mentioned programs and / or methods are implemented as independent software applications for processor execution. In some embodiments, a part or all of the mentioned programs and / or methods are implemented as a software application that is a part of an additional software application. In some embodiments, a part or all of the mentioned programs and / or methods are implemented as a plug-in to a software application. In some embodiments, at least one of the mentioned programs and / or methods is implemented as a software application that is a part of an EDA tool. In some embodiments, a part or all of the mentioned programs and / or methods are implemented as a software application used by an EDA system 600. In some embodiments, a layout diagram including standard cells is generated using a suitable layout generation tool.
[0188] In some embodiments, the program is implemented as a function of a program stored in a non-transitory computer-readable recording medium. Examples of non-transitory computer-readable recording media include, but are not limited to, external / removable and / or internal / embedded storage or memory units, such as one or more of the following: optical disks, such as DVDs; magnetic disks, such as hard disks; semiconductor memories, such as ROM, RAM, memory cards, and the like.
[0189] Figure 7 FIG. 7 is a block diagram of an IC manufacturing system 700 and an IC manufacturing process associated with the IC manufacturing system according to some embodiments. In some embodiments, based on the layout diagram, at least one of the following is manufactured using the IC manufacturing system 700: (A) one or more semiconductor masks, or (B) at least one component in a layer of a semiconductor integrated circuit.
[0190] exist Figure 7In the present invention, the IC manufacturing system 700 includes entities such as a design room 720, a mask room 730, and an IC manufacturer / fabricator ("wafer factory") 750, which interact with each other in the design, development and manufacturing cycle and / or services related to manufacturing IC devices 760. The entities in the IC manufacturing system 700 are connected by a communication network. In some embodiments, the communication network is a single network. In some embodiments, the communication network is a variety of different networks, such as Ethernet and the Internet. The communication network includes wired and / or wireless communication channels. Each entity interacts with one or more of the other entities and provides services to one or more of the other entities and / or receives services from one or more of the other entities. In some embodiments, two or more of the design room 720, the mask room 730, and the IC wafer factory 750 are owned by a single entity. In some embodiments, two or more of the design room 720, the mask room 730, and the IC wafer factory 750 coexist in a common facility and use common resources.
[0191] The design office (or design team) 720 generates an IC design layout 722. The IC design layout 722 includes various geometric patterns, such as an IC layout design for an IC device 760, for example, FIG. 3A to FIG. 3B The planar arrangement 30 discussed. The geometric pattern corresponds to the pattern of the metal, oxide or semiconductor layers that constitute the various elements of the IC device 760 to be manufactured. The various layers are combined to form various IC features. For example, a portion of the IC design layout 722 includes various IC features, such as active regions, gate electrodes, source and drain electrodes, conductive segments or through holes of interlayer interconnects to be formed in a semiconductor substrate (such as a silicon wafer) and disposed in various material layers on the semiconductor substrate. The design room 720 implements an appropriate design program to form the IC design layout 722. The design program includes one or more of a logical design, a physical design, or a placement and routing. The IC design layout 722 is presented in one or more data files with information of the geometric pattern. For example, the IC design layout 722 can be expressed in a GDSII file format or a DFII file format.
[0192] The mask room 730 includes data preparation 732 and mask manufacturing 744. The mask room 730 uses the IC design layout drawing 722 to manufacture one or more masks 745 for manufacturing various layers of the IC device 760 according to the IC design layout drawing 722. The mask room 730 performs mask data preparation 732, wherein the IC design layout drawing 722 is translated into a representative data file ("representative data file, RDF"). The mask data preparation 732 provides RDF to the mask manufacturing 744. The mask manufacturing 744 includes a mask writer. The mask writer converts the RDF into an image on a substrate, such as a mask (master mask) 745 or a semiconductor wafer 753. The IC design layout drawing 722 is manipulated by the (mask) data preparation 732 to conform to the specific characteristics of the mask writer and / or the requirements of the IC wafer factory 750. In Figure 7 , data preparation 732 and mask manufacturing 744 are illustrated as separate elements. In some embodiments, data preparation 732 and mask manufacturing 744 may be collectively referred to as mask data preparation.
[0193] In some embodiments, data preparation 732 includes optical proximity correction (OPC), which uses lithography enhancement techniques to compensate for image errors, such as those that may result from self-diffraction, interference, other process effects, and the like. OPC adjusts IC design layout 722. In some embodiments, data preparation 732 includes other resolution enhancement techniques (RET), such as off-axis illumination, sub-resolution assist features, phase shift masks, other suitable techniques, and the like, or combinations thereof. In some embodiments, inverse lithography technology (ILT) is also used, which treats OPC as an inverse imaging problem.
[0194] In some embodiments, data preparation 732 includes a mask rule checker (MRC) that checks an IC design layout 722 that has been through a process in OPC using a set of mask generation rules that contain certain geometric and / or connectivity constraints to ensure adequate margins, account for variability in semiconductor manufacturing processes, and the like. In some embodiments, the MRC modifies the IC design layout 722 to compensate for the limitations during mask generation 744, which may undo portions of the modifications performed by OPC in order to satisfy the mask generation rules.
[0195] In some embodiments, data preparation 732 includes lithography process checking (LPC), which simulates the process to be performed by IC wafer factory 750 to manufacture IC device 760. LPC simulates this process based on IC design layout 722 to produce simulated manufactured devices, such as IC device 760. Process parameters in LPC simulation may include parameters associated with various processes of IC manufacturing cycle, parameters associated with tools used to manufacture IC, and / or other aspects of manufacturing process. LPC considers various factors, such as virtual image contrast, depth of focus ("depth of focus, DOF"), mask error enhancement factor ("mask error enhancement factor, MEEF"), other suitable factors and the like or combinations thereof. In some embodiments, after the simulated manufactured device has been generated by LPC, if the simulated device shape is not close enough to meet the design rules, OPC and / or MRC are repeated to further refine the IC design layout 722.
[0196] It should be understood that the above description of data preparation 732 has been simplified for the purpose of clarity. In some embodiments, data preparation 732 includes additional features, such as logic operations (LOP) to modify IC design layout 722 according to manufacturing rules. In addition, the procedures applied to IC design layout 722 during data preparation 732 can be performed in a variety of different orders.
[0197] After data preparation 732 and during mask fabrication 744, a mask 745 or a group of masks 745 are fabricated based on the modified IC design layout 722. In some embodiments, mask fabrication 744 includes performing one or more lithography exposures based on the IC design layout 722. In some embodiments, an electron-beam (e-beam) or multiple electron-beam mechanisms are used to form a pattern on a mask (mask or master mask) 745 based on the modified IC design layout 722. Mask 745 can be formed by various techniques. In some embodiments, mask 745 is formed using a binary technique. In some embodiments, the mask pattern includes opaque areas and transparent areas. A radiation beam such as an ultraviolet (UV) beam used to expose an image-sensitive material layer (e.g., a photoresist layer) coated on a wafer is blocked by the opaque areas and transmitted through the transparent areas. In one example, a binary mask version of the mask 745 includes a transparent substrate (e.g., fused silica) of the binary mask and an opaque material (e.g., chromium) coated in the opaque region. In another example, the mask 745 is formed using a phase transfer technique. In a phase shift mask (PSM) version of the mask 745, various features in the pattern formed on the phase shift mask are used to have an appropriate phase difference to enhance resolution and imaging quality. In various examples, the phase shift mask can be an attenuated PSM or an alternating PSM. The masks produced by the mask manufacturing 744 are used in a variety of processes. For example, such masks are used in ion implantation processes to form various doped regions in the semiconductor wafer 753, in etching processes to form various etched regions in the semiconductor wafer 753, and / or in other suitable processes.
[0198] IC fab 750 includes wafer fabrication 752. IC fab 750 is an IC manufacturing business that includes one or more manufacturing facilities for manufacturing a variety of different IC products. In some embodiments, IC fab 750 is a semiconductor foundry. For example, there may be a manufacturing facility for the front-end manufacturing of a variety of IC products (front-end-of-line, FEOL manufacturing), while a second manufacturing facility may provide back-end manufacturing of interconnects and packaging for IC products (back-end-of-line, BEOL manufacturing), and a third manufacturing facility may provide other services for the foundry business.
[0199] IC fab 750 uses mask 745 manufactured by mask chamber 730 to manufacture IC device 760. Therefore, IC fab 750 at least indirectly uses IC design layout 722 to manufacture IC device 760. In some embodiments, semiconductor wafer 753 is manufactured by IC fab 750 using mask 745 to form IC device 760. In some embodiments, IC manufacturing includes performing one or more lithography exposures based at least indirectly on IC design layout 722. Semiconductor wafer 753 includes a silicon substrate, or other appropriate substrate with material layers formed thereon. Semiconductor wafer 753 further includes one or more of various doped regions, dielectric features, multi-level interconnects, and the like (formed in subsequent manufacturing steps).
[0200] As described above, in one embodiment of the present disclosure, a method and system for predicting SIR before the placement stage is provided, which significantly shortens the SIR query time compared to the verification stage in the APR system. With data shift compensation, the SIR prediction results are more accurate with respect to the floorplan data for the prediction. The automated bump optimization for searching for the best bump position avoids more optimization iterations caused by the traditional "trial and error" (manual trial passes) method. It greatly speeds up the execution time of SIR prediction by about 45%, thereby improving the accuracy of the SIR results.
[0201] In some embodiments, a method is provided, the method comprising the following operations: receiving floor plan data of a floor plan layout of a semiconductor device by a static voltage drop (SIR) prediction circuit; generating a first SIR result based on the floor plan data by a machine learning model; generating a first similarity value based on a comparison of the floor plan data with a plurality of training data; generating a second SIR result based on the first SIR result in a mapping table and a first compensation value corresponding to the first similarity value; and generating a bump assignment to update the floor plan data based on a comparison between the second SIR result and a plurality of predetermined SIR values.
[0202] In some embodiments, the second quiescent voltage drop result is equal to a sum of the first quiescent voltage drop result and the first compensation value.
[0203] In some embodiments, the method further includes the following steps: when the first similarity value is less than a threshold value, retraining the machine learning model through the floorplan data and a corresponding static voltage drop result.
[0204] In some embodiments, the method further includes the following steps: generating a bump assignment; generating a bump static voltage drop result corresponding to the bump assignment and plane layout data by a machine learning model; comparing the bump static voltage drop result with a previous bump static voltage drop result to obtain an updated bump assignment; repeating from the step of generating the bump assignment to the step of comparing the bump static voltage drop result until a maximum iteration of optimization reaches a predetermined number; and outputting the updated bump assignment of the bump assignment data.
[0205] In some embodiments, the method further includes the steps of: setting a static voltage drop optimization threshold; and comparing an updated bump static voltage drop result with the static voltage drop optimization threshold to output an updated bump assignment of the bump assignment data.
[0206] In some embodiments, the method further includes the following steps: when the bump static voltage drop result is less than the previous bump static voltage drop result, replacing the updated bump assignment with the bump assignment; and when the bump static voltage drop result is greater than the previous bump static voltage drop result and a difference between the bump static voltage drop result and the previous bump static voltage drop result is within a range, replacing the updated bump assignment with the bump assignment.
[0207] In some embodiments, the method further includes the following steps: generating a third static voltage drop result, the third static voltage drop result being a sum of a fourth static voltage drop result provided by the machine learning model based on the updated floor plan data and a second compensation value, the second compensation value corresponding to a second similarity value associated with a comparison of the updated floor plan data with the training data.
[0208] In some embodiments, the method further includes the step of manufacturing a semiconductor device based on the updated floorplan data.
[0209] In some embodiments, a system is provided, the system comprising: a static voltage drop (SIR) prediction circuit including a machine learning circuit, the machine learning circuit being trained based on a plurality of training data associated with a plurality of electrostatic device designs to predict a SIR result, and being used to: receive floor plan data of a floor plan layout; generate a SIR result and a first similarity result; adjust the SIR result based on the first similarity result; and predict an optimized SIR result based on the SIR result and a plurality of bump assignments before placing a plurality of cells in a floor plan layout.
[0210] In some embodiments, the static voltage drop prediction circuit is further configured to generate a static voltage drop violation map based on the static voltage drop result, wherein the static voltage drop violation map indicates locations of predicted static voltage drop violations in the floorplan layout.
[0211] In some embodiments, the static voltage drop prediction circuit is further configured to perform the following operations: classify the training data for an adversarial detection model to generate a plurality of similarity values based on the classified training data and a plurality of test data; and generate a plurality of compensation values based on a comparison between a plurality of predicted static voltage drop values generated by the machine learning circuit and a plurality of test static voltage drop values corresponding to the test data. The system further includes a mapping file database, which is configured to store the compensation values and the similarity values in a mapping table.
[0212] In some embodiments, the static voltage drop prediction circuit is further configured to search a mapping table for a first compensation value corresponding to a similarity result, thereby adjusting the static voltage drop result by adding the first compensation value to the static voltage drop result.
[0213] In some embodiments, the similarity value is inversely proportional to the compensation value.
[0214] In some embodiments, the system further comprises: a static voltage drop result database for storing static voltage drop results; and a bump assignment database for storing bump assignments. The static voltage drop prediction circuit is communicatively coupled to the static voltage drop result database and the bump assignment database, and is used to: compare the static voltage drop result with a static voltage drop limit value; generate a plurality of bump static voltage drop results based on the bump assignments and the planar layout data; and determine a lowest value of the bump static voltage drop results as the optimized static voltage drop result and update the planar layout data.
[0215] In some embodiments, the system further comprises: a bump assignment adjustment circuit, configured to reduce a bump pitch between a plurality of bumps in the bump assignment in response to the static voltage drop prediction circuit predicting the optimal static voltage drop result.
[0216] In some embodiments, a method is provided, the method comprising the following operations: training a machine learning model in a machine learning circuit based on multiple training data and multiple test data associated with multiple electronic device designs; generating a mapping table based on the multiple training data and the multiple test data, the mapping table comprising multiple compensation values and multiple similarity values; and predicting a SIR result by the machine learning model based on a floor plan data and the mapping table.
[0217] In some embodiments, the method further comprises the steps of: extracting the plurality of training data from a plurality of training floorplan layouts before a placement phase; and extracting the plurality of test data from a plurality of test floorplan layouts before a placement phase.
[0218] In some embodiments, the step of generating the mapping table includes the following steps: classifying the multiple training data to obtain a plurality of classified training data; training an adversarial detection model based on the multiple classified training data and the multiple test data to generate the multiple similarity values; and generating the multiple compensation values based on the difference between the multiple predicted static voltage drop values generated by the machine learning model and the multiple test static voltage drop values corresponding to the multiple test data.
[0219] In some embodiments, the method further includes the following steps: generating a predicted static voltage drop result based on the plane layout data by the machine learning model; generating a similarity value corresponding to the plane layout data; searching for a corresponding compensation value in the mapping table according to the similarity value; and compensating the predicted static voltage drop result by the corresponding compensation value to predict the static voltage drop result.
[0220] In some embodiments, the method further comprises the following steps: when the similarity value is excluded from the mapping table, retraining the machine learning model with the floorplan data and a corresponding static voltage drop result.
[0221] The foregoing summarizes the features of several embodiments so that those skilled in the art may better understand the aspects of an embodiment of the present disclosure. Those skilled in the art will appreciate that they may readily use an embodiment of the present disclosure as a basis for designing or modifying other processes and structures for implementing the same purposes and / or achieving the same advantages of the embodiments introduced herein. Those skilled in the art will also recognize that such equivalent constructions do not depart from the spirit and scope of an embodiment of the present disclosure, and such equivalent constructions may be variously modified, substituted, and replaced herein without departing from the spirit and scope of an embodiment of the present disclosure.
Claims
1. A static voltage drop prediction method, characterized in that: The following steps are involved: A static voltage drop prediction circuit receives a floor plan data of a floor plan layout of a semiconductor device; generating a first static voltage drop result based on the floorplan data by a machine learning model; generating a first similarity value based on a comparison of the floor plan data with a plurality of training data; generating a second static voltage drop result based on the first static voltage drop result in a mapping table and a first compensation value corresponding to the first similarity value; and A bump assignment data is generated based on a comparison between the second static voltage drop result and a plurality of predetermined static voltage drop values to update the floorplan data. 2 . The static voltage drop prediction method as claimed in claim 1 , wherein the second static voltage drop result is equal to a sum of the first static voltage drop result and the first compensation value.
3. The static voltage drop prediction method according to claim 1 further comprises the following steps: When the first similarity value is less than a threshold value, the machine learning model is retrained through the floorplan data and a corresponding static voltage drop result.
4. The static voltage drop prediction method according to claim 1 further comprises the following steps: generating a bump assignment; generating a bump static voltage drop result corresponding to the bump assignment and the floorplan data by the machine learning model; comparing the bump static voltage drop result with a previous bump static voltage drop result to obtain an updated bump assignment; Repeating the steps of generating the bump assignment to comparing the bump static voltage drop results until a maximum number of iterations of optimization reaches a predetermined number; and The updated bump assignment of the bump assignment data is output.
5. The static voltage drop prediction method according to claim 4 further comprises the following steps: Setting a static voltage drop optimization threshold; and An updated bump static voltage drop result is compared with the static voltage drop optimization threshold to output the updated bump assignment of the bump assignment data.
6. The static voltage drop prediction method according to claim 4 further comprises the following steps: When the bump static voltage drop result is less than the previous bump static voltage drop result, replacing the updated bump assignment with the bump assignment; and When the bump static voltage drop result is greater than the previous bump static voltage drop result and a difference between the bump static voltage drop result and the previous bump static voltage drop result is within a range, the updated bump assignment is replaced with the bump assignment.
7. The static voltage drop prediction method according to claim 1 further comprises the following steps: Generate a third static voltage drop result, which is a sum of a fourth static voltage drop result provided by the machine learning model based on the updated floor plan data and a second compensation value, the second compensation value corresponding to a second similarity value associated with a comparison of the updated floor plan data with the plurality of training data.
8. The static voltage drop prediction method according to claim 1 further comprises the following steps: The semiconductor device is manufactured based on the updated floorplan data.
9. A static voltage drop prediction system, characterized in that: Include: A static voltage drop prediction circuit comprising a machine learning circuit is trained based on a plurality of training data associated with a plurality of electronic device designs to predict a static voltage drop result and is used to perform the following operations: receiving floor plan data of a floor plan layout; generating a static voltage drop result and a first similarity result; adjusting the static voltage drop result based on the first similarity result; and An optimized static voltage drop result is predicted based on the static voltage drop result and the plurality of bump assignments before placing the plurality of cells in a floorplan layout.
10. A static voltage drop prediction method, characterized in that: The following steps are involved: training a machine learning model in a machine learning circuit based on a plurality of training data and a plurality of test data associated with a plurality of electronic device designs; generating a mapping table based on the plurality of training data and the plurality of test data, the mapping table comprising a plurality of compensation values and a plurality of similarity values; and A static voltage drop result is predicted by the machine learning model based on a floorplan data and the mapping table.
