A conversion method based on a unified PDK interface specification

Abstract

The application relates to a conversion method based on a unified PDK interface specification, relates to the technical field of electronic design automation, and comprises the following steps: acquiring commercial PDK data; performing semantic analysis and structure reconstruction on a model file based on a machine learning migration learning algorithm to generate a standardized intermediate model; converting the intermediate model into target PDK data conforming to the format of a self-developed simulation platform according to the unified PDK interface specification; and automatically importing the target PDK data into the simulation platform through a templated script and verifying the integrity and consistency of the target PDK data. The method can be applied to a chip system design process, realizes unified access and collaborative simulation of commercial PDK and open source PDK, and improves the integrated circuit design efficiency.

Description

Technical Field

[0001] This invention relates to the field of electronic design automation technology, and specifically to a conversion method based on the unified PDK interface specification. Background Technology

[0002] The Process Design Kit (PDK) serves as a crucial data interface connecting the manufacturing process with EDA (Electronic Design Automation) simulation tools, forming the foundation for chip design, simulation, and verification. However, existing technologies suffer from the following problems:

[0003] Closed and incompatible formats: Mainstream commercial PDKs are developed primarily by international IDMs and only support foreign EDA tools such as Cadence and Synopsys. Their data formats are closed and cannot be directly used on domestically developed platforms.

[0004] Limited model support: For example, the "Hanqing" PDK standard only supports CMOS process and SPICE (Simulation Program with Integrated Circuit Emphasis) / BSIM (Berkeley Short-channel IGFET Model) models, which cannot meet the diverse process requirements of RF, optoelectronics, sensing, advanced packaging and other fields.

[0005] Severe data fragmentation: Traditional PDKs are developed using languages ​​such as SKILL and AEL, resulting in rigid data structures and poor readability, making it difficult for PDK data to be reused, expanded, and migrated across platforms.

[0006] Lack of unified interface specifications: The current lack of a unified PDK interface standard makes it impossible to efficiently integrate data between different processes, devices, and architectures, hindering the collaborative design and optimization of chip systems. Summary of the Invention

[0007] Based on this, the purpose of this invention is to provide a conversion method based on a unified PDK interface specification. By using machine learning transfer learning algorithms, commercial PDK models are parsed and reconstructed. Combined with a unified interface specification and templated scripts, intelligent conversion and efficient import of PDK data are achieved.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A conversion method based on the unified PDK interface specification includes the following steps:

[0010] Acquire commercial PDK data, including model files, layout information, symbols, and CDF configuration;

[0011] Based on machine learning transfer learning algorithms, the model file is semantically parsed and structurally reconstructed to generate a standardized intermediate model;

[0012] According to the unified PDK interface specification, the intermediate model is converted into target PDK data that conforms to the format of the independently developed simulation platform;

[0013] The target PDK data is automatically imported into the simulation platform using a templated script, and its integrity and consistency are verified.

[0014] Optionally, the unified PDK interface specification includes at least one or more of the following: material interface specification, process interface specification, architecture interface specification, and device interface specification;

[0015] The material interface specification is used to enable one-click binding between the material database and the PDK parameterization unit to support the automatic mapping of material properties in process simulation, TCAD or multiphysics.

[0016] The process interface specification is used to abstract process steps into parameterized objects, so as to transform the process model library in PDK into rule files that can be verified and executed by DRC, LVS and PEX;

[0017] The architecture interface specification is used to provide a unified topology, physical, and electrical description of the chip, interconnect, and package to support automatic placement, routing, or simulation.

[0018] The device interface specification is used to achieve unified encapsulation of device models in PDK to support schematic simulation, layout verification and link collaboration.

[0019] Optionally, the unified PDK interface specification adopts a unified naming rule and unit. When the unified PDK interface specification is submitted, it automatically performs a triple check of JSON-Schema (i.e., JSON schema, also known as JSON structure description specification), DRC deck (i.e., Design Rule Check deck), and model syntax. If it fails, it will refuse to merge.

[0020] Optionally, the device model includes optoelectronic, radio frequency, sensing, computing chip, or device model.

[0021] Optionally, the machine learning transfer learning algorithm includes the following steps:

[0022] Parse the syntax tree of a commercial PDK model;

[0023] Extract model parameters, topology, and physical rules;

[0024] The semantic model is mapped to the unified PDK interface specification, and an editable template is generated.

[0025] Optionally, the templated script is developed based on Python, C++, or OA language to support parameterized configuration and batch generation of PDK units.

[0026] Optionally, the templated script supports the following configuration formats:

[0027] XML format, used for the material interface specification;

[0028] JSON format, used for the architecture interface specification and the process interface specification;

[0029] YAML format is used for the device interface specification.

[0030] Optionally, the target PDK data supports at least one or more of the following: schematic-level simulator, TCAD (Technology Computer-Aided Design) process simulator, digital twin platform, or multiphysics coupling simulator.

[0031] Optionally, verifying its integrity and consistency includes the following steps:

[0032] Schematic symbols are checked for consistency with CDF (Component Description Format).

[0033] Verification of the matching between the graphics, ports, and electrical properties generated by the Pcell layout (Parameterized Cell Layout) and the model model;

[0034] Verification and back-annotation of interconnect structure.

[0035] Optionally, it can be applied in the chip system design process to achieve unified access and collaborative simulation of commercial PDK and open source PDK.

[0036] This invention provides a conversion method based on a unified PDK interface specification. It employs machine learning-assisted transfer learning to achieve intelligent parsing and reconstruction of commercial PDK models. It constructs a unified PDK interface specification covering four types of interfaces: materials, processes, architecture, and devices. A templated script engine is developed to support the automated generation, import, and verification of PDK data. Multiple model support (SPICE, Verilog-A, S-parameters) is achieved, covering processes such as RF, optoelectronics, sensing, and packaging. A reconfigurable interface engine is built, supporting XML / JSON / YAML multi-format configuration and connecting TCAD, digital twin, and multiphysics simulation platforms.

[0037] This invention provides a conversion method based on a unified PDK interface specification, which supports multiple processes such as radio frequency, optoelectronics, sensing, and packaging, covering the entire chain of data from materials, processes, architecture, and devices. It breaks the closed limitations of commercial PDK formats, improves PDK development efficiency and simulation platform compatibility, and is suitable for chip system design. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0039] Figure 1 This is a flowchart illustrating the conversion method based on the unified PDK interface specification of the present invention.

[0040] Figure 2 This is a schematic diagram illustrating the creation and storage of the process interface specification for the conversion method of the present invention;

[0041] Figure 3 This is a schematic diagram illustrating the conversion method of the present invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and beneficial effects of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] To address the shortcomings of existing PDK (Process Design Kit) technologies, such as closed and incompatible formats, limited model support, severe data fragmentation, and lack of unified interface specifications, this paper proposes a conversion method based on a unified PDK interface specification to achieve efficient integration between commercial PDKs and independently developed simulation platforms.

[0044] like Figure 1 As shown, it includes the following steps:

[0045] S100, acquires commercial PDK data, including model files, layout information, symbols and CDF configuration;

[0046] S200, based on machine learning transfer learning algorithms, performs semantic parsing and structural reconstruction on the model file to generate a standardized intermediate model;

[0047] S300, according to the unified PDK interface specification, converts the intermediate model into target PDK data that conforms to the format of the independently developed simulation platform;

[0048] S400 automatically imports target PDK data into the simulation platform using templated scripts and verifies its integrity and consistency.

[0049] Specifically, in S100, the system acquires commercial PDK data, including model files, layout information, symbols, and CDF configuration. Specifically, such as... Figure 3 As shown, the system first imports commercial PDK data sources such as ADS PDK Library or Cadence PDK, extracting the model files, layout information, symbol information, and CDF configuration information contained therein. This data is typically stored in a proprietary format and requires further processing before it can be used on the self-developed simulation platform.

[0050] In S200, the system uses machine learning transfer learning algorithms to process commercial PDK model files. The specific implementation includes the following sub-steps:

[0051] 1) Parse the syntax tree of the commercial PDK model;

[0052] 2) Extract model parameters, topology, and physical rules;

[0053] 3) Map to the semantic model of the unified PDK interface specification and generate an editable template.

[0054] The system extracts metadata from ADS PDK and Cadence PDK, and then transforms them based on a unified interface specification. During this process, the system identifies the syntactic structure in the model files, converts it into an abstract syntax tree, and then extracts key model parameters, topology information, and physical rules, ultimately generating an intermediate model representation that conforms to the unified PDK interface specification.

[0055] In S300, the system converts the intermediate model generated in the previous step into target PDK data according to a predefined unified PDK interface specification. The unified PDK interface specification includes at least one or more of the following: material interface specification, process interface specification, architecture interface specification, and device interface specification.

[0056] The material interface specification enables one-click binding of material databases (including at least open-source material databases) with PDK parametric units, supporting automatic mapping of material properties in process simulation, TCAD, or multiphysics. This allows the system to automatically associate material properties with corresponding PDK parametric units, simplifying the application of material properties in different simulation environments.

[0057] The process interface specification is used to abstract process steps into parameterized objects, transforming the process model library in the PDK into rule files that can be validated and executed by DRC, LVS, and PEX. This abstraction allows process steps to be parameterized, facilitating their use in different validation tools. The creation and import of the process interface specification are as follows: Figure 2 As shown.

[0058] Architecture interface specifications provide a unified topology, physical, and electrical description of chips, interconnects, and packages to support automated placement, routing, and simulation. This unified description enables systems to perform placement, routing, and simulation tasks more efficiently in chip design.

[0059] The device interface specification is used to implement a unified encapsulation of device models in the PDK to support schematic simulation, layout verification, and link coordination. The unified encapsulation of device models in the PDK implemented according to the device interface specification can effectively support schematic simulation, layout verification, and link coordination.

[0060] In a preferred embodiment, the unified PDK interface specification adopts a unified naming convention and unit. Upon submission, the specification automatically performs a triple check: JSON-Schema, DRC deck, and model syntax. If any check fails, the merge is rejected. This triple-check mechanism ensures the accuracy and consistency of the transformed PDK data.

[0061] In another preferred embodiment, the device model includes optoelectronic, radio frequency, sensing, computing chip, or device models. These different types of device models can be processed through a unified interface specification to enable application on an independently developed platform.

[0062] In S400, the system uses templated scripts to automatically import the converted target PDK data into the self-developed simulation platform, and performs integrity and consistency verification on the imported data.

[0063] Templated scripts are developed using Python, C++, or OA languages ​​to support parameterized configuration and batch generation of PDK units. These scripts can handle configuration files of different formats, including:

[0064] 1) XML format, used for material interface specifications;

[0065] 2) JSON format, used for architectural interface specifications and process interface specifications;

[0066] 3) YAML format, used for device interface specifications.

[0067] The target PDK data must support at least one of the following simulation platforms: schematic-level simulator, TCAD process simulator, digital twin platform, or multiphysics coupled simulator. This ensures that the converted PDK data can be used in a variety of simulation environments.

[0068] Verifying the integrity and consistency of the target PDK data includes the following steps:

[0069] 1) Check the consistency between schematic symbols and CDF;

[0070] 2) Verification of the matching between the graphics, ports, and electrical properties generated by the Pcell layout and the model;

[0071] 3) Verification and back-annotation of interconnect structure.

[0072] These verification steps ensure that the converted PDK data can function correctly on the self-developed platform and maintain functional consistency with the original commercial PDK data.

[0073] PDK data converted based on the unified PDK interface specification can be applied to chip system design in a self-developed simulation platform. This application enables unified access and collaborative simulation of commercial and open-source PDKs, providing a more flexible and efficient workflow for chip design.

[0074] Through the above steps, the conversion method based on the unified PDK interface specification provided by this invention can effectively convert commercial PDK data into target PDK data that conforms to the format of the independently developed simulation platform, realize interoperability between different PDKs, and improve the efficiency and flexibility of chip design.

[0075] The following is a detailed description using a specific embodiment.

[0076] This embodiment achieves intelligent transformation of commercial PDK models through machine learning-assisted transfer learning, thereby overcoming the limitation of closed commercial tool models. It also combines automated scripts to achieve automated migration of layout information, and finally realizes intelligent import of commercial PDK data. The efficiency of PDK transformation is improved through parameterized and templated scripts, and the efficient expansion of the QLT data interface engine process PDK library is supported.

[0077] Specifically, based on the defined unified interface specification, corresponding interface engine tools were developed for materials, processes, architectures, and devices. This demonstrated the feasibility of the interface specification in software and integrated it into the final software framework. The unified interface specification's interface engine played a crucial role in connecting to project data and supporting efficient simulation and twin operation.

[0078] This embodiment designs interface specifications for data interfaces of materials, processes, architecture, and devices, providing a solid data foundation for the full-chain simulation software. It also provides a series of comprehensive data interfaces and a robust foundation for the efficient design, collaborative optimization, and intelligent evolution of components. Simultaneously, it establishes a PDK library covering various component types, including optoelectronics, radio frequency, computing, sensing, and silicon adapter boards, thus creating a data pathway for the full-chain simulation of typical test devices.

[0079] The specific interface components and their characteristics are as follows:

[0080] 1) Materials Knowledge Base Interface Component: This component connects the open-source MP Project materials database with domestic materials genome databases. It supports access to over 150,000 data entries from the open-source MP Project materials database, with over 50 attribute query APIs for individual material data entries. It also supports access to over 7 million data entries from domestic materials genome databases, with over 30 attribute query APIs for individual material data entries. Furthermore, it supports scheduling, access, and orchestration for various mainstream EDA software using different programming languages ​​(such as Python / C++).

[0081] 2) Process PDK Library Interface Components: This component includes digital interface specifications for process, material, architecture, and device interfaces, supporting schematic-level simulation, layout design, and physical rule verification. The process interface specifications cover PDKs for silicon interposer integration processes, as well as commercial ADS and Cadence PDKs. It also covers four categories of devices: optoelectronics, RF, sensing, and computing, and at least 20 process model interfaces. See Table 1 for details of the specific process model interfaces.

[0082] Table 1. The process PDK library interface components include at least 20 process interface models.

[0083] Serial Number Key process model interface 1 Epitaxial process model interface 2 Positive photoresist process model interface 3 Negative photoresist process model interface 4 Anisotropic etching process model interface 5 Isotropic etching process model interface 6 CMP process model interface 7 Anisotropic deposition process model interface 8 Isotropic deposition process model interface 9 Atomic Layer Deposition (ALD) Process Model Interface 10 O2 oxidation process model interface 11 N2O oxidation process model interface 12 N2 Atmosphere Diffusion Process Model Interface 13 Interface of Thermal Matching and Thermal Mismatch Diffusion Process Model under N2 Atmosphere 14 N2 Atmosphere Thermal Matching Stress Diffusion Process Model Interface 15 Rapid Annealing (RTA) Process Model Interface 16 Ion implantation process model interface 17 Isolated Structure (STI) process model interface 18 Spacer process model interface 19 Trapezoidal etching process model interface 20 Electroplating process model interface 21 TSV process model interface 22 Interface for Micro Bump process model

[0084] 3) Architecture knowledge base interface and device knowledge interface library components: A unified interface was implemented for the chip architecture specification. The architecture and device parts involve architectures such as optoelectronic chips, radio frequency chips, sensing chips, and computing power chips, as well as at least 20 process models, which have been effectively verified. Finally, digital twins achieve precise twins of materials, processes, components, and devices by integrating experimental data and theoretical models.

[0085] The present invention provides a conversion method based on the unified PDK interface specification, which rewrites the core data base based on the standard format of OA and Python, thereby ensuring that process information, Pcells, models and physical verification rules can be efficiently and seamlessly integrated into the software platform.

[0086] Furthermore, to fully meet the needs of various devices, the interface engine integrates three common model formats: BSIM model, S-parameter model, and Verilog-A model. This ensures that the simulation platform can smoothly simulate and optimize various key devices, while also providing users with greater flexibility to customize their own model types. For example, common transistors can use the BSIM model, passive devices can use the S-parameter model, and complex optoelectronic and sensor devices can use the behavioral-level Verilog-A model.

[0087] The conversion method based on the unified PDK interface specification of this invention can intelligently convert PDK through a data interface engine by simply configuring symbol information, CDF information and Pcell information, and specifying the path of ADS / Cadence data. This allows the data to be imported into the self-developed data base and supports the use of downstream simulation platforms, such as XDS and other link simulators, to achieve device design and optimization.

[0088] This invention presents a conversion method based on the unified PDK interface specification. For architecture and device knowledge bases, it develops parsers based on JSON and YAML input formats. The interface engine enables the configuration and modification of arbitrary topologies, cores, and interconnects, maximizing its flexibility. It also supports the import and export of architectures, facilitating knowledge base management and maintenance. The device-related interface engine supports the display, reconstruction, and back-annotation of 3D stack-up information, and also supports formal verification of interconnects to ensure the accuracy of device information.

[0089] This invention provides a conversion method based on a unified PDK interface specification. It employs machine learning-assisted transfer learning to achieve intelligent parsing and reconstruction of commercial PDK models. It constructs a unified PDK interface specification covering four types of interfaces: materials, processes, architecture, and devices. A templated script engine is developed to support the automated generation, import, and verification of PDK data. Multiple model support (SPICE, Verilog-A, S-parameters) is achieved, covering processes such as RF, optoelectronics, sensing, and packaging. A reconfigurable interface engine is built, supporting XML / JSON / YAML multi-format configuration and connecting TCAD, digital twin, and multiphysics simulation platforms.

[0090] This invention provides a conversion method based on a unified PDK interface specification, which supports multiple processes such as radio frequency, optoelectronics, sensing, and packaging, covering the entire chain of data from materials, processes, architecture, and devices. It breaks the closed limitations of commercial PDK formats, improves PDK development efficiency and simulation platform compatibility, and is suitable for chip system design.

[0091] The embodiments described above are merely illustrative of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the present invention.

Claims

1. A conversion method based on the unified PDK interface specification, characterized in that, Includes the following steps: Obtain commercial PDK data, including model files, layout information, symbols, and CDF configuration; Based on machine learning transfer learning algorithms, the model file is parsed using syntax tree parsing to extract model parameters, topology, and physical rules, which are then mapped to the semantic model of the unified PDK interface specification, and an editable template is generated. The unified PDK interface specification includes four types of interfaces: materials, processes, architecture, and devices, and adopts a triple checking mechanism of JSON-Schema, DRC deck, and model syntax. According to the unified PDK interface specification, the intermediate model is converted into target PDK data that conforms to the format of the independently developed simulation platform; the intermediate model is generated by recognizing the syntax structure in the model file, converting it into an abstract syntax tree, and then extracting key model parameters, topology information and physical rules. The target PDK data is automatically imported into the simulation platform using a templated script, and its integrity and consistency are verified. in, Material Interface Specification: Used to enable one-click binding between the material database and PDK parameterized units, supporting automatic mapping of material properties in process simulation, TCAD, or multiphysics. Process Interface Specification: Used to abstract process steps into parameterized objects, enabling the process model library to be transformed into rule files that can be verified and executed by DRC, LVS, and PEX; Architecture Interface Specification: Used to provide a unified topology, physical, and electrical description of chips, interconnects, and packages, supporting automatic placement, routing, or simulation; Device Interface Specification: Used to achieve unified encapsulation of device models in PDK, supporting schematic simulation, layout verification, and link collaboration.

2. The conversion method according to claim 1, characterized in that, The device model includes optoelectronic, radio frequency, sensing, and computing chip models. Its parameters are extracted through syntax tree parsing as described in claim 1 and mapped to the device interface of the unified PDK interface specification.

3. The conversion method according to claim 1, characterized in that, The templated scripts are developed using Python, C++, or OA languages, and support parameterized configuration and batch generation of PDK units. Their configuration format is as follows: XML format is used for material interface specifications; JSON format is used for architectural interface specifications and process interface specifications; YAML format is used for device interface specifications.

4. The conversion method according to claim 1 or 2, characterized in that, The target PDK data supports the following simulation platforms: schematic-level simulator, TCAD process simulator, digital twin platform, or multiphysics coupling simulator.

5. The conversion method according to claim 1 or 2, characterized in that, The steps to verify its integrity and consistency include: Schematic symbol and CDF consistency check; Verify the matching of the graphics, ports, and electrical properties generated by the Pcell layout with the model; Verification of geometric topology consistency of interconnect structure and back-annotation of electrical parameters.

6. The conversion method according to claim 1 or 2, characterized in that, It is applied to the chip system design process to achieve unified access and collaborative simulation of commercial PDK and open source PDK.

Images