A digital chip physical design polymorphic flow platform with feature-first integration
By leveraging the priority integration feature of the digital chip physical design polymorphic process platform, the problems of module design consistency and iteration difficulties in digital chip physical design are solved, achieving efficient process management and rapid iteration, thereby improving design efficiency and quality.
Patent Information
- Application Number
- CN202111134563.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-09-27
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2041-09-27
AI Technical Summary
Existing technologies in digital chip physical design suffer from problems such as inconsistent physical design characteristics when modules are designed in parallel, difficulty in top-level design iteration, complex process management, and frequent software crashes. In particular, they are difficult to iterate quickly and verify uniformly in large-scale team designs.
A digital chip physical design polymorphic process platform with priority integration characteristics is adopted, which is divided into a central monitoring end and a client end. It includes a central monitoring system, a process characteristic component library, a task generator, a process scheduling engine and a graphical interface. Information exchange and process management are realized through a complete configuration table, and it supports the integration of multiple tools and rapid iteration.
It enables efficient control and observation of module-level physical design, simplifies process management, supports rapid iteration and unified verification, reduces the risk of software crashes, and improves design efficiency and quality.
Smart Images

Figure CN113935262B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of computer-aided design process platform, and in particular relates to a digital chip physical design polymorphic process platform with priority integration characteristics. Background Art
[0002] With the rapid development of industrial technology, the functional requirements of modern semiconductor chip design are becoming more and more complex, the process methods are also tending to be more sophisticated, and the integration level of some powerful on-chip systems such as image processing chips is also getting higher and higher. Correspondingly, the automatic layout, routing and sign-off process of the physical design of single digital chips has gradually been limited by computer-aided design tools and computing power. For the physical design of this type of chip, two hierarchical design methodologies, "top-down" and "bottom-up", have gradually emerged in the industry. They rely on splitting and designing large chips according to their scale and structural characteristics and finally integrating them to expand the boundaries of physical design capabilities. However, this split also brings some problems, such as:
[0003] 1) After the split, it is difficult to ensure the consistency of physical design features when designing a large number of modules in parallel, resulting in the failure to meet the timing requirements between modules or the failure to pass physical verification during integration;
[0004] 2) The top-level physical design intentions can be updated and iterated quickly, but when it is implemented at the module level, it is limited by the process time and cannot be iterated quickly;
[0005] 3) The physical design of the module is large and the sign-off requirements are many, which leads to complex process and version management, and sign-off is prone to be overlooked;
[0006] 4) There are many chip physical design tool software, and the file interfaces are complicated. It is very easy for the software to crash or freeze due to different file formats, which greatly affects the design efficiency.
[0007] There are three main types of products on the market with similar functions to this software:
[0008] 1) Standardized processes provided by computer-aided design tool software companies: Compared with such products, this platform has advantages in "polymorphism". This platform is not restricted by the company and requirements of the tool software, and for some new physical design methods and new ideas that go beyond the framework of the design tool software, this platform can be used as a basis for faster development and deployment, and can be efficiently implemented at the underlying module physical design level in a component reuse manner.
[0009] 2) Customized processes for chip design companies that prioritize module quality: Compared with this type of process, this platform has a stronger advantage in the highly customized top-level design method for efficient transmission and penetration of module-level physical design parameters. It can also control the degree of freedom of module-level physical design to varying degrees under the framework of "integration first", with higher flexibility.
[0010] 3) The self-built processes built by chip engineers are generally implemented by customized operating system scripts plus tool languages. They are flexible, but cannot be quickly and uniformly configured for certain integration requirements when used by large-scale teams. At the same time, when the chip module design workload is large and there are many designers, it is difficult to conduct rigorous and unified verification of some key parameters, and management is difficult. Summary of the invention
[0011] The present invention aims at the deficiencies in the prior art and provides a digital chip physical design polymorphic flow platform with priority integration characteristics.
[0012] To achieve the above object, the present invention adopts the following technical solutions:
[0013] A digital chip physical design polymorphic flow platform with priority integration characteristics, characterized by:
[0014] The platform is divided into two levels: a central monitoring end and a client. The central monitoring end includes a central monitoring system, and the client includes a process characteristic component library, a task generator, a process scheduling engine, and a graphical interface.
[0015] The process characteristic component library stores a set of implementation command operations for different computer-aided design tools, and the meta-operations of each tool form components, and the components are combined to form tasks that need to be implemented;
[0016] The task generator packages the task information after user configuration and template normalization, generates an overall process based on task composition with dependency relationships and signal processing mechanisms, and generates a set of task commands for process execution;
[0017] The process scheduling engine sends instructions to the operating system, which then calls the task command set generated by the task generator to complete the background execution of the process;
[0018] The graphical interface is used to monitor the execution status of the process and send instructions to the process scheduling engine to perform task operations;
[0019] The central monitoring system observes and analyzes the process, and the information exchange between the five modules of the central monitoring system, process characteristic component library, task generator, process scheduling engine and graphical interface is carried out through the complete configuration table of the process instance.
[0020] To optimize the above technical solutions, the specific measures taken also include:
[0021] Furthermore, the complete configuration table is generated by adding the user's customized configuration table to the default configuration table of the corresponding process stored in the platform. The priority of the customized configuration table is higher than the default configuration table. The default generated complete configuration table can be overwritten and modified through local configuration. After the complete configuration table of the client process instance is upgraded, the five modules update the platform status according to the complete configuration table and present it to the user.
[0022] Furthermore, the complete configuration table includes the following three parts:
[0023] The first part is the targetRunCfg structure, which is used to define the system information required by the platform when the task is executed;
[0024] The second part is the targetCmdCfg structure, which is used to define the internal parameters and user-defined parameters when generating the task command script in the task command set;
[0025] The third part is a custom structure, which is used to pass complex data structures and corresponding values in task command scripts.
[0026] Furthermore, the task generator first reads in the complete configuration table and checks the integrity of all necessary information, then compiles the template file and reads it into memory according to the task information configured in the complete configuration table, determines the insertion position of the component according to the specific text identifier in the template and inserts the component into the corresponding position of the template buffer, finally searches for the corresponding parameters of the template in the complete configuration table and replaces the parameters, and then writes the tool command script.
[0027] Furthermore, the platform defines the task status as follows: queue, wait, run, stop, suspend and complete; among them, queue, wait and run are defined as awake states, and stop, suspend and complete are defined as sleeping states; the state flow of each task is carried out according to the state machine of the process scheduling engine and the signal received by the task, and the sources of the signal received by the task include: operation signals of the graphical interface, task status change signals of corresponding dependencies, and task running status and file status signals received by the operating system monitoring process.
[0028] Furthermore, each task has a unique task ID. When a task receives a signal to enter a wake-up state, it will start a process for periodic query based on the task ID. This process is used to monitor the running status of the system process corresponding to its own task.
[0029] Furthermore, the central monitoring system uses a two-level directory structure to store the processes initiated by each user: the parent is the process instance group, and the child is the process instance; each parent directory corresponds to a local image of a complete client of a specified version, and multiple sub-process instances of different modules and different processes are allowed under the parent directory. The sub-process instances under the same parent process instance group share the same version of the client code image; specific parameter changes of the sub-process instance are achieved by modifying the complete configuration table.
[0030] Furthermore, the graphical interface is a PyQt-based field of view, which generates task entities represented by graphic entities based on the tasks corresponding to the complete configuration table, and the file dependencies between tasks are represented by the lines between the task entities. The current running status of each task is represented by the color of the task entity, and the color of the task entity changes with the change of the task status; the number of task entities and the number of levels in the field of view are generated according to the number of tasks and dependencies defined in the complete configuration table, and the task entity layout is automatically generated based on the Sugiyama algorithm.
[0031] The beneficial effects of the present invention are:
[0032] 1. The central control and observation function provides a central control platform for physical design project managers, chip architects, top-level logic and physical designers to control and observe module-level physical design;
[0033] 2. Process / module design engineers can configure customized parameters and component designs in a fixed format;
[0034] 3. Process engineers can use code release version control to perform unified process management;
[0035] 4. New design methodologies can be quickly deployed through simple configuration files, and multi-tool process integration is relatively easy;
[0036] 5. Task running status observation is simple and direct, and modular common components help new engineers get started quickly;
[0037] 6. Standardized logging and data publishing processes make the entire design process traceable and reproducible. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 It is a schematic diagram of the platform working mode.
[0039] Figure 2 It is a diagram in configuration table format.
[0040] Figure 3 This is the working mode diagram of the task generator.
[0041] Figure 4It is the process scheduling engine state machine.
[0042] Figure 5 This is the working mode diagram of the central monitoring system.
[0043] Figure 6 It is an example of extracting reports from the central monitoring system.
[0044] Figure 7 It is a diagram illustrating the partition layout of the graphical interface.
[0045] Figure 8 Schematic diagram of graphical interface samples under different process conditions.
[0046] Fig. 9 The graphical interface uses the Sugiyama algorithm to perform a flowchart of task layout. DETAILED DESCRIPTION
[0047] The present invention will now be described in further detail with reference to the accompanying drawings.
[0048] like Figure 1 As shown in the schematic diagram of the platform working mode, the entire system platform is divided into two levels: the central monitoring end and the client. The main users of the central monitoring end are project managers or physical design project layer / architecture engineers / process engineers. In addition to the central monitoring system of the central monitoring end, the client level mainly includes four main modules: feature component library, task automatic planning generator (referred to as task generator), process scheduling engine and process graphic operation interface (referred to as graphic interface).
[0049] There are a large number of command operation sets for different computer-aided software based on different process methodologies in the feature component library, where the meta-operation of each complete tool is referred to as a component (Utility). According to different process methodologies and chip integration characteristics, process designers can freely combine components into a complete task (task) as needed. The process organizer publishes the process by configuring the feature component library and the project default configuration table. Therefore, based on the prescribed format of the platform, feature customization based on different design tools can be freely implemented. Module-level physical design engineers or process users mainly use graphical interfaces to execute, monitor, operate and modify daily tasks. The corresponding task status will also be continuously refreshed locally and saved as a local file to be called by the central monitoring system.
[0050] The task generator and process scheduling engine are the underlying structure of the platform. They have no dependencies on the functional code of the project itself and are not open to platform users. They are mainly responsible for the component installation process and the resource call of the operating system (they can support direct docking with the operating system, or docking with task management systems such as the LSF platform). The main goal of the task is to achieve the initialization-implementation-check closed loop of the basic steps in a methodology. The complexity of the task is defined by the process engineer. The complete information of the task, such as computing resource configuration, dependencies, component sequence, and batch generation, is standardized through user configuration (config) and general templates (template) and passed to the task generator of the platform. The task generator packages the corresponding information and generates an overall process (flow) based on task composition with dependencies and signal processing mechanisms.
[0051] The background execution of the process is completed by sending instructions to the operating system through the task process scheduling engine to call the task command set (including process command set and system command set) generated by the task generator. It can realize two modes: automatic process execution and manual operation execution. The monitoring of the process execution status is completed through the graphical interface. At the same time, the graphical interface can also send instructions to the background process scheduling engine, and then perform operations such as execution, suspension and interruption of tasks.
[0052] The information exchange between all five main modules at the central / client level is carried out through the complete configuration table of the corresponding process instance. The complete configuration table is the core data configuration storage structure of the system process instance. Its configuration can completely determine the operation behavior of the process instance. The complete configuration table is generated by combining the corresponding process default configuration table stored in the platform and the user's custom configuration table. The custom configuration table has a higher priority than the default configuration table. All configuration tables use .json format for data storage and support three data structures: dictionary / list / string. Users can overwrite and modify the default generated complete configuration table through local configuration, and even add variables for customized local code snippets. After the complete configuration table of the client process instance is upgraded, the five main modules corresponding to the platform will update the platform status according to the complete configuration table and display it to the corresponding users.
[0053] like Figure 2The following is the content structure of the complete configuration table, which consists of three main parts. The first part is the targetRunCfg structure. The main function of this structure is to define the information required by the platform system when the task is executed, such as the number of CPU cores, the number of memory, the tool version to be called, the tool startup options, the task dependencies and the generated file list. The second part is the targetCmdCfg structure, which is used to define the internal parameters and user-defined parameters when the task command script is generated in the task command set, such as the module name, the process entry file path and the control switch of the tool software. The third part is the process engineer's custom structure, which is used to pass more complex data structures and corresponding values in the task command script, such as the process corner combination information in the required process library, the corresponding geometric rules of different metal layers of the power grid, and the voltage domain name and voltage value mapping list of the multi-voltage domain module. For any task, all parameters of the targetRunCfg structure and some required parameters in the targetCmdCfg structure need to be fully defined to ensure that the five modules of the platform get the description of the information required for the task. If missing, the task generator will report an error to remind the user. For the custom part and custom structure part of the targetCmdCfg structure, the task generator only performs syntax checking, not logic checking. Future extensions of other module types can be based on compatible configuration tables.
[0054] like Figure 3 As shown in the workflow diagram of the task generator, the task generator first reads in the complete configuration table to check the integrity of the necessary information, and then compiles the template file and reads it into the memory according to the task information configured in the configuration table. The insertion position of the component is determined according to the specific text identifier in the template. At this time, the command file storage buffer after the component is flattened is retained in the memory. The last step is parameter replacement. There are two modes for parameter replacement. The first mode is parameter configuration for the TCL language commonly used in chip design tools. At this time, a parameter definition file will be generated at the beginning of the entire command file, and the file will be displayed when the design tool executes the command file. The second mode is to directly convert the variables in the configuration table into text embedded in the command file, and the corresponding position in the generated command file will be directly displayed as the parameter value. The two parameter conversion modes are distinguished according to the specific text identifiers "$" and "<::>" in the template.
[0055] like Figure 4The following is a state machine of the process scheduling engine using LSF as a task distribution system. The platform defines six task states: inQueue, pending, running, stopped, inValid, and finished. The three states of queue / pending / running are defined as awake, and the stopped / suspended / finished states are defined as sleep. The state flow of each task is based on its Figure 4 The state machine is operated according to the signal received by the task. There are three sources of signals: operation signals of the graphical interface, change signals of the corresponding dependent task status, and task running status and file status signals received by the operating system monitoring process. For example, when a task changes its state, it will send a broadcast signal to its corresponding dependent task to inform it of the change in the state of the dependent task. At this time, the task that receives the signal will jump to another state according to the content of the signal. In addition, each task corresponds to a unique task ID. When a task receives a signal that causes it to enter a wake-up state (queue / wait / run), it will start a process that performs a timed query based on the task ID text. The platform process is used to monitor the operation of the system process corresponding to its own task. In the dormant state, the platform will release the monitoring process to save system computing resources.
[0056] like Figure 5The figure shows the working mode of the central monitoring system. For each platform process started by a user, a two-level directory structure is used for storage: the parent is the process instance group, and the child is the process instance. Each parent directory corresponds to a local image of a complete platform client of a specified version. At the same time, multiple sub-process instances of different modules and processes can be allowed under the parent directory. The sub-process instances under the same parent process instance group share the same version of the platform client code image. Process engineers can regularly release client code version updates for design engineers to choose. The specific parameter changes of the sub-process instance can be achieved by modifying its complete configuration table. When the design engineer uses the platform to create a new process instance, the corresponding directory and platform version information are stored in a record file under the user's root directory. At the same time, the platform provides a script for file copying and archiving based on fixed templates and configuration files. When the design engineer completes a certain task step, he can use the script to publish his design data to the public directory, and automatically send an email to the management personnel when the release is completed (some important task nodes can also be configured to automatically send emails to the designer's own mailbox). When managers need to obtain some specific information, there are two ways to choose from. The first way is to obtain all the process instance directories under monitoring by traversing the record files in all user directories. The second way is to check by viewing the data published by the design engineer in the public directory. Due to the use of a unified directory structure and copy name, the scripts used by the bottom-level engineers to extract reports can also be used by managers to organize data in a unified manner, such as Figure 6 That is, a report result that extracts unified management information of all modules of a certain model of chip.
[0057] Figure 7 This is a graphical interface partition layout description. The graphical interface of this platform is designed based on the PyQt framework and consists of a menu bar, a directory management table, a status bar, and a field of view. The menu bar provides some common functions such as opening files, waking up the interactive interface of a tool, changing task arrangement and help information, etc. The directory management table is based on Figure 5The structure shown allows users to centrally manage a tree table of all processes. Users can use this table to complete operations such as creating a new client, adding and deleting process instances, etc. The stored information is retained in the configuration file under the user directory, and this information will not disappear regardless of whether the interface is closed, opened, or restarted. The menu bar and the directory management table are connected via signal-slot. The field of view is the main body of the process, and tasks are represented by graphic entities in the field of view. The generation of task entities is based on the complete configuration table tasks. The lines between entities are used to represent the file dependencies between tasks described in the configuration table. Based on such a graphical architecture, different task tree forms for different processes can be completely automatically generated through user-defined configuration tables to form a typical directed acyclic graph composed of visual entities (for example Figure 8 The number of entities and levels in the directed acyclic graph can be obtained based on the number of tasks and dependencies defined in the configuration file. To reduce the cross-connection of dependencies, the graphical interface automatically generates the positions of all entities based on the Sugiyama Algorithm. However, due to the characteristics of many parallel tasks in physical design, some folding adjustments are made to parallel tasks that exceed a certain scale, that is, the corresponding hierarchical expansion actions. The process of the Sugiyama algorithm is as follows Fig. 9 As shown in the task layout flowchart. As long as there is no ring structure in the dependency relationship between tasks (if there is, it should be illegal logically), the corresponding process tree diagram can be automatically generated. At the same time, the current running status of each task in the process tree is represented by the color of the entity. As the task status changes, the entity color will also change to inform the user of the current global running status. Because it supports polymorphic processes, the task entities in the field of view also support user operations such as dragging, arranging, and searching.
[0058] The digital chip physical design polymorphic flow platform with priority integration characteristics proposed in the present invention has the following polymorphic features:
[0059] 1. Process polymorphism: As a general technology platform, it can realize processes with different characteristics and functions based on general components through user configuration. For example, the module-level timing analysis OCV (on chip variation) configuration component can also be used for top-level timing analysis. Users only need to configure the corresponding template in the process platform, modify the corresponding macro definition through the configuration file, and pass in the template and component to generate the dependency relationship and command set between tasks corresponding to the top-level timing analysis, which greatly facilitates the rapid configuration and rapid iterative correction required for the introduction of new methodologies.
[0060] 2. Instantiation polymorphism: For the same process, such as the module-level physical design process, according to the design characteristics of different modules, the corresponding macro definitions such as the specific value of congestion_driven_max_utilization or the specific value of max_transition can be retained when designing components, and the corresponding process command set can be easily passed in through the process user's own modification of the overall configuration file. It can also support conditional enabling of some components, that is, setting special tasks for a certain module separately, such as special timing analysis for PCIe modules.
[0061] 3. Task polymorphism: For multiple types of design tools required for the same process, the process platform uniformly uses the task level to encapsulate design tools of different languages and styles, and allows users to independently write them at the template level. At the configuration level, conditional polymorphism with different characteristics such as server requirements, version definitions, etc. under different task conditions is realized. Even for the same tool, different version configurations and wake-up methods can be set for different tasks.
[0062] The digital chip physical design polymorphic flow platform with priority integration characteristics proposed by the present invention has the following priority integration characteristics:
[0063] 1. Top-level controllability: The design intention of the top-level designer can be quickly synchronized with all process users through the establishment and maintenance of corresponding components. Once the component is upgraded, all designers can be quickly required to quickly synchronize to update the corresponding component code through rapid release. In addition, the corresponding user can be individually required to modify the configuration or assembly to conduct rapid model testing.
[0064] 2. Top-level observability: Technical superiors can observe and analyze all processes initiated by this platform through the central monitoring system. The top-level chip physics designer can avoid complicated emails or instant messaging, and directly reference any intermediate data of the platform's process through the central monitoring system and data records.
[0065] It can be seen that the process platform constructed by the present invention standardizes the design behavior of process developers and chip physical design engineers by splitting the physical process into processes, parameterizing the codes, standardizing the file interfaces, and visualizing the operations based on the fundamental principle of "integration first", improves the observability and controllability of module-level physical design by top-level physical design engineers during the module physical design process, reduces the management difficulty caused by chaotic version control and mixed output file formats, thereby effectively ensuring the design quality and design efficiency of the overall chip.
[0066] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.
Claims
1. A digital chip physical design polymorphic flow platform that prioritizes feature integration, Features: The platform is divided into two levels: a central monitoring end and a client. The central monitoring end includes a central monitoring system, and the client includes a process characteristic component library, a task generator, a process scheduling engine, and a graphical interface. The process characteristic component library stores a set of implementation command operations for different computer-aided design tools, and the meta-operations of each tool form components, and the components are combined to form tasks that need to be implemented; The task generator packages the task information after user configuration and template normalization, generates an overall process based on task composition with dependency relationships and signal processing mechanisms, and generates a set of task commands for process execution; The process scheduling engine sends instructions to the operating system, which then calls the task command set generated by the task generator to complete the background execution of the process; The graphical interface is used to monitor the execution status of the process and send instructions to the process scheduling engine to perform task operations; The central monitoring system observes and analyzes the process, and the information exchange between the five modules of the central monitoring system, the process characteristic component library, the task generator, the process scheduling engine and the graphical interface is carried out through the complete configuration table of the process instance; The complete configuration table includes the following three parts: The first part is the targetRunCfg structure, which is used to define the system information required by the platform when the task is executed; The second part is the targetCmdCfg structure, which is used to define the internal parameters and user-defined parameters when generating the task command script in the task command set; The third part is a custom structure, which is used to pass complex data structures and corresponding values in task command scripts; The platform defines the task status as follows: queue, wait, run, stop, suspend and complete; among which, queue, wait and run are defined as awake states, and stop, suspend and complete are defined as dormant states; the state flow of each task is carried out according to the state machine of the process scheduling engine and the signal received by the task, and the sources of the signal received by the task include: the operation signal of the graphical interface, the task status change signal of the corresponding dependency relationship, and the task running status and file status signal received by the operating system monitoring process; The central monitoring system uses a two-level directory structure to store the processes started by each user: the parent is the process instance group, and the child is the process instance; each parent directory corresponds to a local image of a complete client of a specified version, and multiple sub-process instances of different modules and different processes are allowed under the parent directory. The sub-process instances under the same parent process instance group share the same version of the client code image; specific parameter changes of the sub-process instance are achieved by modifying the complete configuration table.
2. A digital chip physical design polymorphic flow platform with priority integration characteristics as claimed in claim 1, Features: The complete configuration table is generated by adding the user's customized configuration table to the default configuration table of the corresponding process stored in the platform. The customized configuration table has a higher priority than the default configuration table. The default generated complete configuration table can be overwritten and modified through local configuration. After the complete configuration table of the client process instance is upgraded, the five modules update the platform status according to the complete configuration table and present it to the user.
3. A digital chip physical design polymorphic flow platform with priority integration characteristics as claimed in claim 1, Features: The task generator first reads in the complete configuration table and checks the integrity of all necessary information. Then, according to the task information configured in the complete configuration table, the template file is compiled and read into the memory. The component insertion position is determined according to the specific text identifier in the template and the component is inserted into the corresponding position of the template buffer. Finally, the corresponding parameters of the template are searched in the complete configuration table and the parameters are replaced, and then the tool command script is written.
4. A digital chip physical design polymorphic flow platform with priority integration characteristics as claimed in claim 1, Features: Each task has a unique task ID. When a task receives a signal to wake it up, it will start a process that performs a timed query based on the task ID. This process is used to monitor the running status of the system process corresponding to its own task.
5. A digital chip physical design polymorphic flow platform with priority integration characteristics as claimed in claim 1, Features: The graphical interface is a PyQt-based field of view, which generates task entities represented by graphic entities based on the tasks corresponding to the complete configuration table. The file dependencies between tasks are represented by the lines between the task entities. The current running status of each task is represented by the color of the task entity, and the color of the task entity changes as the task status changes. The number and number of levels of task entities in the field of view are generated according to the number of tasks and dependencies defined in the complete configuration table, and the task entity layout is automatically generated based on the Sugiyama algorithm.
