Circuit design intelligent modeling method and prediction method based on analog circuit behavior diagram

Abstract

The invention discloses a circuit design intelligent modeling method and prediction method based on an analog circuit behavior diagram. The modeling method comprises the steps that a large signal model and a small signal model are used for representing electrical characteristics of a transistor in an analog circuit, electrical parameters of all electronic components of the analog circuit are extracted through simulation, and the transistor is one of the electronic components; electrical parameters of the electronic components are embedded into a circuit topological graph corresponding to the analog circuit, the electronic components serve as nodes in a circuit behavior graph, physical connection between the electronic components serves as edges of the circuit behavior graph, and the circuit behavior graph is generated; and carrying out graph convolution operation on the behavior characteristics of the circuit behavior graph by utilizing a graph neural network to obtain global characteristic information of the circuit behavior graph. By combining a large signal model, a small signal model, electrical characteristic simulation, behavior graph construction and a graph neural network, electrical characteristics and dynamic behaviors of a circuit are effectively described, and important support is provided for circuit design and optimization.

Description

Technical Field

[0001] This invention relates to the field of analog integrated circuit technology, and in particular to an intelligent modeling and prediction method for circuit design based on analog circuit behavior diagrams. Background Technology

[0002] In the field of electronic design automation (EDA), digital circuit design can now rely heavily on mature computer-aided design (CAD) tools. However, the complexity and variability of analog circuit design mean it still primarily depends on the manual skills of experienced design experts. With the continuous advancement of semiconductor process technology and the increasing performance requirements of various applications, automated and intelligent design tools have become crucial factors driving the development of modern analog integrated circuit (IC) design. In particular, modern analog IC design faces stricter performance, power consumption, and area constraints, demanding a more efficient and intelligent design process.

[0003] Current mainstream approaches typically abstract the analog IC design problem into a "black box," exploring the design space through machine learning (ML) techniques such as reinforcement learning (RL), Bayesian optimization (BO), and evolutionary algorithms (EA). While these techniques offer some solutions for design automation, they primarily rely on substantial computational resources and simulation runs to meet specific design specifications. The main drawback of this approach is its lack of deep understanding of the inherent behavior of circuits; it requires extensive simulations to build accurate alternative models, which is not only time-consuming but also costly. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent modeling and prediction method for circuit design based on analog circuit behavior diagrams, so as to at least solve the aforementioned technical problems. The various technical effects produced by the optional technical solutions among the many technical solutions provided by this invention are detailed below.

[0005] To achieve the above objectives, in a first aspect, this application provides an intelligent circuit design modeling method based on analog circuit behavior diagrams, comprising: The electrical characteristics of transistors in analog circuits are characterized using large-signal and small-signal models, respectively. The electrical parameters of each electronic component in the analog circuit are extracted through simulation, wherein the transistor is one of the electronic components. The electrical parameters of each electronic component are embedded into the circuit topology diagram corresponding to the analog circuit. Each electronic component is used as a node in the circuit behavior diagram, and the physical connections between electronic components are used as edges in the circuit behavior diagram to generate the circuit behavior diagram. By performing graph convolution operations on the behavioral characteristics of the circuit behavior graph using a graph neural network, global feature information of the circuit behavior graph can be obtained.

[0006] In some embodiments, characterizing the electrical characteristics of transistors in an analog circuit using large-signal and small-signal models respectively includes: The transistor's first electrical characteristics in the nonlinear operating region are characterized using a large-signal model, the fundamental parameters of which include: drain-source voltage. and overdrive voltage The drain-source voltage The overdrive voltage characterizes the voltage difference between the source and drain of the transistor. Characterizes the difference between the gate-source voltage and the threshold voltage; The second electrical characteristic of the transistor under AC small-signal excitation is characterized using a small-signal model, the basic parameters of which include transconductance. and turning frequency The transconductor The cutoff frequency characterizes the rate at which the leakage current of the transistor changes with the gate-source voltage. This indicates the high-frequency response capability of the transistor.

[0007] In some embodiments, the step of extracting the electrical parameters of each electronic component of the analog circuit through simulation includes: Identify the terminals of each electronic component in the analog circuit and the operating voltage in the analog circuit. ; Large-signal parameter extraction is performed on the transistor to obtain the DC operating point of the drain-source voltage, the real-time calculated value of the overdrive voltage, and the nonlinear transfer function curve of the transconductance; small-signal parameter extraction is performed simultaneously on the transistor to obtain the frequency-varying characteristics of the gain-bandwidth product and the output impedance based on AC scanning; basic parameter calibration is performed on the resistors, capacitors, and inductors in the analog circuit, including the ohmic characteristics of the resistor, the frequency response of the capacitive reactance of the capacitor, and the critical value of the magnetic saturation of the inductor.

[0008] In some embodiments, treating each of the electronic components as nodes in a circuit topology diagram includes: The terminals of each of the electronic components and the operating voltage in the analog circuit. The mapping is to nodes in the circuit behavior graph.

[0009] In some embodiments, the step of performing graph convolution operations on the behavioral characteristics of the circuit behavior graph using a graph neural network to obtain global feature information of the circuit behavior graph includes: The electrical characteristics of each node are represented by feature vectors and used as input to the graph neural network. Each layer of the graph neural network performs feature aggregation based on the feature vectors of the current node and its neighbors, and performs message passing and feature updates during the feature aggregation process.

[0010] In some embodiments, each layer of the graph neural network performs feature aggregation based on the feature vectors of the current node and its neighboring nodes, including: Using the following formula, in the l-th layer of the graph neural network, the feature vector of the current node v is updated based on the feature vectors of its neighboring nodes u: ; in, This represents the learnable weight matrix. This represents the activation function. This represents an aggregation function used to combine the feature vectors of the current node v. and the feature vector of neighbor node u And the edge attributes between the current node v and its neighbor node u. .

[0011] In some embodiments, the method further includes: Through an attention mechanism, different weights are assigned to each pair of current node u and neighbor node v during feature aggregation and message passing, using the following formula: ; in, This represents the learnable weight matrix used for the query in the attention mechanism. The key represents a learnable weight matrix used for the attention mechanism, the query represents the degree of attention of the current node v to its neighbor node u, and the key represents the feature of the neighbor node u. This represents the feature vector between the current node v and its neighbor node u. and When calculating similarity through transformations of the query and key matrices, the attention weights between the current node v and its neighboring node u are... This represents the function for processing edge attributes, and LeakyReLU represents the activation function.

[0012] Secondly, this application provides a performance prediction method based on analog circuit behavior diagrams, including: Through graph embedding operations, the global feature information obtained by the intelligent circuit design modeling method based on analog circuit behavior graphs in the first aspect is transformed into a graph embedding vector, wherein the graph embedding vector is a vector of fixed dimensions. In the feedforward neural network, the graph embedding vector is subjected to a nonlinear transformation to obtain the performance prediction value of the circuit behavior graph, so as to output the performance prediction result of the circuit behavior graph.

[0013] Thirdly, this application provides an intelligent circuit design device based on analog circuit behavior graphs, comprising: one or more processors; a memory for storing one or more computer programs, wherein the one or more processors are configured to execute the one or more computer programs stored in the memory, so that the one or more processors execute an intelligent circuit design modeling method based on analog circuit behavior graphs as described in any of the first aspects and a performance prediction method based on analog circuit behavior graphs as described in the second aspect.

[0014] Fourthly, this application provides a computer program product stored on a data carrier and designed to execute the intelligent circuit design modeling method based on analog circuit behavior graphs as described in any of the first aspects and the performance prediction method based on analog circuit behavior graphs as described in the second aspect.

[0015] Implementing one of the above-described technical solutions of this application has the following advantages or beneficial effects: This application presents an intelligent modeling and prediction method for circuit design based on analog circuit behavior graphs. It utilizes large-signal and small-signal models to characterize the electrical characteristics of transistors in analog circuits, comprehensively describing these characteristics and effectively separating electrical features from process limitations. This results in a model with high accuracy and robustness under different process conditions. The constructed circuit behavior graph contains the circuit's electrical characteristics, topological information, and the interactions between different electronic components. It not only displays the circuit's static topology but also describes its dynamic characteristics under different operating conditions, providing rich input data for graph neural network modeling and detailed performance analysis and optimization for circuit design. Graph neural networks can capture complex circuit behavior from both global and local perspectives. Through feature propagation at nodes and convolution operations on the graph structure, dynamic characteristics such as signal transmission, gain control, and frequency response can be effectively captured, demonstrating significant advantages in accuracy and performance. Furthermore, the obtained global feature information can be used for circuit performance prediction, which is crucial for circuit design and optimization. It helps engineers evaluate circuit performance under different design conditions and make optimization decisions.

[0016] This application provides a comprehensive and accurate analog circuit modeling and prediction method by combining large-signal models, small-signal models, electrical characteristic simulation, behavioral graph construction, and graph neural networks. It does not require a large number of simulations or thousands of circuit simulations to obtain a sufficiently accurate model, but can effectively describe the electrical characteristics and dynamic behavior of the circuit, providing important support for circuit design and optimization. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of 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 described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a flowchart illustrating an intelligent circuit design modeling method based on analog circuit behavior diagrams according to an embodiment of this application. Figure 2 This is a schematic diagram illustrating the process of converting the current mirror circuit of this application embodiment into a circuit behavior diagram; Figure 3 This is a flowchart illustrating the performance prediction method based on analog circuit behavior diagrams according to an embodiment of this application; Figure 4 This is a schematic diagram of circuit behavior modeling and prediction in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, various exemplary embodiments described below will be referenced to the accompanying drawings, which form part of the exemplary embodiments and depict various exemplary embodiments that may be adopted to implement this application. Unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. It should be understood that they are merely examples of processes, methods, and apparatuses consistent with some aspects of this application disclosed as detailed in the appended claims, and other embodiments may be used, or structural and functional modifications may be made to the embodiments listed herein without departing from the scope and spirit of this application.

[0019] In the description of this application, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. The term "multiple" means two or more. The terms "connected" and "linked" should be interpreted broadly, for example, they can refer to fixed connections, detachable connections, integral connections, mechanical connections, electrical connections, communication connections, direct connections, indirect connections through an intermediate medium, and can refer to the internal connection of two elements or the interaction relationship between two elements. For those skilled in the art, the specific meaning of the above terms in this application can be understood according to the specific circumstances.

[0020] To illustrate the technical solutions described in this application, specific embodiments are provided below, showing only the parts related to the embodiments of this application.

[0021] Example 1: Firstly, such as Figure 1 As shown, this application provides an intelligent circuit design modeling method based on analog circuit behavior diagrams, which includes the following steps S10 to S30.

[0022] S10: The electrical characteristics of transistors in the analog circuit are characterized by large-signal and small-signal models respectively, and the electrical parameters of each electronic component in the analog circuit are extracted by simulation, wherein the transistor is one of the electronic components.

[0023] The analog circuit can be the circuit schematic of a current mirror circuit. The process of converting a current mirror circuit into a circuit behavior diagram is as follows: Figure 2 As shown. The current mirror circuit includes transistors, power supplies, etc. Of course, the analog circuit of this application is not limited to the current mirror circuit, and can also be other circuits with transistors, which are not limited here.

[0024] In some implementations, the electrical characteristics of transistors in analog circuits are characterized using large-signal and small-signal models, respectively, which may include: The transistor's first electrical characteristics in the nonlinear operating region are characterized using a large-signal model, the fundamental parameters of which include drain-source voltage. and overdrive voltage The drain-source voltage The overdrive voltage characterizes the voltage difference between the source and drain of the transistor. Characterizes the difference between the gate-source voltage and the threshold voltage; The second electrical characteristic of a transistor under AC small-signal excitation is characterized using a small-signal model, the fundamental parameters of which include transconductance. and turning frequency The transconductor The cutoff frequency measures the rate at which the leakage current of the transistor changes with the gate-source voltage. This indicates the high-frequency response capability of the transistor.

[0025] Specifically, the electrical characteristics of a transistor include a first electrical characteristic and a second electrical characteristic. The first electrical characteristic is characterized by a large-signal model, representing the electrical characteristics of the transistor in the nonlinear operating range, while the second electrical characteristic is characterized by a small-signal model, representing the electrical characteristics of the transistor under AC small-signal excitation.

[0026] Large-signal parameters describe the performance of a transistor after it is powered on, primarily affecting its swing, gain, and noise performance. Small-signal parameters have a significant impact on key performance characteristics such as bandwidth, slew rate, and power consumption.

[0027] Large-signal models primarily focus on drain-source voltage. and overdrive voltage Impact on transistor performance, overdrive voltage The difference between the gate-source voltage and the threshold voltage represents the portion of the gate-source voltage that exceeds the threshold voltage, and it determines the operating range and switching characteristics of the transistor. Drain-source voltage. The voltage difference between the source and drain of the transistor is characterized, and this voltage difference affects the operating range of the transistor.

[0028] In the large-signal model, the transistor gain tends to saturate at a certain operating voltage, and the overdrive voltage... The increase in amplitude not only affects the switching speed of transistors, but may also introduce greater noise and nonlinear distortion, which is especially important for high-speed digital circuits. Therefore, the focus of large-signal models is to determine the behavior of transistors under large input signals, especially the variation of their operating range, and the impact of these variations on the output characteristics of analog circuits, particularly noise and distortion.

[0029] Compared to large-signal models, small-signal models focus more on the transistor's response to small input changes. They typically evaluate the gain and frequency response of analog circuits by analyzing the effect of small changes in gate-source voltage on the transistor's transconductance and turnaround frequency.

[0030] In small-signal model analysis, transconductance is a core parameter for transistor gain, characterizing the rate at which the transistor's drain current changes with the gate-source voltage. It refers to the rate of change of the drain-source current when the gate-source voltage changes; the larger the transconductance, the stronger the transistor's gain. The cutoff frequency... The high-frequency response capability of the transistor refers to the frequency at which the transistor gain begins to decay. The higher the cutoff frequency, the better the transistor performs in high-frequency applications.

[0031] By using small-signal models, the gain characteristics of transistors under normal operating conditions can be accurately evaluated, and by combining frequency response analysis, the performance of analog circuits in the high-frequency range can be obtained.

[0032] These characteristics not only reflect the linear performance of transistors during operation, but also provide key data for circuit design, especially in high-frequency and radio frequency circuit design, where the cutoff frequency and transconductance are two important factors that determine circuit performance.

[0033] By combining large-signal and small-signal models, the electrical behavior of transistors is comprehensively described, thereby decoupling circuit characteristics from process limitations. Utilizing large-signal and small-signal models to electrically characterize transistors in analog circuits, and to simulate analog circuits, allows for a comprehensive characterization of the electrical characteristics of transistors under different operating conditions.

[0034] After obtaining the electrical characteristics of the transistor, the electrical parameters of each electronic component in the analog circuit are extracted through simulation. It is understood that the transistor is one type of electronic component in this analog circuit. In addition to transistors, electronic components in an analog circuit also include power supplies, resistors, capacitors, and other elements.

[0035] In some embodiments, the step of extracting the electrical parameters of each electronic component of the analog circuit through simulation may include: Identify the terminals of each electronic component in the analog circuit and the operating voltage in the analog circuit. ; Large-signal parameter extraction is performed on the transistor to obtain the DC operating point of the drain-source voltage, the real-time calculated value of the overdrive voltage, and the nonlinear transfer function curve of the transconductance; small-signal parameter extraction is performed simultaneously on the transistor to obtain the frequency-varying characteristics of the gain-bandwidth product and the output impedance based on AC scanning; basic parameter calibration is performed on the resistors, capacitors, and inductors in the analog circuit, including the ohmic characteristics of the resistor, the frequency response of the capacitive reactance of the capacitor, and the critical value of the magnetic saturation of the inductor.

[0036] Specifically, when simulating the electrical parameters of each electronic component in an analog circuit, the terminals and operating voltages of each electronic component are identified. In this context, the terminals of a transistor refer to its gate (G), source (S), and drain (D). The terminals of a power supply include the positive and negative terminals of the power supply voltage. The terminals of a resistor refer to the two ends of the resistor. The terminals of a capacitor refer to the connection points between the two ends of the capacitor, which are the parts connected to external circuits, such as metal leads, solder pads, sockets, etc.

[0037] Terminals and operating voltages of various electronic components Used to map nodes in a circuit behavior graph.

[0038] Simulations are performed on electronic components such as resistors, capacitors, and inductors to calibrate basic parameters, thereby obtaining fundamental electrical parameters such as the ohmic characteristics of resistors, the capacitive reactance frequency response of capacitors, and the magnetic saturation critical value of inductors. Simulations are also performed on the electrical characteristics of transistors, specifically involving large-signal parameter extraction to obtain the drain-source voltage. DC operating point, overdrive voltage Real-time calculated values ​​and transconductance The nonlinear transfer function curve is obtained, and at the same time, small-signal parameters of the transistor are extracted. Based on AC scanning, the frequency-varying characteristics of the gain-bandwidth product and output impedance are obtained, thereby obtaining the electrical characteristics of the transistor and the electrical parameters of each electronic component. These simulation data provide numerical basis for subsequent circuit modeling and can accurately reflect the behavior characteristics of the analog circuit under different operating conditions.

[0039] by Figure 2 Taking a current mirror circuit as an example, when engineers consider using an N-channel metal-oxide-semiconductor field-effect transistor (MOSFET) to construct a current mirror circuit, during simulation, they first set the power supply voltage and adjust the gate-source voltage to control the transistor's operating state. During this process, the transistor's transconductance and cutoff frequency can be detected, and these electrical characteristics can be extracted from the simulation results. Transconductance is typically calculated by measuring the effect of gate-source voltage changes on the source current, while the cutoff frequency describes the rate of gain decay of the transistor at high-frequency operation. By performing simulations under different operating conditions, the specific values ​​of these electrical characteristics of the transistor at different operating points can be obtained.

[0040] Furthermore, since the circuit also contains passive components such as resistors and capacitors, the electrical characteristics of these components also need to be obtained through simulation. For example, the resistance value of the resistor and the capacitance value of the capacitor will have an important impact on the operating characteristics of the current mirror circuit. During the simulation process, the values ​​of these components can be adjusted to further analyze their impact on the circuit performance.

[0041] Accurate circuit simulation not only determines the operating range of transistors, but also obtains the electrical characteristics of other components in the simulated circuit under different frequency and voltage conditions. Furthermore, these electrical characteristics provide important support for the circuit modeling process, ensuring that the behavior of the circuit can be comprehensively described under different electrical conditions.

[0042] S20: Embed the electrical parameters of each electronic component into the circuit topology diagram corresponding to the analog circuit, use each electronic component as a node in the circuit behavior diagram, and use the physical connections between electronic components as edges in the circuit behavior diagram to generate the circuit behavior diagram.

[0043] Electrical parameters of electronic components include the transconductance of transistors. Transition frequency Drain-source voltage and overdrive voltage The basic electrical parameters of electronic components other than the transistors are also included. These basic electrical parameters include the resistance value R of resistors and the capacitance value C of capacitors.

[0044] When constructing the circuit behavior diagram, firstly, the electrical parameters of each electronic component obtained from the simulation are embedded into the circuit topology diagram corresponding to the simulated circuit. Each electronic component (such as transistor, resistor, capacitor, etc.) of the simulated circuit is a node in the circuit behavior diagram.

[0045] In some embodiments, nodes in the circuit behavior diagram The eigenvectors contain the electrical characteristics of the component. For example, for a node corresponding to a transistor, its eigenvectors include transconductance. Transition frequency Drain-source voltage and overdrive voltage For a resistor or capacitor, its characteristic vector includes the resistance value R or the capacitance value C.

[0046] In this way, each electronic component in the circuit is not only connected to other components through its physical connections, but is also accurately represented in the circuit behavior diagram through other electrical characteristics.

[0047] The physical connections between electronic components are represented as edges in the circuit behavior graph, which is then used to generate the circuit behavior graph.

[0048] The circuit behavior diagram can be used to display the electrical characteristics, topology information, and interactions between different components of an analog circuit.

[0049] In some embodiments, treating the various electronic components as nodes in a circuit topology may include: The terminals of each of the electronic components and the operating voltage in the analog circuit. The mapping is to nodes in the circuit behavior graph.

[0050] Specifically, this can be achieved by connecting the transistor's gate (G), source (S), drain (D), the positive and negative terminals of the power supply voltage, the two ends of a resistor, the two ends of a capacitor, and other electronic components, as well as the operating voltage. All of these are mapped to nodes in the circuit behavior diagram. These nodes represent the electrical characteristics of the various electrical components in the circuit behavior diagram.

[0051] Furthermore, the connections between nodes reflect the physical connections between electronic components. These physical connections can be used as edges of the circuit behavior graph, and the topological information of the circuit can be captured through the structure of the circuit behavior graph.

[0052] By constructing circuit behavioral graphs, the electrical characteristics, topology information, and interactions between different electrical components of an analog circuit can be displayed. These graphs not only show the static topology of the circuit but also describe its dynamic characteristics under different operating conditions. For example, the dynamic behaviors of the circuit, such as signal transmission paths, gain changes, and frequency response, can be intuitively presented through the graph structure when operating conditions change. These behavioral graphs provide rich input data for subsequent graph neural network modeling and also offer detailed performance analysis and optimization basis for circuit design.

[0053] by Figure 2 Taking the current mirror circuit as an example, the current mirror circuit is equivalent to an intermediate circuit to obtain the electrical characteristic data of each component. This data is then embedded into the nodes of the circuit topology diagram to generate a circuit behavior graph. In this circuit behavior graph, the feature vector of each node contains the electrical characteristics of the component it represents; that is, the feature vector of a transistor node includes the transconductance of the transistor. Transition frequency Overdrive voltage and drain-source voltage The eigenvectors of resistor nodes include the resistance value. In this way, each node not only represents a component in the circuit but also contains its electrical characteristics. These eigenvectors provide crucial numerical data for subsequent modeling and analysis, enabling the circuit behavior diagram to more accurately describe the actual operating characteristics of the circuit.

[0054] S30: Graph convolution operation is performed on the behavioral characteristics of the circuit behavior graph using a graph neural network to obtain global feature information of the circuit behavior graph.

[0055] This application employs a backbone network structure to assist in circuit modeling, particularly in behavioral-level prediction, by effectively aggregating node information and transforming it into a comprehensive representation of the global graph.

[0056] In some implementations, step S30 may include: The electrical characteristics of each node are represented by feature vectors and used as input to the graph neural network. Each layer of the graph neural network performs feature aggregation based on the feature vectors of the current node and its neighbors, and performs message passing and feature updates during the feature aggregation process.

[0057] Specifically, after constructing the circuit behavior graph, graph neural networks (GNNs) can be used for further learning and modeling. As a deep learning method capable of processing graph-structured data, GNNs allow node features to be propagated based on their connections with neighboring nodes during training, thus gradually learning the circuit's performance characteristics. Through multi-layer node feature propagation and convolution operations, global and local feature information can be extracted from the circuit behavior graph. The process of modeling the circuit behavior graph using GNNs is essentially a graph convolution operation.

[0058] The electrical characteristics of each node are represented by a feature vector, including the feature vector of the node corresponding to the transistor, which includes transconductance. Transition frequency Drain-source voltage and overdrive voltage The feature vectors of resistors or capacitors include the resistance value R or capacitance value C, etc. After representing the electrical characteristics of each node with feature vectors, these vectors are used as input to the graph neural network. Then, each layer of the graph neural network propagates and updates characteristics based on the feature vector information between nodes. This process is carried out through mechanisms such as feature aggregation, message passing, and feature updating to ensure that information from each node in the circuit behavior graph can flow fully, thereby extracting the global feature information of the circuit.

[0059] The core of a graph neural network consists of three parts: feature aggregation, message passing, and feature updating. These steps ensure that node information can be progressively propagated and updated within the multi-layered structure of the graph, ultimately capturing the complex electrical relationship between nodes in the circuit behavior graph.

[0060] In some embodiments, each layer of the graph neural network performs feature aggregation based on the feature vectors of the current node and its neighboring nodes, which may include: Using the following formula, in the l-th layer of the graph neural network, the feature vector of the current node v is updated based on the feature vectors of its neighboring nodes u: ; in, This represents the learnable weight matrix. This represents the activation function, which can be either the ReLU function or the LeakyReLU function. This represents an aggregation function used to combine the feature vectors of the current node v. and the feature vector of neighbor node u And the edge attributes between the current node v and its neighbor node u. Neighbor node u is a neighbor node of the current node v. That is, the feature update of the current node V is based on the information from its neighbor node u.

[0061] The purpose of aggregation functions is to gather information from neighboring nodes together to provide a basis for updating the features of the current node.

[0062] This process uses a multi-layered message passing mechanism to enable information between nodes to be transmitted between different layers of the graph network, thereby capturing the complex electrical characteristic relationships between electronic components in the circuit behavior graph.

[0063] To further improve the performance of graph neural networks, especially to better capture the dependencies between physical parameters during the modeling process, an attention mechanism can be introduced. Therefore, in some embodiments, the method may further include: Through an attention mechanism, different weights are assigned to each pair of current node u and neighbor node v during feature aggregation and message passing, using the following formula: ; in, This represents the learnable weight matrix used for the attention mechanism. The key represents the learnable weight matrix used for the attention mechanism. Both the query and the key are core components of the attention mechanism. The query represents the degree of attention that the current node v pays to its neighbor node u, and the key represents the features of the neighbor node u. This represents the feature vector between the current node v and its neighbor node u. and When calculating similarity through transformations of the query and key matrices, the attention weights between the current node v and its neighboring node u are... This represents a function for processing edge attributes. The function can adjust the weights based on the different physical characteristics of the edges. LeakyReLU represents a commonly used activation function.

[0064] By dynamically adjusting the information transmission weights between nodes, the attention mechanism enables each node to perform more precise information processing based on the electrical characteristics of neighboring nodes and the physical properties of edges when aggregating the features of neighboring nodes, thereby improving the model's predictive ability.

[0065] By introducing an attention mechanism and assigning higher weights to important nodes and edges, the model focuses more on the parts that have a greater impact on circuit performance during the learning process. This enables the graph neural network to effectively capture the complex behavior of the circuit and improve the prediction accuracy of circuit behavior through multi-level feature learning.

[0066] After updating the node features of the graph neural network, the global feature information of the circuit behavior graph is effectively aggregated.

[0067] The embodiments of this application utilize large-signal and small-signal models to characterize the electrical characteristics of transistors in analog circuits, respectively. This comprehensively describes the electrical characteristics of transistors in analog circuits, effectively separating the electrical characteristics and process limitations of the circuit, resulting in models with high accuracy and robustness under different process conditions. The constructed circuit behavior graph contains the circuit's electrical characteristics, topological information, and the interaction relationships between different electronic components. It not only displays the static topological structure of the circuit but also describes its dynamic characteristics under different operating conditions, providing rich input data for graph neural network modeling and detailed performance analysis and optimization basis for circuit design. Using graph neural networks, the complex behavior of circuits can be captured from both global and local perspectives. In graph neural networks, through feature propagation of nodes and convolution operations of the graph structure, dynamic characteristics such as signal transmission, gain control, and frequency response can be effectively captured, exhibiting significant advantages in accuracy and performance. Furthermore, the obtained global feature information can be used for circuit performance prediction, which is of great significance for circuit design and optimization. It can help engineers evaluate circuit performance under different design conditions and make optimization decisions.

[0068] This application provides a comprehensive and accurate analog circuit modeling method by combining large-signal models, small-signal models, electrical characteristic simulation, behavioral graph construction, and graph neural networks. It can effectively describe the electrical characteristics and dynamic behavior of circuits without requiring a large number of simulations or thousands of circuit simulations to obtain a sufficiently accurate model, thus providing important support for circuit design and optimization.

[0069] Secondly, this application also provides a performance prediction method based on analog circuit behavior diagrams, such as... Figure 3 As shown, the performance prediction method based on analog circuit behavior graphs may include steps S100 to S200.

[0070] S100. Through graph embedding operation, the global feature information obtained by the intelligent modeling method for circuit design based on analog circuit behavior graph is transformed into a graph embedding vector, wherein the graph embedding vector is a vector of fixed dimensions. S200. In the feedforward neural network, the graph embedding vector is subjected to a nonlinear transformation to obtain the performance prediction value of the circuit behavior graph, so as to output the performance prediction result of the circuit behavior graph.

[0071] Specifically, after the global feature information obtained by the intelligent circuit design modeling method based on analog circuit behavior graphs is effectively aggregated, graph embedding technology can be used to apply this global feature information for performance prediction. That is, the global feature information of the entire circuit behavior graph is embedded, thereby transforming it into a fixed-dimensional vector, known as a graph embedding vector. This graph embedding vector can be more effectively passed to subsequent models for processing. This subsequent model can be a feedforward neural network, a multilayer perceptron mechanism.

[0072] Furthermore, the graph embedding vector is input into a feedforward neural network and subjected to a series of nonlinear transformations, including activation functions and weight updates, thereby transforming the input graph embedding vector into performance predictions of the circuit behavior graph.

[0073] Finally, after processing by the multilayer sensing mechanism of the feedforward neural network, the feedforward neural network can output performance prediction results of the circuit behavior diagram, such as gain, frequency response, and power consumption. These prediction results are of great significance for circuit design and optimization, and can help engineers evaluate the circuit performance under different design conditions and make optimization decisions.

[0074] Through the above process, graph neural networks can not only model circuits based on their topology and electrical characteristics, but also learn the complex interactions between components within the circuit. Furthermore, they utilize attention mechanisms to amplify the influence of key features, thereby achieving accurate circuit performance prediction. This method has broad application prospects in circuit design and optimization.

[0075] like Figure 4 As shown, Figure 4 This is a schematic diagram illustrating circuit behavior modeling and prediction. (In the context of...) Figure 2 After the current mirror circuit is converted into a circuit behavior graph, it is input into a graph neural network (GNN) for training to obtain the behavior features as global feature information. An attention mechanism is further introduced to aggregate the features to better capture the dependencies between physical parameters. During prediction, the aggregated behavior features are graph embedded and input into a feedforward neural network with a multilayer perceptron mechanism to achieve performance prediction of the circuit behavior graph and obtain the performance prediction results.

[0076] Those skilled in the art will understand that all or part of the features / steps of the above-described method embodiments can be implemented by methods, data processing systems, or computer programs. These features may be implemented without hardware, entirely in software, or in a combination of hardware and software. The aforementioned computer program may be stored in one or more computer-readable storage media. When the computer program is executed (e.g., by a processor), it performs the steps of the above-described embodiments of the intelligent circuit design modeling method based on analog circuit behavior diagrams for robot motion prediction and the robot motion prediction method.

[0077] The aforementioned storage media capable of storing program code include: static hard disks, solid-state hard disks, random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), optical storage devices, magnetic storage devices, flash memory, magnetic disks or optical disks, and / or combinations of the above devices, that is, they can be implemented by any type of volatile or non-volatile storage devices or combinations thereof.

[0078] Example 2: This application also provides an intelligent circuit design device based on analog circuit behavior graphs, comprising: one or more processors; a memory for storing one or more computer programs, wherein the one or more processors are configured to execute the one or more computer programs stored in the memory, so that the one or more processors execute an intelligent circuit design modeling method and a performance prediction method based on analog circuit behavior graphs as described in any one of Embodiment 1.

[0079] The intelligent circuit design device based on analog circuit behavior diagrams of the present invention can be a chip. When the intelligent circuit design device based on analog circuit behavior diagrams is a chip, each module in the chip can be implemented entirely or partially through software, hardware, or a combination thereof. The above modules can be embedded in the processor of the computing device in hardware form or independent of it, or they can be stored in the memory of the computing device in software form, so that the processor can call and execute the operations corresponding to the above modules.

[0080] This application also provides an electronic device, which includes an intelligent circuit design device based on analog circuit behavior diagrams. The electronic device can be a laptop, iPad, mobile terminal, etc.

[0081] Since the implementation of the electronic device is described in detail in Embodiment 1, it will not be repeated here.

[0082] The above description is merely a preferred embodiment of the present invention. Those skilled in the art will understand that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the present invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A circuit design intelligent modeling method based on analog circuit behavior diagram, characterized in that: include: The electrical characteristics of the transistor in the analog circuit are characterized by using a large signal model and a small signal model, respectively, and the electrical parameters of each electronic component of the analog circuit are extracted through simulation, wherein the transistor is one of the electronic components; The electrical parameters of each electronic component are embedded into a circuit topology diagram corresponding to the analog circuit, each electronic component is used as a node in a circuit behavior diagram, and the physical connection between the electronic components is used as an edge of the circuit behavior diagram to generate a circuit behavior diagram; A graph neural network is used to perform graph convolution operations on the behavioral characteristics of the circuit behavior graph to obtain global feature information of the circuit behavior graph.

2. The intelligent modeling method for circuit design based on analog circuit behavior diagram according to claim 1, characterized in that: The method of using a large signal model and a small signal model to respectively characterize the electrical characteristics of transistors in the analog circuit includes: A large signal model is used to characterize the first electrical characteristics of the transistor in the nonlinear operating range. The basic parameters of the large signal model include: drain-source voltage and overdrive voltage ; The drain-source voltage Characterizes the voltage difference between the source and drain of the transistor, the overdrive voltage Characterize the difference between the gate-source voltage and the threshold voltage; A small signal model is used to characterize the second electrical characteristic of the transistor under AC small signal excitation, wherein the basic parameters of the small signal model include transconductance and the turning frequency The transconductance Characterizes the rate at which the drain current of the transistor changes with the gate-source voltage, the turning frequency Indicates the high frequency response capability of the transistor.

3. The intelligent modeling method for circuit design based on analog circuit behavior diagram according to claim 2, characterized in that: The extracting electrical parameters of each electronic component of the analog circuit by simulation includes: Identify the terminals of each electronic component in the analog circuit and the operating voltage in the analog circuit ; The large-signal parameter extraction is performed on the transistor to obtain the DC operating point of the drain-source voltage, the real-time calculated value of the over-drive voltage and the nonlinear transfer function curve of the transconductance; the small-signal parameter extraction of the transistor is performed simultaneously to obtain the frequency-varying characteristics of the gain-bandwidth product and the output impedance based on AC scanning; the basic parameter calibration is performed on the resistor, capacitor and inductor elements in the analog circuit, and the basic parameters include the ohmic characteristics of the resistor, the frequency response of the capacitor and the critical value of the magnetic saturation of the inductor.

4. The intelligent modeling method for circuit design based on analog circuit behavior diagram according to claim 3 is characterized in that: The method of using each of the electronic components as a node in a circuit topology diagram includes: The terminals of the electronic components and the operating voltage in the analog circuit Mapped to nodes in the circuit behavior graph.

5. The intelligent modeling method for circuit design based on analog circuit behavior diagram according to claim 1, characterized in that: The using of a graph neural network to perform a graph convolution operation on the behavior characteristics of the circuit behavior graph to obtain global feature information of the circuit behavior graph includes: The electrical characteristics of each of the nodes are represented by a feature vector as the input of the graph neural network, so that each layer of the graph neural network can perform feature aggregation according to the feature vectors of the current node and neighboring nodes, and perform message transmission and feature update during the feature aggregation process.

6. The intelligent modeling method for circuit design based on analog circuit behavior diagram according to claim 5, characterized in that: Each layer of the graph neural network performs feature aggregation based on the feature vectors of the current node and neighboring nodes, including: Using the following formula, in the lth layer of the graph neural network, the feature vector of the current node v is updated according to the feature vector of the neighboring node u: ； in, represents the learnable weight matrix, represents the activation function, Represents an aggregation function, used to combine the feature vector of the current node v and the feature vector of neighbor node u , and the edge attributes between the current node v and the neighbor node u .

7. The intelligent modeling method for circuit design based on analog circuit behavior diagram according to claim 6, characterized in that: The method further comprises: Through the attention mechanism, different weights are assigned to each pair of current node u and neighbor node v during feature aggregation and message passing through the following formula: ； in, represents the learnable weight matrix for the query used for the attention mechanism, A learnable weight matrix representing a key for the attention mechanism, wherein the query represents the degree of attention of the current node v to the neighbor node u, and the key represents a feature of the neighbor node u; Represents the feature vector between the current node v and the neighbor node u and When similarity is calculated by transforming the query and key matrices, the attention weight between the current node v and the neighbor node u is, It represents the processing function of edge attributes, and LeakyReLU represents the activation function.

8. A performance prediction method based on a simulated circuit behavior diagram, characterized in that: include: Through a graph embedding operation, the global feature information obtained by the intelligent modeling method for circuit design based on the analog circuit behavior graph according to any one of claims 1 to 7 is converted into a graph embedding vector, wherein the graph embedding vector is a vector of fixed dimension; In a feedforward neural network, a nonlinear transformation is performed on the graph embedding vector to obtain a performance prediction value of the circuit behavior graph, so as to output a performance prediction result of the circuit behavior graph.

9. An intelligent circuit design device based on analog circuit behavior diagram, characterized in that: include: one or more processors; A memory for storing one or more computer programs, and one or more processors for executing one or more computer programs stored in the memory, so that one or more processors execute the intelligent modeling method for circuit design based on analog circuit behavior diagrams as described in any one of claims 1 to 7 and the performance prediction method based on analog circuit behavior diagrams as described in claim 8.

10. A computer program product, characterized in that The computer program product is stored on a data carrier and is designed to execute the intelligent modeling method for circuit design based on the analog circuit behavior diagram according to any one of claims 1 to 7 and the performance prediction method based on the analog circuit behavior diagram according to claim 8.

Images