Simulation analysis method and system for tested device

Abstract

The invention discloses a simulation analysis method and system for a tested device. The method comprises the following steps: obtaining a predicted simulation value of each standard circuit unit contained in a tested device under a current process corner through a machine learning model; and obtaining a simulation value of the tested device under the current process corner by using the predicted simulation value of each standard circuit unit contained in the tested device under the current process corner. According to the embodiment of the invention, the machine learning model is applied to the simulation analysis software, the simulation values of the tested device under various process corners can be quickly obtained, the resource and time consumption caused by generating a standard cell library for each process corner is avoided, and the simulation analysis efficiency can be greatly improved.

Description

Technical Field

[0001] The present application relates to the field of integrated circuit design, and in particular to a simulation analysis method and system for a device under test. Background Art

[0002] Standard Cell Design is an important digital integrated circuit design method. Figure 1 As shown, the designer uses the design software to utilize the pre-designed standard circuit units (logic gates, triggers, etc.) in the standard unit library to construct a device under test 110 having a complex device structure, and then uses the simulation analysis software 120 to analyze the circuit structure of the device under test 110, and obtains the simulation value of the device under test 110 (such as the device delay value) based on the various standard circuit units included in the device under test 110. After the device under test is manufactured, the actual value of the device is tested by the machine 130, and the gap between the simulation value and the actual value is compared. The simulation environment parameters and simulation data in the simulation analysis software 120 are adjusted according to the gap, so that the simulation value is close enough to the actual value.

[0003] However, the definition of the simulation value of each standard circuit unit in the standard cell library (for example, the delay value of the standard circuit unit) is based on the change of the process corner, and different process corners correspond to different standard cell libraries. The process corner refers to a set of specific working conditions defined by parameter changes caused by process changes during the semiconductor manufacturing process. These parameters include power supply voltage, temperature and process. Since the parameter values ​​in the process corner are often frequently adjusted during actual measurement by the machine to obtain the actual measured values ​​of the device under test under multiple different process corners, it is necessary to use the K library tool 110 to obtain the standard cell libraries under the corresponding multiple different process corners, and then obtain the simulation value of the device under test 110 according to the standard circuit units under the corresponding multiple different process corners through the simulation software. In this process, since multiple standard cell libraries need to be obtained, the time occupied is often long, resulting in a sharp increase in the time of simulation analysis, and it is impossible to quickly obtain the corresponding simulation value based on the actual measured value. Summary of the invention

[0004] The purpose of the embodiments of the present disclosure is to provide a simulation analysis method and system for a device under test, so as to quickly obtain a simulation value of the device under test.

[0005] In a first aspect, an embodiment of the present disclosure provides a simulation analysis method for a device under test, wherein the device under test is composed of standard circuit cells pre-designed in a standard cell library, including: Obtaining predicted simulation values ​​of each standard circuit unit included in the device under test at the current process angle through a machine learning model; and The simulation value of the device under test at the current process angle is obtained by using the predicted simulation value of each standard circuit unit included in the device under test at the current process angle.

[0006] In some embodiments, the machine learning model is a support vector regression model.

[0007] In some embodiments, the evaluation function used by the machine learning model is a piecewise function related to the device type.

[0008] In some embodiments, the evaluation function used by the machine learning model is equation (3) or (4): = (3) = (4) in, ={ , , , , } represents a data sample, and, Respectively represent the device type, voltage, temperature, process and signal delay value of the corresponding data sample, , Indicates the specific device type.

[0009] In some embodiments, the evaluation function used by the machine learning model is constructed using equation (5): (5) In equation (5), the data samples at the center of the normal distribution under different devices are divided into k equal parts, and the mean square error of the data samples in the jth equal part is , is The corresponding weight, Represents the evaluation function of the i-th device type, where i, j, and k are all positive integers.

[0010] In some embodiments, the method further includes: obtaining a simulation data set for model training by taking the following steps: For each device type, the value ranges of multiple characteristic parameters involved in the current process corner are respectively divided or enumerated, and then multiple data samples in the input data set are formed based on the divided equal parts and the enumerated values; The multiple data samples in the input data set are respectively input into the simulation analysis software to obtain multiple simulation values ​​corresponding to the multiple data samples in the input data set. Each data sample in the input data set and the corresponding simulation value constitute a data sample of the simulation data set.

[0011] In some embodiments, the plurality of characteristic parameters corresponding to the input data set include: power supply voltage, temperature, and process, and the simulation value represents a signal delay value.

[0012] In some embodiments, the method further includes: removing data samples that do not conform to actual working conditions from the input data set.

[0013] In some embodiments, the method further includes: before the step of averaging or enumerating the value ranges of the characteristic parameters involved in the current process angle, normalizing the value ranges of the characteristic parameters involved in the current process angle.

[0014] In a second aspect, an embodiment of the present disclosure provides a simulation analysis system for a device under test, comprising: A prediction unit, used to obtain predicted simulation values ​​of each standard circuit unit included in the device under test under the current process angle through a machine learning model; The simulation unit is used to obtain the simulation value of the device under test at the current process angle by using the predicted simulation value of each standard circuit unit included in the device under test at the current process angle.

[0015] In a third aspect, an embodiment of the present disclosure provides a computer device, including: a memory, the memory being used to store a processing program; A processor, wherein the processor implements any one of the above-mentioned simulation analysis methods when executing the processing program.

[0016] In a fourth aspect, an embodiment of the present disclosure provides a readable storage medium, on which a processing program is stored, and when the processing program is executed by a processor, any one of the above-mentioned simulation analysis methods is implemented.

[0017] The simulation analysis method and system provided by the embodiments of the present disclosure apply machine learning models to simulation analysis, and can quickly obtain simulation values ​​of the device under test under various process corners, avoiding the resource and time consumption caused by generating a standard cell library for each process corner, thereby greatly improving the efficiency of simulation analysis. Furthermore, by training models related to device types in a targeted manner and using models related to device types to give simulation values, it is also helpful to improve the accuracy of simulation values. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and other objects, features and advantages of the embodiments of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which: Figure 1 It is a schematic diagram of the simulation analysis process under the prior art; Figure 2 is a schematic diagram of the simulation analysis process proposed in the embodiment of the present disclosure; Figure 3 It is a flowchart of the training process of the machine learning model provided by the embodiment of the present disclosure.

[0019] Figures 4 to 7 The matching status between the predicted delay values ​​and the actual delay values ​​of four exemplary device types are respectively given. DETAILED DESCRIPTION

[0020] The embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. In each of the accompanying drawings, the same elements are represented by similar reference numerals. For the sake of clarity, the various parts in the accompanying drawings are not drawn to scale. In addition, some well-known parts may not be shown.

[0021] The following describes the embodiments of the present disclosure based on the embodiments, but the embodiments of the present disclosure are not limited to these embodiments. In the detailed description of the embodiments of the present disclosure below, some specific details are described in detail. For those skilled in the art, the embodiments of the present disclosure can be fully understood without the description of these details. In order to avoid confusing the essence of the embodiments of the present disclosure, well-known methods, processes, processes, components and circuits are not described in detail.

[0022] Unless the context clearly requires otherwise, the words "include", "comprising" and the like throughout the specification and claims should be interpreted as inclusive rather than exclusive or exhaustive; that is, as "including but not limited to". In addition, in the description of the embodiments of the present disclosure, unless otherwise specified, "plurality" means two or more.

[0023] Figure 2 Schematic diagram of the simulation analysis process proposed in the embodiment of the present disclosure. Figure 2 As shown, and Figure 1Similarly, the designer uses the design software to construct a device under test 110 with a complex circuit structure. The trained machine learning model 150 can be embedded in the simulation analysis software 120. In the simulation analysis software 120, the machine learning model 150 is used to simulate and analyze the device under test 110 and obtain the simulation value of the device under test 110. Then, the actual measured value of the device under test 110 is obtained through the machine 130. The simulation parameters and data are adjusted by comparing the difference between the simulation value and the actual measured value. When the process angle 140 is adjusted, the adjusted process angle is input into the machine learning model 150 to obtain the simulation value of the device under test under the adjusted process angle. Then, the simulation parameters and data are adjusted by comparing the difference between the new simulation value and the new actual measured value, and so on.

[0024] exist Figure 2 In the example, the device under test 110 is composed of a plurality of standard circuit units, and the simulation prediction value of each standard circuit unit under the current process angle is obtained by the machine learning model 150, and then the simulation analysis software analyzes the simulation prediction value of each standard circuit unit to obtain the simulation value of the device under test 110 under the current process angle. Thus, the simulation analysis software can obtain the simulation values ​​of various circuit structures under various process angles. Since there may be multiple signal transmission paths of the device under test, each signal transmission path includes the signal transmission path of each standard unit included in the device under test (the signal propagation path refers to the path from the input port to the output port, that is, the causal relationship between the two ports in timing), so optionally, when the machine learning model is used to obtain the simulation prediction value of each standard circuit unit, the signal propagation path is also used as a parameter. Optionally, a database is established based on the result data of each standard unit obtained by machine learning, and then the database is searched by a search algorithm in the actual simulation to obtain the corresponding simulation prediction value.

[0025] In some embodiments, the current process angle includes power supply voltage, temperature, and process. The final output simulation value is a signal delay value. After obtaining the simulation prediction value of each standard circuit unit under the current process angle, the simulation analysis software calculates the signal delay value of the device under test in combination with the input transmission delay value (trans) and output load (load) of the actual circuit simulation.

[0026] It should be understood that the device under test 110 can be the standard circuit unit itself. In this case, the simulation analysis software directly uses the simulation prediction value of the standard circuit unit obtained by the machine learning model 150 at the current process angle as the simulation value of the device under test 110.

[0027] Thus, by applying machine learning models to Figure 2In the simulation analysis process of the device under test shown, the simulation value of the device under test can be quickly obtained while ensuring accuracy, and the resource and time occupation caused by repeated generation of standard cell libraries in the prior art can be avoided, thereby significantly improving the simulation analysis efficiency.

[0028] In order to better predict the simulation value of the device under test, the machine learning model 150 may include multiple models related to the device type, and the multiple models have their own independent model structure and model parameters, so that the simulation prediction value under each process angle is obtained by using the model corresponding to the device type corresponding to the standard circuit unit. At the same time, in order to better evaluate the prediction accuracy of the machine learning model, the evaluation function used is a piecewise function related to the device type corresponding to the standard circuit unit, that is, the multiple models included in the machine learning model are evaluated using different piecewise functions.

[0029] Figure 3 It is a flowchart of the training process of the machine learning model provided by the embodiment of the present disclosure.

[0030] In step S302, a simulation data set is constructed.

[0031] In this step, for the specified device type and the specified signal propagation path, an input data set is constructed according to the value range of multiple characteristic parameters involved in the current process angle, and then each data sample in the input data set is input into the simulation analysis software to obtain multiple simulation values, and each data sample in the input data set is respectively combined with the corresponding simulation value in the multiple simulation values ​​to form a data sample in the simulation data set. The specified device type characterizes the specific classification and function of the standard circuit unit. For example, the device type CKBD1 represents a buffer for clock tree construction, and the device type CKND1 represents an inverter for clock tree construction. Standard circuit units often include multiple input and output ports, and the specified signal propagation path indicates the signal transmission path from a specific input port to an output port. The multiple characteristic parameters involved in the current process corner include power supply voltage, temperature, and process. Each characteristic parameter can be set to a specific value range. For example, the power supply voltage range is 1.8~5V, the process range is enumerated as tt (Typ-Typ), ss (Slow-Slow), and ff (Fast-Fast), the temperature range is enumerated as -40c, 25c, 85c, 125c, 150c, and the power consumption range is 0.003~0.297 pF. The continuous value range (such as the power supply voltage range) can be divided into multiple equal parts, and then the equal parts and the enumerated values ​​are combined into multiple one-dimensional vectors, each of which is used as a data sample in the input data set. Moreover, before inputting the input data set into the simulation analysis software to obtain multiple simulation values, the input data set can also be normalized. Through normalization, the magnitude of the characteristic parameters is scaled to the same interval to improve the convergence speed and performance of the algorithm. The specified device type and the specified signal propagation path can be used as or not as elements of the input data set. When the specified device type and the specified signal propagation path are not used as elements of the input data set, different model structures and model parameters are constructed and trained according to different device types and signal propagation paths. In addition, the influence of different signal propagation paths in standard circuit units on the simulation values ​​can also be ignored, and different model structures and model parameters are constructed and trained only according to the device type, so as to improve the efficiency of model training.

[0032] For example, the simulation analysis software is selected as SPICE (Simulation Program with Integrated Circuit Emphasis), which is a software tool for simulating the behavior of integrated circuits and electronic circuits. It describes the characteristics of circuit components (such as resistors, capacitors, inductors, transistors, etc.) through mathematical models and simulates the behavior of circuit components under different conditions. Therefore, when each data sample of the input data set is provided to SPICE, it is equivalent to specifying the working conditions of each circuit component, and SPICE thus outputs the result data of each circuit component under the specified working conditions. As an example, each data sample in the simulation data set includes power supply voltage, temperature, process and signal delay value (output of the simulation analysis software). Further, the input data set in step S302 is preprocessed, mainly including removing data samples that exceed the actual working conditions to optimize the data quality. It is also possible to use a part (for example, 70%) of the simulation data set output in step S302 as a training data set, and another part (for example, 30%) as a verification data set.

[0033] In step S304, the machine learning model is trained according to the simulation data set.

[0034] Through the machine learning model, each simulation prediction value corresponding to each data sample of the simulation data set is obtained, and each simulation prediction value is compared with the simulation value in the simulation data set (for example, the signal delay value obtained by model prediction is compared with the signal delay value obtained by simulation analysis software), and the model parameters are adjusted according to the comparison results. During the training process, MSE and Construct an evaluation function. MSE represents the average value of the square of the difference between the simulated predicted value and the actual simulated value. The smaller the MSE value, the smaller the prediction error of the model and the better the prediction effect. The formula of MSE is shown in formula (1). The formula of (key coefficient) is shown in formula (2).

[0035] Formula (1) Where n is the number of samples, is the ith sample actually simulated, is the i-th sample predicted by the model.

[0036] Formula (2) in, represents the deviation between the actual simulation and the average sample, represents the residual sum of squares (prediction variance), is the average of the actual simulation.

[0037] In a further embodiment, the machine learning model used in the above embodiment includes multiple models related to device types, and the evaluation function takes a piecewise function related to the device type of the device under test, as shown in equations (3) and (4): = (3) = (4) ={ , , , , } represents a data sample, and, Respectively represent the device type, voltage, temperature, process and signal delay value of the corresponding data sample, , represents the specific device type. Formulas (3) and (4) can also be combined (for example / ) is used as the evaluation function of the machine learning model used in the above embodiment.

[0038] In a further embodiment, since it is found through actual measurement that the signal delay value has a normal distribution law under different processes (SS / FF / TT), temperature, and voltage, the evaluation function used by the machine learning model used in the above embodiment can also be constructed in combination with equation (5): (5) In equation (5), the data samples at the center of the normal distribution under different devices are divided into k equal parts, and the mean square error of the data samples in the jth equal part is , is The corresponding weight, Represents the evaluation function of the i-th device type, where i, j, and k are all positive integers.

[0039] According to the above evaluation function, when training the model, the model parameters that best suit the simulation data set are calculated for each device type, and the model parameters with the best score are obtained through each independent evaluation function. This design method is more targeted and has better relevance.

[0040] In some embodiments, the machine learning model in the above embodiments is Support Vector Regression (SVR). Support Vector Regression (SVR) is the application of Support Vector Machine (SVM) to regression problems. SVR performs regression by mapping the input space to a high-dimensional feature space and finding the optimal hyperplane in the high-dimensional space. Its goal is to find a hyperplane that minimizes the difference between the mapping value of the training sample on the hyperplane and the target value, while keeping the error within a certain range. SVR provides a powerful nonlinear regression analysis tool that can maintain prediction accuracy while having a certain tolerance for noise and outliers in the data. As an example, the open source SVR module is used for machine learning training, where GridSearchCV is used for autonomous iteration, the parameters are set to five-fold cross validation, the score evaluation is a piecewise function of "neg_mean_squared_error" related to the device type, n_jobs=regularization coefficient optimization range is regressor__C:[0.1, 1, 3, 5, 7, 10, 100], the gamma parameter range is selected from regressor__gamma: [0.01, 0.05,0.1, 0.5, 1], and the kernel function selection range is regressor__kernel: [rbf, sigmoid, linear]. neg_mean_squared_error represents the negative value of MSE, and the model parameters are solved by maximizing neg_mean_squared_error during the optimization process. The GridsearchCV process autonomously cross-trains the model according to the MSE piecewise function, thereby obtaining the model parameters with the best score and saving them as the best model related to the device type of the device under test. Table 1 shows an example of the results obtained in this way.

[0041] Table 1 Training parameters and training scores Device Type MSE (covariance) <![CDATA[R 2 (Key coefficient)]]> Best Parameters BestScore CKBD1 0.002518266 0.998199431 {'regressor__C': 1, 'regressor__gamma':0.1, 'regressor__kernel': 'rbf'} 0.002556383 CKMUX2D1 0.002590002 0.998455824 {'regressor__C': 10, 'regressor__gamma':0.1, 'regressor__kernel': 'rbf'} 0.002637318 CKND1 0.002318747 0.998707048 {'regressor__C': 10, 'regressor__gamma':0.1, 'regressor__kernel': 'rbf'} 0.002336784 ND2D1 0.00228704 0.998785511 {'regressor__C': 10, 'regressor__gamma':0.1, 'regressor__kernel': 'rbf'} 0.002300996

[0042] Figures 4 to 7 The matching status between the predicted delay values ​​predicted by the machine learning model and the simulated delay values ​​by the simulation analysis software for the four device types in Table 1 is given respectively. In the figure, the horizontal axis represents the simulated delay value and the vertical axis represents the predicted delay value. As shown in the figure, the matching status between the simulated delay values ​​and the predicted delay values ​​of the four device types can meet the target requirements.

[0043] According to the same concept as the above-mentioned embodiment, the embodiment of the present disclosure further provides a simulation analysis system, which includes a prediction unit and a simulation unit. The prediction unit is used to obtain the predicted simulation value of each standard circuit unit contained in the device under test at the current process angle through a machine learning model for a device under test. The simulation unit is used to obtain the simulation value of the device under test at the current process angle using the predicted simulation value of each standard circuit unit contained in the device under test at the current process angle.

[0044] In summary, the simulation analysis method and system provided above apply the machine learning model to the simulation analysis software, which can quickly obtain the simulation values ​​of the device under test under various process corners, avoiding the resource and time consumption caused by generating a standard cell library for each process corner, thereby greatly improving the simulation analysis efficiency. Furthermore, by training models related to device types in a targeted manner and using models related to device types to give simulation values, it is also helpful to improve the accuracy of simulation values.

[0045] In addition, an embodiment of the present disclosure further provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and each step in the above method is implemented when the processor executes the computer program.

[0046] Furthermore, the embodiments of the present disclosure further provide a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method are implemented.

[0047] The embodiments according to the embodiments of the present disclosure are described above, and these embodiments do not describe all the details in detail, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and changes can be made based on the above description. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of the present disclosure, so that those skilled in the art can make good use of the embodiments of the present disclosure and the modified use based on the embodiments of the present disclosure. The embodiments of the present disclosure are limited only by the claims and their full scope and equivalents.

Claims

1. A simulation analysis method for a device under test, wherein the device under test is composed of standard circuit cells pre-designed in a standard cell library, comprising: Obtaining predicted simulation values ​​of each standard circuit unit included in the device under test at the current process angle through a machine learning model; as well as The simulation value of the device under test at the current process angle is obtained by using the predicted simulation value of each standard circuit unit included in the device under test at the current process angle.

2. The simulation analysis method according to claim 1, wherein: The machine learning model is a support vector regression model.

3. The simulation analysis method according to claim 1, wherein: The evaluation function adopted by the machine learning model is a piecewise function related to the device type.

4. The simulation analysis method according to claim 3, wherein: The evaluation function used by the machine learning model is equation (3) or (4): = （3） = （4） in, ={ , , , , } represents a data sample, and, Respectively represent the device type, voltage, temperature, process and signal delay value of the corresponding data sample, , Indicates the specific device type.

5. The simulation analysis method according to claim 3, wherein: The evaluation function used in the machine learning model is constructed using equation (5): （5） In equation (5), the data samples at the center of the normal distribution under different devices are divided into k equal parts, and the mean square error of the data samples in the jth equal part is , is The corresponding weight, Represents the evaluation function of the i-th device type, where i, j, and k are all positive integers.

6. The simulation analysis method according to claim 1, further comprising: Use the following steps to obtain a simulation data set for model training: For each device type, the value ranges of multiple characteristic parameters involved in the current process corner are respectively divided or enumerated, and then multiple data samples in the input data set are formed based on the divided equal parts and the enumerated values; The multiple data samples in the input data set are respectively input into the simulation analysis software to obtain multiple simulation values ​​corresponding to the multiple data samples in the input data set. Each data sample in the input data set and the corresponding simulation value constitute a data sample of the simulation data set.

7. The simulation analysis method according to any one of claims 1 to 6, wherein: The multiple characteristic parameters corresponding to the input data set include: power supply voltage, temperature, and process, and the simulation value represents a signal delay value.

8. The simulation analysis method according to claim 6, further comprising: Data samples that do not conform to actual working conditions are removed from the input data set.

9. A simulation analysis system for a device under test, comprising: A prediction unit, used to obtain predicted simulation values ​​of each standard circuit unit included in the device under test under the current process angle through a machine learning model; The simulation unit is used to obtain the simulation value of the device under test at the current process angle by using the predicted simulation value of each standard circuit unit included in the device under test at the current process angle.

10. A computer device comprising: A memory, the memory being used to store a processing program; A processor, wherein when the processor executes the processing program, the simulation analysis method according to any one of claims 1 to 8 is implemented.

11. A readable storage medium having a processing program stored thereon, wherein the processing program, when executed by a processor, implements the simulation analysis method according to any one of claims 1 to 8.

Images