Machine learning-based de-embedding method for radio frequency circuit and device and related equipment
By using a machine learning-based method and a neural network prediction model, the problem of high-frequency and high-precision de-embedding methods in the existing technology relying on a large number of calibration components is solved, achieving a high-precision, low-cost and widely applicable de-embedding effect.
Patent Information
- Application Number
- CN202510631897.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-16
- Publication Date
- 2025-09-26
AI Technical Summary
The high-frequency and high-precision de-embedding methods in the existing technology rely on a large number of calibration components, which increases processing costs and testing difficulty, and has poor versatility and cannot be applied to specific circuit structures and processes.
A machine learning-based approach is used to design parameterized models of the test structure and the structure to be predicted, generate data sets, and train a neural network prediction model. A small number of calibration components are used to achieve high-frequency and high-precision de-embedding, which is applicable to a variety of circuit structures and processes.
It achieves high-frequency and high-precision de-embedding, reduces the number of calibration parts, lowers processing and testing costs, and is widely applicable to different circuit structures and processes, improving test efficiency and accuracy.
Smart Images
Figure CN120706340A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of instruments and measurements, and in particular to a de-embedding method for radio frequency circuits and devices based on machine learning and related equipment. Background Art
[0002] Accurate de-embedding methods are crucial for achieving precise modeling of devices and circuits. This is because, whether in board-level or on-chip circuit testing scenarios, the designed device under test (DUT) is not directly connected to the test instrument's ports. Instead, it requires additional fixture circuitry (fixtures) to connect to the test instrument. Therefore, during testing, the effects of the fixtures must be removed to restore the DUT's true values. This process is known as de-embedding.
[0003] Conventional de-embedding methods primarily include the equivalent circuit method and the cascade method. The former primarily includes the open-short method and its derivatives, such as the open-short-load method and the open-open-short-short method. The latter primarily includes the through-method and the TRL method. However, both methods have numerous shortcomings. First, their de-embedding effectiveness is closely related to the number of calibration components. The larger the de-embedding frequency range and the higher the accuracy, the more calibration components are required, which undoubtedly significantly increases the processing cost of high-frequency, high-precision de-embedding. Second, the large number of calibration components increases testing difficulty, requiring testers to repeatedly insert needles to ensure consistent test results, which consumes considerable time and increases testing costs. Finally, the circuit equivalence assumptions made by conventional de-embedding methods regarding fixtures often only apply to specific circuit structures or processes, making them less universal. Summary of the Invention
[0004] To at least partially address one of the technical issues existing in the prior art, the present invention aims to provide a machine learning-based de-embedding method and related equipment for RF circuits and devices. This method combines high-frequency, high-precision de-embedding capabilities with strong versatility, while requiring only a small number of calibration components, significantly reducing processing and testing costs. Furthermore, the present invention has broad applicability and is not limited to specific circuit structures or processes.
[0005] The first technical solution adopted by the present invention is:
[0006] A de-embedding method for radio frequency circuits and devices based on machine learning, comprising the following steps:
[0007] Design test structures and structures to be predicted;
[0008] Establish parametric models of the test structure and the structure to be predicted;
[0009] Perform parameter sweep or Monte Carlo analysis on the established test structure and the structure model to be predicted to generate a data set;
[0010] Build a neural network prediction model and train the model using the generated dataset;
[0011] Input the measured data of the test structure into the trained neural network prediction model to output the measured data of the structure to be predicted;
[0012] The measured data of the test structure and the structure to be predicted are processed according to the type of the structure to be predicted to obtain the measured results of the DUT.
[0013] Furthermore, in addition to the DUT test structure, the test structure also includes one or more of a through test structure, a line test structure, a load test structure, or a reflection test structure. For example, when the structure to be predicted is the DUT, only the DUT test structure is usually required; when the structure to be predicted is the fixture, only the DUT test structure and the through test structure are usually required.
[0014] Furthermore, the structure to be predicted is a fixture or a DUT.
[0015] Furthermore, the parameterized models of the test structure and the structure to be predicted are a combination of one or more of an electromagnetic simulation model, an equivalent circuit model, a mathematical fitting model or a neural network model.
[0016] Furthermore, the generated data set consists of test structure data samples and to-be-predicted structure data samples.
[0017] Furthermore, the measured data of the test structure and the structure to be predicted are processed according to the type of the structure to be predicted, including:
[0018] When the structure to be predicted is a fixture, the cascade de-embedding formula is used to calculate the measured results of the DUT: Among them, T DUT 、T LF 、T RF and T MEAS Represents the T parameter matrix of DUT, left fixture, right fixture and DUT test structure respectively;
[0019] When the structure to be predicted is DUT, the output data of the neural network prediction model is the actual measurement result of DUT.
[0020] The second technical solution adopted by the present invention is:
[0021] A neural network prediction model is obtained by training the method described above, wherein the input is the test structure data and the output is the structure data to be predicted.
[0022] The third technical solution adopted by the present invention is:
[0023] A computer program product or computer program includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the aforementioned machine learning-based de-embedding method for radio frequency circuits and devices.
[0024] The fourth technical solution adopted by the present invention is:
[0025] A computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement a machine learning-based de-embedding method for radio frequency circuits and devices as described above; or the storage medium storing the neural network prediction model as described above.
[0026] The fifth technical solution adopted by the present invention is:
[0027] An electronic device, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the above-mentioned machine learning-based de-embedding method for radio frequency circuits and devices; or the storage medium stores the above-mentioned neural network prediction model.
[0028] The beneficial effects of the present invention are: the present invention provides a complete set of de-embedding solutions for RF circuits and devices based on machine learning, covering data set generation and neural network prediction model training. By training the neural network model to learn the intrinsic mapping relationship between test structure data and structure data to be predicted, high-frequency and high-precision de-embedding can be achieved without relying on a large number of calibration parts, and it has a wide range of applicable scenarios and is not limited to specific circuit structures and processes. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following introduction is made to the drawings of the embodiments of the present invention or the related technical solutions in the prior art. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative work.
[0030] Figure 1 This is a flow chart of a de-embedding method for radio frequency circuits and devices based on machine learning in Example 1 of the present invention;
[0031] Figure 2 Schematic diagram of a through-test structure and a fixture structure to be predicted in Example 2 of the present invention;
[0032] Figure 3 Schematic diagram of the equivalent circuit topology of the through-test structure in Example 2 of the present invention;
[0033] Figure 4 This is a diagram showing the S-parameter fitting effect of the equivalent circuit model of the through-test structure in Example 2 of the present invention;
[0034] Figure 5 This is a diagram showing the S-parameter fitting effect of the equivalent circuit model of the structure to be predicted for Fixture in Example 2 of the present invention;
[0035] Figure 6 Schematic diagram of the distribution of data sets generated by Monte Carlo analysis of the equivalent circuit model of the through-test structure in Example 2 of the present invention;
[0036] Figure 7 1 is a diagram showing the prediction effect of the neural network prediction model for the fixture structure to be predicted in Example 2 of the present invention, wherein the solid line is the model prediction result and the dotted line is the electromagnetic simulation result;
[0037] Figure 8 Schematic diagram of the prediction accuracy of the neural network prediction model for the fixture structure to be predicted in Example 2 of the present invention;
[0038] Figure 9 Schematic diagram of a through-test structure and a fixture to-be-predicted structure in Example 3 of the present invention;
[0039] Figure 10 Schematic diagram of the network cascade topology of the through test structure in Example 3 of the present invention;
[0040] Figure 11 This is a diagram showing the S-parameter fitting effect of the polynomial model of the through-test structure in Example 3 of the present invention;
[0041] Figure 12 This is a diagram showing the S-parameter fitting effect of the polynomial model of the structure to be predicted for Fixture in Example 3 of the present invention;
[0042] Figure 13 3. This is a schematic diagram of the distribution of a data set generated by Monte Carlo analysis of an equivalent circuit of a through-test structure in Example 3 of the present invention;
[0043] Figure 143 is a diagram showing the prediction effect of the neural network prediction model in Example 3 of the present invention, wherein the solid line is the model prediction result and the dotted line is the electromagnetic simulation result;
[0044] Figure 15 Schematic diagram of the prediction accuracy of the neural network prediction model in Example 3 of the present invention. DETAILED DESCRIPTION
[0045] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application and are not to be construed as limiting the present application. For the step numbers in the following embodiments, they are provided only for the convenience of explanation and are not intended to limit the order of the steps. The order of execution of the steps in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.
[0046] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments and are not intended to limit the embodiments of the present application. The singular forms of "a", "said", and "the" used in the embodiments of the present application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. In addition, unless otherwise clearly defined, words such as setting, installing, and connecting should be understood in a broad sense, and those skilled in the art can reasonably determine the specific meanings of the above words in the present invention in combination with the specific content of the technical solution.
[0047] In the description of this application, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, they cannot be understood as limitations on this application.
[0048] In the description of this application, "several" means one or more, "many" means more than two, "greater than," "less than," and "exceed" are understood to exclude the number itself, while "above," "below," and "within" are understood to include the number itself. The terms "first" and "second" are used solely to distinguish technical features and are not to be construed as indicating or implying relative importance, or as implicitly specifying the number or order of the technical features indicated.
[0049] In the description of this application, "and / or" describes the relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally indicates that the associated objects are in an "or" relationship.
[0050] Example 1:
[0051] like Figure 1 As shown, the embodiments of the present application provide a de-embedding method for RF circuits and devices based on machine learning. By training a neural network model to learn the intrinsic mapping relationship between test and predicted structure data, high-frequency and high-precision de-embedding can be achieved using a small number of calibration components. The method is applicable to a wide range of scenarios and is not limited to specific circuit structures and processes. The method includes the following steps:
[0052] S1. Design the test structure and the structure to be predicted.
[0053] Specifically, in addition to the DUT test structure, the test structure may also include one or more of a through test structure, a line test structure, a load test structure, and a reflection test structure; for example, when the structure to be predicted is the DUT, usually only the DUT test structure is required; when the structure to be predicted is the Fixture, usually only the DUT test structure and the through test structure are required.
[0054] The structure to be predicted is a fixture or a DUT.
[0055] S2. Establish parameterized models of the test structure and the structure to be predicted.
[0056] Optionally, the parameterized models of the test structure and the structure to be predicted may be a combination of one or more of an electromagnetic simulation model, an equivalent circuit model, a mathematical fitting model, and a neural network model.
[0057] S3. Perform parameter scanning or Monte Carlo analysis on the established test structure and the structure model to be predicted to generate a data set.
[0058] Specifically, the generated data set consists of test structure data samples and to-be-predicted structure data samples.
[0059] S4. Establish a neural network prediction model and use the generated data set to train the model.
[0060] S5. Input the measured data of the test structure into the trained neural network prediction model to output the measured data of the structure to be predicted.
[0061] S6. Process the measured data of the test structure and the structure to be predicted according to the type of the structure to be predicted to obtain the measured result of the DUT.
[0062] In order to obtain the actual measurement results of the device under test, the data processing is divided into two situations:
[0063] When the structure to be predicted is a fixture, the cascade de-embedding formula is used to calculate the measured results of the DUT: Among them, T DUT 、T LF 、T RF and T MEAS They represent the T parameter matrix of DUT, left fixture, right fixture and DUT test structure respectively; when the structure to be predicted is DUT, the output data of the neural network prediction model is the actual measured result of DUT.
[0064] As an implementation method, the present application embodiment selects Fixture as the structure to be predicted, and the predicted data is the S parameter frequency response, with the frequency range selected as 0.1 to 67 GHz. In order to verify the feasibility of the method, two prediction examples of the fixture to be predicted structure are disclosed. After the fixture to be predicted structure is predicted, the cascaded de-embedding formula can be further used. Calculate the measured results of the DUT, where T DUT 、T LF 、T RF and T MEAS The T parameter matrices represent the DUT, left fixture, right fixture, and DUT test structure respectively. The T parameters can then be converted into S parameters using the network parameter change formula.
[0065] Example 2:
[0066] Please also refer to Figure 2-8 This embodiment discloses a prediction case of a transmission line type fixture structure to be predicted, and the test structure is a straight-through structure.
[0067] like Figure 2 As shown, the transmission line fixture to be predicted structure is half of the through test structure, wherein the line width W1 of the transmission line fixture is 20um, the line length L1 is 240um, the substrate is gallium arsenide, the dielectric constant is 12.9, and the substrate thickness is 75um.
[0068] In terms of parameterized modeling, this embodiment establishes a lumped equivalent circuit model. Figure 3 The lumped equivalent circuit model topology of the through-test structure is shown. i and z i The π-type network formed is the transmission line fixture equivalent circuit, y em and z emThe π-type network formed is the equivalent circuit of the additional electromagnetic field in the through-test structure. Electromagnetic simulation is used to extract the network parameters of the additional electromagnetic field in the fixture to be predicted structure and the through-test structure, and vector fitting technology is used to establish the y i 、z i 、y em and z em The lumped equivalent circuit model of Figure 4 and Figure 5 As shown in the figure, the established equivalent circuit model has a good fitting effect on both the through-test structure and the fixture structure to be predicted.
[0069] In terms of data set generation, this embodiment uses Monte Carlo analysis. After applying + / -10% and uniformly distributed Monte Carlo analysis to all parameters in the lumped equivalent circuit model, the distribution of 1001 through-test structure samples is as follows: Figure 6 The deviations of these samples gradually increase with increasing frequency, which is consistent with actual physical phenomena.
[0070] In terms of the neural network prediction model, this embodiment uses a deep neural network. The deep neural network has a total of 9 hidden layers, each of which is a fully connected layer, and the activation function uses the hyperbolic tangent function. The input of the deep neural network is the S parameters of the through-test structure, and the output is the S parameters of the fixture structure to be predicted. Before training, the data set is first normalized to the interval [-1, 1]. During the training process, the optimizer uses the Adam optimizer, the initial learning rate is set to 0.001, the number of epochs is set to 10,000, and the learning rate is reduced to 0.98 times the original after every 50 epochs. The validation set is generated by electromagnetic simulation parameter scanning. The W1 parameter scanning range is 18-22um, the step is 0.5um, and the L1 parameter scanning range is 216-264um, the step is 6um. Therefore, the validation set has a total of 81 samples.
[0071] Figure 7 and Figure 8 The S-parameter prediction performance and prediction accuracy of the fixture structure to be predicted in this embodiment are demonstrated. Prediction accuracy is represented by the coefficient of determination; the closer the coefficient of determination is to 1, the higher the prediction accuracy. Considering the bilaterally symmetrical structure of the transmission line fixture, only S11 and S12 need to be verified. It can be seen that this embodiment achieves prediction accuracy exceeding 0.99 for both the real and imaginary parts, achieving accurate prediction.
[0072] Example 3:
[0073] Please also refer to Figure 9-15 This embodiment discloses a prediction case of a GSG-type fixture structure to be predicted, and the test structure is straight-through.
[0074] like Figure 9 As shown, the GSG type fixture structure to be predicted is half of the through test structure, where the line width W2 of the GSG type fixture is 20um, the line length L2 is 120um, the width of the signal pad is W3=60um, the width of the ground pad is W4=80um, the substrate is gallium arsenide, the dielectric constant is 12.9, and the substrate thickness is 75um.
[0075] In terms of parameterized modeling, this embodiment establishes a polynomial mathematical fitting model. Figure 10 The network cascade topology of the through-test structure is shown. Due to the discontinuity between the GSG pad and the metal interconnection line, each fixture consists of a cascade of the GSG network (GSG), the discontinuity network (Dis.) and the metal interconnection network (Int.). The through-test structure is composed of the Left Fixture, the extra electromagnetic field (Extra EM) in the through-test structure and the Right Fixture cascade. Electromagnetic simulation is used to extract the network parameters of the GSG pad, discontinuity, metal interconnection line and extra electromagnetic field, and a polynomial fitting algorithm is used to extract the polynomial coefficients, such as the polynomial_fit() function in Python. Figure 11 and Figure 12 As shown in the figure, the established polynomial mathematical fitting model has a good fitting effect on both the through-test structure and the fixture structure to be predicted.
[0076] In terms of data set generation, this embodiment uses Monte Carlo analysis. After applying + / -10% and uniformly distributed Monte Carlo analysis to all parameters in the lumped equivalent circuit model, the distribution of 1001 through-test structure samples is as follows: Figure 13 The deviations of these samples gradually increase with increasing frequency, which is consistent with actual physical phenomena.
[0077] In terms of the neural network prediction model, this embodiment uses a one-dimensional convolutional neural network. This one-dimensional convolutional neural network has two layers: a one-dimensional convolutional layer as the input layer and a fully connected layer as the output layer. The activation function is the hyperbolic tangent function. The input of this deep neural network is the S parameters of the through-test structure, and the output is the S parameters of the fixture structure to be predicted. Before training, the dataset is normalized to the interval [-1, 1]. During training, the Adam optimizer is used as the optimizer, with an initial learning rate set to 0.001 and the number of epochs set to 10,000. The learning rate is reduced by 0.98 times after every 50 epochs. The validation set is generated by electromagnetic simulation parameter sweeps. The parameter sweep range of W2 is 18-22 μm with a step of 0.5 μm, the parameter sweep range of W3 is 55-65 μm with a step of 5 μm, and the parameter sweep range of W4 is 72-88 μm with a step of 8 μm. L2 is fixed at 120 μm, resulting in a total of 81 samples in the validation set.
[0078] Figure 14 and Figure 15 The S-parameter prediction performance and prediction accuracy of the fixture structure to be predicted in this embodiment are demonstrated. Prediction accuracy is represented by the coefficient of determination; the closer the coefficient of determination is to 1, the higher the prediction accuracy. Considering the GSG fixture's asymmetrical structure, S11, S12, and S22 need to be verified. It can be seen that this embodiment achieves prediction accuracy exceeding 0.99 for both the real and imaginary parts of all three parameters, achieving accurate prediction.
[0079] In summary, the method of the present invention can achieve accurate prediction of the fixture network parameters when using a small number of calibration parts (only one through-part), and the fixture network parameters can be further used to de-embed the network parameters of the test structure. On the other hand, the method proposed in the present invention shows excellent robustness, that is, the trained neural network prediction model can achieve accurate prediction under a wide range of fluctuations in the physical parameters of the test structure and the structure to be predicted. The de-embedding method for RF circuits and devices based on machine learning provided by the present invention has both high-frequency and high-precision de-embedding capabilities and strong versatility, while only requiring a small number of calibration parts, which greatly reduces the processing and testing costs.
[0080] Example 4:
[0081] A neural network prediction model is obtained by training the method described above, wherein the input is the test structure data and the output is the structure data to be predicted.
[0082] In this embodiment, the neural network prediction model can be implemented using a deep neural network, a convolutional neural network, or other neural networks. This neural network prediction model can accurately predict the physical parameters of the test structure and the structure to be predicted under a wide range of fluctuations, and has the same functions and technical effects as the above-mentioned embodiments, which are not described in detail here.
[0083] Example 5:
[0084] In some possible implementations, various aspects of the method of the embodiments of the present invention may also be implemented in the form of a program product, which includes program code. When the program product is run on a computer device, the program code is used to cause the computer device to perform the steps of a machine learning-based de-embedding method for radio frequency circuits and devices according to various exemplary embodiments of the present application as described above in this specification. The executable computer program code or "code" used to perform various embodiments may be written in a high-level programming language such as C, C++, Python, Smalltalk, Java, JavaScript, Visual Basic, structured query language (e.g., Transact-SQL), Perl, or in various other programming languages.
[0085] Example 6:
[0086] An embodiment of the present invention further provides a computer-readable storage medium, wherein the storage medium stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by a processor to implement the following Figure 1 A de-embedding method for RF circuits and devices based on machine learning is shown.
[0087] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program. The program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.
[0088] Since the storage medium is a storage medium corresponding to a de-embedding method for radio frequency circuits and devices based on machine learning in an embodiment of the present invention, and the principle of solving the problem by the storage medium is similar to that of the method, the implementation of the storage medium can refer to the implementation process of the above-mentioned method embodiment, and the repeated parts will not be repeated.
[0089] Example 7:
[0090] An embodiment of the present invention further provides an electronic device, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the following Figure 1 A de-embedding method for RF circuits and devices based on machine learning is shown.
[0091] It is understood that the memory may include random access memory (RAM) or read-only memory (ROM). Optionally, the memory includes a non-transitory computer-readable storage medium. The memory may be used to store instructions, programs, codes, code sets, or instruction sets. The memory may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function, instructions for implementing the various method embodiments described above, etc.; the data storage area may store data created based on the use of the server, etc.
[0092] The processor may include one or more processing cores. The processor utilizes various interfaces and circuits to connect various components within the server. It executes various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory, as well as accessing data stored in memory. Optionally, the processor may be implemented using at least one of the following hardware forms: digital signal processing (DSP), field-programmable gate array (FPGA), and programmable logic array (PLA). The processor may integrate one or a combination of a central processing unit (CPU) and a modem. The CPU primarily processes the operating system and application programs, while the modem handles wireless communications. It is understood that the modem may not be integrated into the processor and may be implemented separately via a single chip.
[0093] Since the electronic device is an electronic device corresponding to a de-embedding method for radio frequency circuits and devices based on machine learning in an embodiment of the present invention, and the principle of solving the problem by the electronic device is similar to that of the method, the implementation of the electronic device can refer to the implementation process of the above-mentioned method embodiment, and the repeated parts will not be repeated.
[0094] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0095] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.
[0096] The above embodiments are intended only to illustrate the technical concepts and features of the present invention. Their purpose is to enable those skilled in the art to understand the contents of the present invention and implement them accordingly. They are not intended to limit the scope of protection of the present invention. Any equivalent changes or modifications made based on the essence of the present invention are intended to be covered by the scope of protection of the present invention.
Claims
1. A de-embedding method for radio frequency circuits and devices based on machine learning, characterized in that: The following steps are involved: Design test structures and structures to be predicted; Establish parametric models of the test structure and the structure to be predicted; Perform parameter sweep or Monte Carlo analysis on the established test structure and the structure model to be predicted to generate a data set; Build a neural network prediction model and train the model using the generated dataset; Input the measured data of the test structure into the trained neural network prediction model to output the measured data of the structure to be predicted; The measured data of the test structure and the structure to be predicted are processed according to the type of the structure to be predicted to obtain the measured results of the DUT.
2. The de-embedding method for radio frequency circuits and devices based on machine learning according to claim 1, characterized in that: In addition to the DUT test structure, the test structure also includes one or more of a through test structure, a line test structure, a load test structure or a reflection test structure.
3. The de-embedding method for radio frequency circuits and devices based on machine learning according to claim 1, characterized in that: The structure to be predicted is a fixture or a DUT.
4. The de-embedding method for radio frequency circuits and devices based on machine learning according to claim 1, characterized in that: The parameterized models of the test structure and the structure to be predicted are a combination of one or more of an electromagnetic simulation model, an equivalent circuit model, a mathematical fitting model or a neural network model.
5. The de-embedding method for radio frequency circuits and devices based on machine learning according to claim 1, characterized in that: The generated data set consists of test structure data samples and to-be-predicted structure data samples.
6. The de-embedding method for radio frequency circuits and devices based on machine learning according to claim 1, characterized in that: The measured data of the test structure and the structure to be predicted are processed according to the type of the structure to be predicted, including: When the structure to be predicted is a fixture, the cascade de-embedding formula is used to calculate the measured results of the DUT: Among them, T DUT 、T LF 、T RF and T MEAS Represents the T parameter matrix of DUT, left fixture, right fixture and DUT test structure respectively; When the structure to be predicted is DUT, the output data of the neural network prediction model is the actual measurement result of DUT.
7. A neural network prediction model, characterized in that: The method is obtained by training according to any one of claims 1 to 5, wherein the input is the test structure data and the output is the structure data to be predicted.
8. A computer program product, characterized in that The computer program product comprises computer instructions, which are used to perform the method according to any one of claims 1 to 6 when executed by a processor.
9. A computer-readable storage medium, characterized in that The storage medium stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by the processor to implement the method according to any one of claims 1 to 6; or the storage medium stores the neural network prediction model according to claim 7.
10. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by the processor to implement the method according to any one of claims 1 to 6; or, the memory stores the neural network prediction model according to claim 7.