A parameter adaptive extraction method and system for a GaN HEMT equivalent circuit model

Abstract

The application discloses a kind of GaN HEMT equivalent circuit model parameter self-adapting extraction method and system, comprising: extracting parasitic parameter value by small signal equivalent circuit parasitic parameter extraction method, establishes initial population;Stripping parasitic parameter in initial population obtains each bias eigenvalue by network analysis, and simulation equivalent circuit S parameter data is obtained by calculation;Judgment is calculated according to S parameter data individual fitness;New individual is generated to replace singular individual in population until population is non-singular;According to the search range of adjustment algorithm of non-singular population;According to the individual selection, crossover, variation rule set to the individual in initial population is carried out genetic calculation, and new genetic individual is obtained;Whether the preset target is satisfied is judged, and the element parameter corresponding to optimal individual is output as the final parameter of small signal model;I-V model is optimized by genetic algorithm, and the weight matrix of fitness function is updated according to model error, and optimization is carried out again to meet the preset target, and the small signal model parameter is combined to establish large signal equivalent circuit model.The application greatly improves the modeling speed of small signal model and the accuracy of large signal model without sacrificing accuracy, which provides key technical support for reliable design of high-power, high-frequency microwave circuit.

Description

Technical Field

[0001] This invention relates to the equivalent circuit modeling technology of third-generation semiconductor material GaN HEMT, and in particular to an adaptive extraction method and system for GaN HEMT equivalent circuit model parameters. Background Technology

[0002] Semiconductor device models have a significant impact on improving circuit design accuracy. The more complex the circuit structure and the higher the power of the operating frequency band, the higher the requirements for device models. Establishing accurate semiconductor device models is crucial for improving the success rate of RF microwave / millimeter-wave monolithic integrated circuit design and shortening the circuit development cycle. The most important aspect of establishing an empirical basis equivalent model of the device is selecting a suitable numerical algorithm to accurately and efficiently extract the component parameters in the equivalent circuit. In existing technologies, scholars have proposed various optimization algorithms and modeling methods, such as traditional genetic algorithms, particle swarm optimization algorithms, and scaling models based on electromagnetic simulation. However, they still have significant drawbacks: traditional genetic algorithms do not consider the dynamic adjustment of the parameter search range, resulting in a high probability of singular values ​​and low optimization efficiency; large-signal model optimization does not focus on optimizing large error bias points, limiting the overall fitting accuracy. Summary of the Invention

[0003] The purpose of this invention is to provide an adaptive extraction method and system for GaN HEMT equivalent circuit model parameters, which solves the problems of low efficiency in small-signal model parameter extraction and insufficient accuracy in large-signal model in traditional GaN HEMT modeling, and achieves efficient and accurate extraction of model parameters.

[0004] The technical solution for achieving the objective of this invention is as follows:

[0005] An adaptive method for extracting parameters of a GaN HEMT equivalent circuit model includes:

[0006] Step 1: De-embedding the S-parameter test data of GaN HEMT;

[0007] Step 2: Extract GaN HEMT parasitic parameter values ​​using the small-signal equivalent circuit parasitic parameter extraction method to establish an initial population for the genetic algorithm;

[0008] Step 3: Obtain the measured intrinsic Y parameters by stripping parasitic parameters through S-parameter testing, and obtain the intrinsic parameters through network analysis;

[0009] Step 4: Determine whether the intrinsic parameters of all individuals in the population are less than zero, i.e., determine whether the individual is singular. If the individual is singular, add a penalty factor to the individual; otherwise, calculate its fitness.

[0010] Step 5: Determine whether the population contains a strange individual carrying a penalty factor. If so, generate a new individual within the parasitic parameter range to replace the strange individual and execute Step 3. If there is no strange individual in the population, update the parasitic parameter range based on the existing population, and then perform genetic algorithm calculations on the individuals in the initial population according to the set individual selection, crossover, and mutation rules to obtain a new genetic population and execute Step 6.

[0011] Step 6: Determine whether the genetic algorithm meets the preset objective. If the termination condition is not met, the initial population is updated and optimized by updating the genetic individuals. If the condition is met, the optimal individual is output based on the genetic individuals, and the component parameters corresponding to the optimal individual are output as the final parameters of the small-signal equivalent circuit. Otherwise, proceed to step 7.

[0012] Step 7: Initialize the weight matrix to 1, optimize the IV model using a genetic algorithm, calculate the current error matrix based on the current optimization result, and update the weight matrix of the fitness function according to the error matrix.

[0013] Step 8: Determine if the genetic algorithm meets the preset objective: If the termination condition is not met, continue to optimize the IV model with the updated weight matrix; if it is met, output the current best result as the final parameters of the IV model.

[0014] Step 9: Combine the small signal model parameters and the IV model parameters to construct the large signal model.

[0015] Furthermore, in step 1, the open-short de-embedding method is used to de-embedding the S-parameter test data of GaN HEMT.

[0016] Furthermore, the method for extracting parasitic parameters of the small-signal equivalent circuit in step 2 is the cutoff condition method.

[0017] Furthermore, in step 4, the fitness is calculated using an error function, which is:

[0018]

[0019]

[0020]

[0021]

[0022]

[0023] in This is the weighted value of the error between the real and imaginary parts of each element in the S matrix. Indicates the weighting factor. This represents the S-parameter error between simulation and testing. Indicates frequency sorting, This indicates the total number of frequencies considered.

[0024] Furthermore, in step 5, the parasitic parameter range is updated based on the existing population, and then a genetic algorithm is used to calculate the individuals in the initial population according to the set individual selection, crossover, and mutation rules to obtain a new genetic population, specifically including:

[0025] Count all individuals in the current population, select the individual with the lowest fitness in the current population, and then select individuals whose fitness meets the condition by taking twice the fitness of that individual as the maximum value; take the parasitic parameters of the selected individuals as a set, calculate the maximum and minimum values ​​of each parasitic parameter, and expand the range by 10% to form a new range of parasitic parameters.

[0026] Based on the rules of selection, crossover, and mutation, the reproduction and evolution of biological populations are simulated, and genetic calculations are performed on individuals in the population to obtain new genetic individuals.

[0027] Furthermore, the termination condition in step 6 is: reaching the maximum number of evolutionary generations and the minimum target value.

[0028] Furthermore, the fitness function in step 7 is:

[0029]

[0030]

[0031] in, For fitness, The error between the test value and the fitted value of Ids for the nth bias is represented by N biases. This is the current measurement value for the nth bias. The simulated current value for the nth bias is... The weight is the nth bias, and all bias weights are initialized to 1.

[0032] An adaptive parameter extraction system for GaN HEMT equivalent circuit model includes:

[0033] De-embedding units are used to de-embedding the S-parameter test data of GaN HEMT;

[0034] The population initialization unit extracts GaN HEMT parasitic parameter values ​​using the small-signal equivalent circuit parasitic parameter extraction method to establish the initial population for the genetic algorithm.

[0035] The intrinsic parameter acquisition unit obtains the measured intrinsic Y parameters by stripping parasitic parameters through testing S parameters and obtains intrinsic parameters through network analysis.

[0036] The intrinsic parameter judgment unit determines whether the intrinsic parameters of all individuals in the population are less than zero, i.e., whether the individual is singular. If the individual is singular, a penalty factor is added to the individual; otherwise, its fitness is calculated.

[0037] The singular individual judgment unit determines whether the population contains singular individuals carrying a penalty factor. If so, it generates a new individual within the parasitic parameter range to replace the singular individual, and then determines whether the new individual is singular. If there are no singular individuals in the population, it updates the parasitic parameter range based on the existing population, and then performs genetic algorithm calculations on the individuals in the initial population according to the set individual selection, crossover, and mutation rules to obtain a new genetic population.

[0038] The first judgment unit of the genetic algorithm determines whether the genetic algorithm meets the preset goal: if the termination condition is not met, the initial population is updated and iterated with genetic individuals for optimization; if it is met, the optimal individual is output according to the genetic individuals, and the component parameters corresponding to the optimal individual are output as the final parameters of the small-signal equivalent circuit; otherwise, the genetic algorithm continues to be executed.

[0039] The fitness function update unit initializes the weight matrix to 1, optimizes the IV model through a genetic algorithm, calculates the current error matrix based on the current optimization result, and updates the weight matrix of the fitness function according to the error matrix.

[0040] The second judgment unit of the genetic algorithm determines whether the genetic algorithm meets the preset objective: if the termination condition is not met, the IV model is further optimized with the updated weight matrix; if the condition is met, the current best result is output as the final parameters of the IV model.

[0041] The large-signal model determination unit combines the small-signal model parameters and the IV model parameters to form the large-signal model.

[0042] Compared with the prior art, the significant advantages of this invention are:

[0043] (1) The adaptive range genetic algorithm proposed in this invention filters the reasonable search range of parasitic parameters through non-singular intrinsic parameters and updates it dynamically, effectively suppressing the occurrence of singular values ​​such as negative capacitance and negative resistance of intrinsic parameters, avoiding the waste of computation caused by singular values ​​in traditional algorithms, and greatly improving the efficiency and accuracy of parameter extraction of small signal models.

[0044] (2) The adaptive weighted genetic algorithm proposed in this invention updates the weight matrix of the error function iteratively, assigns higher weights to bias points with larger errors, and enables the optimization algorithm to focus on fitting key bias points. This solves the problem that traditional genetic algorithms are prone to getting trapped in local optima and have excessive errors at some bias points, and accurately characterizes the nonlinear characteristics of GaN HEMT devices. Attached Figure Description

[0045] Figure 1 This is the topology diagram of the small-signal equivalent circuit model.

[0046] Figure 2 This is the equivalent circuit topology under low-frequency pinch-off conditions.

[0047] Figure 3 This is the equivalent circuit diagram of inductance and resistance in the "cold-forward" state of a high-frequency GaN HEMT.

[0048] Figure 4 The topology is the small-signal equivalent circuit model for the pinch-off state.

[0049] Figure 5 Flowchart for adaptive range genetic algorithm optimization of small-signal model for GaN HEMT devices.

[0050] Figure 6 Flowchart for removing external parasitic parameters.

[0051] Figure 7 Generate a topological map of 1000 non-singular individuals for two parameter ranges.

[0052] Figure 8 A comparison chart of adaptive and non-adaptive algorithm optimizations for different initial parameter ranges is shown. Figure 8 (a) is 0.3 to 3 times the initial value. Figure 8 (b) is 0.15 to 6 times the initial value. Figure 8 (c) is 0.1 to 9 times the initial value.

[0053] Figure 9 The graph shows a comparison between the measured (circles) and simulated (solid lines) S-parameters of a 4×200 μm small-signal model. Figure 9 (a)-(e) are biased with Vgs = -4~0 V and Vds = 32 V. Figure 9 (f)-(j) The bias is Vgs = -4~0 V, Vds = 28 V. Figure 9 (k)-(o) biased with Vgs = -4~0 V, Vds = 24 V, Figure 9 The biases (p)-(t) are Vgs = -4~0 V and Vds = 20 V.

[0054] Figure 10 This is a circuit topology diagram for a large-signal model.

[0055] Figure 11 The flowchart shows the adaptive error function method.

[0056] Figure 12 The weight matrix diagram is for adaptive updates.

[0057] Figure 13 This is a comparison chart of the errors of the traditional genetic algorithm (bar chart) and the adaptive weighted genetic algorithm (dot chart).

[0058] Figure 14 This is a comparison chart of the optimization results of the traditional genetic algorithm and the adaptive weight genetic algorithm.

[0059] Figure 15 The graph shows a comparison between the simulation results (solid line) and the test results (dots) of the 4×200μm model of the large-signal output characteristics of GaN HEMT. Figure 15 (a) represents the maximum power load at 3GHz. Figure 15 (b) is the maximum efficiency load at 3GHz. Figure 15 (c) represents the maximum power load at 6GHz. Figure 15 (d) represents the maximum efficiency load at 6GHz. Detailed Implementation

[0060] This invention proposes an adaptive method for extracting parameters of the GaN HEMT equivalent circuit model. First, the method performs open-short de-embedding on the test S-parameters to obtain pure S-parameters. Then, it extracts initial values ​​of parasitic elements from the small-signal model based on the cutoff condition method. An adaptive range genetic algorithm is then used to optimize the parasitic parameters of the small-signal model, dynamically adjusting the search range of the optimization parameters based on the parasitic parameters corresponding to the non-singular eigenvalues, resulting in a high-precision small-signal model. Next, an adaptive weighted genetic algorithm is used to optimize the drain current model parameters of the large-signal model. By iteratively updating the weight matrix of the error function, the optimization focus on bias conditions with large errors is dynamically strengthened, improving the overall fitting accuracy of the drain current model. Finally, the small-signal model and the IV model are combined to construct the large-signal model. This invention significantly improves the modeling speed of the small-signal model and the accuracy of the large-signal model without sacrificing accuracy, providing key technical support for the reliable design of high-power, high-frequency microwave circuits.

[0061] The present invention will now be described in further detail with reference to the accompanying drawings.

[0062] Combination Figure 5 and Figure 11 This invention provides an adaptive method for extracting parameters of a GaN HEMT equivalent circuit model, comprising the following steps:

[0063] The first step is to use the open-short de-embedding method, which effectively eliminates the interference of external test structures on the characteristics of the device itself by stripping the parasitic capacitance of the pads and the parasitic inductance and resistance of the connection lines from the test S-parameters, thus providing a clean data foundation for subsequent parameter extraction and model scaling.

[0064] The second step involves using the cutoff condition method to extract the initial values ​​of parasitic and intrinsic elements step by step by taking advantage of the difference in high and low frequency characteristics under pinch-off conditions.

[0065] The third step involves testing the S-parameters to remove parasitic parameters, obtaining the measured intrinsic Y-parameters, and then using network analysis to obtain the intrinsic parameters. The process for removing parasitic parameters from the S-parameter test values ​​is as follows: Figure 6 As shown. The expression for the intrinsic Y parameter is as follows:

[0066]

[0067] Based on the formula for the intrinsic Y parameters, the expressions for the eight intrinsic parameters can be derived as follows:

[0068]

[0069]

[0070]

[0071]

[0072]

[0073]

[0074]

[0075]

[0076]

[0077]

[0078] The fourth step is to determine whether the intrinsic parameters of all individuals in the population are less than zero. If an individual's intrinsic parameter has a value less than zero, a penalty factor is added to that individual; otherwise, its S-parameter error is calculated using the following formula:

[0079]

[0080]

[0081]

[0082]

[0083]

[0084] Step 5: Determine if the population contains individuals carrying the penalty factor. If so, generate new individuals within the parasitic parameter range to replace them, and skip to step 3. If there are no singular individuals in the population, update the parasitic parameter range based on the existing population, and then perform genetic algorithm calculations on the individuals in the initial population according to the set individual selection, crossover, and mutation rules to obtain a new genetic population, and proceed to step 6. Specifically, this includes: (The steps are repeated in the original text.)

[0085] Step 5-1: Count all individuals in the current population, select the individual with the lowest fitness, and then use twice the fitness of that individual as the maximum value. Select individuals whose fitness meets this condition. Using the parasitic parameters of the selected individuals as a set, calculate the maximum and minimum values ​​of each parasitic parameter, and expand this range by 10% to form a new range of parasitic parameters.

[0086] Step 5-2: Based on the selection, crossover, and mutation rules, simulate the reproduction and evolution of a biological population and perform genetic calculations on the individuals in the population to obtain new genetic individuals. Skip to step 3 until the termination condition of step 5 is met.

[0087] The sixth step is to determine whether the termination conditions of the genetic algorithm are met. There are two termination conditions: the maximum number of generations and the minimum objective value. If these conditions are met, the small-signal model adaptive optimization is complete; otherwise, proceed to the next step.

[0088] The ninth step is to perform adaptive optimization of the IV model, initialize all values ​​of the weight matrix to 1, and then run the genetic algorithm once to obtain the Ids result calculated by the first genetic algorithm.

[0089] Step 10: Select the weight matrix for the portion with the largest error, and add adaptive weights based on how much the error exceeds the average error. Then proceed with the next iteration. Continue this process until the termination condition is met, at which point the loop ends.

[0090] Step 11, based on Figure 10 The large-signal model topology shown is used to construct a large-signal equivalent circuit model.

[0091] In the second step of the process, the small-signal model topology used in this invention is as follows: Figure 1 As shown, when f < 5 GHz, , At that time, the low-frequency pinch-off state can be equivalent to Figure 2 The topology shown is simple, and after derivation, it can be expressed using Y parameters as follows:

[0092]

[0093] In the formula, The Y parameter represents the transformation of the S-parameter test value. It is angular frequency. The capacitance value is an unknown quantity. Since the number of unknowns exceeds the number of equations, a direct solution is clearly not possible. Therefore, [the following is a more detailed explanation:] and Set it as a scan parameter, and use the scan value as the value for subsequent steps.

[0094] Then, in an awkward silence ( Make it as large as possible without damaging the die. ) can be obtained as follows Figure 3 The equivalent circuit is shown. For the high-frequency range above 20 GHz, the capacitance in the circuit is small, and the unstripped intrinsic capacitance can be ignored. The Z-parameter matrix equation characterizing this network can be derived:

[0095]

[0096] Therefore, inductance L g L d L s and resistance R g R d R s They can be represented as follows:

[0097]

[0098]

[0099] During the scanning of Cpd and Cpd, the values ​​of all parasitic parameters can be obtained through the above process. Combined with the S-parameter test data under pinch-off conditions, the equivalent circuit corresponding to the full-frequency band test is as follows: Figure 4 As shown, parasitic parameters can be extracted layer by layer from the circuit diagram, and then the intrinsic parameters at each frequency can be calculated using the intrinsic parameter formula. The intrinsic parameters should not change with frequency; therefore, the average value across the entire frequency band is taken as the intrinsic parameter value. This yields all component parameters under pinch-off conditions. Figure 4 The circuit topology allows for the calculation of simulated S-parameter values. By continuously scanning Cpg and Cpd, the final set of component values ​​with the minimum error in the corresponding S-parameters is selected as the initial values.

[0100] After updating the parasitic parameter range according to the method in step seven, the probability of singular individuals appearing in the generated new individuals can be greatly reduced. 1000 non-singular individuals were generated using both parameter ranges, and the results are as follows. Figure 7As shown, generating 1000 non-singular individuals using the initial parasitic parameter range requires 150 iterations, while the adaptively updated range only requires 15 iterations. Each new individual requires S-parameter test values ​​to remove parasitic parameters before calculating intrinsic parameters, and this step is where the computational time is primarily consumed. To save computational time, each iteration only calculates the singular individuals from the previous iteration; therefore, the total computational cost should be the sum of the number of singular individuals appearing in all iterations.

[0101] Table 1 compares the computation time and computational cost required by the two algorithms to generate 1000 new individuals. Specifically, the number of calculations for the intrinsic parameters after adaptive update is only 2604, which is only 14% of the 18754 calculations required for the initial parasitic parameter range. The computation time required for adaptive update is 165 seconds, while the initial parasitic parameter range takes 1092 seconds. It can be seen that the adaptively updated parameter range saves a significant amount of computation and time.

[0102] Table 1. Comparison of 1000 individuals generated by adaptive genetic algorithm and traditional genetic algorithm.

[0103]

[0104] Parasitic parameter optimization of a GaN HEMT with a gate length of 0.25 μm and a gate width of 4 × 200 μm was performed using a small-signal model. Using the same initial parasitic parameter values ​​and genetic algorithm parameters, the initial ranges of the algorithm were set to 0.3–3 times, 0.15–6 times, and 0.1–9 times the initial parasitic parameter values, respectively. Optimization was performed on the same device, and the resulting optimal fitness variation curves are shown below. Figure 8 As shown.

[0105] As shown in Table 2, after both algorithms have completed their evolution, the computation time of the adaptive algorithm is much shorter than that of the traditional genetic algorithm. When the initial ranges are 0.3 to 3 times, 0.15 to 6 times, and 0.1 to 9 times, respectively, the time taken by the adaptive range genetic algorithm is only 27%, 13%, and 9% of that of the traditional genetic algorithm, which greatly saves the optimization time of the small signal model.

[0106] Table 2. Comparison of adaptive genetic algorithm and traditional genetic algorithm under different initial parasitic parameter ranges.

[0107]

[0108] The parameters in the small signal model were optimized according to this method, and the values ​​of each parasitic parameter are shown in Table 3.

[0109] Table 3 Parasitic parameter values

[0110]

[0111] The S-parameters obtained from model simulation were compared with the actual measured S-parameters of the device. Specific results are as follows: Figure 9 As shown.

[0112] The bias errors are shown in Table 4. Within the tested biases, the S-parameter error is less than 5%, which shows that the small-signal model established by this method can accurately characterize the scattering parameters of GaN HEMT devices.

[0113] Table 4. S-parameter errors of the 4×200 μm small-signal model

[0114]

[0115] The process for step ten is as follows: Figure 11 As shown, the algorithm first initializes all values ​​of the weight matrix to 1, then runs the genetic algorithm once to obtain the Ids result calculated in the first iteration. Based on the error calculation formula, the weight matrix with the largest error is selected, and adaptive weights are added according to the degree to which its error exceeds the average error. Then, the next iteration is performed. This process continues until the termination condition is met, at which point the loop ends. Using this adaptive weight matrix, the optimization algorithm can achieve better results.

[0116] Figure 12 It is a matrix updated by an adaptive weighting algorithm, whose weights evolve and iterate according to the error, increasing the bias weights with larger errors. Figure 13 A comparison of the optimization errors of the adaptive weight genetic algorithm and the traditional genetic algorithm shows that the error of most biases is smaller than that of the traditional genetic algorithm. The optimization results of the traditional genetic algorithm and the adaptive weight genetic algorithm are as follows: Figure 14 As shown, this method reduces the error of the Ids model by 36%.

[0117] To verify the large-signal output characteristics of the model, load-pull simulations were performed in ADS, comparing the output power and additional efficiency at the maximum power impedance point and the maximum efficiency impedance point at 3GHz and 6GHz frequencies, respectively. The results are as follows: Figure 15 As shown, the average error of Pout at 3GHz is 0.256 and the average error of PAE is 1.997%; the average error of Pout at 6GHz is 0.151 and the average error of PAE is 2.877%, which verifies the accuracy of the model.

[0118] This invention also provides an adaptive extraction system for GaN HEMT equivalent circuit model parameters, comprising:

[0119] De-embedding units are used to de-embedding the S-parameter test data of GaN HEMT;

[0120] The population initialization unit extracts GaN HEMT parasitic parameter values ​​using the small-signal equivalent circuit parasitic parameter extraction method to establish the initial population for the genetic algorithm.

[0121] The intrinsic parameter acquisition unit obtains the measured intrinsic Y parameters by stripping parasitic parameters through testing S parameters and obtains intrinsic parameters through network analysis.

[0122] The intrinsic parameter judgment unit determines whether the intrinsic parameters of all individuals in the population are less than zero, i.e., whether the individual is singular. If the individual is singular, a penalty factor is added to the individual; otherwise, its fitness is calculated.

[0123] The singular individual judgment unit determines whether the population contains singular individuals carrying a penalty factor. If so, it generates a new individual within the parasitic parameter range to replace the singular individual, and then determines whether the new individual is singular. If there are no singular individuals in the population, it updates the parasitic parameter range based on the existing population, and then performs genetic algorithm calculations on the individuals in the initial population according to the set individual selection, crossover, and mutation rules to obtain a new genetic population.

[0124] The first judgment unit of the genetic algorithm determines whether the genetic algorithm meets the preset goal: if the termination condition is not met, the initial population is updated and iterated with genetic individuals for optimization; if it is met, the optimal individual is output according to the genetic individuals, and the component parameters corresponding to the optimal individual are output as the final parameters of the small-signal equivalent circuit; otherwise, the genetic algorithm continues to be executed.

[0125] The fitness function update unit initializes the weight matrix to 1, optimizes the IV model through a genetic algorithm, calculates the current error matrix based on the current optimization result, and updates the weight matrix of the fitness function according to the error matrix.

[0126] The second judgment unit of the genetic algorithm determines whether the genetic algorithm meets the preset objective: if the termination condition is not met, the IV model is further optimized with the updated weight matrix; if the condition is met, the current best result is output as the final parameters of the IV model.

[0127] The large-signal model determination unit combines the small-signal model parameters and the IV model parameters to form the large-signal model.

[0128] The adaptive extraction method for the GaN HEMT equivalent circuit model proposed in this invention avoids manual adjustment of the optimization algorithm. When extracting parasitic parameters, it adaptively and dynamically adjusts the search range of the optimization parameters based on the parasitic parameters corresponding to the non-singular intrinsic parameters, greatly improving the extraction efficiency. When extracting IV parameters, it effectively improves the overall fitting accuracy of the drain current model by adaptively iteratively updating the weight matrix of the error function. The maximum errors in the model's output power and added efficiency are less than 0.4 dB and 3%, respectively, providing a reliable solution for high-precision parameter extraction of large-signal models.

[0129] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An adaptive method for extracting parameters of a GaN HEMT equivalent circuit model, characterized in that, include: Step 1: De-embedding the S-parameter test data of GaN HEMT; Step 2: Extract GaN HEMT parasitic parameter values ​​using the small-signal equivalent circuit parasitic parameter extraction method to establish an initial population for the genetic algorithm; Step 3: Obtain the measured intrinsic Y parameters by stripping parasitic parameters through S-parameter testing, and obtain the intrinsic parameters through network analysis; Step 4: Determine whether the intrinsic parameters of all individuals in the population are less than zero, i.e., determine whether the individual is singular. If the individual is singular, add a penalty factor to the individual; otherwise, calculate its fitness. Step 5: Determine whether the population contains a strange individual carrying a penalty factor. If so, generate a new individual within the parasitic parameter range to replace the strange individual and execute Step 3. If there is no strange individual in the population, update the parasitic parameter range based on the existing population, and then perform genetic algorithm calculations on the individuals in the initial population according to the set individual selection, crossover, and mutation rules to obtain a new genetic population and execute Step 6. Step 6: Determine if the genetic algorithm meets the preset objective: If the termination condition is not met, then use genetic individuals to update the initial population for optimization. If the conditions are met, the optimal individual is output based on the genetic individual, and the component parameters corresponding to the optimal individual are output as the final parameters of the small-signal equivalent circuit; otherwise, step 7 is executed. Step 7: Initialize the weight matrix to 1, optimize the IV model using a genetic algorithm, calculate the current error matrix based on the current optimization result, and update the weight matrix of the fitness function according to the error matrix. Step 8: Determine if the genetic algorithm meets the preset objective: If the termination condition is not met, continue to optimize the IV model with the updated weight matrix; if it is met, output the current best result as the final parameters of the IV model. Step 9: Combine the small signal model parameters and the IV model parameters to construct the large signal model.

2. The adaptive extraction method for GaN HEMT equivalent circuit model parameters according to claim 1, characterized in that, Step 1 uses the open-short de-embedding method to de-embedding the S-parameter test data of GaN HEMT.

3. The adaptive extraction method for GaN HEMT equivalent circuit model parameters according to claim 1, characterized in that, The method for extracting parasitic parameters of the small-signal equivalent circuit in step 2 is the cutoff condition method.

4. The adaptive extraction method for GaN HEMT equivalent circuit model parameters according to claim 1, characterized in that, In step 4, the fitness is calculated using an error function, which is: in This is the weighted value of the error between the real and imaginary parts of each element in the S matrix. Indicates the weighting factor. This represents the S-parameter error between simulation and testing. Indicates frequency sorting, This indicates the total number of frequencies considered.

5. The adaptive extraction method for GaN HEMT equivalent circuit model parameters according to claim 1, characterized in that, Step 5 updates the parasitic parameter range based on the existing population, and then performs genetic algorithm calculations on the individuals in the initial population according to the set individual selection, crossover, and mutation rules to obtain a new genetic population, specifically including: Count all individuals in the current population, select the individual with the lowest fitness in the current population, and then select individuals whose fitness meets the condition by taking twice the fitness of that individual as the maximum value; take the parasitic parameters of the selected individuals as a set, calculate the maximum and minimum values ​​of each parasitic parameter, and expand the range by 10% to form a new range of parasitic parameters. Based on the rules of selection, crossover, and mutation, the reproduction and evolution of biological populations are simulated, and genetic calculations are performed on individuals in the population to obtain new genetic individuals.

6. The adaptive extraction method for GaN HEMT equivalent circuit model parameters according to claim 1, characterized in that, The termination condition in step 6 is: reaching the maximum number of evolutionary generations and the minimum target value.

7. The adaptive extraction method for GaN HEMT equivalent circuit model parameters according to claim 1, characterized in that, The fitness function in step 7 is: in, For fitness, The error between the test value and the fitted value of Ids for the nth bias is represented by N biases. This is the current measurement value for the nth bias. The simulated current value for the nth bias is... The weight is the nth bias, and all bias weights are initialized to 1.

8. An adaptive parameter extraction system for GaN HEMT equivalent circuit model, characterized in that, include: De-embedding units are used to de-embedding the S-parameter test data of GaN HEMT; The population initialization unit extracts GaN HEMT parasitic parameter values ​​using the small-signal equivalent circuit parasitic parameter extraction method to establish the initial population for the genetic algorithm. The intrinsic parameter acquisition unit obtains the measured intrinsic Y parameters by stripping parasitic parameters through testing S parameters and obtains intrinsic parameters through network analysis. The intrinsic parameter judgment unit determines whether the intrinsic parameters of all individuals in the population are less than zero, i.e., whether the individual is singular. If the individual is singular, a penalty factor is added to the individual; otherwise, its fitness is calculated. The singular individual judgment unit determines whether the population contains singular individuals carrying a penalty factor. If so, it generates a new individual within the parasitic parameter range to replace the singular individual, and then determines whether the new individual is singular. If there are no singular individuals in the population, it updates the parasitic parameter range based on the existing population, and then performs genetic algorithm calculations on the individuals in the initial population according to the set individual selection, crossover, and mutation rules to obtain a new genetic population. The first judgment unit of the genetic algorithm determines whether the genetic algorithm meets the preset goal: if the termination condition is not met, the initial population is updated and iterated with genetic individuals for optimization; if it is met, the optimal individual is output according to the genetic individuals, and the component parameters corresponding to the optimal individual are output as the final parameters of the small-signal equivalent circuit; otherwise, the genetic algorithm continues to be executed. The fitness function update unit initializes the weight matrix to 1, optimizes the IV model through a genetic algorithm, calculates the current error matrix based on the current optimization result, and updates the weight matrix of the fitness function according to the error matrix. The second judgment unit of the genetic algorithm determines whether the genetic algorithm meets the preset objective: if the termination condition is not met, the IV model is further optimized with the updated weight matrix; if the condition is met, the current best result is output as the final parameters of the IV model. The large-signal model determination unit combines the small-signal model parameters and the IV model parameters to form the large-signal model.

Images