Multi-scale full-parameter large-signal modeling system for ultra-wideband high-power transistor

Abstract

The invention discloses a multi-scale full-parameter large-signal modeling system for an ultra-wideband high-power transistor, which relates to the field of multi-scale full-parameter large-signal modeling and comprises a data module, an extraction module, a model module, an adjustment module and a generation module. Parasitic stripping and parameter extraction are carried out to obtain a frequency domain feature matrix, feature tensor construction and double-branch network training are carried out based on the frequency domain feature matrix and physical size parameters and electrical parameters of a transistor, mapping to an ASM model is carried out to obtain a multi-dimensional fine calibration scalable ASM model MASM, and electrothermal coupling and polarization scattering effect dynamic correction are carried out to obtain a polarization scattering effect. According to the method, the all-working-condition electrode drain physical base ASM model M * ASM-T is obtained, polarization scattering modulation analysis is carried out based on the all-working-condition electrode drain physical base ASM model M * ASM-T, the finally optimized all-working-condition electrode scattering drain physical base ASM model M * ASM-T-S is obtained, and the reliability of system modeling is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of multi-scale full-parameter large-signal modeling, and in particular to a multi-scale full-parameter large-signal modeling system for ultra-wideband high-power transistors. Background Art

[0002] In modern communications, radar, satellites, and other fields, ultra-wideband, high-power transistors are core components whose performance directly affects the overall efficiency of the system. With the development of 5G / 6G communication technology, higher requirements are placed on transistors with wide bandwidth (e.g., DC–100GHz), high power (Vds = 0–50V), high linearity, and reliability. However, existing modeling methods have significant limitations:

[0003] 1. Insufficient processing of parasitic effects: Existing technologies mostly use fixed parasitic parameter models, which make it difficult to accurately describe the dynamic modulation effect of bias voltage (Vgs, Vds) on parasitic inductance Lp and capacitance Cp, resulting in large modeling errors in high-frequency bands (especially millimeter-wave bands).

[0004] 2. Lack of multi-physics field coupling modeling: Traditional models usually only consider electrical characteristics, ignoring electrothermal coupling effects (such as the influence of junction temperature on transconductance and output resistance) and polarization scattering effects (such as the modulation of polarization Coulomb field intensity on carrier mobility), and cannot truly reflect the nonlinear behavior of devices under high-power conditions.

[0005] 3. Inefficient parameter extraction and model calibration: Manual calibration relies on experience and is time-consuming. Especially under wide-band and multi-bias conditions, it is difficult to achieve cross-scale mapping of parameters (such as changes in gate length and channel width). In addition, traditional methods lack regularization constraints on the frequency response of broadband scattering parameters, resulting in insufficient physical interpretability of the model at high frequencies.

[0006] 4. Poor adaptability to wide bandwidth and high power: The existing model is difficult to accurately characterize the dispersion characteristics within a wide bandwidth (such as the frequency dependence of the gate-source capacitance) and the thermal dissipation characteristics under high power, which limits its application in scenarios such as 5G / 6G power amplifiers.

[0007] In order to solve the above-mentioned defects, a technical solution is now provided. Summary of the Invention

[0008] In order to solve the technical problems raised by the above background technology, the present invention is proposed. The embodiments of the present invention provide an ultra-wideband high-power transistor multi-scale full-parameter large-signal modeling system.

[0009] The object of the present invention can be achieved by the following technical solutions: an ultra-wideband high-power transistor multi-scale full-parameter large-signal modeling system, comprising a data module, an extraction module, a model module, an adjustment module, and a generation module;

[0010] The data module constructs a multi-dimensional data matrix based on the physical size parameters and electrical parameters of the transistor;

[0011] The extraction module performs parasitic stripping and parameter extraction based on the multidimensional data matrix to obtain the frequency domain feature matrix;

[0012] The model module constructs the feature tensor and performs dual-branch network training based on the frequency domain feature matrix, the physical size parameters and electrical parameters of the transistor, and maps it to the ASM model to obtain a multi-dimensional, calibrated and scalable ASM model M. ASM ;

[0013] The adjustment module is based on the multi-dimensional fine-tuned scalable ASM model M ASM Perform dynamic corrections on electrothermal coupling and polarization scattering effects to obtain the full-condition electrode drain physical basis ASM model M* ASM-T ;

[0014] The generation module is based on the full-condition electrode drain physical base ASM model M* ASM-T Perform polarization scattering modulation analysis to obtain the final optimized full-condition electrode scattering drain physical basis ASM model M* ASM-T-S .

[0015] Furthermore, the multidimensional data matrix analysis steps are as follows:

[0016] The physical size parameters and electrical parameters of the transistor are obtained through on-chip measurement. The specific physical size parameters include the gate length L, gate width W and dielectric layer thickness D of the transistor. The electrical parameters include gate-source voltage Vgs, drain-source voltage Vds, drain current Id, broadband scattering parameters S11 and S22, where S11 represents the input reflection coefficient and S22 represents the output reflection coefficient. A multidimensional data matrix Gparam = [L, W, D, Vgs, Vds, Id, S11, S22] is constructed.

[0017] Furthermore, the frequency domain feature matrix analysis steps are as follows:

[0018] The gate-source bias voltage processed by the hyperbolic tangent function tanh is multiplied by the angular frequency and the gate-source capacitance to obtain the nonlinear dispersion compensation term ΔYnon. The parasitic admittance Yparas is subtracted from the measured admittance matrix Ymeas, and the nonlinear dispersion compensation term ΔYnon is added to obtain the stripped parasitic dispersion intrinsic admittance Yintr.

[0019] Strip off the parasitic complementary intrinsic admittance, calculate the transconductance gm(f) by taking the real part of the partial derivative of the forward transfer admittance Y21 with respect to the gate-source bias voltage, take the real part of the output admittance Y22 and perform the inverse operation to obtain the output resistance Rout(f), take the imaginary part of the input admittance Y11 and divide it by the angular frequency to obtain the gate-source capacitance Cgs(f), and obtain the broadband intrinsic parameter matrix Mintri;

[0020] Based on the wide-band intrinsic parameter matrix Mintri, the parameter value of each frequency point fi is extracted to generate the eigenvector F(fi), and the frequency domain feature matrix Fintri is obtained.

[0021] Furthermore, the analysis steps of the measured admittance matrix Ymeas are as follows:

[0022] Based on the multidimensional data matrix Gparam, a dynamic modulation factor is introduced to establish a bias-dependent parasitic model and obtain the dynamic parasitic parameter matrix Aparas(Vgs, Vds);

[0023] The parasitic inductance Lp (Vgs) and the parasitic capacitance Cp (Vds) in the dynamic parasitic parameter matrix Aparas form a series RLC network. For each frequency point f, the impedance Zparas of the parasitic element is calculated and converted into the parasitic admittance Yparas.

[0024] A vector network analyzer is used to measure the scattering parameter S of the transistor in the target frequency band. The bias point (Vgs, Vds) is fixed during measurement, and the S parameter is converted into an admittance matrix using the matrix conversion formula to obtain the measured admittance matrix Ymeas.

[0025] Furthermore, the multi-dimensional precision scalable ASM model M ASM The analysis steps are as follows:

[0026] The physical size, bias voltage and frequency domain feature matrix Fintri are spliced ​​and input into the neural network to obtain the dynamic bias frequency domain feature tensor Xipt. The dynamic bias frequency domain feature tensor is input through a two-branch neural network and the intrinsic parameters are predicted. The S parameter frequency response is introduced as a regularization constraint. Through the loss function descent optimization training, the parameters output by the neural network are mapped to the scalable equation framework of the ASM model to obtain the multi-dimensional precision scalable ASM model M. ASM .

[0027] Furthermore, the full-operation-condition electrode-drain physical-based ASM model M* ASM-T The analysis steps are as follows:

[0028] Based on the calculation of the channel inversion layer charge and the polarization doping modulated surface potential, the potential gradient charge migration density is obtained, and the dynamic channel potential gradient charge distribution is obtained through the channel charge continuity equation with the polarization corrected mobility;

[0029] Based on the preliminary drain current model, the instantaneous power consumption Pdiss is corrected to obtain the corrected power consumption. The polarization-corrected mobility and threshold voltage are corrected by combining the electrothermal coupling response temperature data set Tj(Vds, Pdiss) and the dynamic channel potential gradient charge distribution Qi(y, t). The corrected polarization-corrected mobility and threshold voltage are then substituted into the preliminary drain current model to obtain the temperature potential gradient charge-corrected drain current.

[0030] Based on the electrothermal coupling response temperature data set Tj and the time-frequency gate charge modulation capacitance Cgg(f), the broadband gate capacitance model is temperature-corrected to obtain the temperature-dependent broadband time-frequency gate capacitance.

[0031] Based on the dynamic channel potential gradient charge distribution Qi(y, t) and the temperature-dependent broadband time-frequency gate capacitance Cgg*(f, Tj), the temperature potential gradient charge correction drain current is capacitively modulated and the potential gradient charge distribution is adjusted to obtain the electrothermal dynamic charge polarization drain current;

[0032] Based on the electrothermal dynamic charge polarization drain current, the electrothermal coupling response temperature data set Tj and the dynamic channel potential gradient charge distribution, the heat conduction equation is reversely coupled with the charge distribution to obtain the spatiotemporal electrothermal drain potential gradient data set;

[0033] Based on the spatiotemporal electrothermal drain potential gradient temperature data set Tj* and electrothermal dynamic charge polarization drain current A scalable ASM model for transient and steady-state electrothermal coupling ASM-T The corrected transconductance and corrected output resistance in the dynamic adjustment are obtained to obtain the full-condition electrode drain physical basis ASM model M* ASM-T .

[0034] Furthermore, the preliminary drain current model analysis steps are as follows:

[0035] Obtain the polarization electric field Epolar and the carrier concentration n, couple the square term of the polarization electric field with the square root term of the carrier concentration, calculate the polarization Coulomb scattering rate, convert the polarization Coulomb scattering rate into the polarization-corrected mobility upol, obtain the gate voltage Vgs, oxide layer thickness tox, channel doping concentration NA, and obtain the polarization-corrected mobility through the closed-form expression of the surface potential to obtain the polarization-doping modulated surface potential;

[0036] Obtain the channel inversion layer charge and distribute it evenly to the gate. Calculate the oxide layer capacitance charge by using the polarization doping modulated surface potential and the gate-source voltage, flat band voltage, and oxide layer capacitance. Add the oxide layer capacitance charge and the distribution value of the inversion layer charge to obtain the polarization modulated gate charge.

[0037] Obtain the frequency f, channel resistance Rch and polarization modulated gate charge Qg to establish a broadband gate capacitance model and obtain the time-frequency gate charge modulated capacitance;

[0038] Based on the calculation of drain-source voltage and gate length, the voltage per unit length is obtained, and the short channel modulation mechanism is introduced to obtain the short channel polarization effective electric field. The nonlinear enhancement control of the inversion layer by the gate-source voltage is integrated with the polarization mobility and the width-to-length ratio. The short channel polarization effective electric field is modulated by the tanh function to perform drain voltage saturation treatment to obtain a preliminary drain current model.

[0039] Furthermore, the transient steady-state electrothermal coupling scalable ASM model M ASM-T The analysis steps are as follows:

[0040] Through TCAD tool geometric modeling, the physical size parameters of the crystal are defined, including the transistor gate length L, gate width W, and dielectric layer thickness D. The 3D structural framework of the device is constructed, and the material thermal property parameters of the transistor are obtained, including thermal conductivity κ and specific heat capacity Cr. Key thermal properties are configured for each material layer, and a steady-state transient model for thermal simulation is constructed. The heat dissipation boundary is set and the convection coefficient is defined.

[0041] Apply bias at time step tn, calculate drain current, calculate instantaneous power dissipation Pdiss(tn), input instantaneous power dissipation Pdiss(tn) into the heat conduction equation as a heat source, solve the temperature distribution T(x,y,z,tn) by finite element method, update time tn+1=tn+Δt, repeat the above operation until Stop when it is less than the set threshold, select the point with the highest temperature in the device, such as the channel center or near the drain contact, and extract the temperature value Tj(tn) at each time step tn, forming a time series Tj(t), scan and record it, and obtain the electrothermal coupling response temperature data set Tj(Vds,Pdiss);

[0042] Multi-dimensional precision calibration of scalable ASM model M based on electrothermal coupled response dataset ASM The self-heating effect is dynamically corrected to obtain the transient steady-state electrothermal coupling scalable ASM model M ASM-T .

[0043] Furthermore, the final optimized full-operation electrode scattering drain physical basis ASM model M* ASM-T-S The analysis steps are as follows:

[0044] The frequency-temperature coupling modulation factor S based on the polarization scattering field intensity leakage voltage PC (Vds, f, Tj*), for the full-condition electrode-drain physical-based ASM model M* ASM-T Polarization scattering adjustment is performed to obtain the final optimized full-condition electrode scattering drain physical basis ASM model M* ASM-T-S .

[0045] Furthermore, the polarization scattering field intensity leakage voltage frequency temperature coupling modulation factor S PC The (Vds, f, Tj*) analysis steps are as follows:

[0046] A temperature-dependent field-modulated polarization model is established based on the drain-source voltage Vds and the spatiotemporal electrothermal drain potential gradient temperature dataset Tj* to obtain the polarization Coulomb field intensity.

[0047] Based on the polarized Coulomb field intensity, the polarized Coulomb field scattering probability is obtained through the Born approximation theory. The polarized Coulomb field scattering probability is normalized to obtain the dimensionless scattering factor.

[0048] Substituting the polarized Coulomb field intensity into the dimensionless scattering factor, the relationship between the drain-source voltage, frequency, and spatiotemporal electrothermal drain potential gradient temperature data set is analyzed to obtain the polarized scattering field intensity drain voltage frequency temperature coupling modulation factor S. PC (Vds, f, Tj*).

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. The present invention constructs a multidimensional data matrix based on the physical size parameters and electrical parameters of the transistor through the data module. The extraction module performs parasitic stripping and parameter extraction based on the multidimensional data matrix to obtain a frequency domain feature matrix. The model module performs feature tensor construction and dual-branch network training based on the frequency domain feature matrix, the physical size parameters and electrical parameters of the transistor, and maps them to the ASM model to obtain a multidimensional precision scalable ASM model M. ASM , the adjustment module is based on the multi-dimensional fine-tuned scalable ASM model M ASM Perform dynamic corrections on electrothermal coupling and polarization scattering effects to obtain the full-condition electrode drain physical basis ASM model M* ASM-T , can accurately describe the dynamic modulation effect of bias voltage (Vgs, Vds) on parasitic inductance Lp and capacitance Cp, reduce the modeling error in high frequency bands (especially millimeter wave bands), and enable multi-physics field coupling modeling.

[0051] 2. The present invention generates a module based on the full-condition electrode drain physical base ASM model M* ASM-T Perform polarization scattering modulation analysis to obtain the final optimized full-condition electrode scattering drain physical basis ASM model M* ASM-T-S, not only electrical characteristics are considered, but also electrothermal coupling effects (such as the influence of junction temperature on transconductance and output resistance) and polarization scattering effects (such as the modulation of polarization Coulomb field intensity on carrier mobility), truly reflecting the nonlinear behavior of the device under high-power conditions, and increasing the efficiency of parameter extraction and model calibration: under wide-band and multi-bias conditions, cross-scale mapping of parameters (such as changes in gate length and channel width) can be achieved. In addition, the regularization constraints of the frequency response of broadband scattering parameters can be applied, resulting in high physical interpretability of the model in the high-frequency band. The dispersion characteristics within a wide-band range (such as the frequency dependence of gate-source capacitance) and thermal dissipation characteristics under high power are accurately characterized, which can be applied in scenarios such as 5G / 6G power amplifiers. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. The following drawings are not intentionally scaled to the actual size, and the focus is on illustrating the main purpose of the present invention.

[0053] Figure 1 is a system block diagram of the present invention;

[0054] Figure 2 This is a flow chart of the extraction module of the present invention;

[0055] Figure 3 This is a flow chart of the generation module of the present invention. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts also fall within the scope of protection of the present invention.

[0057] like Figure 1 As shown, the ultra-wideband high-power transistor multi-scale full-parameter large-signal modeling system includes a data module, an extraction module, a model module, an adjustment module, and a generation module.

[0058] The data module constructs a multi-dimensional data matrix based on the physical size parameters and electrical parameters of the transistor;

[0059] The extraction module performs parasitic stripping and parameter extraction based on the multidimensional data matrix to obtain the frequency domain feature matrix;

[0060] The model module constructs the feature tensor and performs dual-branch network training based on the frequency domain feature matrix, the physical size parameters and electrical parameters of the transistor, and maps it to the ASM model to obtain a multi-dimensional, calibrated and scalable ASM model M. ASM ;

[0061] The adjustment module is based on the multi-dimensional fine-tuned scalable ASM model M ASM Perform dynamic corrections on electrothermal coupling and polarization scattering effects to obtain the full-condition electrode drain physical basis ASM model M* ASM-T ;

[0062] The generation module is based on the full-condition electrode drain physical base ASM model M* ASM-T Perform polarization scattering modulation analysis to obtain the final optimized full-condition electrode scattering drain physical basis ASM model M* ASM-T-S .

[0063] Specifically, the analysis steps of the data module are as follows:

[0064] The physical size parameters and electrical parameters of the transistor are obtained through on-chip measurement. The specific physical size parameters include the gate length L, gate width W and dielectric layer thickness D of the transistor. The electrical parameters include gate-source voltage Vgs, drain-source voltage Vds, drain current Id, broadband scattering parameters S11 and S22, where S11 represents the input reflection coefficient and S22 represents the output reflection coefficient. A multidimensional data matrix Gparam = [L, W, D, Vgs, Vds, Id, S11, S22] is constructed.

[0065] Specifically, such as Figure 2 As shown in Figure 2, the analysis steps of the extraction module are as follows:

[0066] Based on the multidimensional data matrix Gparam, a dynamic modulation factor is introduced to establish a bias-dependent parasitic model, including:

[0067] Lp(Vgs)=ηL+βW+ɑ×ln(Vgs+1);

[0068] Cp(Vds)=∈ef×A / T×(1+φ×Vds 2 )

[0069] Where Lp(Vgs) represents the parasitic inductance related to the gate-source bias voltage Vgs, η and β represent fitting coefficients, ɑ and φ represent voltage modulation factors, Cp(Vds) represents the parasitic capacitance related to the drain-source bias voltage Vds, ∈ef represents the effective dielectric constant, and A represents the capacitor area. The dynamic parasitic parameter matrix Aparas(Vgs, Vds) is obtained.

[0070] The parasitic inductance Lp (Vgs) and the parasitic capacitance Cp (Vds) in the dynamic parasitic parameter matrix Aparas form a series RLC network. For each frequency point f, the impedance Zparas of the parasitic element is calculated and converted into the parasitic admittance Yparas.

[0071] Zparas=jωLp+1 / (jωCp)+Rp

[0072] Yparas=1 / Zparas, where j represents the imaginary unit, ω represents the angular frequency, and Rp represents the resistance;

[0073] Use a vector network analyzer to measure the scattering parameters S (S11, S12, S21, S22) of the transistor in the target frequency band, such as DC–100 GHz, which represents the frequency range of DC frequency from 0 to 100 GHz. S21 represents the forward transmission coefficient and S12 represents the reverse transmission coefficient. The bias point (Vgs, Vds) needs to be fixed during measurement. The S parameters are converted into an admittance matrix using the matrix conversion formula to obtain the measured admittance matrix Ymeas. Specifically:

[0074] Ymeas=Y0×(IS)(I+S) -1

[0075] Specifically, Y0 is the admittance corresponding to the characteristic admittance of 50Ω, I is the unit matrix, the gate-source bias voltage processed by the hyperbolic tangent function tanh is multiplied by the angular frequency and the gate-source capacitance to obtain the nonlinear dispersion compensation term ΔYnon, the parasitic admittance Yparas is subtracted from the measured admittance matrix Ymeas, and the nonlinear dispersion compensation term ΔYnon is added to obtain the stripped parasitic dispersion intrinsic admittance Yintr. The specific formula is as follows:

[0076] ΔYnon=jω×Cg×tanh(Vgs / VT)

[0077] Yintr=Ymeas-Yparas+ΔYnon

[0078] Where Cg represents the gate-source capacitance reference value, VT represents the threshold voltage, the critical voltage at which the transistor turns on;

[0079] Stripping parasitic bulk intrinsic admittance Y11 represents the input admittance, Y21 represents the forward transfer admittance, Y22 represents the output admittance, and Y12 represents the backward transfer admittance. The transconductance gm(f) is obtained by taking the real part of the partial derivative of the forward transfer admittance Y21 with respect to the gate-source bias voltage. The output admittance Y22 is taken and the real part is calculated and the inverse is processed to obtain the output resistance Rout(f). The imaginary part of the input admittance Y11 is taken and divided by the angular frequency to obtain the gate-source capacitance Cgs(f). The broadband intrinsic parameter matrix Mintri[gm(f), Rout(f), Cgs(f)] is obtained.

[0080] Based on the wide-band intrinsic parameter matrix Mintri[gm(f), Rout(f), Cgs(f)], the parameter value of each frequency point fi is extracted to generate the eigenvector F(fi)=[gm(fi), Rout(fi), Cgs(fi)], and the frequency domain feature matrix Fintri=[F(f1), F(f2), ..., F(fn)] is obtained, where n represents the number of frequency points and i represents the serial number of the frequency point.

[0081] Specifically, the analysis steps of the model module are as follows:

[0082] The physical dimensions L, W, D, bias voltages Vgs, Vds and frequency domain feature matrix Fintri are spliced ​​and input into the neural network to obtain the dynamic bias frequency domain feature tensor Xipt = [L, W, D, Vgs, Vds, Fintri]. The dynamic bias frequency domain feature tensor is input into the dual-branch neural network and the intrinsic parameters g^m and R^out are predicted. The S parameter frequency response is introduced as a regularization constraint. Through loss function descent optimization training, the parameters output by the neural network are mapped to the scalable equation framework of the ASM model to obtain the multi-dimensional precision scalable ASM model M. ASM , specifically as follows:

[0083] g^m=NN1(Xipt;θ1)

[0084] R^out=NN2(Xipt;θ2)

[0085] L=∑ fi (||g^m(fi)-gm(fi)|| 2 +||R^out(fi)-Rout(fi)|| 2 )+ψ×TV(S11,S22)

[0086] TV(S11, S22)=∑ fi (|S11(fi+1)-S11(fi)|+|S22(fi+1)-S22(fi)|)

[0087] M ASM =FASM°NN(L,W,T,Vgs,Vds;θ * )

[0088] Specifically, NN1 represents the first branch neural network, θ1 represents the parameters of the first branch neural network, g^m represents the predicted value of transconductance, R^out represents the predicted value of output resistance, NN2 represents the second branch neural network, θ2 represents the parameters of the second branch neural network, ||*|| 2represents the square of the Euclidean norm, gm(fi) represents the true value of the transconductance at frequency fi, Rout(fi) represents the true value of the output resistance at frequency fi, TV(S11, S22) represents the total variation term, S11(fi+1) represents the input reflection coefficient at frequency fi+1, S22(fi+1) represents the output reflection coefficient at frequency fi+1, L represents the loss function, NN represents the trained neural network, θ * represents the optimal parameter set of the neural network after training, ψ represents the total variation regularization coefficient, which is used to adjust the weight of the frequency domain smoothing constraint term, FASM represents the physical equation mapping of the ASM model, and ° represents the function composite symbol, which represents the cascade of two mappings: first, the parameters are generated by the neural network, and then the parameters are substituted into the ASM model equation;

[0089] Specifically, the ASM model is a scalable device model framework based on semiconductor physics mechanisms. Designed for ultra-wideband, high-power transistors, it uses parameterized equations to describe device behavior from DC to millimeter-wave frequencies. It supports cross-scale mapping of physical dimensions (gate length, channel width, etc.) and electrical properties, and combines multi-physics integration capabilities (electrothermal coupling, polarization effects, etc.) with system-level simulation adaptability. The parameter mapping is generated through a neural network. Its core function is to establish an efficient mapping relationship between "input features (physical dimensions, bias conditions, frequency domain response) - intrinsic parameters - model equations." The neural network learns complex nonlinear relationships from massive operating condition data in a data-driven manner, accurately predicting the key parameters required for the ASM model. It also introduces measured S-parameter frequency responses as regularization constraints to ensure that the parameters conform to physical laws (such as high-frequency parasitic effects and thermal dissipation characteristics). This integration enables the ASM model to transcend the limitations of traditional manual calibration, retaining physical interpretability while significantly improving modeling accuracy under all operating conditions (wideband, high power, and high temperature).

[0090] Specifically, the analysis steps of the adjustment module are as follows:

[0091] Through TCAD tool geometric modeling, the physical size parameters of the crystal, such as the gate length L, gate width W, and dielectric layer thickness D of the transistor, are defined. The three-dimensional structural framework of the device is constructed, and the material thermal property parameters of the transistor are obtained, which are thermal conductivity κ and specific heat capacity Cr. Key thermal properties are configured for each material layer, and a thermal simulation steady-state transient model is constructed. The heat dissipation boundary Tsink = Tamb is set, where Tsink represents the temperature of the heat sink and Tamb represents the ambient temperature. The convection coefficient is defined, specifically air convection h = 10W / m 2 ×K, the model includes the steady-state equation and the transient equation of heat conduction:

[0092]

[0093] in represents the gradient operator, T represents the temperature distribution, Pdiss represents the different power consumption under Vds, Pdiss = Vds × Id, t represents time, and ρ represents the material density;

[0094] Apply bias Vgs(tn) and Vds(tn) at time step tn, calculate the drain current Id(tn), and calculate the instantaneous power dissipation Pdiss(tn), Pdiss(tn) = Vds(tn) × Id(tn) / (L × W × D). Input the instantaneous power dissipation Pdiss(tn) as the heat source into the heat conduction equation, solve the temperature distribution T(x, y, z, tn) by the finite element method, update the time tn+1 = tn+Δt, and repeat the above operation until Stop when it is less than the set threshold. Select the point with the highest temperature in the device, such as the channel center or near the drain contact, and extract the temperature value Tj(tn) at each time step tn. Form a time series Tj(t). Scan Vds (0-50V) and Vgs (0-5V), record Id, Pdiss, and Tj under each voltage pair, and obtain the electrothermal coupling response temperature data set Tj(Vds, Pdiss).

[0095] Multi-dimensional precision calibration of scalable ASM model M based on electrothermal coupled response dataset ASM The self-heating effect is dynamically corrected to obtain the transient steady-state electrothermal coupling scalable ASM model M ASM-T , specifically:

[0096] g*m=gm×exp(-(Tj-T0) / τ(Vds))×(1-Φ×Pdiss / Pmax)

[0097] R * out=Rout×(1+δ×(Tj-T0))×(1+ζ×ln(1+Pdiss / Pref))

[0098] Where gm represents the multidimensional fine-tuned scalable ASM model M ASM The raw transconductance is the ability of the gate voltage to control the drain current at the reference temperature. T0 represents the reference temperature, τ(Vds) represents the bias-dependent thermal time constant, τ(Vds) = τ0 × (1 + γVds), τ0 represents the reference value of the thermal time constant at the reference voltage, Pmax represents the maximum power consumption, Pref represents the reference power consumption, δ represents the temperature coefficient, Φ represents the power consumption correction factor, ζ represents the power consumption logarithmic correction factor, and Rout represents the multi-dimensional precision scalable ASM model M. ASM The original output resistance, g*m and R * out represents the corrected transconductance and output resistance, for the multi-dimensional scalable ASM model M ASMPerform dynamic correction to obtain the transient steady-state electrothermal coupling scalable ASM model M ASM-T ;

[0099] Obtain the polarization electric field Epolar and the carrier concentration n, couple the square term of the polarization electric field with the square root term of the carrier concentration, calculate the polarization Coulomb scattering rate, convert the polarization Coulomb scattering rate into the polarization-corrected mobility upol, obtain the gate voltage Vgs, oxide layer thickness tox, channel doping concentration NA, and obtain the polarization-corrected mobility through the closed-form expression of the surface potential to obtain the polarization-doping modulated surface potential;

[0100] The above calculation formula is as follows:

[0101]

[0102] upol=qτ sca / m *

[0103]

[0104] in It represents the scattering rate, which describes the scattering frequency of carriers affected by the polarization electric field and concentration. represents the inverse of the intrinsic scattering time, the scattering benchmark when there is no polarization field, ψ represents the polarization coupling coefficient, which quantifies the degree of coupling between the polarization electric field and scattering, upol represents the polarization-corrected mobility, which considers the carrier mobility after polarization scattering, and m * It represents the effective mass of carriers, reflecting the mass of carrier movement in the semiconductor. q represents the elementary charge, the amount of carrier charge. VFB represents the flat band voltage, the voltage when the gate makes the surface energy band flat. t co Indicates the oxide layer thickness, the physical thickness of the gate oxide layer, ε co represents the dielectric constant of the oxide layer, N′a represents the effective doping concentration, Na represents the channel doping concentration, φs represents the polarization doping modulated surface potential, and u0 represents the intrinsic mobility, which is the carrier mobility without considering the polarization Coulomb field scattering effect.

[0105] Obtain the channel inversion layer charge and distribute it evenly to the gate. Calculate the oxide layer capacitance charge by using the polarization doping modulated surface potential and the gate-source voltage, flat band voltage, and oxide layer capacitance. Add the oxide layer capacitance charge and the distribution value of the inversion layer charge to obtain the polarization modulated gate charge.

[0106] The specific processing process is: Qg=C co ×(Vgs-VFB-φs)+Qin / 2

[0107] C co =ε co / t co, where Qin represents the channel inversion layer charge, VFB represents the flat band voltage, C co represents the oxide layer capacitance, C co ×(Vgs-VFB-φs) represents the oxide layer capacitance charge, Qg represents the polarization modulation gate charge;

[0108] Obtain the frequency f, channel resistance Rch and polarization modulated gate charge Qg to establish a broadband gate capacitance model and obtain the time-frequency gate charge modulated capacitance. The model includes:

[0109]

[0110] τ ca =Rch×C co , where Cgg(f) represents the time-frequency gate charge modulation capacitance, τ ca represents the charging time constant;

[0111] Based on the calculation of drain-source voltage and gate length, the voltage per unit length is obtained, and the short-channel modulation mechanism is introduced to obtain the short-channel polarization effective electric field. By nonlinearly enhancing the inversion layer through the gate-source voltage, this control result is integrated with the polarization mobility and aspect ratio. The short-channel polarization effective electric field is modulated by the tanh function to perform drain voltage saturation treatment, and a preliminary drain current model is obtained.

[0112] Specifically, Eef = Vds / L × (1 + δln(L / Lcri))

[0113]

[0114] Where Eef represents the effective electric field of short channel polarization, Vds / L represents the voltage per unit length, δ represents the short channel modulation coefficient, which quantifies the penetration effect of the leakage field on the short channel and is applicable to nanometer-scale gate lengths, and Lcri represents the critical length. represents the preliminary drain current in the preliminary drain current model, β represents a constant related to the geometry and oxide capacitance, Vth represents the threshold voltage, the minimum gate voltage at which the transistor turns on, tanh represents the hyperbolic tangent function, It represents the modulation of the current by the quantized drain-source voltage Vds, and describes the saturation characteristics of the drain current;

[0115] Based on the calculation of the channel inversion layer charge and the polarization doping modulated surface potential, the potential gradient charge migration density is obtained, and the dynamic channel potential gradient charge distribution is obtained through the channel charge continuity equation with the polarization corrected mobility;

[0116] Specifically, the channel charge continuity equation: Where ∈Si represents the dielectric constant of silicon, y represents the channel direction coordinate, The charge migration density of the potential gradient is expressed by solving the partial differential equation by the finite element method to obtain the dynamic channel potential gradient charge distribution Qi(y, t). Represents the rate of change of charge distribution of dynamic channel potential gradient;

[0117] Based on the preliminary drain current model, the instantaneous power consumption Pdiss is corrected to obtain the corrected power consumption. The polarization-corrected mobility and threshold voltage are corrected by combining the electrothermal coupling response temperature data set Tj(Vds, Pdiss) and the dynamic channel potential gradient charge distribution Qi(y, t). The corrected polarization-corrected mobility and threshold voltage are then substituted into the preliminary drain current model to obtain the temperature potential gradient charge-corrected drain current.

[0118] Specifically, Among them upol * represents the corrected polarization-corrected mobility, exp represents the exponential function with the natural constant e as the base, It represents the rate of change of the charge distribution of the dynamic channel potential gradient, reflecting the non-steady-state characteristics of carrier transport, α represents the modulation coefficient; Vth * =Vth+κ×(Tj-T0), where Vth * represents the corrected threshold voltage, κ represents the threshold voltage temperature coefficient, and the corrected polarization-corrected mobility and the corrected threshold voltage are substituted into the preliminary drain current model to obtain the temperature potential gradient charge-corrected drain current

[0119] Based on the electrothermal coupling response temperature data set Tj and the time-frequency gate charge modulation capacitance Cgg(f), the broadband gate capacitance model is temperature-corrected to obtain the temperature-dependent broadband time-frequency gate capacitance.

[0120] Specifically, Where λ is the thermal expansion coefficient of the oxide layer, C* CO represents the corrected oxide layer capacitance, which is obtained by correcting the oxide layer capacitance in the broadband gate capacitance model to obtain the temperature-dependent broadband time-frequency gate capacitance Cgg*(f, Tj);

[0121] Based on the dynamic channel potential gradient charge distribution Qi(y, t) and the temperature-dependent broadband time-frequency gate capacitance Cgg*(f, Tj), the temperature potential gradient charge correction drain current is capacitively modulated and the potential gradient charge distribution is adjusted to obtain the electrothermal dynamic charge polarization drain current;

[0122] Specifically,

[0123] in, represents the electrothermal dynamic charge polarization drain current, represents the coupling coefficient, Cgg0 represents the initial gate capacitance, and Qi0 represents the charge migration density of the reference potential gradient;

[0124] Based on the electrothermal dynamic charge polarization drain current, the electrothermal coupling response temperature data set Tj and the dynamic channel potential gradient charge distribution, the heat conduction equation is reversely coupled with the charge distribution to obtain the spatiotemporal electrothermal drain potential gradient data set;

[0125] Specifically,

[0126] Where κ represents thermal conductivity, Cr represents specific heat capacity, represents the gradient operator, χ represents the calibration coefficient, which quantifies the enhancement weight of the charge dynamics on the heat source, and max represents the function that takes the maximum value. The partial differential equation is solved by the finite element method to obtain the spatiotemporal electrothermal drain potential gradient temperature dataset Tj*.

[0127] Based on the spatiotemporal electrothermal drain potential gradient temperature data set Tj* and electrothermal dynamic charge polarization drain current A scalable ASM model for transient and steady-state electrothermal coupling ASM-T The corrected transconductance and corrected output resistance in the dynamic adjustment are obtained to obtain the full-condition electrode drain physical basis ASM model M* ASM-T ;

[0128]

[0129] Where ξ represents the current dependence coefficient, I d0 represents the reference drain current, τ(Tj*) represents the dynamic time constant related to temperature, ν represents the temperature modulation coefficient, μ represents the current saturation effect coefficient, I sat represents the saturation current, Tcrit represents the critical temperature, g*m1 represents the full-condition electrode drain physical base ASM model M* ASM-T The corrected transconductance dynamic adjustment value, R * out1 represents the electrode-drain physical model M* under all working conditions ASM-T The corrected output resistance dynamic adjustment value is obtained to obtain the full-condition electrode drain physical basis ASM model M* ASM-T .

[0130] Specifically, such as Figure 3 As shown, the analysis steps for generating the module are as follows:

[0131] A temperature-dependent field-modulated polarization model is established based on the drain-source voltage Vds and the spatiotemporal electrothermal drain potential gradient temperature dataset Tj* to obtain the polarized Coulomb field intensity. The specific model includes:

[0132] E PC=χe×∈0×(Eex+Pspo / ∈0), where χe represents the material polarizability, χe=χe0×exp(-Tj* / Tcrit), χe0 represents the initial polarizability, ∈0 represents the vacuum dielectric constant, Eex represents the external electric field, which is obtained by dividing the drain-source voltage Vds by the gate length L, Pspo represents the spontaneous polarization intensity, E PC represents the polarization Coulomb field intensity;

[0133] Based on the polarized Coulomb field intensity, the polarized Coulomb field scattering probability is obtained through the Born approximation theory. The polarized Coulomb field scattering probability is normalized to obtain the dimensionless scattering factor.

[0134] Substituting the polarized Coulomb field intensity into the dimensionless scattering factor, the relationship between the drain-source voltage, frequency, and spatiotemporal electrothermal drain potential gradient temperature data set is analyzed to obtain the polarized scattering field intensity drain voltage frequency temperature coupling modulation factor S. PC (Vds, f, Tj*);

[0135]

[0136] Among them, Γ PC represents the polarized Coulomb field scattering probability, q represents the electron charge, m* represents the effective mass, and n inv represents the carrier density in the inversion layer, represents the reduced Planck constant, f represents the frequency, fc represents the cutoff frequency, the carrier dynamic response threshold, ∝ represents the proportional relationship, k0 represents the material characteristic constant, E ref represents the reference electric field strength, k represents the comprehensive proportional coefficient, S PC (Vds, f, Tj*) represents the polarization scattering field intensity leakage voltage frequency temperature coupling modulation factor;

[0137] The frequency-temperature coupling modulation factor S based on the polarization scattering field intensity leakage voltage PC (Vds, f, Tj*), for the full-condition electrode-drain physical-based ASM model M* ASM-T Polarization scattering adjustment is performed to obtain the final optimized full-condition electrode scattering drain physical basis ASM model M* ASM-T-S ;

[0138] g*m2=g*m1×(1+Я×S PC (Vds, f, Tj*));

[0139] R * out2=R * out1×(1-Φ×S PC (Vds, f, Tj*) / Scrit));

[0140] Where Я represents the modulation coefficient of the scattering factor on the transconductance, g*m2 represents the final optimized full-condition electrode scattering drain physical basis ASM model M* ASM-T-S The transconductance adjustment value, R * out2 represents the final optimized full-condition electrode scattering drain physical-based ASM model M* ASM-T-S The output resistance adjustment value is Scrit, which is used to limit over-modulation. Φ is the scattering modulation output resistance proportional coefficient.

[0141] Specifically, the multi-scale, full-parameter, large-signal modeling system for ultra-wideband high-power transistors is constructed through a multi-dimensional data matrix (integrating physical size parameters, bias voltage, drain current and broadband scattering parameters), combined with a dynamic parasitic model (bias-dependent modeling) and parasitic stripping technology (frequency domain admittance matrix conversion, nonlinear dispersion compensation) to achieve precise extraction of intrinsic parameters (transconductance, output resistance, gate-source capacitance); based on a dual-branch neural network architecture (physical branch and frequency domain branch) and S-parameter frequency response regularization constraints, the scalable ASM model parameters (mobility, threshold voltage, parasitic parameters) are optimized to construct a multi-dimensional precision MASM model; through electrothermal coupling modeling (solving the heat conduction equation, generating junction temperature sequences) and integrating polarization scattering effects (polarization Coulomb field strength, scattering probability, dynamic channel potential gradient charge distribution), the transient and steady-state electrothermal coupling model MASM-T and the full-operating-condition polarization scattering model M*ASM-TS are generated in turn. The system realizes the full-process modeling of the electric-thermal-polarization-scattering multi-field coupling effects, supports the accurate characterization of the nonlinear behavior of devices under wide bandwidth, high power, and multiple physical scales, and provides a modeling tool with explainable physical mechanisms, adaptive parameter scaling, and accurate response to all working conditions, thereby improving the engineering efficiency of semiconductor devices in high-frequency efficiency (PAE), linearity (ACPR) and reliability optimization.

[0142] The above is an illustration of the present invention and should not be considered as limiting thereof. Although several exemplary embodiments of the present invention have been described, it will be readily understood by those skilled in the art that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present invention. Therefore, all such modifications are intended to be included within the scope of the present invention as defined by the claims. It should be understood that the above is an illustration of the present invention and should not be considered as being limited to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The present invention is defined by the claims and their equivalents.

Claims

1. An ultra-wideband high-power transistor multi-scale full-parameter large-signal modeling system, comprising a data module, an extraction module, a model module, an adjustment module, and a generation module, characterized in that: The data module constructs a multi-dimensional data matrix based on the physical size parameters and electrical parameters of the transistor; The extraction module performs parasitic stripping and parameter extraction based on the multidimensional data matrix to obtain the frequency domain feature matrix; The model module constructs the feature tensor and performs dual-branch network training based on the frequency domain feature matrix, the physical size parameters and electrical parameters of the transistor, and maps it to the ASM model to obtain a multi-dimensional, calibrated and scalable ASM model M. ASM ; The adjustment module is based on the multi-dimensional fine-tuned scalable ASM model M ASM Perform dynamic corrections on electrothermal coupling and polarization scattering effects to obtain the full-condition electrode drain physical basis ASM model M* ASM-T ; The generation module is based on the full-condition electrode drain physical base ASM model M* ASM-T Perform polarization scattering modulation analysis to obtain the final optimized full-condition electrode scattering drain physical basis ASM model M* ASM-T-S .

2. The ultra-wideband high-power transistor multi-scale full-parameter large-signal modeling system according to claim 1 is characterized in that: The multidimensional data matrix analysis steps are as follows: The physical size parameters and electrical parameters of the transistor are obtained through on-chip measurement. The physical size parameters include the gate length L, gate width W and dielectric layer thickness D of the transistor. The electrical parameters include the gate-source voltage Vgs, drain-source voltage Vds, drain current Id, broadband scattering parameters S11 and S22, where S11 represents the input reflection coefficient and S22 represents the output reflection coefficient. A multidimensional data matrix Gparam = [L, W, D, Vgs, Vds, Id, S11, S22] is constructed.

3. The ultra-wideband high-power transistor multi-scale full-parameter large-signal modeling system according to claim 1 is characterized in that: The frequency domain characteristic matrix analysis steps are as follows: The gate-source bias voltage processed by the hyperbolic tangent function tanh is multiplied by the angular frequency and the gate-source capacitance to obtain the nonlinear dispersion compensation term ΔYnon. The parasitic admittance Yparas is subtracted from the measured admittance matrix Ymeas, and the nonlinear dispersion compensation term ΔYnon is added to obtain the stripped parasitic dispersion intrinsic admittance Yintr. By taking the real part of the partial derivative of the forward transfer admittance Y21 with respect to the gate-source bias voltage in the stripped parasitic compensated scattered intrinsic admittance, the transconductance gm(f) is obtained. The real part of the output admittance Y22 in the stripped parasitic compensated scattered intrinsic admittance is taken and the inverse is processed to obtain the output resistance Rout(f). The imaginary part of the input admittance Y11 in the stripped parasitic compensated scattered intrinsic admittance is taken and divided by the angular frequency to obtain the gate-source capacitance Cgs(f), and the broadband intrinsic parameter matrix Mintri is obtained. Based on the wide-band intrinsic parameter matrix Mintri, the parameter value of each frequency point fi is extracted to generate the eigenvector F(fi), and the frequency domain feature matrix Fintri is obtained.

4. The ultra-wideband high-power transistor multi-scale full-parameter large-signal modeling system according to claim 3 is characterized in that: The analysis steps of the measured admittance matrix Ymeas are as follows: Based on the multidimensional data matrix Gparam, a dynamic modulation factor is introduced to establish a bias-dependent parasitic model and obtain the dynamic parasitic parameter matrix Aparas(Vgs, Vds); The parasitic inductance Lp (Vgs) and the parasitic capacitance Cp (Vds) in the dynamic parasitic parameter matrix Aparas form a series RLC network. For each frequency point f, the impedance Zparas of the parasitic element is calculated and converted into the parasitic admittance Yparas. A vector network analyzer is used to measure the scattering parameter S of the transistor in the target frequency band. The bias point (Vgs, Vds) is fixed during measurement, and the S parameter is converted into an admittance matrix using the matrix conversion formula to obtain the measured admittance matrix Ymeas.

5. The ultra-wideband high-power transistor multi-scale full-parameter large-signal modeling system according to claim 1, characterized in that: The multi-dimensional refined scalable ASM model M ASM The analysis steps are as follows: The physical size, bias voltage and frequency domain feature matrix Fintri are spliced ​​and input into the neural network to obtain the dynamic bias frequency domain feature tensor Xipt. The dynamic bias frequency domain feature tensor is input through a two-branch neural network and the intrinsic parameters are predicted. The S parameter frequency response is introduced as a regularization constraint. Through the loss function descent optimization training, the parameters output by the neural network are mapped to the scalable equation framework of the ASM model to obtain the multi-dimensional precision scalable ASM model M. ASM .

6. The ultra-wideband high-power transistor multi-scale full-parameter large-signal modeling system according to claim 1, characterized in that: The full-condition electrode-drain physics-based ASM model M* ASM-T The analysis steps are as follows: Based on the calculation of the channel inversion layer charge and the polarization doping modulated surface potential, the potential gradient charge migration density is obtained, and the dynamic channel potential gradient charge distribution is obtained through the channel charge continuity equation with the polarization corrected mobility; Based on the preliminary drain current model, the instantaneous power consumption Pdiss is corrected to obtain the corrected power consumption. The polarization-corrected mobility and threshold voltage are corrected by combining the electrothermal coupling response temperature data set Tj(Vds, Pdiss) and the dynamic channel potential gradient charge distribution Qi(y, t). The corrected polarization-corrected mobility and threshold voltage are then substituted into the preliminary drain current model to obtain the temperature potential gradient charge-corrected drain current. Based on the electrothermal coupling response temperature data set Tj and the time-frequency gate charge modulation capacitance Cgg(f), the broadband gate capacitance model is temperature-corrected to obtain the temperature-dependent broadband time-frequency gate capacitance. Based on the dynamic channel potential gradient charge distribution Qi(y, t) and the temperature-dependent broadband time-frequency gate capacitance Cgg*(f, Tj), the temperature potential gradient charge correction drain current is capacitively modulated and the potential gradient charge distribution is adjusted to obtain the electrothermal dynamic charge polarization drain current; Based on the electrothermal dynamic charge polarization drain current, the electrothermal coupling response temperature data set Tj and the dynamic channel potential gradient charge distribution, the heat conduction equation is reversely coupled with the charge distribution to obtain the spatiotemporal electrothermal drain potential gradient data set; Based on the spatiotemporal electrothermal drain potential gradient temperature data set Tj* and electrothermal dynamic charge polarization drain current A scalable ASM model for transient and steady-state electrothermal coupling ASM-T The corrected transconductance and corrected output resistance in the dynamic adjustment are obtained to obtain the full-condition electrode drain physical basis ASM model M* ASM-T .

7. The ultra-wideband high-power transistor multi-scale full-parameter large-signal modeling system according to claim 6, characterized in that: The steps for analyzing the preliminary drain current model are as follows: Obtain the polarization electric field Epolar and the carrier concentration n, couple the square term of the polarization electric field with the square root term of the carrier concentration, calculate the polarization Coulomb scattering rate, convert the polarization Coulomb scattering rate into the polarization-corrected mobility upol, obtain the gate voltage Vgs, oxide layer thickness tox, channel doping concentration NA, and obtain the polarization-corrected mobility through the closed-form expression of the surface potential to obtain the polarization-doping modulated surface potential; Obtain the channel inversion layer charge and distribute it evenly to the gate. Calculate the oxide layer capacitance charge by using the polarization doping modulated surface potential and the gate-source voltage, flat band voltage, and oxide layer capacitance. Add the oxide layer capacitance charge and the distribution value of the inversion layer charge to obtain the polarization modulated gate charge. Obtain the frequency f, channel resistance Rch and polarization modulated gate charge Qg to establish a broadband gate capacitance model and obtain the time-frequency gate charge modulated capacitance; Based on the calculation of drain-source voltage and gate length, the voltage per unit length is obtained, and the short channel modulation mechanism is introduced to obtain the short channel polarization effective electric field. The nonlinear enhancement control of the inversion layer by the gate-source voltage is integrated with the polarization mobility and the width-to-length ratio. The short channel polarization effective electric field is modulated by the tanh function to perform drain voltage saturation treatment to obtain a preliminary drain current model.

8. The ultra-wideband high-power transistor multi-scale full-parameter large-signal modeling system according to claim 7, characterized in that: The transient and steady-state electrothermal coupled scalable ASM model M ASM-T The analysis steps are as follows: Through TCAD tool geometric modeling, the physical size parameters of the crystal are defined, including the transistor gate length L, gate width W, and dielectric layer thickness D. The 3D structural framework of the device is constructed, and the material thermal property parameters of the transistor are obtained, including thermal conductivity κ and specific heat capacity Cr. Key thermal properties are configured for each material layer, and a steady-state transient model for thermal simulation is constructed. The heat dissipation boundary is set and the convection coefficient is defined. Apply bias at time step tn, calculate drain current, calculate instantaneous power dissipation Pdiss(tn), input instantaneous power dissipation Pdiss(tn) into the heat conduction equation as a heat source, solve the temperature distribution T(x,y,z,tn) by finite element method, update time tn+1=tn+Δt, repeat the above operation until Stop when it is less than the set threshold, select the point with the highest temperature in the device, such as the channel center or near the drain contact, and extract the temperature value Tj(tn) at each time step tn, forming a time series Tj(t), scan and record it, and obtain the electrothermal coupling response temperature data set Tj(Vds,Pdiss); Multi-dimensional precision calibration of scalable ASM model M based on electrothermal coupled response dataset ASM The self-heating effect is dynamically corrected to obtain the transient steady-state electrothermal coupling scalable ASM model M ASM-T .

9. The ultra-wideband high-power transistor multi-scale full-parameter large-signal modeling system according to claim 1, characterized in that: The final optimized full-operation electrode scattering drain physical-based ASM model M* ASM-T-S The analysis steps are as follows: The frequency-temperature coupling modulation factor S based on the polarization scattering field intensity leakage voltage PC (Vds, f, Tj*), for the full-condition electrode-drain physical-based ASM model M* ASM-T Polarization scattering adjustment is performed to obtain the final optimized full-condition electrode scattering drain physical basis ASM model M* ASM-T-S .

10. The ultra-wideband high-power transistor multi-scale full-parameter large-signal modeling system according to claim 9, characterized in that: The polarization scattering field intensity leakage voltage frequency temperature coupling modulation factor S PC The (Vds, f, Tj*) analysis steps are as follows: A temperature-dependent field-modulated polarization model is established based on the drain-source voltage Vds and the spatiotemporal electrothermal drain potential gradient temperature dataset Tj* to obtain the polarization Coulomb field intensity. Based on the polarized Coulomb field intensity, the polarized Coulomb field scattering probability is obtained through the Born approximation theory. The polarized Coulomb field scattering probability is normalized to obtain the dimensionless scattering factor. Substituting the polarized Coulomb field intensity into the dimensionless scattering factor, the relationship between the drain-source voltage, frequency, and spatiotemporal electrothermal drain potential gradient temperature data set is analyzed to obtain the polarized scattering field intensity drain voltage frequency temperature coupling modulation factor S. PC (Vds, f, Tj*).

Images