CN-120217983-B - DNN-based large signal model modeling method, system, equipment and storage medium

CN120217983BCN 120217983 BCN120217983 BCN 120217983BCN-120217983-B

Abstract

The application relates to a DNN-based large signal model modeling method, a DNN-based large signal model modeling system, DNN-based large signal model modeling equipment and DNN-based large signal model modeling storage medium, which relate to the technical field of integrated circuits, wherein the method is based on a GaN microwave device comprising a GaN transistor, and the method comprises the steps of performing large signal test on the GaN transistor to obtain traveling wave characteristics; processing the input layer characteristic and the output layer characteristic in the traveling wave characteristic to obtain input layer data and output layer data, extracting real part data and imaginary part data, training by using DNN to obtain a high-power region deep neural network model, extracting amplitude data and phase data, training by using DNN to obtain a low-power region deep neural network model, training the traveling wave characteristic according to the high-power region deep neural network model and the low-power region deep neural network model to obtain a training traveling wave characteristic, converting the training traveling wave characteristic into a voltage current characteristic, and constructing according to the voltage current characteristic to obtain a large-signal behavior model. The method has the technical effect of improving the accuracy of a large-signal modeling model of the GaN microwave device.

Inventors

ZHENG JIAXIN
XIE BIN
XU SHENG
GUO YONGXIN

Assignees

长三角集成电路工业应用技术创新中心
江苏集萃集成电路应用技术管理有限公司
江苏集萃集成电路应用技术创新中心有限公司

Dates

Publication Date: 20260512
Application Date: 20250311

Claims (10)

1. A DNN-based large signal model modeling method based on a GaN microwave device comprising a GaN transistor, the method comprising: Carrying out large-signal test on the GaN transistor to obtain traveling wave characteristics, wherein the traveling wave characteristics comprise input layer characteristics and output layer characteristics; processing the input layer characteristics to obtain input layer data, and processing the output layer characteristics to obtain output layer data; Extracting real part data and imaginary part data of the input layer data and the output layer data, and training by using DNN according to the real part data and the imaginary part data to obtain a high-power region deep neural network model; extracting amplitude data and phase data of the input layer data and the output layer data, and training by using DNN according to the amplitude data and the phase data to obtain a low-power region deep neural network model; training the traveling wave characteristics according to the high-power region deep neural network model and the low-power region deep neural network model to obtain training traveling wave characteristics, and converting the training traveling wave characteristics into voltage and current characteristics; And constructing and obtaining a large signal behavior model according to the voltage-current characteristics.
2. The method of claim 1, wherein performing a large signal test on the GaN transistor to obtain the traveling wave characteristic comprises: acquiring fundamental wave conditions and harmonic wave conditions for testing; Carrying out input power scanning on the GaN transistor under the fundamental wave condition and the harmonic wave condition through a vector network analyzer to obtain a scanning result; And reading the scanning result through a coupler to obtain the traveling wave characteristic.
3. The method of claim 1, wherein the input layer characteristics comprise incident waves and the output layer characteristics comprise reflected waves, wherein processing the input layer characteristics to obtain input layer data and processing the output layer characteristics to obtain output layer data comprises: calculating a target phase of an input port fundamental wave incident wave; obtaining input layer data according to the residual incident waves of the target phase normalization processing; and normalizing the reflected wave according to the target phase to obtain output layer data.
4. A method according to claim 3, wherein said extracting real and imaginary data of said input and output layer data comprises: extracting a target amplitude of an input port fundamental wave incident wave, extracting real part data and imaginary part data of the input layer data, and extracting the real part data and the imaginary part data of the output layer data; The training to obtain the high-power region deep neural network model according to the real part data and the imaginary part data by using DNN comprises the following steps: Normalizing the target amplitude, the real part data and the imaginary part data of the input layer data, and the real part data and the imaginary part data of the output layer data to obtain high-power region training data; The training data of the high power area are imported into a preset first neural network to train and calculate a first loss value; judging whether the first loss value is smaller than a preset value or not; and if the first loss value is smaller than the preset value, stopping training and obtaining a high-power region deep neural network model according to the trained first neural network.
5. The method of claim 4, wherein the extracting amplitude data and phase data of the input layer data and the output layer data comprises: Extracting the dB amplitude of an input port fundamental wave incident wave, wherein the dB amplitude is in dB, extracting the amplitude data and the phase data of the input layer data, and extracting the amplitude data and the phase data of the output layer data; the training by DNN according to the amplitude data and the phase data to obtain a low-power region deep neural network model comprises the following steps: normalizing the dB amplitude, the amplitude data and the phase data of the input layer data, and the amplitude data and the phase data of the output layer data to obtain low-power region training data; The training data of the low power area are imported into a preset second neural network to train and calculate a second loss value; judging whether the second loss value is smaller than a second preset value or not; And if the second loss value is smaller than a second preset value, stopping training and obtaining a low-power region deep neural network model according to the trained second neural network.
6. The method of claim 1, wherein the training the traveling wave characteristics from the high power region-depth neural network model and the low power region-depth neural network model to obtain training traveling wave characteristics comprises: setting a judging point of a high-low power area; Constructing a zoned deep neural network model according to the judgment point, the high-power regional deep neural network model and the low-power regional deep neural network model; Training the traveling wave characteristics according to the zonal deep neural network model and performing inverse normalization processing to obtain training traveling wave characteristics.
7. The method according to claim 1, wherein after the large signal behavior model is constructed according to the voltage-current characteristics, the method further comprises: Receiving verification traveling wave characteristics, wherein the verification traveling wave characteristics comprise actual data of the verification traveling wave characteristics; Simulating the characteristic of the verification traveling wave by using the large signal behavior model to obtain simulation data: Judging whether the errors of the actual data and the simulation data are within an allowable range or not; If the error is within the allowable range, judging that the large-signal behavior model can be put into use; If the error is not within the allowable range, the large signal behavior model is judged to be unavailable.
8. A DNN-based large signal model modeling system, the system being based on a GaN microwave device comprising a GaN transistor, the system comprising: A traveling wave characteristic acquisition module (1501) for performing a large signal test on the GaN transistor to obtain traveling wave characteristics, wherein the traveling wave characteristics include input layer characteristics and output layer characteristics; The traveling wave characteristic processing module (1502) is used for processing the input layer characteristics to obtain input layer data and processing the output layer characteristics to obtain output layer data; a high-power region model training module (1503) for extracting real part data and imaginary part data of the input layer data and the output layer data, and training by using DNN according to the real part data and the imaginary part data to obtain a high-power region deep neural network model; The low-power region model training module (1504) is used for extracting amplitude data and phase data of the input layer data and the output layer data, and obtaining a low-power region deep neural network model by utilizing DNN training according to the amplitude data and the phase data; A voltage-current characteristic conversion module (1505) for training the traveling wave characteristic according to the high-power region deep neural network model and the low-power region deep neural network model to obtain a training traveling wave characteristic, and converting the training traveling wave characteristic into a voltage-current characteristic; And the behavior model construction module (1506) is used for constructing a large-signal behavior model according to the voltage-current characteristics.
9. A computer device comprising a memory and a processor, the memory having stored thereon a computer program capable of being loaded by the processor and performing the method according to any of claims 1 to 7.
10. A computer readable storage medium, characterized in that a computer program is stored which can be loaded by a processor and which performs the method according to any one of claims 1to 7.

Description

DNN-based large signal model modeling method, system, equipment and storage medium Technical Field The application relates to the technical field of integrated circuits, in particular to a DNN-based large signal model modeling method, a DNN-based large signal model modeling system, DNN-based large signal model modeling equipment and a DNN-based large signal model modeling storage medium. Background DNN is an abbreviation for deep neural network (Deep Neural Network), which is a multi-layer neural network structure aimed at enhancing the expressive power of models and the ability to learn complex features by stacking multiple hidden layers. With the development of the present infrastructure, communication industry and aerospace technology, the microwave system of the equipment has higher and higher requirements on miniaturization, high temperature resistance, radiation resistance, high power, ultrahigh frequency, suitability for working in severe environments and the like. Wide band gap semiconductor materials and devices typified by GaN and SiC have become a hot spot of research, and research and development of high-performance semiconductor materials and devices that can operate at higher frequencies and with higher power have great significance. Along with the continuous improvement of the quality of epitaxial materials, the device process is continuously perfected, and the development of AlGaN/GaN HEMT devices is very rapid. Due to spontaneous polarization and piezoelectric polarization effects, the AlGaN/GaN heterojunction can generate two-dimensional electron gas with high mobility and can be controlled by a gate voltage. In recent years, the characteristic index of the device is rapidly developed, and especially the microwave power characteristic of the AlGaN/GaN HEMT device is greatly improved. In addition to manufacturing process techniques and device characteristics, modeling work of AlGaN/GaN HEMTs has been the focus of research. Due to the outstanding application in the radio frequency microwave field, the research of AlGaN/GaN HEMT device models is always an important component in the device research field. The device models include a small signal model and a large signal model. For small signal models, the small signal modeling work has been advanced to a certain extent, both domestic and foreign, with the S-parameter method being most widely used. The large signal modeling is a difficulty in microwave power device analysis, and the commonly used GaN HEMT device model can be generally divided into an empirical analysis model, a physical model, a form base model and a behavior model. The performance curves of the existing model transistors in low power and high power areas have mutation and other reasons, so that the accuracy of a large-signal modeling model is seriously influenced, and the model accuracy is low. Disclosure of Invention In order to improve accuracy of a large signal modeling model of a GaN microwave device, the application provides a DNN-based large signal modeling method, a DNN-based large signal modeling system, DNN-based large signal modeling equipment and a DNN-based large signal modeling storage medium. In a first aspect, the present application provides a DNN-based large signal model modeling method, which adopts the following technical scheme: Carrying out large-signal test on the GaN transistor to obtain traveling wave characteristics, wherein the traveling wave characteristics comprise input layer characteristics and output layer characteristics; processing the input layer characteristics to obtain input layer data, and processing the output layer characteristics to obtain output layer data; Extracting real part data and imaginary part data of the input layer data and the output layer data, and training by using DNN according to the real part data and the imaginary part data to obtain a high-power region deep neural network model; extracting amplitude data and phase data of the input layer data and the output layer data, and training by using DNN according to the amplitude data and the phase data to obtain a low-power region deep neural network model; training the traveling wave characteristics according to the high-power region deep neural network model and the low-power region deep neural network model to obtain training traveling wave characteristics, and converting the training traveling wave characteristics into voltage and current characteristics; And constructing and obtaining a large signal behavior model according to the voltage-current characteristics. According to the technical scheme, the large signal characteristics of the low power and the high power range can be fitted at the same time by using the sectional modeling, the weight of traveling wave parameters of the high power area and the low power area is highlighted by adopting the deep neural network to construct the behavior model, the learning sensitivity of the neural network is enhanced, and the acc