CN-122017511-A - AI-assisted high-frequency broadband circuit parasitic effect eliminating method and device

CN122017511ACN 122017511 ACN122017511 ACN 122017511ACN-122017511-A

Abstract

The application provides an AI-assisted high-frequency broadband circuit parasitic effect eliminating method and device, wherein the method comprises the steps of obtaining a first measurement parameter, a second measurement parameter and a working condition parameter of a target device, wherein the second measurement parameter is a measurement parameter of the target device in an open circuit, short circuit and direct connection structure; based on the second measurement parameter, the first measurement parameter is subjected to preliminary de-embedding processing by a de-embedding algorithm to obtain a preliminary de-embedding parameter, and the data set of the first measurement parameter, the second measurement parameter, the working condition parameter and the preliminary de-embedding parameter is input into a target neural network model so as to output the target device parameter through the target neural network model. According to the method, the traditional algorithm is used as a part of data preprocessing, so that the neural network can learn and correct by utilizing the preliminary de-embedding parameters generated by the traditional algorithm, the difficulty that the neural network learns directly from complex original data is avoided, and the training efficiency is improved.

Inventors

DU YUAN
TANG ZHENGCHAO
ZHANG JINHAI
ZHOU WENXIA
Meng Shuohong
GUO JIACHENG
DU LI

Assignees

南京大学

Dates

Publication Date: 20260512
Application Date: 20260211

Claims (10)

1. An AI-assisted high frequency broadband circuit parasitics rejection method, comprising: Acquiring a first measurement parameter, a second measurement parameter and a working condition parameter of a target device, wherein the second measurement parameter is a measurement parameter of the target device in an open-circuit structure, a short-circuit structure and a straight-through structure; based on the second measurement parameters, performing preliminary de-embedding processing on the first measurement parameters by adopting a de-embedding algorithm to obtain preliminary de-embedding parameters; and inputting the data sets of the first measurement parameter, the second measurement parameter, the working condition parameter and the preliminary de-embedding parameter into a target neural network model so as to output a target device parameter through the target neural network model.
2. The AI-assisted high frequency broadband circuit parasitics rejection method of claim 1, wherein the target neural network model is a pre-trained neural network model; the method further comprises the steps of: Acquiring a training sample set, wherein the training sample set comprises first measurement parameters, second measurement parameters, working condition parameters, preliminary de-embedding parameters and real network parameters of a plurality of sample devices; constructing a basic neural network model; And training the basic neural network model by taking the training sample set as an input characteristic and the real network parameter as a training target so as to obtain the neural network model.
3. The AI-assisted high frequency broadband circuit parasitics rejection method of claim 2, wherein the real network parameters are obtained by simulation software under target de-embedding conditions.
4. The AI-assisted high frequency broadband circuit parasitics rejection method of claim 1, wherein the outputting of target device parameters by the target neural network model is followed by the method further comprising: performing jump detection and smoothing on the target device parameters to obtain correction parameters; and determining the correction parameter as the target device parameter.
5. The AI-assisted high frequency broadband circuit parasitics rejection method of claim 4, wherein performing jump detection and smoothing on the target device parameter to obtain a correction parameter comprises: Detecting the parameter variation gradient of each frequency point in the sequence of the target device parameter variation along with the frequency; When the parameter change gradient of a certain frequency point exceeds a target gradient, marking the frequency point as an abnormal jump point, wherein the target gradient is a threshold value determined by the overall change gradient of the sequence or the change gradient of an adjacent frequency point; and carrying out smooth correction on the parameter value of the abnormal jump point to control the frequency response curve of the target device parameter to be continuously smoothed so as to obtain a correction parameter.
6. The AI-assisted high frequency broadband circuit parasitics rejection method of claim 1, wherein the preliminary de-embedding of the first measurement parameter with a de-embedding algorithm based on the second measurement parameter comprises: based on the measurement parameters of the open circuit structure, the short circuit structure and the through structure, constructing a parasitic network model corresponding to the open circuit structure, the short circuit structure and the through structure; and performing de-embedding operation on the first measurement parameter by using the parasitic network model so as to strip parasitic effects contained in the first measurement parameter and obtain the preliminary de-embedding parameter.
7. The AI-assisted high frequency broadband circuit parasitics rejection method of claim 1, wherein the target neural network model is GRNN.
8. The AI-assisted high frequency broadband circuit parasitics rejection method of claim 1, wherein the operating condition parameters include operating frequency and bias voltage; the outputting the target device parameters through the target neural network model comprises the following steps: And taking the working frequency and the bias voltage as input features in the data set, and inputting the first measurement parameter, the second measurement parameter and the preliminary de-embedding parameter into the target neural network model, wherein the target neural network model corrects the parameters of the target device according to the working frequency and the bias voltage so as to output the parameters of the target device through the target neural network model.
9. An AI-assisted high frequency broadband circuit parasitics rejection apparatus, comprising: The device comprises an acquisition module, a control module and a control module, wherein the acquisition module is used for acquiring a first measurement parameter, a second measurement parameter and a working condition parameter of a target device, wherein the second measurement parameter is a measurement parameter of the target device in an open-circuit structure, a short-circuit structure and a straight-through structure; the processing module is used for carrying out preliminary de-embedding processing on the first measurement parameters by adopting a de-embedding algorithm based on the second measurement parameters so as to obtain preliminary de-embedding parameters; And the de-embedding module is used for inputting the data set of the first measurement parameter, the second measurement parameter, the working condition parameter and the preliminary de-embedding parameter into a target neural network model so as to output the target device parameter through the target neural network model.
10. A computing device comprising a processor and a memory, the processor configured to execute instructions stored in the memory to cause the computing device to perform the method of any one of claims 1 to 8.

Description

AI-assisted high-frequency broadband circuit parasitic effect eliminating method and device Technical Field The application relates to the technical field of semiconductor testing and radio frequency microwave measurement, in particular to an AI-assisted high-frequency broadband circuit parasitic effect eliminating method and device. Background In rf microwave and semiconductor testing, high frequency device feature extraction presents challenges. The parasitic effect is introduced into the structures such as the test clamp and the bonding pad, so that the measurement result deviates from the intrinsic characteristics of the device, and the real characteristics of the target device are accurately separated from the overall result containing the clamp effect, so that the modeling key requirement is realized. For this requirement, de-embedding algorithms are used to strip parasitic effects, which can be classified into one, two, three, etc. steps depending on the number of test structures used. One-step method is an open-structure-based "open" method, two-step method is an open-short "method combining open and short circuits, and three-step method includes" open-short-load "method and the like. Furthermore, neural network technology has been introduced into this field to improve accuracy. For example, there are methods to achieve de-embedding directly based on artificial neural networks, without the need for traditional calibration pieces. In addition, the deep neural network is combined with the TRL de-embedding method to reduce the requirements on chip area and probe placement. However, the above solution combined with neural networks has not been effectively integrated with the "open-short-thru" de-embedding procedure. This results in a decrease in accuracy of the conventional "open-short-thru" method due to model limitations in the high frequency band, while the neural network also fails to fully utilize the optimized data generated by the conventional algorithm to improve its own training and correction efficiency. Disclosure of Invention The application provides an AI-assisted high-frequency broadband circuit parasitic effect eliminating method and device, which are used for solving the problem that a neural network and an open-short-thru cannot be effectively combined. In a first aspect, the present application provides an AI-assisted high frequency broadband circuit parasitics rejection method, including: Acquiring a first measurement parameter, a second measurement parameter and a working condition parameter of a target device, wherein the second measurement parameter is a measurement parameter of the target device in an open-circuit structure, a short-circuit structure and a straight-through structure; based on the second measurement parameters, performing preliminary de-embedding processing on the first measurement parameters by adopting a de-embedding algorithm to obtain preliminary de-embedding parameters; and inputting the data sets of the first measurement parameter, the second measurement parameter, the working condition parameter and the preliminary de-embedding parameter into a target neural network model so as to output a target device parameter through the target neural network model. In some possible embodiments, the target neural network model is a pre-trained neural network model; the method further comprises the steps of: Acquiring a training sample set, wherein the training sample set comprises first measurement parameters, second measurement parameters, working condition parameters, preliminary de-embedding parameters and real network parameters of a plurality of sample devices; constructing a basic neural network model; And training the basic neural network model by taking the training sample set as an input characteristic and the real network parameter as a training target so as to obtain the neural network model. In some possible embodiments, the real network parameters are obtained by simulation software under target de-embedding conditions. In some possible embodiments, the outputting, by the target neural network model, the target device parameter, after which the method further includes: performing jump detection and smoothing on the target device parameters to obtain correction parameters; and determining the correction parameter as the target device parameter. In some possible embodiments, the performing jump detection and smoothing on the target device parameter to obtain a correction parameter includes: Detecting the parameter variation gradient of each frequency point in the sequence of the target device parameter variation along with the frequency; When the parameter change gradient of a certain frequency point exceeds a target gradient, marking the frequency point as an abnormal jump point, wherein the target gradient is a threshold value determined by the overall change gradient of the sequence or the change gradient of an adjacent frequency point; and carrying out smo