Deep Learning-Based Design Methods and Devices for Millimeter-Wave and Terahertz Amplifiers

Abstract

This invention discloses a deep learning-based design method and related apparatus for millimeter-wave and terahertz amplifiers, belonging to the field of millimeter-wave and terahertz integrated circuit design. The amplifier design method includes the following steps: pixelating multiple arbitrary embedded compensation network structures and constructing corresponding two-dimensional matrices; inputting the two-dimensional matrices into a network parameter prediction model to obtain the corresponding embedded network S-parameters; constructing the relationship between the embedded network S-parameters and the amplifier's Y-parameters; globally optimizing the embedded network S-parameters to obtain the optimal embedded compensation network structure; and designing the amplifier based on the optimal embedded compensation network structure. This invention can solve the problems of limited solution space and strong correlation with operating frequency in embedded compensation networks.

Description

Technical Field

[0001] This invention belongs to the field of millimeter-wave and terahertz integrated circuit design, specifically relating to a deep learning-based millimeter-wave and terahertz amplifier design method and related devices. Background Technology

[0002] In the Asia-Pacific Hertz band, the limited gain of CMOS devices is a key factor restricting the design of CMOS Asia-Pacific Hertz amplifiers. Due to the high-frequency characteristics of the Asia-Pacific Hertz band, parasitic parameters within CMOS devices, such as resistance, capacitance, and inductance, have a particularly significant impact on device gain. These parasitic parameters not only reduce device gain but also introduce additional phase delay and noise, thus affecting the overall performance of the amplifier.

[0003] Currently, the common techniques used to reduce the impact of parasitic parameters are device physical characteristic analysis and layout optimization. In-depth analysis of the physical characteristics of CMOS devices reveals the sources and influencing factors of parasitic parameters, allowing for the optimization of device structure and layout to reduce their values. However, while this method can improve device gain to some extent, it remains limited by the inherent physical characteristics of the device itself, making it difficult to achieve significant performance improvements.

[0004] To improve the gain performance of CMOS Asia-Pacific Hertz amplifiers, researchers have introduced embedded compensation theory and methods. Embedded compensation network technology uses additional compensation elements within the network to counteract the effects of parasitic parameters, thereby restoring and improving the device's gain. Patent CN114024514B discloses an overneutralized amplifier structure with a series-type lossy embedded network. By introducing a series-type lossy embedded network into a traditional two-port amplifier network, the unidirectional power gain is increased under preset conditions. Based on this, a Y / Z type LLREN is used to further improve the maximum usable gain, effectively enhancing the amplifier's power gain. However, because embedded compensation network technology relies on network parameter theory analysis and the embedded network typically uses a transmission line structure, the solution space is limited. Furthermore, the transmission line network parameters are strongly correlated with the operating frequency, which limits the performance of the embedded compensation network in the Asia-Pacific Hertz band, making it difficult to simultaneously achieve the requirements of wide bandwidth and high gain. Summary of the Invention

[0005] The purpose of this invention is to provide a design method and related apparatus for millimeter-wave and terahertz amplifiers based on deep learning, which solves the problem of limited solution space and strong correlation with operating frequency of embedded compensation networks, so as to achieve both wide bandwidth and high gain of the amplifier.

[0006] To address the aforementioned technical problems, embodiments of this application provide a deep learning-based design method for millimeter-wave and terahertz amplifiers, comprising the following steps:

[0007] Multiple arbitrary embedded compensation network structures are pixelated and their corresponding two-dimensional matrices are constructed.

[0008] The two-dimensional matrix is ​​input into the network parameter prediction model to obtain the corresponding embedded network S-parameters;

[0009] After establishing the relationship between the S-parameters of the embedded network and the Y-parameters of the amplifier, the S-parameters of the embedded network are globally optimized to obtain the optimal embedded compensation network structure. The amplifier is then designed based on the optimal embedded compensation network structure.

[0010] In some implementations, the process of establishing the network parameter prediction model is as follows: electromagnetic simulation is performed on multiple arbitrary embedded compensation network structures after pixelation to obtain a two-dimensional matrix and the corresponding embedded network S-parameters. The two-dimensional matrix and the embedded network S-parameters are used as the input and output of a convolutional neural network (CNN) respectively. The CNN is then trained and verified to obtain the network parameter prediction model.

[0011] In some implementations, the convolutional neural network includes n convolutional layers and m fully connected layers, where n is the number of convolutional layers and m is the number of fully connected layers. Each convolutional layer is followed by a normalization layer and an activation layer in sequence. The first m-1 fully connected layers are followed by a normalization layer, an activation layer, and a random deactivation layer in sequence. The activation layer uses the Leaky ReLU function.

[0012] In some implementations, a genetic algorithm is used to globally optimize the S-parameters of the embedding network to obtain the optimal embedding compensation network structure, specifically including the following steps:

[0013] The two-dimensional matrices corresponding to the multiple arbitrary embedded compensation network structures are used as the parent generation, and the mean square error between the embedded network S-parameters and the expected S-parameters is used as the cost function.

[0014] The error between the embedded network S-parameters and the expected S-parameters is calculated based on the cost function, and the two-dimensional matrices corresponding to the minimum errors are selected as the first subset.

[0015] Randomly select several two-dimensional matrices from the parent generation and filter out the two two-dimensional matrices corresponding to the minimum error to generate a child generation. Repeat this step until the number of child generations is equal to the number of parent generations. The child generations constitute a second subset.

[0016] The first subset and the second subset are merged into a new population to iterate over the parent generation. The above steps are repeated until a preset number of iterations is reached to obtain the optimal embedding compensation network structure.

[0017] In some implementations, the step of constructing the relationship between the S-parameters of the embedded network and the Y-parameters of the amplifier specifically includes:

[0018] The amplifier's Y-parameters are composed of the Y-parameters of the embedded compensation network structure and the Y-parameters of the transistor;

[0019] The Y parameters of the embedded compensation network structure are obtained based on the S parameters of the embedded network and the identity matrix.

[0020] Based on the Y-parameters and identity matrix of the embedded compensation network structure, and in conjunction with the Y-parameters of the transistor, the relationship between the S-parameters of the embedded network and the Y-parameters of the amplifier is obtained.

[0021] In some implementations, after performing electromagnetic simulation on multiple pixelated arbitrary embedded compensation network structures to obtain a two-dimensional matrix and the corresponding embedded network S-parameters, the method further includes: expanding the two-dimensional matrix and the corresponding embedded network S-parameters using a generative adversarial network.

[0022] To address the aforementioned technical problems, embodiments of this application provide a deep learning-based millimeter-wave and terahertz amplifier design system, comprising:

[0023] An embedded compensation network pixelation module is used to pixelate multiple arbitrary embedded compensation network structures and construct corresponding two-dimensional matrices;

[0024] The embedded network S-parameter prediction module inputs the two-dimensional matrix into the network parameter prediction model to obtain the corresponding embedded network S-parameters.

[0025] The amplifier design module establishes the relationship between the S-parameters of the embedded network and the Y-parameters of the amplifier, then performs global optimization on the S-parameters of the embedded network to obtain the optimal embedded compensation network structure, and designs the amplifier based on the optimal embedded compensation network structure.

[0026] To address the aforementioned technical problems, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable in the processor. When the processor executes the computer program, it implements the steps of the deep learning-based millimeter-wave and terahertz amplifier design method.

[0027] To address the aforementioned technical problems, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the deep learning-based millimeter-wave and terahertz amplifier design method.

[0028] To address the aforementioned technical problems, this application provides a computer program product, which includes a computer program. The computer program, when executed by a processor, implements the steps of the deep learning-based millimeter-wave and terahertz amplifier design method.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] This invention provides a deep learning-based design method for millimeter-wave and terahertz amplifiers. By pixelating an arbitrary embedded compensation network, a two-dimensional matrix corresponding to the pixelated embedded compensation network is constructed. This two-dimensional matrix is ​​then input into a network parameter prediction model to obtain the embedded network's S-parameters. After establishing the relationship between the embedded network's S-parameters and the amplifier's Y-parameters, the embedded network's S-parameters are globally optimized to obtain the optimal embedded compensation network structure. Finally, the amplifier is designed based on the optimal embedded compensation network. This method offers a large solution space, and the optimal embedded compensation network structure enables the acquisition of wideband and high-gain amplifiers, solving the problems of limited solution space and strong correlation with operating frequency in embedded compensation networks. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the amplifier topology using an embedded compensation network in an embodiment of the present invention;

[0032] Figure 2 A schematic diagram of the overall scheme of the deep learning-based millimeter-wave and terahertz amplifier design method provided in the embodiments of the present invention;

[0033] Figure 3 A flowchart illustrating a deep learning-based millimeter-wave and terahertz amplifier design method provided for embodiments of the present invention;

[0034] Figure 4 A schematic diagram of the structure of a deep learning-based millimeter-wave and terahertz amplifier design system provided in an embodiment of the present invention;

[0035] Figure 5 This is a schematic diagram illustrating the process of optimizing the S-parameters of the embedding network using a genetic algorithm in an embodiment of the present invention;

[0036] Figure 6 The following is a simulation diagram of a broadband amplifier design provided in an embodiment of the present invention. In the diagram, (a) is the circuit schematic of the broadband amplifier; and (b) is the simulation result of the broadband amplifier. Detailed Implementation

[0037] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings of this application are intended to cover non-exclusive inclusion.

[0038] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0039] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0040] like Figure 2 and Figure 3 As shown, this embodiment provides a deep learning-based design method for millimeter-wave and terahertz amplifiers, including the following steps:

[0041] S1: Pixelate multiple arbitrary embedded compensation network structures and construct corresponding two-dimensional matrices. Specifically, using the idea of ​​image pixelation, 400,000 20×20 two-dimensional matrices are randomly generated using MATLAB for arbitrary embedded compensation network structures.

[0042] S2, the two-dimensional matrix is ​​input into the network parameter prediction model to obtain the corresponding embedded network S-parameters. Specifically, on the PyTorch platform, the two-dimensional matrix and the embedded network S-parameters generated by electromagnetic simulation are used as the input and output of the convolutional neural network, respectively. The convolutional neural network is trained and validated to obtain the network parameter prediction model. The function fitting relationship between the embedded network S-parameters and the embedded compensation network structure is constructed through the network parameter prediction model.

[0043] This convolutional neural network architecture includes 12 convolutional layers and 5 fully connected layers. Each convolutional layer is followed by a normalization layer and an activation layer. The first four fully connected layers are followed by a normalization layer, an activation layer, and a random deactivation layer. The activation layer uses the Leaky ReLU function to prevent gradient vanishing, ensuring good learning performance of the convolutional neural network. The random deactivation layer reduces the possibility of overfitting during learning. In this embodiment, each convolutional layer has 102 channels. The kernel sizes of the first to twelfth convolutional layers are 16×16, 12×12, 10×10, 6×6, 5×5, 5×5, 4×4, 4×4, 4×4, 3×3, 3×3, and 3×3, respectively. The neuron data of the first to fifth fully connected layers are 1000, 1000, 800, 800, and 120, respectively.

[0044] The 20×20 two-dimensional matrix in S1 is input into the CNN above, where the first row, first column, eighteenth row and eighteenth column of the two-dimensional matrix represent the excitation port positions, respectively. The CNN output is the embedded network S-parameters of the electromagnetic simulation of the embedded compensation network structure represented by the two-dimensional matrix.

[0045] Preferably, during electromagnetic simulation, a generative adversarial network (GAN) is used to expand the dataset by using the two-dimensional matrix and the embedded network S-parameters as the input and output of the GAN, respectively. This reduces the time required for data generation. Then, on the PyTorch platform, the two-dimensional matrix and the embedded network S-parameters are used as the input and output of a convolutional neural network (CNN), and the CNN is trained and validated to obtain a network parameter prediction model. This GAN can take the form of Deep Convolutional Generative Adversarial Network (DCGAN), Progressive Growing Generative Adversarial Network (PGAN), or Wasserstein Generative Adversarial Network (WGAN), etc.

[0046] S3. After establishing the relationship between the S-parameters of the embedded network and the Y-parameters of the amplifier, the S-parameters of the embedded network are globally optimized to obtain the optimal embedded compensation network structure. The amplifier is then designed based on the optimal embedded compensation network.

[0047] like Figure 1 As shown, the Y-parameters of the amplifier are composed of the Y-parameters of the embedded compensation network structure and the Y-parameters of the transistor, i.e., the following equation (1):

[0048] (1)

[0049] In equation (1), For the amplifier's Y parameters, For the Y-parameters of the transistor, The Y parameters are used to embed the compensation network structure.

[0050] The relationship between the S-parameters of the embedded network and the Y-parameters of the embedded compensation network structure is established, as shown in equation (2):

[0051] (2)

[0052] In equation (2), It is the identity matrix. To embed the network S-parameters.

[0053] Based on equations (1) and (2), the relationship between the S-parameters of the embedded network and the Y-parameters of the amplifier is obtained, namely equation (3):

[0054] (3)

[0055] By combining the relationship between the S-parameters of the embedding network and the Y-parameters of the amplifier, the optimal embedding compensation network structure is obtained by globally optimizing the S-parameters of the embedding network using a genetic algorithm. This genetic algorithm can be replaced by other global optimization algorithms, such as: Empire competition algorithm, particle swarm optimization algorithm, simulated annealing algorithm, greedy algorithm, and neighborhood search algorithm.

[0056] like Figure 5 As shown, the specific optimization process of the genetic algorithm is as follows:

[0057] S3.1 First, initialize the population. Use MATLAB to randomly generate 2048 two-dimensional matrices as parents as described in S1. Use the CNN trained in S2 to predict the embedding network S parameters of the embedding compensation network represented by the 2048 two-dimensional matrices.

[0058] S3.2, with the expected S-parameters as the objective, uses the mean squared error between the embedded network S-parameters and the expected S-parameters as the cost function, as shown in the following equation:

[0059]

[0060]

[0061]

[0062] In the above formula, (Cost Function) is the cost function. To embed the network S-parameters, For the expected S-parameters, This represents the functional fit between the S-parameters of the embedded network and the structure of the embedded compensation network.

[0063] The error between the embedded network S-parameters and the expected S-parameters is calculated using the above cost function. The minimum error between the expected S-parameters and the embedded network S-parameters is used as the objective function, and the two-dimensional matrices corresponding to the 16 minimum errors are directly inherited by the offspring.

[0064] S3.3: Randomly select 256 two-dimensional matrices from the parent generation, read the corresponding errors, and select the two two-dimensional matrices corresponding to the minimum errors as parent individuals. After generating a child generation through crossover, repeat S3.3 until the number of child generations is equal to the number of parent generations described in S3.1.

[0065] S3.4 updates the population using the offspring generated in S3.3 and the two-dimensional matrix corresponding to the 16 minimum errors in S3.2. The new population also includes the offspring generated by crossover in S3.3 and the two-dimensional matrix corresponding to the 16 minimum errors in S3.2.

[0066] Repeat steps S3.2 to S3.4 until the preset number of iterations is 100, then generate the 100th generation population. Finally, select the two-dimensional matrix corresponding to the minimum error as the optimal solution to obtain the optimal embedding compensation network structure.

[0067] Finally, the amplifier is designed based on the optimal embedded compensation network structure, such as... Figure 6 As shown, electromagnetic simulation results demonstrate that the proposed method designs an Asia-Pacific Hertz amplifier with an operating frequency range of 180 GHz to 220 GHz, a gain greater than 1.8 dB, and good stability, consuming a total power of approximately 7.8 mA. These simulation results validate the amplifier design method based on convolutional neural networks and genetic algorithms provided in this embodiment. This method effectively addresses the limitations of existing embedded compensation networks in terms of solution space and their strong correlation with operating frequency, achieving a broadband and high-gain amplifier.

[0068] The deep learning-based millimeter-wave and terahertz amplifier design method provided in this embodiment has the following advantages:

[0069] (1) Using the idea of ​​image pixelation, the arbitrary embedded compensation network structure is decomposed into pixels. The arbitrary embedded compensation network structure is represented by a two-dimensional matrix. Then, the fitting relationship between the embedded compensation network structure and the S-parameters of the embedded network is constructed using a convolutional neural network. The solution space is large. Finally, the broadband and high-gain millimeter-wave and terahertz amplifiers can be obtained through the iteration of the genetic algorithm. This method can solve the problem that the solution space of the embedded compensation network is limited and strongly correlated with the operating frequency. It can explore the ability of active devices to provide gain in the Asia-Pacific Hertz band, so that broadband and high-gain amplifiers can be designed.

[0070] (2) Using generative adversarial networks to expand the embedded compensation network structure and the dataset formed by its corresponding embedded network S-parameters can reduce the time required for electromagnetic simulation and improve design efficiency.

[0071] (3) According to the different requirements of the amplifier, by modifying the cost function of the genetic algorithm, the optimal embedding compensation network structure that meets the performance requirements can be generated by combining the fitting relationship between the arbitrary pixelated embedding compensation network structure constructed by the convolutional neural network algorithm and the S-parameters of the embedding network. It has the advantages of strong applicability and efficient design.

[0072] like Figure 4 As shown, this embodiment also provides a deep learning-based millimeter-wave and terahertz amplifier design system, including:

[0073] An embedded compensation network pixelation module is used to pixelate multiple arbitrary embedded compensation network structures and construct corresponding two-dimensional matrices;

[0074] The embedded network S-parameter prediction module inputs the two-dimensional matrix into the network parameter prediction model to obtain the corresponding embedded network S-parameters.

[0075] The embedded network S-parameter prediction module is specifically used to input the two-dimensional matrix into a convolutional neural network and output the embedded network S-parameters to obtain a dataset consisting of the embedded network S-parameters corresponding to the two-dimensional matrix. The convolutional neural network is then trained and validated using this dataset to obtain a network parameter prediction model. The convolutional neural network includes n convolutional layers and m fully connected layers, where n is the number of convolutional layers and m is the number of fully connected layers. Each convolutional layer is followed by a normalization layer and an activation layer. The first m-1 fully connected layers are followed by a normalization layer, an activation layer, and a random deactivation layer. The activation layer uses the Leaky ReLU function.

[0076] The embedded network S-parameter prediction module further includes:

[0077] A generative adversarial module is used to expand the two-dimensional matrix and its corresponding embedded network S-parameters using a generative adversarial network.

[0078] The amplifier design module establishes the relationship between the S-parameters of the embedded network and the Y-parameters of the amplifier, then performs global optimization on the S-parameters of the embedded network to obtain the optimal embedded compensation network structure, and designs the amplifier based on the optimal embedded compensation network structure.

[0079] The amplifier design module is specifically used to obtain the Y parameters of the embedded compensation network structure based on the S parameters of the embedded network and the identity matrix; and to obtain the relationship between the S parameters of the embedded network and the Y parameters of the amplifier based on the Y parameters of the embedded compensation network structure and the identity matrix, combined with the Y parameters of the transistor.

[0080] The amplifier design module is further configured to: combine the relationship between the embedded network S-parameters and the amplifier Y-parameters, construct a parent generation from the two-dimensional matrices corresponding to the multiple arbitrary embedded compensation network structures, and use the mean square error between the embedded network S-parameters and the expected S-parameters as a cost function; calculate the error between the embedded network S-parameters and the expected S-parameters according to the cost function, and select several two-dimensional matrices corresponding to the minimum error as a first subset; randomly select several two-dimensional matrices from the parent generation and select two two-dimensional matrices corresponding to the minimum error to cross-generate a child generation, repeating this step until the number of child generations equals the number of parent generations, and the child generations constitute a second subset; merge the first subset and the second subset as a new population to iterate the parent generation, repeating the above steps until a preset number of iterations is reached to obtain the optimal embedded compensation network structure, and design the amplifier according to the optimal embedded compensation network structure.

[0081] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0082] This embodiment also provides a computer device, which includes a processor and a memory. The memory is used to store a computer program (in this embodiment, the computer program includes computational components and iterative components, capable of model calculation and model updating). The computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to realize the corresponding method flow or corresponding function. The processor described in this embodiment can be used for the operation of millimeter-wave and terahertz amplifier design methods based on deep learning.

[0083] This embodiment also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the deep learning-based millimeter-wave and terahertz amplifier design method in the above embodiment.

[0084] This embodiment also provides a computer program product, which includes a computer program that, when executed by a processor, implements the corresponding steps of the deep learning-based millimeter-wave and terahertz amplifier design method described in the above embodiment.

[0085] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0086] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0087] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0088] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A deep learning-based design method for millimeter-wave and terahertz amplifiers, characterized in that, Includes the following steps: Multiple arbitrary embedded compensation network structures are pixelated and their corresponding two-dimensional matrices are constructed. The two-dimensional matrix is ​​input into the network parameter prediction model to obtain the corresponding embedded network S-parameters; After establishing the relationship between the S-parameters of the embedded network and the Y-parameters of the amplifier, the S-parameters of the embedded network are globally optimized to obtain the optimal embedded compensation network structure. The amplifier is then designed based on the optimal embedded compensation network structure. The process of establishing the network parameter prediction model is as follows: electromagnetic simulation is performed on multiple arbitrary embedded compensation network structures after pixelation to obtain a two-dimensional matrix and the corresponding embedded network S-parameters. The two-dimensional matrix and the embedded network S-parameters are used as the input and output of the convolutional neural network, respectively. The convolutional neural network is trained and verified to obtain the network parameter prediction model. The optimal embedding compensation network structure is obtained by globally optimizing the S-parameters of the embedding network using a genetic algorithm, specifically including the following steps: The two-dimensional matrices corresponding to the multiple arbitrary embedded compensation network structures are used as the parent generation, and the mean square error between the embedded network S-parameters and the expected S-parameters is used as the cost function. The error between the embedded network S-parameters and the expected S-parameters is calculated based on the cost function, and the two-dimensional matrices corresponding to the minimum errors are selected as the first subset. Randomly select several two-dimensional matrices from the parent generation and filter out the two two-dimensional matrices corresponding to the minimum error to generate a child generation. Repeat this step until the number of child generations is equal to the number of parent generations. The child generations constitute a second subset. The first subset and the second subset are merged into a new population to iterate the parent generation. The above steps are repeated until a preset number of iterations are reached to obtain the optimal embedding compensation network structure. The steps for constructing the relationship between the S-parameters of the embedded network and the Y-parameters of the amplifier specifically include: The amplifier's Y-parameters are composed of the Y-parameters of the embedded compensation network structure and the Y-parameters of the transistor; The Y parameters of the embedded compensation network structure are obtained based on the S parameters of the embedded network and the identity matrix. Based on the Y-parameters and identity matrix of the embedded compensation network structure, and in conjunction with the Y-parameters of the transistor, the relationship between the S-parameters of the embedded network and the Y-parameters of the amplifier is obtained.

2. The deep learning-based millimeter-wave and terahertz amplifier design method according to claim 1, characterized in that, The convolutional neural network includes n convolutional layers and m fully connected layers, where n is the number of convolutional layers and m is the number of fully connected layers. Each convolutional layer is followed by a normalization layer and an activation layer in sequence. The first m-1 fully connected layers are followed by a normalization layer, an activation layer, and a random deactivation layer in sequence. The activation layer uses the Leaky ReLU function.

3. The deep learning-based millimeter-wave and terahertz amplifier design method according to claim 1, characterized in that, After performing electromagnetic simulation on multiple pixelated arbitrary embedded compensation network structures to obtain a two-dimensional matrix and the corresponding embedded network S-parameters, the method further includes: expanding the two-dimensional matrix and its corresponding embedded network S-parameters using a generative adversarial network.

4. A deep learning-based millimeter-wave and terahertz amplifier design system, characterized in that, include: An embedded compensation network pixelation module is used to pixelate multiple arbitrary embedded compensation network structures and construct corresponding two-dimensional matrices; The embedded network S-parameter prediction module inputs the two-dimensional matrix into the network parameter prediction model to obtain the corresponding embedded network S-parameters. The amplifier design module constructs the relationship between the S-parameters of the embedded network and the Y-parameters of the amplifier, then performs global optimization on the S-parameters of the embedded network to obtain the optimal embedded compensation network structure, and designs the amplifier based on the optimal embedded compensation network structure. The process of establishing the network parameter prediction model is as follows: electromagnetic simulation is performed on multiple arbitrary embedded compensation network structures after pixelation to obtain a two-dimensional matrix and the corresponding embedded network S-parameters. The two-dimensional matrix and the embedded network S-parameters are used as the input and output of the convolutional neural network, respectively. The convolutional neural network is trained and verified to obtain the network parameter prediction model. The optimal embedding compensation network structure is obtained by globally optimizing the S-parameters of the embedding network using a genetic algorithm, specifically including the following steps: The two-dimensional matrices corresponding to the multiple arbitrary embedded compensation network structures are used as the parent generation, and the mean square error between the embedded network S-parameters and the expected S-parameters is used as the cost function. The error between the embedded network S-parameters and the expected S-parameters is calculated based on the cost function, and the two-dimensional matrices corresponding to the minimum errors are selected as the first subset. Randomly select several two-dimensional matrices from the parent generation and filter out the two two-dimensional matrices corresponding to the minimum error to generate a child generation. Repeat this step until the number of child generations is equal to the number of parent generations. The child generations constitute a second subset. The first subset and the second subset are merged into a new population to iterate the parent generation. The above steps are repeated until a preset number of iterations are reached to obtain the optimal embedding compensation network structure. The steps for constructing the relationship between the S-parameters of the embedded network and the Y-parameters of the amplifier specifically include: The amplifier's Y-parameters are composed of the Y-parameters of the embedded compensation network structure and the Y-parameters of the transistor; The Y parameters of the embedded compensation network structure are obtained based on the S parameters of the embedded network and the identity matrix. Based on the Y-parameters and identity matrix of the embedded compensation network structure, and in conjunction with the Y-parameters of the transistor, the relationship between the S-parameters of the embedded network and the Y-parameters of the amplifier is obtained.

5. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable in the processor, wherein the processor executes the computer program to implement the steps of the deep learning-based millimeter-wave and terahertz amplifier design method according to any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the deep learning-based millimeter-wave and terahertz amplifier design method according to any one of claims 1 to 3.

7. A computer program product, said computer product comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the deep learning-based millimeter-wave and terahertz amplifier design method according to any one of claims 1 to 3.

Images