Method for screening two-dimensional semiconductor material based on density functional theory and machine learning

Abstract

The invention discloses a method for screening a two-dimensional semiconductor material based on a density functional theory and machine learning, and belongs to the field of machine learning, and the method comprises the steps: S1, calculating the electronic structure and energy band characteristics of a novel low-dimensional electronic material through density functional theory calculation software VASP; utilizing a machine learning algorithm to select a novel low-dimensional electronic material with potential; s2, designing a novel field effect transistor and photoelectric device structure based on the material selected in the step S1; s3, optimizing device performance parameters by using a machine learning algorithm; s4, designing a heterojunction structure, and analyzing the influence of an interface state on carrier transportation; s5, utilizing a machine learning algorithm to guide and regulate interface states and defects; and S6, applying machine learning to assist in researching the electronic structure and band gap characteristic change of the two-dimensional material, and designing a strain sensor and an adjustable band gap device. Through the combination of DFT and machine learning, the electronic transport and photoelectric properties of the novel low-dimensional electronic material are analyzed, and the design of a high-performance semiconductor device is optimized.

Description

Technical Field

[0001] The present invention relates to the field of machine learning technology, and in particular to a method for screening two-dimensional semiconductor materials based on density functional theory and machine learning. Background Art

[0002] With the rapid development of information technology and new energy technology, research in the field of materials science has been continuously promoted. Low-dimensional electronic materials have become a hot topic in current scientific research due to their unique physical properties and broad application prospects. In particular, two-dimensional semiconductor materials, with their controllable thickness, high carrier mobility, and sensitive photoelectric response, have shown great potential in the next generation of electronic devices, photoelectric conversion devices, and thermoelectric devices. Studying the transport properties and photoelectric characteristics of low-dimensional electronic materials will not only help reveal their basic physical mechanisms, but also provide theoretical guidance for the design and optimization of new semiconductor devices.

[0003] Low-dimensional materials such as graphene, transition metal dichalcogenides (TMDs), black phosphorus, etc., due to their unique two-dimensional structure and quantum effects, show physical properties different from traditional three-dimensional materials. These materials have significant advantages in electronic structure, energy band characteristics, thermal conductivity, photoelectric response, etc., making them have broad application prospects in the future fields of nanoelectronics, optoelectronics and energy conversion. However, the research on these new low-dimensional materials is still in its early stages, and many basic physical mechanisms are still not fully understood, especially the intrinsic connection between the microstructure of the material and the transport properties needs to be further explored. Summary of the invention

[0004] In order to solve the technical problems existing in the above-mentioned prior art that there is currently little research on new low-dimensional materials, many basic physical mechanisms are still not fully understood, and especially the intrinsic connection between the material microstructure and the transport properties is still unclear, the present invention provides a method for screening two-dimensional semiconductor materials based on density functional theory and machine learning.

[0005] To implement the above technical solution, the present invention provides a method for screening two-dimensional semiconductor materials based on density functional theory and machine learning, comprising the steps of:

[0006] S1: Use density functional theory calculation software VASP to calculate the electronic structure and band characteristics of new low-dimensional electronic materials; combine first-principles calculations to deeply analyze the relationship between the material's microstructure and electronic transport properties, and

[0007] Use machine learning algorithms to select new low-dimensional electronic materials with potential;

[0008] S2: Based on the materials selected in step S1, use the semiconductor device simulation software Sentaurus or TCAD to design

[0009] Design and simulate semiconductor devices;

[0010] S3: Optimizing semiconductor device performance parameters using machine learning algorithms;

[0011] S4: Design heterojunction structures and analyze the effects of interface states on carrier transport;

[0012] S5: Through surface modification and doping techniques, machine learning algorithms are used to guide the regulation of interface states and defects;

[0013] S6: Introduce strain control and external electric field, use machine learning to assist in studying the changes in the electronic structure and band gap characteristics of two-dimensional materials, design strain sensors and adjustable band gap devices, and optimize the device operating parameters under electric field control.

[0014] Further, in step S1, the steps of using density functional theory to calculate the electronic structure and energy band characteristics of the new two-dimensional semiconductor material are as follows:

[0015] S11: Electron density calculation: The Kohn-Sham equation is used to solve the single electron wave function and energy of the system, and then the electron density is calculated:

[0016]

[0017] S12: Exchange-correlation energy calculation: Based on the local density approximation LDA or the generalized gradient approximation GGA, the exchange-correlation energy is calculated:

[0018] E xc [n(r)]=∫n(r)∈ xc [n(r)]dr

[0019] S13: Band structure calculation: The band structure of the two-dimensional material is obtained by calculating the periodic structure of the system.

[0020] Furthermore, in step S1, the first-principles calculation includes the calculation of the electronic structure of the material, the calculation of the relationship between the energy band structure and the electron mobility, the calculation of the relationship between the density of states and the conductivity, and the calculation of the influence of the microstructure on the electron transport.

[0021] Furthermore, in step S1, the steps of selecting potential new low-dimensional electronic materials using a machine learning algorithm are as follows:

[0022] A1: Data input, using DFT calculation results as training data;

[0023] Training data input: Use DFT calculation to obtain the electronic structure, band structure and state density of the material as training data;

[0024] Parameter input: input the material physical parameters of electron density distribution, band gap size, electron mobility and state density obtained through DFT calculation;

[0025] Dataset construction: Use the data obtained through the above steps to establish a training set representing material properties as a

[0026] Inputs to machine learning models;

[0027] A2: Model training, through machine learning model training, extract the relationship between material performance and structure;

[0028] Model selection, using machine learning models suitable for processing complex data relationships to establish a mapping between the material microstructure and its physical properties;

[0029] Feature extraction, finding the key variables that affect material properties from a large amount of input data;

[0030] A3: Performance prediction, predicting the performance of uncalculated materials; the model predicts the performance parameters of band gap, mobility and conductivity based on its structural information;

[0031] A4: Screening and sorting, screening materials according to performance indicators;

[0032] Screening basis, screening criteria include band gap width, electron mobility and photoelectric response speed;

[0033] Feedback optimization: material data showing excellent potential is fed back into DFT calculations to obtain higher-precision calculation verification.

[0034] Further, in step S2, the steps of designing and simulating the semiconductor device using semiconductor device simulation software Sentaurus or TCAD include:

[0035] S21: Use semiconductor device simulation software TCAD or Sentaurus to construct the geometric structure of transistors or optoelectronic devices. The specific component procedures include:

[0036] Material selection: using two-dimensional semiconductor materials selected by DFT calculation and machine learning as channel materials; device geometry design: building a physical model of transistors or optoelectronic devices and setting device geometry parameters, including gate length, channel length, channel thickness, source and drain electrode positions, and gate dielectric thickness;

[0037] Doping design, setting the doping type and concentration of semiconductor devices during simulation;

[0038] S22: Define the material physical model. The specific process is as follows:

[0039] Electron Transport Model: Set up the transport model used to simulate electron migration in the material;

[0040] Material parameter setting: input material parameters obtained from DFT calculation; the material parameters include band gap, electron mobility, state density and electron effective mass;

[0041] Interface state and defect model: define interface state density and defect density parameters, and simulate the effects of these factors on the device

[0042] Performance impact;

[0043] S23: Conduct electrical and optical property simulations to simulate the working characteristics of the device;

[0044] The electrical characteristic simulation includes: simulating current-voltage characteristics (IV curve); simulating key parameters such as on-state current, leakage current, switching ratio and subthreshold swing;

[0045] The optical characteristic simulation includes: simulating photocurrent-voltage characteristics (IV characteristics);

[0046] S24: Perform design optimization and iteration. The optimization and iteration process includes:

[0047] Parameter optimization: adjust the key geometric parameters and material physical parameters of the device based on the simulation results;

[0048] Performance comparison, comparative analysis with devices made of traditional materials;

[0049] Defect control: by simulating the impact of interface defects and doping concentration on device performance, specific optimization measures are proposed;

[0051] S25: Output and visualization results.

[0052] Furthermore, in step S3, the device performance parameters include gate voltage, channel length, and source and drain electrode materials.

[0053] Further, in step S3, density functional theory DFT and machine learning ML are combined to form a material design and optimization platform, including the steps of:

[0054] S31: The data obtained by DFT calculation can be used as a training set;

[0055] S32: Using trained machine learning models, it is possible to learn implicit patterns and laws from a large number of material properties, thereby quickly predicting the performance of new materials;

[0056] S33: adjusting its model parameters according to the calculation results of DFT;

[0057] S34: By analyzing the DFT calculation results, we use machine learning algorithms to optimize the key parameters of the transistor’s gate voltage and channel length.

[0058] Further, in step S4, the heterojunction structure includes graphene / molybdenum disulfide and black phosphorus / tungsten disulfide WS2.

[0059] Furthermore, in step S6, the specific implementation process of optimizing the device operating parameters under electric field regulation includes:

[0060] S61: Machine learning adjusts model parameters based on DFT calculation results;

[0061] S611: Calculate the electronic structure, band gap and physical properties of the material using DFT software (such as VASP);

[0062] S612: inputting the material property data obtained by DFT calculation into the machine learning model as a training set;

[0063] S613: When the machine learning model receives new material data, it can make more accurate predictions about the properties of the new material based on the properties previously learned from DFT calculations. As more DFT data is input, the model's predictive ability continues to improve;

[0064] S62: DFT selects computational parameters based on machine learning feedback;

[0065] S621: Before starting the DFT calculation, the machine model uses its fast prediction function to perform preliminary screening of large batches of materials;

[0066] S622: Based on the preliminary prediction results of machine learning, DFT calculations select promising materials for detailed electronic structure calculations;

[0067] S623: Machine learning provides more appropriate calculation parameter suggestions based on past DFT calculation results;

[0068] S624: After completing the DFT calculation, the new data is fed back to the machine learning model to continuously optimize the model parameters.

[0069] Furthermore, the preliminary screening includes predicting the band gap and mobility performance of the material through simple structural parameters.

[0070] In summary, the present invention has the following beneficial effects compared with the prior art:

[0071] This invention aims to deeply analyze the electronic transport and optoelectronic properties of new low-dimensional electronic materials through the combination of density functional theory (DFT) and machine learning, and then provide guidance for the design and optimization of high-performance semiconductor devices. This invention will not only help promote the frontier development of materials science in theory, but also provide key technical support for the industrialization of next-generation electronic and optoelectronic devices, and has important academic and practical application value.

[0072] The present invention makes full use of the advantages of density functional theory (DFT) in accurately describing the electronic structure and properties of materials, and provides a reliable and high-quality data basis for machine learning. At the same time, the powerful data processing and pattern recognition capabilities of machine learning algorithms are used to quickly analyze and predict a large amount of material data, thus realizing the organic combination of theoretical calculation and data-driven.

[0073] The present invention uses advanced machine learning algorithms, such as convolutional neural networks (CNN) and recurrent neural networks (RNN) in deep learning, as well as reinforcement learning algorithms, to predict and optimize the performance of new low-dimensional electronic materials. These algorithms can automatically learn complex patterns and laws from data, thereby more accurately revealing the relationship between material performance and structure.

[0074] The present invention establishes a collaborative optimization mechanism between density functional theory and machine learning algorithms, and achieves mutual promotion and optimization between the two through continuous iteration and feedback. For example, the machine learning algorithm can adjust the model parameters according to the results of density functional theory calculations to improve the accuracy of predictions; and the density functional theory (DFT) calculation can also select more targeted calculation parameters and methods according to the feedback of the machine learning algorithm, thereby improving the calculation efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0076] Figure 1 A flow chart of a method for screening two-dimensional semiconductor materials based on density functional theory and machine learning provided by the present invention;

[0077] Figure 2 Flowchart for predicting and screening material properties for machine learning algorithms;

[0078] Figure 3 Flowchart for designing and simulating semiconductor devices based on new low-dimensional electronic materials using professional semiconductor device simulation software (Sentaurus or TCAD);

[0079] Figure 4Flowchart for predicting and optimizing the properties of new low-dimensional electronic materials through machine learning algorithms;

[0080] Figure 5 Flowchart of the feedback optimization mechanism for machine learning and density functional theory (DFT). DETAILED DESCRIPTION

[0081] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0082] See also Figure 1 As shown, the present invention provides a method for screening two-dimensional semiconductor materials based on density functional theory and machine learning, comprising the steps of:

[0083] S1: Use density functional theory (DFT) calculation software such as VASP to accurately calculate the electronic structure and band characteristics of new low-dimensional electronic materials. Combined with first-principles calculations, deeply analyze the relationship between the microstructure of the material and the electronic transport performance. Use machine learning algorithms such as neural networks and support vector machines to predict and screen the performance of materials and select new low-dimensional electronic materials with potential for semiconductor device design.

[0084] S2: Based on the characteristics of the selected materials, use professional semiconductor device simulation software such as Sentaurus, TCAD, etc. to design new field effect transistor (FET) and optoelectronic device structures to design and simulate semiconductor devices based on new low-dimensional electronic materials. Optimize device performance by adjusting device structural parameters and external conditions such as gate voltage, channel length, light intensity, etc. Simulate the interface characteristics between different two-dimensional materials, analyze the impact of interface states on carrier transport, and regulate through surface modification and doping techniques.

[0085] S3: Optimize device performance parameters using machine learning algorithms to optimize key parameters such as current on / off ratio and subthreshold swing;

[0086] S4: Study the interface characteristics between different two-dimensional materials, design heterojunction structures, and analyze the influence of interface states on carrier transport;

[0087] S5: Through surface modification and doping techniques, machine learning algorithms are used to guide the regulation of interface states and defects to improve the electronic and optoelectronic performance of devices;

[0088] S6: Introduce strain control and external electric field, use machine learning to assist in studying the changes in the electronic structure and band gap characteristics of two-dimensional materials, design strain sensors and adjustable band gap devices, and optimize the device operating parameters under electric field control.

[0089] Systematically analyze and process the data obtained from theoretical calculations, simulations, and experiments. Use statistical methods and data mining techniques to extract key information from the data and reveal the laws of material performance and device characteristics. Verify the reliability and consistency of the research results by comparing the data obtained by different research methods.

[0090] About first principles:

[0091] In step S1, first-principles, also known as ab initio calculations, are a calculation method based on the basic principles of quantum mechanics and independent of experimental data. The most commonly used first-principles calculation method is density functional theory (D FT), which predicts the basic properties of materials, including electron density, band structure, state density, etc., by solving the electronic structure equation of the system.

[0092] The first principles method is based on the Schrödinger equation, which describes the movement of electrons in the nuclear potential field to calculate the total energy, electron distribution and other physical properties of the system. This method does not require empirical parameters, so it can be used in theory to study the basic physical and chemical properties of new materials.

[0093] The core of its calculation is to solve the Schrödinger equation for multi-electron systems. Since it is very complicated to directly solve the Schrödinger equation for multi-electron systems, DFT simplifies this problem by introducing electron density and replaces the original Schrödinger equation with the Kohn-Sham equation to calculate the electronic structure of the system. This can more efficiently describe the ground state properties of the material.

[0094] As a preferred embodiment, in step S1, the steps of using density functional theory (DFT) to calculate the electronic structure and energy band characteristics of the new two-dimensional semiconductor material are as follows:

[0095] Density functional theory (DFT) is a quantum mechanical calculation method widely used in condensed matter physics, materials science and quantum chemistry. It describes multi-electron systems based on electron density rather than wave functions, which greatly reduces the complexity of calculations. DFT can accurately predict the electronic structure, band structure and other physical and chemical properties of materials by calculating the spatial distribution of electrons. The basis of DFT is the Hohenberg-Kohn theorem and the Kohn-Sham equation:

[0096] Hohenberg-Kohn theorem: states that the ground state energy of a system is a functional of the electron density, and that the electron density uniquely determines all physical properties of the system.

[0097] Kohn-Sham equation: By constructing an imaginary single-electron system, the total electron density of the single-electron orbital is used to approximate the electron density of the real system. The specific equation is:

[0098]

[0099] Among them, V eff (r) is the effective potential, including external potential, Coulomb interaction between electrons, and exchange correlation potential, etc. By solving this equation, the ground state energy and electron density of the system can be obtained.

[0100] In the present invention, density functional theory is mainly used to calculate the electronic structure and energy band characteristics of the new two-dimensional semiconductor material. Specifically, the steps of density functional theory (DFT) calculation are as follows:

[0101] (1) Calculation of electron density: The Kohn-Sham equation is used to solve the single electron wave function and energy of the system, and then the electron density is calculated:

[0102]

[0103] Among them, ψ i (r) is the Kohn-Sham orbital, and n(r) represents the electron density.

[0104] (2) Exchange-correlation energy calculation: Based on the local density approximation (LDA) or the generalized gradient approximation (GGA), the exchange-correlation energy is calculated:

[0105] E xc [n(r)]=∫n(r)∈ xc [n(r)]dr

[0106] Among them, ∈ xc is the exchange-correlation energy density.

[0107] (3) Band structure calculation: The band structure of two-dimensional materials is obtained by calculating the periodic structure of the system. The band calculation results can be used to predict the conductivity and semiconductor properties of the material.

[0108] As a preferred method, in step S1, the first principles are used to deeply analyze the relationship between the microstructure (such as atomic arrangement, bonding characteristics) of the two-dimensional material and its electronic transport properties (such as electron mobility, conductivity). The specific steps are as follows:

[0109] Calculate the electronic structure of a material:

[0110] Use first-principles calculation tools (such as VASP) to solve the Kohn-Sham equation and calculate the material's electron density, band structure, state density, etc. These electronic structure characteristics can reveal important physical properties of the material, such as conductivity and band gap.

[0111] Through these calculations, we can understand how the electrons in the two-dimensional material are distributed at different energy levels and further determine the material's conductivity. For example, the smaller the band gap of a material, the easier it is for electrons to pass through the material, thereby improving its conductivity.

[0112] Relationship between band structure and electron mobility:

[0113] The energy band structure directly reflects the conductive properties of the material. For conductors, the electrons in the conduction band have a higher state density and can move freely; while for semiconductors, there is a band gap between the conduction band and the valence band, and electrons need to cross the band gap to participate in conduction.

[0114] By calculating the band structure from first principles, the electron mobility of the material can be accurately predicted. If the effective mass of the material is small, the electron mobility is usually higher, which makes the material have better electron transport performance.

[0115] Relationship between state density and conductivity:

[0116] Density of States (DOS) describes the number of electronic states at each energy level. By calculating the density of states, we can understand the distribution of electrons in a material at different energy levels. Materials with good conductivity usually have a higher density of states near the Fermi level, which means that more electrons can participate in conduction.

[0117] Analysis of the density of states can reveal changes in the material's conductive properties under different temperatures and external conditions.

[0118] Effect of microstructure on electron transport:

[0119] The first principles method can accurately calculate the effects of microstructural changes such as defects and doping in materials on electron transport. For example, defects or interface states in two-dimensional materials can scatter electrons, thereby reducing electron mobility.

[0120] Analyzing a material's crystal structure, interlayer interactions, and defect distribution can reveal changes in its electron transport properties. Such analysis is crucial for designing high-performance semiconductor materials.

[0121] As a preferred embodiment, Figure 2 As shown, in step S1, the machine learning algorithm combines the density functional theory (DFT) calculation results to predict and screen the performance of new low-dimensional electronic materials, thereby selecting potential materials for the design of semiconductor devices. This process is achieved through the following steps:

[0122] (1) Data input: DFT calculation results as training data

[0123] In the present invention, high-quality data such as the electronic structure, band structure, and state density of the material are first obtained through DFT calculations. These data represent the basic physical properties of the material and are important inputs for training machine learning models.

[0124] Input data: Physical parameters of materials such as electron density distribution, band gap size, electron mobility, and state density come from DFT calculations.

[0125] Dataset construction: Based on these data, a training set representing material properties is established as input for the machine learning model.

[0126] (2) Machine learning model training: extracting the relationship between material properties and structure

[0127] Machine learning models are able to extract the relationship between a material’s microstructure and its properties by learning implicit patterns in input data.

[0128] Model selection: Use machine learning models suitable for processing complex data relationships, such as neural networks (NN), support vector machines (SVM), etc., to establish a mapping between the material microstructure and its physical properties.

[0129] Feature extraction: Machine learning finds key variables that affect material properties from a large amount of input data by automating feature extraction. These variables may include band gap width, shape of state density distribution, electron mobility, etc.

[0130] (3) Performance prediction: Performance prediction for uncalculated materials

[0131] The trained machine learning model can quickly predict the properties of new materials without time-consuming DFT calculations.

[0132] Prediction process: For two-dimensional materials that have not been calculated, the model can predict their band gap, mobility, conductivity and other performance parameters based on their structural information (such as crystal structure, bond length, bond angle, etc.).

[0133] Efficient screening: Compared with direct DFT calculations, machine learning can quickly screen a large number of potential high-performance materials, thereby reducing the amount of complex calculations.

[0134] (4) Screening and sorting: Screening materials based on performance indicators

[0135] Based on the prediction results of the machine learning model, the most promising low-dimensional materials can be screened out for subsequent semiconductor device design.

[0136] Screening criteria: Screening criteria may include key performance indicators of semiconductor devices such as band gap width, electron mobility, photoelectric response speed, etc. Machine learning models can rank materials by potential based on these indicators.

[0137] Feedback optimization: If certain materials show excellent potential, this data can be fed back into the DFT calculation for more accurate calculation verification. This two-way feedback optimization ensures that the selected materials not only have high predicted performance, but also maintain excellent performance in accurate calculations.

[0138] (5) Advantages and superiority

[0139] Compared with the traditional DFT global calculation method, machine learning has the following advantages in predicting and screening material properties:

[0140] Efficiency: Machine learning algorithms can quickly process large amounts of data and make predictions, avoiding tedious DFT global calculations and greatly improving the efficiency of material screening.

[0141] Accuracy: By learning from DFT calculation results, machine learning can capture the complex relationship between material properties and microstructure and make accurate performance predictions.

[0142] Automated screening: Machine learning automatically selects the most promising materials from large amounts of data, reducing the time and errors of manual screening.

[0143] As a preferred embodiment, in step S2, if Figure 3 As shown, the design of new field effect transistor (FET) and optoelectronic device structures using professional semiconductor device simulation software (such as Sentaurus or TCAD) can effectively simulate the device performance based on new low-dimensional electronic materials. The following is a detailed description of the process:

[0144] (1) Device structure construction

[0145] Using professional semiconductor device simulation software such as TCAD or Sentaurus, you first need to build the geometric structure of the transistor or optoelectronic device. This process mainly includes:

[0146] Material selection: Two-dimensional semiconductor materials (such as MoS) selected based on DFT calculations and machine learning 2 , WS 2 etc.) as channel materials.

[0147] Device geometry design: Build a physical model of a transistor or optoelectronic device and set the device’s geometric parameters, including gate length, channel length, channel thickness, source and drain electrode positions, gate dielectric thickness, etc.

[0148] Doping design: During the simulation process, the doping type and concentration of semiconductor devices are set, such as n-type or p-type doping, to optimize the conductivity of the channel.

[0149] (2) Material physical model definition

[0150] After the device structure is constructed, the physical model of the material needs to be defined to simulate charge transport, electronic energy bands, and carrier behavior.

[0151] Electron transport model: Set up transport models for simulating electron migration in materials, such as drift-diffusion model, quantum tunneling model, etc. These models can accurately capture the motion characteristics of electrons in low-dimensional materials.

[0152] Material parameter setting: By inputting material parameters obtained from DFT calculations, such as band gap, electron mobility, state density, electron effective mass, etc., the physical properties of the material can be accurately reflected.

[0153] Interface state and defect model: For two-dimensional materials, the interface state between the material and the metal and the effect of defects on carriers are crucial. By defining parameters such as interface state density and defect density, the impact of these factors on device performance can be simulated.

[0154] (3) Working characteristics of analog devices

[0155] After the device structure and material parameters are set, simulation software is used to analyze and simulate the device's operating characteristics.

[0156] Electrical characteristics simulation:

[0157] For field effect transistors (FETs), the current-voltage characteristics (IV curves) are simulated to analyze the conductive performance under different gate voltages and source-drain voltages.

[0158] The performance of the device is evaluated by simulating key parameters such as on-state current, leakage current, switching ratio, and subthreshold swing.

[0159] By adjusting parameters such as gate voltage and channel length, the conductivity and switching characteristics of the transistor can be optimized.

[0160] Optical property simulation:

[0161] For optoelectronic devices, the photocurrent response is simulated and the photoelectric conversion efficiency under illumination conditions of different wavelengths and intensities is studied.

[0162] By simulating the photocurrent-voltage characteristics (IV characteristics), the response speed and sensitivity of the photodetector can be evaluated and the light response characteristics of the photoelectric device can be optimized.

[0163] (4) Design optimization and iteration

[0164] After obtaining preliminary results from the simulation, the design can be optimized through multiple iterations to improve device performance.

[0165] Parameter optimization: Based on the simulation results, adjust the key geometric parameters of the device (such as gate length, channel thickness, etc.) and material physical parameters (such as doping concentration, carrier mobility, etc.) to optimize performance.

[0166] Performance comparison: Compare with devices made of traditional materials and analyze the improvements in conductivity, photoelectric response, power consumption, etc. of devices based on new two-dimensional materials.

[0167] Defect control: By simulating the impact of interface defects and doping concentration on device performance, specific optimization measures are proposed, such as reducing interface state density or optimizing contact resistance.

[0168] (5) Output and visualization results

[0169] After the optimization is completed, the simulation software can output specific performance parameters and visual results to demonstrate the device's operating characteristics and improvement effects.

[0170] IV curve: The electrical characteristic curve of the output field effect transistor, showing performance indicators such as on-state current, switching ratio, and subthreshold swing.

[0171] Photoresponse curve: Output the photocurrent response of the optoelectronic device under different lighting conditions to evaluate the photoelectric conversion efficiency and detection sensitivity.

[0172] Visual graphics: Displays visual results such as carrier distribution, electron density distribution, and photocurrent distribution to help better understand the physical process of the device.

[0173] Preferably, in step S3, the device performance parameters include gate voltage, channel length and source / drain electrode materials.

[0174] As a preferred embodiment, Figure 4 As shown, in step S3, density functional theory (DFT) and machine learning (ML) complement each other to form a powerful material design and optimization platform. The specific combination is as follows:

[0175] (1) DFT provides high-quality training data:

[0176] DFT calculation is a high-precision material calculation method based on first principles, which can accurately predict the electronic structure, band structure, state density and other information of the material. However, due to its complexity, DFT calculation has a slow calculation speed and cannot be completed quickly in large-scale material screening.

[0177] The data obtained through DFT calculations can be used as training sets to provide high-quality input for machine learning algorithms. For example, the band structure, electron density, band gap, state density and other physical parameters of two-dimensional semiconductor materials can be obtained through DFT calculations, and then these data can be used to train machine learning models.

[0178] (2) Machine learning to quickly predict material properties:

[0179] The trained machine learning model can learn implicit patterns and laws from a large number of material properties, so as to quickly predict the performance of new materials. Compared with direct DFT calculations, machine learning can greatly improve the screening efficiency, especially in the screening process of large-scale material libraries.

[0180] Machine learning models such as neural networks (NN) and support vector machines (SVM) can predict the band gap, electron mobility, photoresponse and other properties of materials by learning DFT calculation results. Such rapid predictions can reduce the number of materials that need to be calculated by DFT.

[0181] (3) Collaborative optimization process:

[0182] In the present invention, the machine learning algorithm is not only used to predict material properties, but also to adjust its model parameters according to the DFT calculation results. This collaborative optimization mechanism ensures the high accuracy of the machine learning model and further improves the efficiency of the DFT calculation. Specifically, machine learning can adjust the initial conditions of the DFT calculation through a feedback loop, such as selecting a more appropriate material model or calculation parameters, thereby accelerating the DFT calculation process.

[0183] This feedback loop greatly improves the accuracy of material property predictions while avoiding lengthy recalculation.

[0184] (4) Through the combination of DFT and machine learning, the present invention has significant advantages in the following aspects:

[0185] Significantly improve computing efficiency:

[0186] Traditional DFT calculations require a lot of computing resources and time, especially when dealing with complex two-dimensional semiconductor materials. However, through the rapid prediction function of machine learning models, potential two-dimensional semiconductor materials can be screened out in a short time, thereby reducing the number and time of DFT calculations.

[0187] Data-driven optimization:

[0188] The present invention uses machine learning algorithms to extract effective information from a large amount of DFT calculation data, optimizing the process of material screening and device design. For example, through convolutional neural networks (CNN) and support vector machines (SVM), the complex relationship between material structure and its physical properties can be effectively identified, thereby optimizing material performance.

[0189] This data-driven optimization process avoids the complex derivation and parameter debugging in pure theoretical calculations, and can discover the potential of new materials faster and more accurately.

[0190] Improve forecast accuracy:

[0191] Machine learning can capture more complex structure-performance relationships by learning from DFT calculation results. This learning ability enables machine learning models to predict material properties with higher accuracy with less input data. Compared with simple theoretical predictions, this invention combines the high accuracy of DFT with the efficient prediction of machine learning, and can more accurately discover two-dimensional semiconductor materials with excellent performance.

[0192] Supports complex device designs:

[0193] The present invention not only screens materials through DFT calculation and machine learning, but also optimizes the design of two-dimensional semiconductor devices through the combination of the two. For example, by analyzing the DFT calculation results, the machine learning algorithm can optimize the key parameters of the transistor such as the gate voltage and channel length, significantly improving the performance of the device.

[0194] Preferably, in step S4, the heterojunction structure includes graphene / molybdenum disulfide and black phosphorus / tungsten disulfide WS2.

[0195] As a preferred embodiment, Figure 5 As shown, in step S6, the combination of the machine learning algorithm and the density functional theory (DFT) adopts a collaborative optimization feedback mechanism. This mechanism enhances the computational efficiency and prediction accuracy of both through bidirectional feedback. The specific implementation process is as follows:

[0196] (1) Machine learning adjusts model parameters based on DFT calculation results

[0197] DFT calculations provide machine learning with high-precision data on physical properties such as electronic structure and energy bands of materials. By learning this data, machine learning algorithms can optimize their internal model parameters, thereby improving the accuracy of predictions of new material properties.

[0198] Step 1: DFT calculation of material properties:

[0199] The electronic structure, band gap, density of states and other physical properties of the material are calculated by DFT software (such as VASP). Since DFT can accurately describe the electronic behavior of the material from the first principles, these data are highly accurate.

[0200] Step 2: Train the machine learning model:

[0201] The material property data obtained by DFT calculation is input into the machine learning model (such as neural network, support vector machine) as a training set. Machine learning learns the relationship between the physical properties and structure of the material by continuously optimizing its model parameters.

[0202] Step 3: Model optimization and verification:

[0203] When the machine learning model is fed with new material data, it can make more accurate predictions about the properties of the new material based on the properties previously learned from DFT calculations. The model’s predictive power continues to improve as more DFT data is fed in.

[0204] Through such learning and adjustment, the machine learning model is able to automatically adjust its parameters to adapt to the complex physical properties of the material, thereby improving prediction accuracy.

[0205] (2) DFT selects calculation parameters based on machine learning feedback

[0206] DFT calculations are usually computationally intensive, so through machine learning feedback, DFT can select more targeted calculation parameters before calculation, thereby optimizing calculation efficiency and avoiding unnecessary calculations.

[0207] Step 1: Machine learning to initially predict material properties:

[0208] Before starting DFT calculations, machine learning models can be used to perform preliminary screening of a large number of materials through their rapid prediction capabilities. For example, they can predict the band gap, mobility and other properties of materials through simple structural parameters.

[0209] Step 2: Feedback to DFT to select calculation priority:

[0210] Based on the preliminary predictions from machine learning, DFT calculations can select promising materials for detailed electronic structure calculations. The predictions from machine learning provide a preliminary basis for judgment, allowing DFT calculations to focus on the most promising materials, thus saving computing time and resources.

[0211] Step 3: Adjust DFT calculation parameters:

[0212] Machine learning can also suggest more appropriate calculation parameters based on past DFT calculation results. For example, if a certain type of material is found to have a small band gap, DFT can give priority to high-precision calculation methods (such as hybrid functionals) to further improve the calculation accuracy.

[0213] Step 4: Feedback loop optimization:

[0214] After the DFT calculation is completed, the new data is fed back to the machine learning model to continuously optimize the model parameters. Through this two-way feedback loop, DFT and machine learning promote each other and achieve a double improvement in calculation accuracy and efficiency.

[0215] Based on the above machine learning and density functional theory (DFT) feedback optimization mechanism, the following advantages can be achieved:

[0216] Improve computational efficiency: The preliminary prediction and screening capabilities of machine learning enable DFT calculations to focus on the most promising materials, avoiding redundant calculations on large-scale materials.

[0217] Improve prediction accuracy: DFT high-precision calculations provide training data for machine learning, which continuously adjusts model parameters to more accurately predict material properties.

[0218] Bidirectional feedback optimization: The bidirectional feedback loop between DFT and machine learning achieves the coordinated optimization of the two, enabling the present invention to screen out two-dimensional semiconductor materials with excellent performance in a shorter time.

[0219] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for screening two-dimensional semiconductor materials based on density functional theory and machine learning, characterized in that: Includes steps: S1: Use the density functional theory calculation software VASP to calculate the electronic structure and energy band characteristics of new low-dimensional electronic materials; combine first-principles calculations to deeply analyze the relationship between the microstructure of the material and the electronic transport performance, and use machine learning algorithms to select new low-dimensional electronic materials with potential; S2: Based on the materials selected in step S1, design and simulate the semiconductor device using semiconductor device simulation software Sentaurus or TCAD; S3: Optimizing semiconductor device performance parameters using machine learning algorithms; S4: Design heterojunction structures and analyze the effects of interface states on carrier transport; S5: Through surface modification and doping technology, machine learning algorithms are used to guide the regulation of interface states and defects; S6: Introduce strain control and external electric field, use machine learning to assist in studying the changes in the electronic structure and band gap characteristics of two-dimensional materials, design strain sensors and adjustable band gap devices, and optimize the device operating parameters under electric field control.

2. The method for screening two-dimensional semiconductor materials based on density functional theory and machine learning according to claim 1, characterized in that: In step S1, the steps of using density functional theory to calculate the electronic structure and band characteristics of the new two-dimensional semiconductor material are as follows: S11: Electron density calculation: The Kohn-Sham equation is used to solve the single electron wave function and energy of the system, and then the electron density is calculated: S12: Exchange-correlation energy calculation: Based on the local density approximation LDA or the generalized gradient approximation GGA, the exchange-correlation energy is calculated: AND xc [n(r)]=∫n(r)∈ xc [n(r)]dr S13: Band structure calculation: The band structure of the two-dimensional material is obtained by calculating the periodic structure of the system.

3. The method for screening two-dimensional semiconductor materials based on density functional theory and machine learning according to claim 1, characterized in that: In step S1, the first-principles calculation includes the calculation of the electronic structure of the material, the calculation of the relationship between the energy band structure and the electron mobility, the calculation of the relationship between the density of states and the conductivity, and the calculation of the influence of the microstructure on the electron transport.

4. The method for screening two-dimensional semiconductor materials based on density functional theory and machine learning according to claim 1, characterized in that: In step S1, the steps of selecting potential new low-dimensional electronic materials using a machine learning algorithm are as follows: A1: Data input, using DFT calculation results as training data; Training data input: Use DFT calculation to obtain the electronic structure, band structure and state density of the material as training data; Parameter input: input the material physical parameters of electron density distribution, band gap size, electron mobility and state density obtained through DFT calculation; Dataset construction, using the data obtained through the above steps to establish a training set representing material properties as input for the machine learning model; A2: Model training, through machine learning model training, extract the relationship between material performance and structure; Model selection, using machine learning models suitable for processing complex data relationships to establish a mapping between the material microstructure and its physical properties; Feature extraction, finding the key variables that affect material properties from a large amount of input data; A3: Performance prediction, predicting the performance of uncalculated materials; the model predicts the performance parameters of band gap, mobility and conductivity based on its structural information; A4: Screening and sorting, screening materials according to performance indicators; Screening basis, screening criteria include band gap width, electron mobility and photoelectric response speed; Feedback optimization: material data showing excellent potential is fed back into DFT calculations to obtain higher-precision calculation verification.

5. The method for screening two-dimensional semiconductor materials based on density functional theory and machine learning according to claim 1, characterized in that: In step S2, the steps of designing and simulating a semiconductor device using semiconductor device simulation software Sentaurus or TCAD include: S21: Use semiconductor device simulation software TCAD or Sentaurus to construct the geometric structure of transistors or optoelectronic devices. The specific component procedures include: Material selection: Two-dimensional semiconductor materials selected based on DFT calculations and machine learning as channel materials; Device geometry design, building a physical model of a transistor or optoelectronic device, and setting the device's geometric parameters, including gate length, channel length, channel thickness, source and drain electrode positions, and gate dielectric thickness; Doping design, setting the doping type and concentration of semiconductor devices during simulation; S22: Define the material physical model. The specific process is as follows: Electron Transport Model: Set up the transport model used to simulate electron migration in the material; Material parameter setting: input material parameters obtained from DFT calculation; the material parameters include band gap, electron mobility, state density and electron effective mass; Interface state and defect model: define interface state density and defect density parameters, and simulate the impact of these factors on device performance; S23: Conduct electrical and optical property simulations to simulate the working characteristics of the device; The electrical characteristic simulation includes: simulating current-voltage characteristics (IV curve); simulating key parameters such as on-state current, leakage current, switching ratio and subthreshold swing; The optical characteristic simulation includes: simulating photocurrent-voltage characteristics (IV characteristics); S24: Perform design optimization and iteration. The optimization and iteration process includes: Parameter optimization: adjust the key geometric parameters and material physical parameters of the device based on the simulation results; Performance comparison, comparative analysis with devices made of traditional materials; Defect control: by simulating the impact of interface defects and doping concentration on device performance, specific optimization measures are proposed; S25: Output and visualization results.

6. The method for screening two-dimensional semiconductor materials based on density functional theory and machine learning according to claim 1, characterized in that: In step S3, the device performance parameters include gate voltage, channel length, and source and drain electrode materials.

7. The method for screening two-dimensional semiconductor materials based on density functional theory and machine learning according to claim 6, characterized in that: In step S3, density functional theory DFT and machine learning ML are combined to form a material design and optimization platform, including the following steps: S31: The data obtained by DFT calculation can be used as a training set; S32: Using trained machine learning models, it is possible to learn implicit patterns and laws from a large number of material properties, thereby quickly predicting the performance of new materials; S33: adjusting its model parameters according to the calculation results of DFT; S34: By analyzing the DFT calculation results, we use machine learning algorithms to optimize the key parameters of the transistor’s gate voltage and channel length.

8. The method for screening two-dimensional semiconductor materials based on density functional theory and machine learning according to claim 1, characterized in that: In step S4, the heterojunction structure includes graphene / molybdenum disulfide and black phosphorus / tungsten disulfide WS2.

9. The method for screening two-dimensional semiconductor materials based on density functional theory and machine learning according to claim 1, characterized in that: In step S6, the specific implementation process of optimizing the device operating parameters under electric field regulation includes: S61: Machine learning adjusts model parameters based on DFT calculation results; S611: Calculate the electronic structure, band gap and physical properties of the material using DFT software (such as VASP); S612: inputting the material property data obtained by DFT calculation into the machine learning model as a training set; S613: When the machine learning model receives new material data, it can make more accurate predictions about the properties of the new material based on the properties previously learned from DFT calculations. As more DFT data is input, the model's predictive ability continues to improve; S62: DFT selects computational parameters based on machine learning feedback; S621: Before starting the DFT calculation, the machine model uses its fast prediction function to perform preliminary screening of large batches of materials; S622: Based on the preliminary prediction results of machine learning, DFT calculations select promising materials for detailed electronic structure calculations; S623: Machine learning provides more appropriate calculation parameter suggestions based on past DFT calculation results; S624: After completing the DFT calculation, the new data is fed back to the machine learning model to continuously optimize the model parameters.

10. The method for screening two-dimensional semiconductor materials based on density functional theory and machine learning according to claim 9, characterized in that: The preliminary screening includes predicting the band gap and mobility performance of the material through simple structural parameters.

Images