Construction method of potential function model of yttrium-doped antimony-tellurium phase change memory material and potential function model

Abstract

The application relates to the technical field of material science, in particular to a construction method of a potential function model of a yttrium-doped antimony-tellurium phase change storage material and the potential function model. The construction method of the potential function model of the yttrium-doped antimony-tellurium phase change storage material comprises the following steps: obtaining a data set, wherein the data set comprises a plurality of training data, each piece of training data comprises a Y-Sb-Te model and a material property corresponding to the Y-Sb-Te model; training a preset NEP model by using the data set to obtain the potential function model of the yttrium-doped antimony-tellurium phase change storage material. The method solves the problem of low efficiency of current machine learning potential, can accelerate the calculation under a graphics computing card, realizes the calculation efficiency close to the classical potential function, simultaneously makes up for the shortcoming that the classical potential function cannot be applied to the complex potential energy surface scene of multiple phases and multiple configurations, and realizes the molecular dynamics simulation close to the first principle accuracy.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of material science, and particularly relates to a construction method of a potential function model of a yttrium-doped antimony-tellurium phase change memory material and the potential function model. BACKGROUND

[0002] Sulfur family phase change memory materials are considered as a strong candidate material for integrating new nonvolatile memory, computing and memory integrated chips and neural network chips. Sulfur family phase change materials encode and record binary information through fast reversible phase change between crystal and amorphous phases and significant differences in electrical properties between the two phases. In a phase change memory device, by applying a strong and short electric pulse, the crystalline phase change material is melted and rapidly cooled (cooling speed is above 10 10 K / s) to form an amorphous state with high resistivity; on the contrary, by applying a weak and long electric pulse, the amorphous phase change material is rapidly crystallized (crystallization speed is in the order of nanoseconds) to form a crystalline state with low resistivity. Due to the unique information storage mode of the phase change material, it has the characteristics of non-volatility, high integration and radiation resistance.

[0003] At present, the commonly used sulfur family phase change material is a Ge-Sb-Te ternary compound, such as pseudo-binary GeTe-Sb2Te3, Sb2Te and Ge 15 Sb 85 , etc. The most commonly used is a series of compounds on the GeTe-Sb2Te3 composition line, including GeTe, Ge1Sb2Te4, Ge1Sb4Te7, Ge2Sb2Te5 and Sb2Te3 compounds. The higher the content of GeTe, the better the thermal stability of the phase change material but the slower the phase change speed and the higher the phase change power consumption; on the contrary, the higher the content of Sb2Te3, the higher the phase change speed of the phase change material and the lower the phase change power consumption, but the thermal stability of the phase change material will be poor. At present, memory devices are striving to develop in the direction of high integration, and the problems of power consumption and speed are the first problems to be solved. Therefore, Sb2Te3 phase change material without Ge element has become a hot material in the current phase change memory field.

[0004] Due to poor thermal stability of amorphous phase, intrinsic Sb2Te3 is difficult to apply in phase change memory devices. In order to solve this problem, researchers combine first-principles calculation and material gene high-throughput method to screen candidate dopants such as Sc, Y and Ti in transition metals. In recent years, it has been proved that these dopants greatly improve the performance of Sb2Te3 phase change materials. Not only effectively improve the thermal stability of amorphous phase, but also improve the power consumption and phase change speed of phase change material to a certain extent. However, the internal mechanism of the dopant to improve the thermal stability of amorphous phase and crystallization performance of phase change memory material has not been fully revealed. In experimental research, it is difficult to characterize the structure characteristics and dynamic mechanism of phase change materials. Researchers try to explore the mechanism in depth through theoretical calculation method. Through the first-principle method, the electronic structure information of the crystalline state of the phase change material and the local structure characteristics of the amorphous phase can be revealed, but in larger scale dynamic simulation, due to the high cost of first-principle calculation, it cannot be effectively carried out. Through the traditional classical potential function method, large-scale simulation of phase change materials can be realized, but the poor reproduction ability of the classical potential function to the potential energy surface of the material makes the error in the simulation process larger, which cannot effectively and reasonably study.

[0005] The development of artificial intelligence machine learning method provides an opportunity for high-precision large-scale dynamic exploration of phase change materials. By combining first-principle data and machine learning model, a potential function model can be obtained, which can accurately describe the potential energy surface of the material system, and in a multi-phase multi-component system, the machine learning potential function model can still maintain good accuracy. For example, the matrix tensor potential MTP and the Gaussian approximation potential GAP machine learning model. However, the current potential function has a large model volume, and the efficiency is very slow in the process of molecular dynamics simulation. Compared with the traditional classical potential function, it has a great disadvantage in efficiency, and it is difficult to carry out large-scale long-time simulation process, which cannot meet the needs of phase change material exploration of phase change dynamics, and still needs to be further improved. Therefore, a potential function model with simulation accuracy and efficiency is needed to realize the dynamic simulation of phase change materials in device scale to realize the exploration of device performance and material performance. SUMMARY

[0006] Therefore, the present application provides a method for constructing a yttrium-doped antimony-tellurium phase change memory material potential function model and a potential function model to solve the problem of low efficiency of the current machine learning potential function model.

[0007] In a first aspect, the present application provides a method for constructing a potential function model of a yttrium-doped antimony-tellurium phase change storage material, comprising the following steps: obtaining a data set, wherein the data set comprises a plurality of training data, each training data comprising a Y-Sb-Te model and a material property corresponding to the Y-Sb-Te model; training a preset NEP model using the data set to obtain a potential function model of the yttrium-doped antimony-tellurium phase change storage material.

[0008] The method for constructing a potential function model of a yttrium-doped antimony-tellurium phase change storage material provided by the present application trains a preset NEP model using a data set to obtain a potential function model of the yttrium-doped antimony-tellurium phase change storage material, which significantly reduces the calculation cost compared to traditional machine learning potential functions, realizes large-scale and long-time molecular dynamics simulation, and provides a reliable theoretical tool for in-depth understanding of the phase change mechanism of doped Sb2Te3 materials and optimization of device performance.

[0009] In an optional embodiment, the Y-Sb-Te model comprises a Y-Sb-Te crystal phase model, a Y-Sb-Te crystal phase perturbation model, a Y-Sb-Te liquid phase model, a Y-Sb-Te liquid phase perturbation model, a Y-Sb-Te amorphous model, and a Y-Sb-Te amorphous perturbation model; obtaining the data set comprises: obtaining the Y-Sb-Te crystal phase model; performing a melting-rapid cooling simulation on the Y-Sb-Te crystal phase model to obtain the Y-Sb-Te liquid phase model and the Y-Sb-Te amorphous model; perturbing the Y-Sb-Te crystal phase model, the Y-Sb-Te liquid phase model, and the Y-Sb-Te amorphous model using a preset random perturbation algorithm to obtain the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase perturbation model, and the Y-Sb-Te amorphous perturbation model; and performing first-principle calculations on the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model, and the Y-Sb-Te amorphous perturbation model to obtain material properties corresponding to the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model, and the Y-Sb-Te amorphous perturbation model.

[0010] This embodiment significantly expands the coverage of the training data in the phase space and the configuration space by constructing a diversified structure system containing crystal phase, liquid phase, amorphous phase, and their perturbation models, and obtaining high-precision energy, force, and stress material properties through first-principle calculations, enhances the representativeness and diversity of the data set, effectively improves the generalization ability and prediction accuracy of the trained potential function model, and enables it to accurately describe the atomic interactions in the Y-Sb-Te system in different phases and dynamic processes, laying a solid foundation for high-fidelity large-scale phase change dynamics simulation.

[0011] In an optional embodiment, the Y-Sb-Te crystal phase model, the Y-Sb-Te liquid phase model and the Y-Sb-Te amorphous model are perturbed by a preset random perturbation algorithm respectively to obtain a Y-Sb-Te crystal phase perturbation model, a Y-Sb-Te liquid phase perturbation model and a Y-Sb-Te amorphous perturbation model, including: the Y-Sb-Te crystal phase model, the Y-Sb-Te liquid phase model and the Y-Sb-Te amorphous model are subjected to cell shape perturbation and atomic position perturbation by a random perturbation algorithm respectively to obtain the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase perturbation model and the Y-Sb-Te amorphous perturbation model; wherein the cell shape perturbation condition is that the lengths of the a, b and c axes are randomly stretched or compressed within ±(4%~6%), and the three angles α, β and γ are randomly deformed within ±(4%~5%); the atomic position perturbation condition is that the three coordinates x, y and z are randomly offset from the initial position within ±(4%~5%).

[0012] This embodiment effectively simulates the complex environment that the material may experience in actual application, such as stress, deformation and thermal disturbance, by applying controllable random perturbation in two dimensions of cell shape and atomic position, significantly expands the coverage of training data in the structural phase space, enhances the diversity and physical authenticity of the data set, not only improves the description ability of the potential function model for non-equilibrium structures, but also enables it to simulate the dynamic response behavior of the material under the action of external fields (such as mechanical deformation, impact effect, etc.), further expands the applicability and prediction reliability of the model under multiple scenarios and multiple physical conditions.

[0013] In an optional embodiment, the first-principle calculation of the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model and the Y-Sb-Te amorphous perturbation model includes: the first-principle calculation of the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model and the Y-Sb-Te amorphous perturbation model, wherein the parameters of the first-principle calculation are: the plane cutoff energy is 550~650 eV, the K point density in the Brillouin zone is set by the KSPACING parameter to be 0.15~0.25 Å -1 , the pseudo potential is selected as PAW-PBE (Projector Augmented Wave-Perdew-Burke-Ernzerhof), and the van der Waals dispersion interaction is selected as the DFT-D3 (Dispersion correction-version 3) method.

[0014] The embodiment significantly improves the calculation accuracy of key material properties such as energy, force and stress by adopting the first principle calculation parameters of high truncation energy, high K point density, accurate pseudo potential (PAW-PBE) and considering van der Waals correction (D3), effectively reduces the systematic error of theoretical simulation, ensures the high quality and high reliability of the training data, so that the potential function model constructed can more accurately reflect the real physical interaction of the system, and significantly enhances the prediction accuracy and stability of the model under complex phase transition process and multi-state structure.

[0015] In an optional embodiment, the preset NEP model is trained by using the data set to obtain a potential function model of a yttrium-doped antimony tellurium phase change storage material, and the training of the selected parameters includes that the cutoff radius of the radial and angular descriptors is 6-8 Å and 5-7 Å, the basis set number of the radial and angular descriptors is 8-12 and 6-10, the expansion coefficient of the angular descriptor is (4 2 0) or (4 2 1), and the number of hidden layer neurons is 60-100.

[0016] The embodiment balances the expression ability and complexity of the model by reasonably setting the descriptor cutoff radius, basis function number, angular momentum expansion coefficient and neural network structure parameters of the NEP model, effectively avoids overfitting, significantly improves the training efficiency and convergence stability, and thus constructs a potential function model with high precision, strong generalization ability and good calculation efficiency, which provides a reliable guarantee for accurately simulating the phase change dynamics and multi-phase structure evolution of the yttrium-doped antimony tellurium material.

[0017] In a second aspect, the present application also provides a potential function model, which is constructed by using the construction method of the yttrium-doped antimony tellurium phase change storage material potential function model in the first aspect or any of the corresponding embodiments thereof.

[0018] In a third aspect, the present application also provides a construction device of a yttrium-doped antimony tellurium phase change storage material potential function model, which includes an acquisition module and a training module. The acquisition module is used to acquire a data set, and the data set includes a plurality of training data, and each training data includes a Y-Sb-Te model and a material property corresponding to the Y-Sb-Te model. The training module is used to train a preset NEP model by using the data set to obtain a potential function model of a yttrium-doped antimony tellurium phase change storage material.

[0019] In a fourth aspect, the present application provides a computer device, comprising a memory and a processor, the memory and the processor being connected with each other in communication, the memory storing computer instructions, and the processor executing the computer instructions to perform the method for constructing a potential function model of a yttrium-doped antimony-tellurium phase change storage material according to the first aspect or any of the corresponding embodiments.

[0020] In a fifth aspect, the present application provides a computer readable storage medium, the computer readable storage medium storing computer instructions, the computer instructions being used to make a computer perform the method for constructing a potential function model of a yttrium-doped antimony-tellurium phase change storage material according to the first aspect or any of the corresponding embodiments.

[0021] In a sixth aspect, the present application provides a computer program product, comprising computer instructions, the computer instructions being used to make a computer perform the method for constructing a potential function model of a yttrium-doped antimony-tellurium phase change storage material according to the first aspect or any of the corresponding embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed in the specific embodiments or the prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0023] Figure 1 is a flow chart of the method for constructing a potential function model of a yttrium-doped antimony-tellurium phase change storage material according to an embodiment of the present application;

[0024] Figure 2 is a flow chart of another method for constructing a potential function model of a yttrium-doped antimony-tellurium phase change storage material according to an embodiment of the present application;

[0025] Figure 3 is a flow chart of an example of the method for constructing a potential function model of a yttrium-doped antimony-tellurium phase change storage material according to an embodiment of the present application;

[0026] Figure 4 is a training fitting result of a potential function of a yttrium-doped antimony-tellurium phase change material according to an embodiment of the present application;

[0027] Figure 5 is a prediction result of the potential function on an energy-volume curve according to an embodiment of the present application;

[0028] Figure 6 is a prediction result of the potential function on a radial distribution function of a liquid phase and an amorphous phase according to an embodiment of the present application;

[0029] Figure 7is a prediction result of the potential function on the phase transition material body phase melting point in the embodiment of the present application;

[0030] Figure 8 is a comparison of the NEP potential function with the efficiency of the first-principle calculation and the GAP potential function model in the embodiment of the present application;

[0031] Figure 9 is a structural block diagram of a potential function model construction device of a yttrium-doped antimony-tellurium phase change storage material according to the embodiment of the present application;

[0032] Figure 10 is a hardware structure schematic diagram of a computer device of the embodiment of the present application. DETAILED DESCRIPTION

[0033] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0034] According to the embodiment of the present application, a construction method embodiment of a potential function model of a yttrium-doped antimony-tellurium phase change storage material is provided. It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0035] In the present embodiment, a construction method of a potential function model of a yttrium-doped antimony-tellurium phase change storage material is provided, which can be used in a computer device. Figure 1 is a flowchart of a construction method of a potential function model of a yttrium-doped antimony-tellurium phase change storage material according to the embodiment of the present application, as shown in Figure 1 The flowchart includes the following steps:

[0036] Step S101: Obtain a data set, wherein the data set includes a plurality of training data, and each training data includes a Y-Sb-Te model and a material property corresponding to the Y-Sb-Te model.

[0037] Specifically, the material property includes at least one of the following: unit cell energy, atomic force, and lattice stress.

[0038] Step S102: Train a preset NEP model using the data set to obtain a potential function model of a yttrium-doped antimony-tellurium phase change storage material.

[0039] NEP is a machine learning potential function that aims to approximate the interatomic interactions of complex material systems through a neural network model. Compared with traditional empirical potential functions, NEP can provide higher accuracy and can be applied to a wide range of material systems. NEP uses a set of descriptors to represent the local environment around each atom and learns the relationship between these descriptors and energy, force through training.

[0040] Specifically, the NEP model in the GPUMD (Graphics Processing Units Molecular Dynamics) software package can be used. GPUMD is a GPU-accelerated molecular dynamics program that is used with NEP and is developed by Fan Zheyong et al. It supports NEP potential function driven MD (Molecular Dynamics) simulation, efficient thermal conductivity calculation based on Green-Kubo or NEMD (Non-Equilibrium Molecular Dynamics), structure optimization, phonon spectrum calculation, interface thermal conductivity calculation, etc.

[0041] The construction method of the yttrium-doped antimony-tellurium phase change storage material potential function model provided in this embodiment uses a data set to train a preset NEP model to obtain a potential function model of the yttrium-doped antimony-tellurium phase change storage material. Compared with traditional machine learning potential functions, the calculation cost is greatly reduced, and large-scale and long-time molecular dynamics simulation is realized, providing a reliable theoretical tool for in-depth understanding of the phase change mechanism of the doped Sb2Te3 material and optimizing the device performance.

[0042] In this embodiment, a construction method of a yttrium-doped antimony-tellurium phase change storage material potential function model is provided, which can be used in a computer device. Figure 2 Another construction method of a yttrium-doped antimony-tellurium phase change storage material potential function model according to an embodiment of the present application is shown in the flowchart of Figure 2 as shown, the flowchart includes the following steps:

[0043] Step S201: Obtain a data set, wherein the data set includes a plurality of training data, and each training data includes a Y-Sb-Te model and a material property corresponding to the Y-Sb-Te model.

[0044] The Y-Sb-Te model includes a Y-Sb-Te crystal phase model, a Y-Sb-Te crystal phase perturbation model, a Y-Sb-Te liquid phase model, a Y-Sb-Te liquid phase perturbation model, a Y-Sb-Te amorphous model, and a Y-Sb-Te amorphous perturbation model.

[0045] In an optional embodiment, the step S2011 to the step S2014 are included in the obtaining the dataset.

[0046] The step S2011 is obtaining a Y-Sb-Te crystal phase model.

[0047] Specifically, the Y-Sb-Te crystal phase model is obtained by constructing an Sb-Te crystal phase model using a preset material database, wherein the Sb-Te crystal phase model includes an Sb crystal phase model, an Sb2Te crystal phase model, an SbTe crystal phase model, an Sb2Te3 crystal phase model, and a Te crystal phase model; doping Y elements into the Sb2Te crystal phase model using a preset algorithm to obtain a first model; doping Y elements into the SbTe crystal phase model using a preset algorithm to obtain a second model; doping Y elements into the Sb2Te3 crystal phase model using a preset algorithm to obtain a third model; and using the first model, the second model, and the third model to constitute the Y-Sb-Te crystal phase model.

[0048] Specifically, the Sb-Te metastable and stable crystal phase models under different components can be constructed by the Materials Project material database. The selected metastable crystal phase structure is a rock salt structure, and the selected stable crystal phase is a zincite structure. The lattice constant is provided by the Materials Project database.

[0049] The method for obtaining the Y-Sb-Te crystal phase model in the embodiment fully considers the influence of different chemical ratios and doping concentrations on the material structure and performance by constructing crystal phase models covering various Sb-Te compounds and Y element doping systems, significantly improves the coverage and diversity of the training data in the composition space, and makes the constructed potential function model have excellent component adaptability and universality. Not only can the model accurately describe the phase transition behavior under a specific ratio, but also can be widely used for predicting the structure evolution, thermodynamic and kinetic properties of various Y-Sb-Te alloy systems, thereby expanding the application potential of the model in more extensive scenarios such as material design and doping optimization.

[0050] The step S2012 is performing a melting-rapid cooling simulation on the Y-Sb-Te crystal phase model to obtain a Y-Sb-Te liquid phase model and a Y-Sb-Te amorphous model.

[0051] Specifically, the Y-Sb-Te crystal phase model is subjected to a melting-rapid cooling simulation to obtain a Y-Sb-Te liquid phase model and a Y-Sb-Te amorphous model, including: subjecting the Y-Sb-Te crystal phase model to a melting-rapid cooling simulation by using a preset simulation process, wherein the simulation process includes: randomization at a temperature of 2000-2500 K, followed by incubation at a liquidus temperature of 1000-1200 K, then cooling at a cooling rate of 10-20 K / ps, and finally incubation at 275-325 K; obtaining a trajectory of the melting-rapid cooling simulation, and extracting frames from the trajectory at a preset time interval to obtain the Y-Sb-Te liquid phase model and the Y-Sb-Te amorphous model.

[0052] For example, the simulation process includes: randomization at a temperature of 2000 K, followed by incubation at a liquidus temperature of 1100 K, then cooling at a cooling rate of 20 K / ps, and finally incubation at 300 K; obtaining a trajectory of the melting-rapid cooling simulation, and extracting frames from the trajectory at a preset time interval to obtain the Y-Sb-Te liquid phase model and the Y-Sb-Te amorphous model.

[0053] The determination method of the Y-Sb-Te liquid phase model and the Y-Sb-Te amorphous model provided in this embodiment accurately reproduces the structural evolution path of the Y-Sb-Te system from a crystal phase to a liquid phase and then to an amorphous phase by simulating the melting and rapid cooling thermal history in the actual phase change process, and the obtained liquid phase and amorphous model are closer to the atomic configuration in the actual physical process; the structural rationality and stability of the amorphous model are effectively improved by sufficient relaxation in a key temperature range and cooling at a cooling rate consistent with the operating characteristics of a phase change memory, and the configuration sampling of the liquid phase and the amorphous state is further enriched by the trajectory frame extraction strategy, thereby providing a structure close to the real condition for constructing a high-fidelity training data set and subsequent accurate simulation of phase change dynamics.

[0054] Step S2013: perturbing the Y-Sb-Te crystal phase model, the Y-Sb-Te liquid phase model, and the Y-Sb-Te amorphous model by using a preset random perturbation algorithm to obtain a Y-Sb-Te crystal phase perturbed model, a Y-Sb-Te liquid phase perturbed model, and a Y-Sb-Te amorphous perturbed model.

[0055] Specifically, the Y-Sb-Te crystal phase model, the Y-Sb-Te liquid phase model and the Y-Sb-Te amorphous model are perturbed by a preset random perturbation algorithm to obtain Y-Sb-Te crystal phase perturbation model, Y-Sb-Te liquid phase perturbation model and Y-Sb-Te amorphous perturbation model, including: the Y-Sb-Te crystal phase model, the Y-Sb-Te liquid phase model and the Y-Sb-Te amorphous model are perturbed by a random perturbation algorithm to obtain Y-Sb-Te crystal phase perturbation model, Y-Sb-Te liquid phase perturbation model and Y-Sb-Te amorphous perturbation model; wherein the cell shape perturbation condition is that the a, b, c three-axis length is randomly stretched or compressed in the range of ± (4% ~ 6%), and the α, β, γ three angles are randomly deformed in the range of ± (4% ~ 5%); the atomic position perturbation condition is that the x, y, z three coordinates are randomly offset away from the initial position in the range of ± (4% ~ 5%).

[0056] For example, the cell shape perturbation condition is that the a, b, c three-axis length is randomly stretched or compressed in the range of ± 5%, and the α, β, γ three angles are randomly deformed in the range of ± 3%; the atomic position perturbation condition is that the x, y, z three coordinates are randomly offset away from the initial position in the range of ± 5%.

[0057] For example, the Y-Sb-Te crystal phase model, the Y-Sb-Te liquid phase model and the Y-Sb-Te amorphous model can be perturbed by the mcsqs algorithm in the Alloy Theoretic Automated Toolkit. The Y doping concentration is 3.7% and 8.7%, and the Y doping mode is to replace the Sb cation position in the model.

[0058] Wherein, mcsqs is a core tool in Alloy Theoretic Automated Toolkit (ATAT), which is used to generate special quasi-random structure (Special Quasirandom Structure, SQS) to simulate the local structure of disordered alloy or multi-component doped system. It is based on Monte Carlo optimization algorithm, by matching the RDF (Radial Distribution Function) or coordination number statistics of real alloy, a finite size supercell is generated, which is as close as possible to the disordered state of infinite solid solution in statistics.

[0059] The random perturbation algorithm of the embodiment effectively simulates the complex environment that the material may experience in actual application, such as stress, deformation and thermal disturbance, significantly expands the coverage of the training data in the structure phase space, enhances the diversity and physical authenticity of the data set, not only improves the description ability of the potential function model for the non-equilibrium structure, but also enables it to simulate the dynamic response behavior (such as mechanical deformation, impact effect, etc.) of the material under the action of external field, further expands the applicability and prediction reliability of the model under multiple scenarios and multiple physical conditions.

[0060] Step S2014: performing first-principle calculation on the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model and the Y-Sb-Te amorphous perturbation model to obtain material properties corresponding to the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model and the Y-Sb-Te amorphous perturbation model.

[0061] Specifically, the first-principle calculation on the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model and the Y-Sb-Te amorphous perturbation model includes: performing first-principle calculation on the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model and the Y-Sb-Te amorphous perturbation model, wherein the parameters of the first-principle calculation are: the plane cutoff energy is 550-650 eV, the K-point density in the Brillouin zone is set by the KSPACING parameter to be 0.15-0.25 Å -1 , the pseudo potential is selected as PAW-PBE, and the van der Waals dispersion effect is selected as D3 method.

[0062] For example, the parameters of the first-principle calculation are: the plane cutoff energy is 600 eV, the K-point density in the Brillouin zone is set by the KSPACING parameter to be 0.2 Å -1 , the pseudo potential is selected as PAW-PBE, and the van der Waals dispersion effect is selected as D3 method.

[0063] The method for determining material properties provided in the embodiment significantly improves the calculation accuracy of key material properties such as energy, force and stress by using first principle calculation parameters such as high cutoff energy, high K-point density, accurate pseudo potential (PAW-PBE) and considering van der Waals correction (D3), effectively reduces the systematic error of theoretical simulation, ensures the high quality and high reliability of the training data, so that the potential function model constructed can more accurately reflect the real physical interaction of the system, and significantly enhances the prediction accuracy and stability of the model under complex phase change process and multi-state structure. The data set acquisition method of the embodiment significantly expands the coverage range of the training data in the phase space and the configuration space by constructing a diversified structure system including crystal phase, liquid phase, amorphous phase and perturbation model, and combining first principle calculation to obtain high-precision material properties such as energy, force and stress, enhances the representativeness and diversity of the data set, effectively improves the generalization ability and prediction accuracy of the trained potential function model, so that it can accurately describe the atomic interaction of Y-Sb-Te system in different phases and dynamic processes, and lays a solid foundation for realizing high-fidelity large-scale phase change dynamics simulation.

[0064] Specifically, the material properties include at least one of the following: unit cell energy, atomic force, and stress.

[0065] Therefore, the embodiment can use unit cell energy, atomic force and stress as training targets, so that the potential function model can accurately learn the atomic interaction characteristics of the Y-Sb-Te system from multiple aspects such as energy, force field and stress response, not only improving the model's ability to describe the thermodynamic stability and dynamic behavior of the system, but also significantly enhancing its ability to simulate complex physical phenomena such as structural evolution, interface effect and mechanical response during phase change, thereby improving the accuracy and reliability of the model in predicting the phase change performance and mechanical stability of the material.

[0066] Step S202: training the preset NEP model using the data set to obtain a potential function model of the yttrium-doped antimony tellurium phase change storage material.

[0067] Specifically, training the preset NEP model using the data set to obtain a potential function model of the yttrium-doped antimony tellurium phase change storage material includes: training the preset NEP model using the data set to obtain a potential function model of the yttrium-doped antimony tellurium phase change storage material, wherein the selected parameters for training include: the cutoff radius of the radial and angular descriptors is 6-8 Å and 5-7 Å, the basis set number of the radial and angular descriptors is 8-12 and 6-10, the expansion coefficient of the angular descriptor is (4 2 0) or (4 2 1), and the number of hidden layer neurons is 60-100.

[0068] For example, the training of the selected parameters includes: the cutoff radius of the radial and angular descriptors is 7 Å and 6 Å, the number of basis sets of the radial and angular descriptors is 10 and 8, the expansion coefficient of the angular descriptor is 4 2 0, and the number of hidden layer neurons is 70.

[0069] The training method of the potential function model provided in the embodiment balances the expression ability and complexity of the model, avoids overfitting, significantly improves the training efficiency and convergence stability, and thus constructs a potential function model with high precision, strong generalization ability and good calculation efficiency, thereby providing a reliable guarantee for accurately simulating the phase transition dynamics and multi-phase structure evolution of the yttrium-doped antimony tellurium material. Further, after obtaining the potential function model, the prediction ability of the potential function model is verified, including numerical accuracy verification and physical property prediction verification: the numerical accuracy verification is performed by comparing the RMSE between the NEP prediction and the VASP calculation, and the physical property prediction is performed by comparing the lattice constant, the disordered phase radial distribution function and the bulk melting point.

[0070] The method for constructing the potential function model of the yttrium-doped antimony tellurium phase change storage material provided in the embodiment has the following beneficial effects:

[0071] (1) The dataset is constructed based on the results of the first principle calculation, and the relationship between different structure models and energy, force and potential force can be obtained, thereby providing high-quality data with good accuracy for the training of artificial intelligence machine learning.

[0072] (2) The problem of low efficiency of the current machine learning potential is solved, the calculation can be accelerated under the graphics calculation card, the calculation efficiency close to the classical potential function is realized, the shortcomings of the classical potential function that cannot be applied to the complex potential energy surface scene of multiple phases and multiple configurations are made up, and the molecular dynamics simulation close to the first principle accuracy is realized.

[0073] (3) The method provides a theoretical model and a necessary basis for exploring the amorphous phase structure evolution of the yttrium antimony tellurium phase change storage material, and has a wide application prospect in optimizing the performance of the phase change memory device and exploring the mechanism of the phase change material.

[0074] To describe the method for constructing the potential function model of the yttrium-doped antimony tellurium phase change storage material provided in the embodiment in more detail, a specific example is given. Figure 3 As shown in the example, the example includes the following steps:

[0075] Step 1, constructing metastable and stable crystal phase structure models of Sb-Te binary compounds under different compositions by the Materials Project material database, the compound compositions involved include Sb, Sb2Te, SbTe, Sb2Te3 and Te five kinds, the metastable crystal phase structure selected is rock salt structure, the stable crystal phase selected is zinc blende structure, and the lattice constant is provided by the Materials Project database. For the five compositions, the number of atoms in the stable state model is 4, 9, 6, 15 and 6 respectively, and the number of atoms in the metastable state model is 32, 180, 180, 180 and 8;

[0076] Step 2, using the mcsqs algorithm in Alloy Theoretic Automated Toolkit (ATAT) to dope Y element in the Sb2Te, SbTe and Sb2Te3 three Sb-Te crystal phase models obtained in step 1, the selected Y doping concentration is 3.7% and 8.7%, and the doping mode of Y is to replace the Sb cation position in the Sb-Te model. ATAT mainly generates random doping models through cluster expansion algorithm. First, read the cell structure characteristics, i.e. zinc blende and rock salt structure, through corrdump. The selected parameters are: -noe and -nop remove vacancy and point cluster characteristics, -2=8.62 consider the near-neighbor relationship of 2 atoms within 8.62 Å cutoff value, -3=11.41 consider the near-neighbor relationship of 3 atoms within 11.41 Å cutoff value, -clus only use corrdump algorithm to generate cluster characteristics. After completing the feature analysis of the initial model, use mcsqs to generate doped models. The selected parameters are: -n specifies the number of atoms in the generated doped model, -tol=0.01 sets the error tolerance of the generated doped model to 0.01, and finally generates multiple Y-doped crystal models;

[0077] Step 3, using random perturbation algorithm to perturb the cell shape and atomic position of the Y-Sb-Te crystal phase model generated in step 2: for the cell shape, the a, b, c three-axis length is stretched or compressed within ±5%, and the α, β, γ three angles are deformed within ±3%; for the atomic position, the x, y, z three coordinates are offset away from the initial position within ±5%, and the crystal phase model with certain distortion is generated based on the above method;

[0078] Step 4, melt-quench simulation of the Y-Sb-Te phase change material model generated in step 2 by first-principles molecular dynamics simulation: the initial model uses a crystalline phase model with a lattice constant of a = b = c = 18.75 Å and 180 atoms, a molecular dynamics step of 2 fs, randomization at a temperature of 2000 K for 20 ps, incubation at a liquid phase temperature of 1100 K for 30 ps, cooling at a rate of 20 K / ps to 300 K, and finally incubation at a room temperature of 300 K for 20 ps to obtain an amorphous phase model. Sample the trajectory of the simulation process and extract the liquid phase and amorphous model of Y-Sb-Te at fixed intervals of 1 ps;

[0079] Step 5, perturb the cell shape and atomic position of the Y-Sb-Te liquid and amorphous phase model generated in step 4 by a random perturbation algorithm to generate an amorphous phase model with certain distortion. The perturbation algorithm used is consistent with that in step 3;

[0080] Step 6, perform high-precision first-principles calculations on the Y-Sb-Te model generated in steps 1-5 by the first-principles software VASP. The parameters selected for the first-principles calculation are: a plane cutoff energy of 600 eV, K-point density in the Brillouin zone set by the KSPACING parameter to 0.2 Å -1 , PAW-PBE for pseudopotential, and D3 method for van der Waals interaction. After the calculation is completed, extract the cell energy, atomic force, and material properties such as the stress from the VASP calculation output as the dataset for the neural network potential function model. The number of structures in the training set is 2608, which is divided into a training set of 2080 and a test set of 528 according to an 8:2 ratio;

[0081] Step 7, train and test the dataset in step 6 using the neural evolution potential function model NEP in GPUMD. The parameters selected for the NEP potential function training are: NEP version 4, radial and angular descriptor cutoff radius 7 Å and 6 Å, radial and angular descriptor basis set 10 and 8, angular descriptor expansion coefficient 4 2 0, and hidden layer neuron number 70. The machine learning interatomic potential model for describing the Y-Sb-Te system is obtained. Subsequently, the prediction ability of the NEP potential function is verified, including numerical accuracy verification and physical property prediction verification: numerical accuracy verification by comparing the RMSE between NEP prediction and VASP calculation, and physical property prediction by comparing the lattice constant, disordered phase radial distribution function, and bulk melting point.

[0082] The yttrium-doped antimony telluride phase change memory material artificial intelligence neural network potential function developed in this embodiment is shown by calculations to have:

[0083] (1) The artificial intelligence neural network potential function of the yttrium-doped antimony tellurium phase change storage material developed in this embodiment has good fitting convergence characteristics, and the potential function has good numerical fitting accuracy on atomic energy, atomic force and potential stress, as shown in Figure 4 .

[0084] (2) The energy-volume curve and lattice constant obtained by comparing the first principle calculation with the artificial intelligence neural network potential function of the yttrium-doped antimony tellurium phase change storage material developed in this embodiment are shown in Figure 5 and Table 1. The results show that the potential function obtained by the present application is very close to the result of the first principle calculation, and the potential function model meets the expected effect.

[0085] Table 1 Comparison of equilibrium volume and lattice constant between NEP prediction and DFT calculation

[0086]

[0087] (3) The liquid phase and amorphous phase radial distribution function obtained by comparing the first principle molecular dynamics with the artificial intelligence neural network potential function of the yttrium-doped antimony tellurium phase change storage material developed in this embodiment is shown in Figure 6 . The results show that the radial distribution function obtained by the potential function of the present application is very close to the result of the first principle calculation, and the potential function model meets the expected effect.

[0088] (4) The melting point of the bulk phase change material is calculated by using the artificial intelligence neural network potential function of the yttrium-doped antimony tellurium phase change storage material developed in this embodiment by the molecular dynamics solid-liquid equilibrium method, and compared with the experimental value, as shown in Figure 7 . The results show that the potential function of the present application is more accurate in predicting the melting point of the yttrium antimony tellurium phase change material, and the error with the experimental value is within a reasonable range, and the potential function model meets the expected effect.

[0089] As can be seen, the artificial intelligence neural network potential function of the yttrium-doped antimony tellurium phase change storage material developed in the present application can accurately describe the lattice constant, radial distribution function and bulk melting point of the yttrium antimony tellurium phase change material, can accurately reproduce the short-range order characteristics and crystallization performance of the yttrium antimony tellurium phase change storage material, and is far higher than the first principle molecular dynamics and the traditional machine learning potential function (taking the GAP model as an example) in calculation efficiency, as shown in Figure 8 .

[0090] Therefore, the application adopts the construction method of the artificial intelligence neural network potential function of the yttrium-doped antimony tellurium phase change memory material, combines first principle high-throughput calculation, artificial intelligence neural network model and large-scale molecular dynamics simulation method, takes the Y-Sb-Te phase change material as the research object, fits the atomic energy, atomic force and position force stress structure characteristics based on the neural evolution potential framework, and fits the molecular dynamics potential function with the first principle precision.

[0091] In the embodiment, a construction device of a potential function model of a yttrium-doped antimony tellurium phase change memory material is also provided, and the device is used for realizing the above-mentioned embodiments and preferred embodiments, and will not be described again.

[0092] The embodiment provides a construction device of a potential function model of a yttrium-doped antimony tellurium phase change memory material, as shown in the following formula (1) and the following formula (2). Figure 9 The device comprises the following modules.

[0093] The acquisition module 901 is configured to acquire a data set, wherein the data set comprises a plurality of training data, and each training data comprises a Y-Sb-Te model and a material property corresponding to the Y-Sb-Te model.

[0094] The training module 902 is configured to train a preset NEP model by using the data set, to obtain the potential function model of the yttrium-doped antimony tellurium phase change memory material.

[0095] In some optional embodiments, the acquisition module 901 comprises a crystal phase model acquisition unit, a melting and rapid cooling simulation unit, a perturbation unit, and a calculation unit. The crystal phase model acquisition unit is configured to acquire a Y-Sb-Te crystal phase model. The melting and rapid cooling simulation unit is configured to perform melting and rapid cooling simulation on the Y-Sb-Te crystal phase model to obtain a Y-Sb-Te liquid phase model and a Y-Sb-Te amorphous model. The perturbation unit is configured to perturb the Y-Sb-Te crystal phase model, the Y-Sb-Te liquid phase model, and the Y-Sb-Te amorphous model respectively using a preset random perturbation algorithm to obtain a Y-Sb-Te crystal phase perturbation model, a Y-Sb-Te liquid phase perturbation model, and a Y-Sb-Te amorphous perturbation model. The calculation unit is configured to perform first-principle calculation on the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model, and the Y-Sb-Te amorphous perturbation model to obtain material properties corresponding to the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model, and the Y-Sb-Te amorphous perturbation model.

[0096] In some optional embodiments, the perturbation unit is configured to perform cell shape perturbation and atomic position perturbation on the Y-Sb-Te crystal phase model, the Y-Sb-Te liquid phase model, and the Y-Sb-Te amorphous model respectively using a random perturbation algorithm to obtain the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase perturbation model, and the Y-Sb-Te amorphous perturbation model. The cell shape perturbation is performed under the condition that the lengths of the a, b, and c axes are randomly stretched or compressed within ±(4%~6%), and the angles of α, β, and γ are randomly deformed within ±(4%~5%). The atomic position perturbation is performed under the condition that the x, y, and z coordinates are randomly offset from the initial positions within ±(4%~5%).

[0097] In some optional embodiments, the calculation unit is configured to perform first-principle calculation on the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model, and the Y-Sb-Te amorphous perturbation model. The parameters of the first-principle calculation are as follows: the plane cutoff energy is 550~650 eV, the K-point density in the Brillouin zone is set by the KSPACING parameter to be 0.15~0.25 Å-1, the pseudo potential is selected as PAW-PBE, and the van der Waals dispersion interaction is selected as D3 method.

[0098] In some optional embodiments, the training module is specifically configured to train the preset NEP model by using the data set to obtain the potential function model of the yttrium-doped antimony tellurium phase change memory material, wherein the training selected parameters include: the cutoff radius of the radial and angular descriptors is 6-8 Å and 5-7 Å, the basis set number of the radial and angular descriptors is 8-12 and 6-10, the expansion coefficient of the angular descriptor is (4 2 0) or (4 2 1), and the number of hidden layer neurons is 60-100.

[0099] Further function descriptions of the above-mentioned various modules and units are the same as those of the above-mentioned corresponding embodiments, and will not be described here again.

[0100] The construction device of the potential function model of the yttrium-doped antimony tellurium phase change memory material in the embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit, Application Specific Integrated Circuit) circuit, a processor and a memory executing one or more software or fixed programs, and / or other devices that can provide the above-mentioned functions.

[0101] The embodiment of the application also provides a computer device having the above-mentioned Figure 9 construction device of the potential function model of the yttrium-doped antimony tellurium phase change memory material.

[0102] Please refer to Figure 10 , Figure 10 is a structural schematic diagram of a computer device provided by an optional embodiment of the application, as Figure 10 shown, the computer device includes one or more processors 10, a memory 20, and an interface for connecting various components, including a high-speed interface and a low-speed interface. Various components are communicatively connected to each other by using different buses, and can be installed on a common mainboard or in other ways as needed. The processor can process instructions executed in the computer device, including instructions stored in the memory or on the memory to display graphical information on a GUI on an external input / output device (such as a display device coupled to the interface). In some optional embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memories, if necessary. Similarly, multiple computer devices can be connected, each providing part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 10 In the figure, the processor 10 is taken as an example.

[0103] The processor 10 can be a central processing unit, a network processing unit, or a combination thereof. The processor 10 can further include a hardware chip. The hardware chip can be an application specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device can be a complex programmable logic device, a field programmable logic device, a general array logic, or any combination thereof.

[0104] The memory 20 stores instructions executable by the at least one processor 10 to cause the at least one processor 10 to perform the methods illustrated in the above embodiments.

[0105] The memory 20 can include a program storage area and a data storage area. The program storage area can store an operating system and application programs required by at least one function. The data storage area can store data created according to the use of the computer device, and the like. In addition, the memory 20 can include a high-speed random access memory, and can further include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some alternative embodiments, the memory 20 can optionally include a memory disposed remotely from the processor 10, which can be connected to the computer device through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0106] The memory 20 can include a volatile memory, such as a random access memory, and can also include a non-volatile memory, such as a flash memory, a hard disk, or a solid state disk. The memory 20 can further include a combination of the above-mentioned types of memories.

[0107] The computer device further includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30, and the output device 40 can be connected by a bus or other means, Figure 10 For example, by a bus.

[0108] The input device 30 can receive input digital or character information, and generate key signal inputs related to the user settings and function controls of the computer device, such as a touch screen, a keypad, a mouse, a trackpad, a touchpad, a pointing stick, one or more mouse buttons, a trackball, a joystick, and the like. The output device 40 can include a display device, an auxiliary lighting device (e.g., an LED), a tactile feedback device (e.g., a vibration motor), and the like. The display device includes, but is not limited to, a liquid crystal display, a light emitting diode, a display, and a plasma display. In some alternative embodiments, the display device can be a touch screen.

[0109] The embodiments of the present application further provide a computer readable storage medium, and the method according to the embodiments of the present application can be implemented in hardware, firmware, or recorded in a storage medium, or stored in a remote storage medium or a non-transitory machine readable storage medium and downloaded to a local storage medium through network, so that the method described herein can be processed by such software on a storage medium using a general purpose computer, a special purpose processor, or programmable or special hardware. The storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid state disk, etc. Further, the storage medium can also include a combination of the above-mentioned memories. It can be understood that the computer, the processor, the microprocessor controller, or the programmable hardware includes a storage component that can store or receive software or computer code, when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the method shown in the above embodiments is implemented.

[0110] Part of the present application can be applied as a computer program product, for example, computer program instructions, when executed by a computer, the operation of the computer can invoke or provide the method and / or technical solutions according to the present application. Those skilled in the art should understand that the form of computer program instructions in computer readable medium includes but is not limited to source file, executable file, installation package file, etc. Correspondingly, the way of computer program instructions executed by computer includes but is not limited to: the computer directly executes the instructions, or the computer compiles the instructions and then executes the corresponding compiled program, or the computer reads and executes the instructions, or the computer reads and installs the instructions and then executes the corresponding installed program. Here, the computer readable medium can be any available computer readable storage medium or communication medium accessible to the computer.

[0111] Although the embodiments of the present application are described in conjunction with the accompanying drawings, various modifications and changes can be made by those skilled in the art without departing from the spirit and scope of the present application, and such modifications and changes fall within the scope defined by the appended claims.

Claims

1. A method for constructing a potential function model for yttrium-doped antimony-tellurium phase transition memory materials, characterized in that, include: Obtain a dataset, wherein the dataset includes multiple training data sets, each training data set including a Y-Sb-Te model and material properties corresponding to the Y-Sb-Te model; wherein the material properties include stress. The dataset was used to train a pre-defined NEP model to obtain a potential function model for yttrium-doped antimony tellurium phase change memory material. The Y-Sb-Te model includes a Y-Sb-Te crystalline phase model, a Y-Sb-Te crystalline phase perturbation model, a Y-Sb-Te liquid phase model, a Y-Sb-Te liquid phase perturbation model, a Y-Sb-Te amorphous model, and a Y-Sb-Te amorphous perturbation model; the acquired dataset includes: Obtain the Y-Sb-Te crystal phase model; The Y-Sb-Te crystal phase model was subjected to melting-rapid cooling simulation to obtain the Y-Sb-Te liquid phase model and the Y-Sb-Te amorphous model. The Y-Sb-Te crystalline phase model, the Y-Sb-Te liquid phase model, and the Y-Sb-Te amorphous model are perturbed using a preset random perturbation algorithm to obtain the Y-Sb-Te crystalline phase perturbation model, the Y-Sb-Te liquid phase perturbation model, and the Y-Sb-Te amorphous perturbation model, respectively. First-principles calculations were performed on the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model, and the Y-Sb-Te amorphous perturbation model to obtain the material properties corresponding to the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model, and the Y-Sb-Te amorphous perturbation model. The step of perturbing the Y-Sb-Te crystalline phase model, the Y-Sb-Te liquid phase model, and the Y-Sb-Te amorphous model using a preset random perturbation algorithm to obtain the Y-Sb-Te crystalline phase perturbation model, the Y-Sb-Te liquid phase perturbation model, and the Y-Sb-Te amorphous perturbation model includes: The Y-Sb-Te crystalline phase model, the Y-Sb-Te liquid phase model, and the Y-Sb-Te amorphous model are subjected to perturbation of cell shape and atomic position using the random perturbation algorithm, respectively, to obtain the Y-Sb-Te crystalline phase perturbation model, the Y-Sb-Te liquid phase perturbation model, and the Y-Sb-Te amorphous perturbation model. The conditions for the cell shape perturbation are: the lengths of the three axes a, b, and c are randomly stretched or compressed within ±(4%~6%), and the included angles α, β, and γ are randomly deformed within ±(4%~5%). The conditions for the atomic position perturbation are: the three coordinates x, y, and z are randomly offset away from the initial position within ±(4%~5%).

2. The method according to claim 1, characterized in that, The first-principles calculations performed on the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model, and the Y-Sb-Te amorphous perturbation model include: First-principles calculations were performed on the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model, and the Y-Sb-Te amorphous perturbation model. The parameters used in these first-principles calculations were: a plane cutoff energy of 550–650 eV, and a K-point density in the Brillouin zone set to 0.15–0.25 Å using the KSPACING parameter. -1 The pseudopotential was selected as PAW-PBE, and the van der Waals dispersion method was selected as D3.

3. The method according to claim 1, characterized in that, The step of training a pre-defined NEP model using the dataset to obtain the potential function model for the yttrium-doped antimony tellurium phase transition memory material includes: The dataset is used to train a preset NEP model to obtain a potential function model for yttrium-doped antimony tellurium phase transition memory material. The parameters selected for training include: the cutoff radius of the radial and angular descriptors is 6~8 Å and 5~7 Å, the number of basis sets of the radial and angular descriptors is 8~12 and 6~10, the expansion coefficient of the angular descriptor is (4 2 0) or (4 2 1), and the number of hidden layer neurons is 60~100.

4. A potential function model, characterized in that, The potential function model of the yttrium-doped antimony tellurium phase transition memory material as described in any one of claims 1 to 3 was constructed.

5. A device for constructing a potential function model for yttrium-doped antimony-tellurium phase transition memory materials, characterized in that, include: The acquisition module is used to acquire a dataset, wherein the dataset includes multiple training data, each training data including a Y-Sb-Te model and material properties corresponding to the Y-Sb-Te model; wherein the material properties include stress. The training module is used to train the preset NEP model using the dataset to obtain the potential function model of the yttrium-doped antimony tellurium phase change memory material. The Y-Sb-Te model includes a Y-Sb-Te crystal phase model, a Y-Sb-Te crystal phase perturbation model, a Y-Sb-Te liquid phase model, a Y-Sb-Te liquid phase perturbation model, a Y-Sb-Te amorphous model, and a Y-Sb-Te amorphous perturbation model; the acquisition module includes a crystal phase model acquisition unit, a melting rapid cooling simulation unit, a perturbation unit, and a calculation unit. The crystal phase model acquisition unit is used to acquire the Y-Sb-Te crystal phase model; The melting-rapid cooling simulation unit is used to perform melting-rapid cooling simulation on the Y-Sb-Te crystal phase model to obtain the Y-Sb-Te liquid phase model and the Y-Sb-Te amorphous model. The perturbation unit is used to perturb the Y-Sb-Te crystal phase model, the Y-Sb-Te liquid phase model, and the Y-Sb-Te amorphous model using a preset random perturbation algorithm, respectively, to obtain the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase perturbation model, and the Y-Sb-Te amorphous perturbation model; The calculation unit is used to perform first-principles calculations on the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model, and the Y-Sb-Te amorphous perturbation model to obtain the material properties corresponding to the Y-Sb-Te crystal phase model, the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase model, the Y-Sb-Te liquid phase perturbation model, the Y-Sb-Te amorphous model, and the Y-Sb-Te amorphous perturbation model. The perturbation unit is specifically used to: perform cell shape perturbation and atomic position perturbation on the Y-Sb-Te crystal phase model, the Y-Sb-Te liquid phase model, and the Y-Sb-Te amorphous model respectively using the random perturbation algorithm to obtain the Y-Sb-Te crystal phase perturbation model, the Y-Sb-Te liquid phase perturbation model, and the Y-Sb-Te amorphous perturbation model; wherein the cell shape perturbation condition is: the lengths of the three axes a, b, and c are randomly stretched or compressed within ±(4%~6%), and the included angles α, β, and γ are randomly deformed within ±(4%~5%); the atomic position perturbation condition is: the three coordinates x, y, and z are randomly offset away from the initial position within ±(4%~5%).

6. A computer device, characterized in that, include: The system includes a memory and a processor, which are interconnected and communicate with each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the method for constructing the potential function model of the yttrium-doped antimony tellurium phase transition memory material as described in any one of claims 1 to 3.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to execute the method for constructing the potential function model of the yttrium-doped antimony tellurium phase transition memory material as described in any one of claims 1 to 3.

8. A computer program product, characterized in that, The invention includes computer instructions for causing a computer to execute the method for constructing the potential function model of the yttrium-doped antimony tellurium phase transition memory material as described in any one of claims 1 to 3.

Images