Bayesian optimization-based hafnium zirconium oxide film process parameter optimization method and system

Abstract

The invention provides a Bayesian optimization-based hafnium zirconium oxide film process parameter optimization method and system. The method comprises the following steps: preparing a hafnium zirconium oxide film; obtaining the performance of the hafnium-zirconium-oxide thin film, and constructing a data set of process parameter conditions and corresponding performance of the hafnium-zirconium-oxide thin film; based on the data set, adopting a Bayesian optimization method to optimize the hafnium zirconium oxide film process parameters, comprising the steps of sequentially constructing an agent model, defining an acquisition function, iteratively optimizing the acquisition function and updating the agent model, and repeating the steps until a termination condition is met; and determining an optimal process parameter combination for preparing the hafnium zirconium oxide thin film according to a Bayesian optimization result. According to the method, the optimal process parameters of the hafnium zirconium oxide film can be quickly determined, and the cost of developing a semiconductor process is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of hafnium zirconium oxide thin film preparation, and in particular to a method and system for optimizing process parameters of hafnium zirconium oxide thin films based on Bayesian optimization. Background Art

[0002] Artificial intelligence is developing rapidly, which has put forward huge demands on computing power. In this context, storage-computing integration has been proposed to solve the problem of storage wall in traditional computing architecture. Ferroelectric devices have great potential in realizing storage-computing integration, but semiconductor processes are extremely complex. The key processes for developing these AI-supported chips are still completed by process engineers using intuition and experience, which usually requires repeated trials. The application of artificial intelligence in process engineering for manufacturing new chips has aroused widespread interest.

[0003] Artificial Intelligence There are many examples of computer algorithms outperforming humans in complex tasks, such as playing board games such as chess and go. However, in these cases, computers can only make decisions after training or generating large amounts of cheap data. In contrast, it is expensive to collect process data for silicon wafers, so engineers typically test only hundreds (trillions) of different combinations of parameters such as pressure, power, reactive gas flow, and wafer temperature when developing semiconductor processes. Unlike a board game with well-defined rules, a wafer reactor system is governed by an immeasurable number of microscopic physical and chemical interactions between wafer materials, plasma, and reactor components. The lack of sufficient data in specific areas of interest makes it difficult to form computer models with atomic-level accuracy, which is known as the “small” data problem. Therefore, how to reduce the target cost of developing semiconductor processes (i.e., minimize the amount of data that needs to be collected) relative to experienced process engineers poses new challenges for artificial intelligence. Summary of the invention

[0004] In view of the defects in the prior art, the object of the present invention is to provide a method and system for optimizing process parameters of hafnium zirconium oxide thin films based on Bayesian optimization.

[0005] According to one aspect of the present invention, a method for optimizing process parameters of a hafnium zirconium oxide thin film based on Bayesian optimization is provided, the method comprising:

[0006] Preparation of hafnium zirconium oxide thin films;

[0007] Obtaining the performance of the hafnium zirconium oxide film and constructing a data set of process parameter conditions and corresponding performance of the hafnium zirconium oxide film;

[0008] Based on the data set, the process parameters of the hafnium zirconium oxide film are optimized by using a Bayesian optimization method, including: sequentially performing the steps of building a proxy model, defining an acquisition function, iteratively optimizing the acquisition function, and updating the proxy model, and repeating the above steps until a termination condition is met;

[0009] According to the results of Bayesian optimization, the optimal process parameter combination for preparing hafnium zirconium oxide thin films was determined.

[0010] Further, the hafnium zirconium oxide thin film is prepared, wherein: the hafnium zirconium oxide thin film is formed by depositing a thin film by ALD;

[0011] The ALD film deposition involves at least one metal precursor and at least one reaction gas. The process parameters in the ALD film deposition process include deposition temperature, gas pressure and gas flow ratio.

[0012] Furthermore, the obtaining of the performance of the hafnium zirconium oxide film and constructing a data set of process parameter conditions and corresponding performance of the hafnium zirconium oxide film includes:

[0013] Obtain the test results of hysteresis performance of hafnium zirconium oxide film;

[0014] According to the hysteresis performance test results, the Ec and 2Pr parameters corresponding to each process parameter condition are extracted to form a data set corresponding to the process parameter condition and the performance.

[0015] Furthermore, the steps of constructing a proxy model, defining an acquisition function, iteratively optimizing, and updating the proxy model are performed in sequence, wherein the constructing of the proxy model includes:

[0016] Define an objective function;

[0017] Determine the parameter space of factors affecting the electrical properties of hafnium zirconium oxide thin films;

[0018] The Gaussian process is used as a surrogate model to approximate the objective function, and the uncertainty of the predicted value of the objective function is calculated to form a confidence interval.

[0019] Furthermore, the optimization goal of the objective function is to maximize the 2pr value; the parameter space includes deposition temperature, gas pressure and gas flow ratio.

[0020] Furthermore, the steps of constructing a proxy model, defining an acquisition function, iteratively optimizing, and updating the proxy model are performed in sequence, wherein the acquisition function is selected from any one of expected improvement, probability improvement, and upper confidence bound, and the acquisition function is determined based on a predefined exploration and exploitation balance criterion.

[0021] Furthermore, the above steps are repeated until a termination condition is met, wherein the termination condition is any one of reaching a predetermined number of iterations, an improvement amount being less than a certain threshold, and reaching a preset performance indicator.

[0022] Furthermore, the method of determining the optimal process parameter combination for preparing the hafnium zirconium oxide film based on the results of Bayesian optimization includes: measuring the electrical properties of the film prepared using the process parameter results of Bayesian optimization, and feeding the measurement results back to the Bayesian optimization step, and using statistical analysis to evaluate the robustness of the optimal process parameter combination.

[0023] According to another aspect of the present invention, a system for optimizing process parameters of a hafnium zirconium oxide thin film based on Bayesian optimization is provided, the system comprising:

[0024] A hafnium zirconium oxide thin film preparation device, used for preparing a hafnium zirconium oxide thin film;

[0025] A data set construction module, used to obtain the performance of the hafnium zirconium oxide film and construct a data set of process parameter conditions and corresponding performance of the hafnium zirconium oxide film;

[0026] A Bayesian optimization module, including an agent model building unit, a collection function definition unit, a collection function optimization unit, and a model updating unit;

[0027] The parameter determination module is used to determine the optimal process parameter combination for preparing the hafnium zirconium oxide film according to the result of the Bayesian optimization module.

[0028] Furthermore, the system also includes an evaluation module and / or a display module, wherein the evaluation module is used to measure the electrical properties of the film prepared using the process parameter results of Bayesian optimization, and to feed back the measurement results to the Bayesian optimization module; the display module serves as an input interface for the process parameters of the hafnium zirconium oxide film preparation, and is used to display the Bayesian optimization results and the determined optimal process parameter combination.

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

[0030] 1. The present invention prepares a hafnium zirconium oxide film and constructs a data set of electrical properties and process parameters of the hafnium zirconium oxide film, and optimizes the process parameters based on a Bayesian optimization method, thereby solving the problems of a large amount of collected data and high target cost in developing semiconductor processes. The present invention can minimize the amount of data that needs to be collected and reduce the cost of developing semiconductor processes.

[0031] 2. The present invention adopts Bayesian optimization to simplify complex optimization problems into a series of iterative processes. Each step selects the next sampling point based on the current best estimate, aiming to find the global optimal solution of the objective function and approach the optimal solution within the fewest possible iterations. It is particularly suitable for situations where function evaluation is costly or the number of evaluations is limited, thereby quickly determining the optimal process parameters for preparing hafnium zirconium oxide thin films. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0033] Figure 1 Schematic diagram of a flow chart of a method for optimizing process parameters of a hafnium zirconium oxide thin film based on Bayesian optimization in one embodiment of the present invention. DETAILED DESCRIPTION

[0034] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several variations and improvements may be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0035] An embodiment of the present invention provides a method for optimizing process parameters of a hafnium zirconium oxide thin film based on Bayesian optimization, referring to Figure 1 , the method comprising:

[0036] S1, preparing a hafnium zirconium oxide thin film;

[0037] S2, obtaining the performance of the hafnium zirconium oxide film, and constructing a data set of process parameter conditions and corresponding performance of the hafnium zirconium oxide film;

[0038] S3. Based on the data set, the Bayesian optimization method is used to optimize the process parameters of the hafnium zirconium oxide film, including: sequentially constructing a proxy model, defining an acquisition function, iteratively optimizing the acquisition function, and updating the proxy model, and repeating the above steps until the termination condition is met;

[0039] S4. According to the results of Bayesian optimization, the optimal process parameter combination for preparing hafnium zirconium oxide thin film is determined.

[0040] In an embodiment of the present invention, the preparation process of the hafnium zirconium oxide capacitor device includes sputtering a bottom electrode, ALD depositing a thin film, and sputtering a top electrode in sequence, and a hafnium zirconium oxide thin film is formed by ALD depositing the thin film. Hafnium zirconium oxide thin films are widely used in fields such as data storage, transistor gate dielectrics, and dynamic random access memory capacitor dielectrics. However, its preparation process is complicated and involves multiple parameters. For example, ALD depositing a thin film involves at least one metal precursor and at least one reaction gas. The process parameters in the ALD film deposition process include but are not limited to deposition temperature, gas pressure, and gas flow ratio. Slight changes in these parameters may have a significant effect on the performance of the film. Therefore, it is necessary to improve the performance of the film by optimizing the process parameters to meet the use requirements.

[0041] In some embodiments, in step S2, in order to construct a data set for the process of hafnium zirconium oxide thin film, the data generated in the experiment are collected and the data are preprocessed; it should be noted that although Bayesian optimization performs well on small data sets, the small amount of data may cause the model to predict inaccurately in the initial stage; especially in high-dimensional space, data sparsity may lead to unstable optimization results. The more data, the more accurate the model prediction results. Specifically, nearly one hundred process data and test results of corresponding thin films are collected. For example, the test results of hysteresis performance of hafnium zirconium oxide thin film are obtained; according to the test results of hysteresis performance, the coercive electric field Ec and 2Pr parameters corresponding to each process parameter condition are extracted to form a data set corresponding to the process parameter condition and the performance. Among them, the coercive electric field is the electric field strength corresponding to the polarization P is zero; Pr represents the residual polarization, which is the polarization strength that the material still maintains when the electric field E is zero, and the multiple of the residual polarization represents the difference between positive and negative polarization on the hysteresis loop.

[0042] In a specific embodiment, the prepared hafnium zirconium oxide thin film device includes 30nm TiN as a top electrode (TE), HZO as a dielectric layer, and 30nm TiN as a bottom electrode (BE). The device is manufactured on an n-doped silicon substrate, and the top electrode and the bottom electrode are deposited using a magnetron sputtering machine. The HZO ferroelectric layer is deposited by atomic layer deposition. The process conditions of atomic layer deposition are controlled, and hafnium zirconium oxide thin films with different process conditions are prepared, and their electrical properties are tested.

[0043] The measured hysteresis curve is processed to extract the parameters that determine the film properties, such as Ec and 2Pr, corresponding to the film process parameters, and construct a data set corresponding to the process conditions, thereby constructing a hafnium zirconium oxide film data set.

[0044] In order to optimize the process effect of hafnium zirconium oxide thin film, for example, to optimize the parameters in the ALD film deposition process for the electrical properties such as the PV curve performance of the prepared hafnium zirconium oxide thin film, the embodiment of the present invention introduces a Bayesian optimization method. Bayesian optimization is an efficient global optimization strategy, which is particularly suitable for optimizing black box functions with high evaluation costs. This method is based on the machine learning method of Bayesian theory and Gaussian process (GP). It approximates the objective function by constructing a proxy model (usually a Gaussian process), and uses the probability model of this objective function to guide the search process, so as to find the optimal solution of the function in as few steps as possible. The core steps of Bayesian optimization include: constructing a proxy model: defining the acquisition function; optimizing the acquisition function; updating the model; repeating the above steps until a certain stopping criterion is met.

[0045] In some embodiments, in step S3, a proxy model is constructed using existing data points (input-output pairs of functions), which is a probabilistic representation of the objective function. Constructing the proxy model includes: defining an objective function for quantifying electrical properties such as PV curves, specifically, the optimization goal of the objective function is to maximize the 2pr value, thereby clarifying the optimization goal; identifying and defining the range of all adjustable parameters, which affect the electrical properties of the hafnium zirconium oxide film, and determining the parameter space of factors affecting the electrical properties of the hafnium zirconium oxide film, and the parameter space should include all variables that may affect the objective function. Exemplarily, the parameter space includes but is not limited to deposition temperature, gas pressure, and gas flow ratio; using Gaussian process (GP) as a proxy model to approximate the objective function, and calculating the uncertainty of the predicted value of the objective function to form a confidence interval; Gaussian process can not only provide a prediction of the objective function, but also give the uncertainty of the prediction (confidence interval).

[0046] The acquisition function such as expected improvement (EI), probability of improvement (PI) or knowledge gradient (KG) is used to determine the location of the most valuable query point for the next step according to the current state of the proxy model, and find the best balance between exploration (finding areas with high uncertainty) and utilization (finding areas with high predicted values). In some embodiments, in step S3, the acquisition function is selected from any one of expected improvement (EI), probability improvement (PI) and upper confidence bound (UCB), and is used to determine the most valuable query point for the next step based on the current state of the proxy model, and the acquisition function is determined based on a predefined exploration and utilization balance standard.

[0047] In some embodiments, in step S3, in the process of optimizing the acquisition function, an input point that maximizes the acquisition function value is found, and this point is the next candidate point for evaluating the objective function. The iterative optimization process includes using a small number of randomly selected parameter combinations to perform a preliminary evaluation of the objective function, and using the acquisition function to determine the next set of parameters based on the proxy model to evaluate the objective function. In the process of updating the model, the objective function is evaluated at the new candidate point, and then the proxy model is updated using the new data point (i.e., the new input-output pair). The termination condition is any one of reaching a predetermined number of iterations, the improvement amount being less than a certain threshold, and reaching a preset performance indicator.

[0048] In the above-mentioned embodiments of the present invention, Bayesian optimization mainly involves the following key contents: (1) Objective function: The objective function is the function to be optimized, where X is defined on the input space and the specific form of f is unknown. Bayesian optimization is particularly suitable for processing objective functions with high evaluation costs. (2) Prior distribution: Before any observation data, the knowledge of the objective function is represented by the prior distribution. Using the Gaussian process as the prior in Bayesian optimization can well capture the relationship between function values ​​(i.e., covariance) and can update the prior to form the posterior distribution. (3) Likelihood function: The likelihood function measures the likelihood that the objective function will generate these data given the observation data. In Bayesian optimization, the likelihood function is usually implicit in the update of the Gaussian process. (4) Posterior distribution: Combining the prior distribution and the new observation data, the posterior distribution can be calculated by Bayes' theorem. The posterior distribution reflects the new understanding of the objective function under the observation data. (5) Acquisition function: The acquisition function defines a score on the candidate point based on the current posterior distribution, which is used to guide where the next step should be sampled, thereby balancing exploration (sampling in areas with high uncertainty) and exploitation (sampling in areas with high expected function values). (6) Optimization process: Use the acquisition function to select the next evaluation point, and then use the objective function to evaluate at this point. The evaluation result is used to update the Gaussian process model (i.e., the posterior distribution). This process is repeated until the stopping condition is met (such as reaching the maximum number of iterations or the improvement is less than a certain threshold). (7) Stopping criterion: Determine when to stop the optimization process, which may be based on a variety of factors, such as budget constraints, performance thresholds that have been reached, or diminishing marginal returns of improvements.

[0049] In some embodiments, in step S4, after completing the optimization iteration, the electrical characteristics of the film prepared using the process parameter results of Bayesian optimization are measured, the optimal parameter combination obtained and its corresponding electrical performance are analyzed, and the measurement results are fed back to the Bayesian optimization step, and the robustness of the optimal process parameter combination is evaluated using statistical analysis. In addition, additional experiments are performed to verify the optimal process parameter combination determined by Bayesian optimization.

[0050] In the above-mentioned embodiment of the present invention, a hafnium zirconium oxide film is first prepared and a data set of the hafnium zirconium oxide film, process parameters and corresponding electrical properties based on experimental data is constructed, and Bayesian optimization is used to optimize and predict the film performance based on the data set, thereby optimizing the process effect based on the Bayesian optimization method, solving the problems of a large number of collected data and high target costs in developing semiconductor processes, and can minimize the number of data that need to be collected, thereby reducing the cost of developing semiconductor processes.

[0051] In the above embodiment of the present invention, the complex optimization problem is simplified into a series of iterative processes through Bayesian optimization, and each step selects the next sampling point based on the current best estimate, aiming to find the global optimal solution of the objective function. Through this strategy, Bayesian optimization can approximate the optimal solution in as few iterations as possible, which is particularly suitable for situations where the function evaluation cost is high or the number of evaluations is limited. It can efficiently explore and utilize the process parameter space to find the global optimal solution. The performance of the hafnium zirconium oxide film may show a highly nonlinear and complex dependence on the process parameters. In the embodiment of the present invention, Bayesian optimization approximates the objective function through Gaussian process (GP) or other proxy models. Bayesian optimization does not require an explicit expression of the objective function, that is, it does not require the objective function to have a clear analytical expression, which is very suitable for complex process parameter optimization problems, especially when the objective function can only be obtained through experiments or simulations, thereby solving the problem of objective function complexity. For the process parameter optimization of hafnium zirconium oxide film, the method of the above embodiment of the present invention can quickly determine the optimal process parameters for preparing hafnium zirconium oxide film, and find the optimized parameter combination faster than traditional grid search or random search.

[0052] Based on the same inventive concept, another embodiment of the present invention provides a hafnium zirconium oxide film process parameter optimization system based on Bayesian optimization, the system comprising a hafnium zirconium oxide film preparation device, a data set construction module, a Bayesian optimization module and a parameter determination module, wherein: the hafnium zirconium oxide film preparation device is used to prepare a hafnium zirconium oxide film, including an atomic layer deposition device; the data set construction module is used to obtain the performance of the hafnium zirconium oxide film, and to construct a data set of process parameter conditions and corresponding performance of the hafnium zirconium oxide film; the Bayesian optimization module comprises an agent model construction unit, an acquisition function definition unit, an acquisition function optimization unit, and a model update unit; the Bayesian optimization module is configured with hardware and software resources for executing the Bayesian optimization algorithm, including an implementation mechanism of the agent model and the acquisition function, and is further configured with a mechanism for determining the optimization termination condition; the parameter determination module is used to determine the optimal process parameter combination for preparing the hafnium zirconium oxide film according to the result of the Bayesian optimization module.

[0053] In some embodiments, the system further includes an evaluation module and / or a display module, wherein the evaluation module includes at least one instrument and software tool related to electrical property measurement, and is used to measure the electrical properties of a thin film prepared using the process parameter results of Bayesian optimization, and feed the measurement results back to the Bayesian optimization module, such as measuring and analyzing the PV curve of a hafnium zirconium oxide thin film, and feeding the measurement results back to the Bayesian optimization module for optimization; the display module serves as an input interface for the process parameters for preparing the hafnium zirconium oxide thin film, and can input and adjust the process parameters, as well as display the Bayesian optimization results and the determined optimal process parameter combination.

[0054] Regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0055] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various modifications or variations within the scope of the claims, which does not affect the essence of the present invention. The above preferred features can be used in any combination without conflicting with each other.

Claims

1. A method for optimizing process parameters of hafnium zirconium oxide thin films based on Bayesian optimization, characterized in that: include: Preparation of hafnium zirconium oxide thin films; Obtaining the performance of the hafnium zirconium oxide film and constructing a data set of process parameter conditions and corresponding performance of the hafnium zirconium oxide film; Based on the data set, the process parameters of the hafnium zirconium oxide film are optimized by using a Bayesian optimization method, including: sequentially performing the steps of building a proxy model, defining an acquisition function, iteratively optimizing the acquisition function, and updating the proxy model, and repeating the above steps until a termination condition is met; According to the results of Bayesian optimization, the optimal process parameter combination for preparing hafnium zirconium oxide thin films was determined.

2. The method for optimizing process parameters of hafnium zirconium oxide thin films based on Bayesian optimization according to claim 1, characterized in that: The method comprises the following steps: forming the hafnium zirconium oxide thin film by depositing a thin film by ALD; The ALD film deposition involves at least one metal precursor and at least one reaction gas. The process parameters in the ALD film deposition process include deposition temperature, gas pressure and gas flow ratio.

3. The method for optimizing process parameters of hafnium zirconium oxide thin films based on Bayesian optimization according to claim 1, characterized in that: The obtaining of the performance of the hafnium zirconium oxide film and constructing a data set of process parameter conditions and corresponding performance of the hafnium zirconium oxide film includes: Obtain the test results of hysteresis performance of hafnium zirconium oxide film; According to the hysteresis performance test results, the Ec and 2Pr parameters corresponding to each process parameter condition are extracted to form a data set corresponding to the process parameter condition and the performance.

4. The method for optimizing process parameters of hafnium zirconium oxide thin films based on Bayesian optimization according to claim 1, characterized in that: The steps of constructing a proxy model, defining an acquisition function, iteratively optimizing, and updating the proxy model are performed in sequence, wherein the constructing of the proxy model includes: Define an objective function; Determine the parameter space of factors affecting the electrical properties of hafnium zirconium oxide thin films; The Gaussian process is used as a surrogate model to approximate the objective function, and the uncertainty of the predicted value of the objective function is calculated to form a confidence interval.

5. The method for optimizing process parameters of hafnium zirconium oxide thin films based on Bayesian optimization according to claim 4, characterized in that: The optimization goal of the objective function is to maximize the 2pr value; the parameter space includes deposition temperature, gas pressure and gas flow ratio.

6. The method for optimizing process parameters of hafnium zirconium oxide thin films based on Bayesian optimization according to claim 1, characterized in that: The steps of constructing a proxy model, defining an acquisition function, iteratively optimizing, and updating the proxy model are performed in sequence, wherein the acquisition function is selected from any one of expected improvement, probability improvement, and upper confidence bound, and the acquisition function is determined based on a predefined exploration and utilization balance standard.

7. The method for optimizing process parameters of hafnium zirconium oxide thin films based on Bayesian optimization according to claim 1, characterized in that: The above steps are repeated until a termination condition is met, wherein the termination condition is any one of reaching a predetermined number of iterations, an improvement amount being less than a certain threshold, and reaching a preset performance indicator.

8. The method for optimizing process parameters of hafnium zirconium oxide thin films based on Bayesian optimization according to claim 1, characterized in that: The method of determining the optimal process parameter combination for preparing the hafnium zirconium oxide film based on the results of Bayesian optimization includes: measuring the electrical properties of the film prepared using the process parameter results of Bayesian optimization, feeding the measurement results back to the Bayesian optimization step, and using statistical analysis to evaluate the robustness of the optimal process parameter combination.

9. A hafnium zirconium oxide thin film process parameter optimization system based on Bayesian optimization, characterized in that: include: A hafnium zirconium oxide thin film preparation device, used for preparing a hafnium zirconium oxide thin film; A data set construction module, used to obtain the performance of the hafnium zirconium oxide film and construct a data set of process parameter conditions and corresponding performance of the hafnium zirconium oxide film; A Bayesian optimization module, including an agent model building unit, a collection function definition unit, a collection function optimization unit, and a model updating unit; The parameter determination module is used to determine the optimal process parameter combination for preparing the hafnium zirconium oxide film according to the result of the Bayesian optimization module.

10. The process parameter optimization system for hafnium zirconium oxide thin films based on Bayesian optimization according to claim 9, characterized in that: Also includes an evaluation module and / or a display module, wherein the evaluation module is used to measure the electrical characteristics of the film prepared using the process parameter results of the Bayesian optimization, and feed back the measurement results to the Bayesian optimization module; The display module serves as an input interface for the process parameters of the hafnium zirconium oxide thin film preparation, and is used to display the Bayesian optimization results and the determined optimal process parameter combination.

Images