A low-damage double-sided ion beam etching method for high-frequency quartz crystal resonators

Abstract

The present invention relates to the field of quartz crystal etching technology, and discloses a low-damage double-sided ion beam etching method for a high-frequency quartz crystal resonator, comprising: based on a real-time feedback control mechanism, achieving precise control of the etching process of the high-frequency quartz crystal resonator. By starting the upper and lower symmetrical ion sources in a vacuum chamber, a double-sided etching field is formed to ensure balanced etching of both sides of the wafer. The system obtains the wafer thickness and the working resonant frequency in real time, and compares them with the preset frequency to obtain the frequency offset. If the offset exceeds the threshold, the energy and gas correction factors are determined according to the offset amplitude, and the ion beam energy and gas ratio are dynamically adjusted to achieve the linkage optimization of etching intensity and damage control; if the offset is lower than the threshold, etching is terminated. The present invention realizes the transformation of the etching process from "fixed parameter control" to "adaptive feedback control" through embedded real-time perception and parameter linkage optimization, thereby improving the intelligence and precision of the process.

Description

Technical Field

[0001] The present invention relates to the technical field of quartz crystal etching, and in particular to a low-damage double-sided ion beam etching method for a high-frequency quartz crystal resonator. Background Art

[0002] Quartz crystal resonators are widely used in communications, navigation, radar, and high-precision clock systems due to their high frequency stability and excellent quality factor. To achieve the target resonant frequency, quartz wafers typically undergo high-precision thinning during processing. Ion beam etching, with its advantages of high purity, low contamination, and strong etching directionality, has become an important method for achieving micron-level thickness control. It is particularly suitable for microstructural adjustment and frequency precision control of high-frequency crystal resonators.

[0003] However, existing ion beam etching methods still have many technical bottlenecks, such as uncontrollable thermal effects during the etching process, uneven beam distribution, and the easy superposition of surface damage and residual stress. These factors will directly affect the frequency stability of quartz crystals and the consistency of finished products. Especially in the manufacture of high-frequency devices, the thickness of the quartz wafer is highly correlated with the resonant frequency, and any slight deviation will cause frequency drift. Because traditional processes lack real-time feedback and dynamic adjustment mechanisms for the etching process, frequency control usually relies on preset parameters or human intervention, resulting in limited etching depth accuracy, unstable device performance, and difficulty in meeting the manufacturing requirements of high consistency and high yield.

[0004] Therefore, there is an urgent need to propose a low-damage etching method with real-time resonant frequency detection function. The existing ion beam etching method lacks real-time feedback and dynamic adjustment mechanism, resulting in poor frequency control accuracy, easy surface damage, and poor performance consistency of quartz crystal resonators during the etching process, making it difficult to meet the manufacturing requirements of high-frequency devices. Summary of the Invention

[0005] In view of this, the present invention proposes a low-damage double-sided ion beam etching method for high-frequency quartz crystal resonators, aiming to solve the problem that the ion beam etching method in the current technology lacks real-time feedback and dynamic adjustment mechanism, resulting in poor frequency control accuracy, easy surface damage, poor performance consistency of the quartz crystal resonator during the etching process, and difficulty in meeting the manufacturing requirements of high-frequency devices.

[0006] The present invention provides a low-damage double-sided ion beam etching method for a high-frequency quartz crystal resonator, comprising:

[0007] A quartz wafer is placed on a rotating stage in a vacuum chamber, and symmetrically arranged upper and lower ion sources are activated to form a double-sided etching field;

[0008] Acquire the real-time quartz crystal thickness and the operating resonant frequency of the quartz crystal resonator, and determine the frequency offset based on the relationship between the operating resonant frequency of the quartz crystal resonator and the configured preset operating resonant frequency;

[0009] Initialize the etching parameters and determine the etching operation of the quartz wafer based on the relationship between the frequency offset and the configured preset frequency offset threshold, where:

[0010] If the frequency offset exceeds the preset frequency offset threshold, the output energy correction factor and the output gas correction factor are determined according to the relationship between the frequency offset and the configured preset frequency offset threshold, and the etching parameters are adjusted according to the output energy correction factor and the output gas correction factor;

[0011] If the frequency offset is lower than a preset frequency offset threshold, the etching operation is terminated according to the determination.

[0012] Furthermore, determining the output energy correction factor and the output gas correction factor based on the relationship between the frequency offset and the configured preset frequency offset threshold includes:

[0013] Determining an etching intensity level based on the amplitude of the frequency offset and a configured preset amplitude;

[0014] The pre-configured damage control model is called according to the etching intensity level, and the energy correction factor for adjusting the ion beam energy and the gas correction factor for adjusting the reaction gas mixing ratio are output.

[0015] Furthermore, determining the etching intensity level based on the amplitude of the frequency offset and the configured preset amplitude includes:

[0016] The etching intensity level is determined according to the relationship between the amplitude of the frequency offset and the configured first preset amplitude and second preset amplitude:

[0017] When the amplitude of the frequency offset is lower than or equal to the first preset amplitude, the etching intensity level is determined to be L1;

[0018] When the amplitude of the frequency offset is higher than the first preset amplitude and the amplitude of the frequency offset is lower than or equal to the second preset amplitude, the etching intensity level is determined to be L2;

[0019] When the amplitude of the frequency offset is higher than the second preset amplitude, the etching intensity level is determined to be L3;

[0020] Wherein, 50 ppm is smaller than the first preset amplitude and smaller than the second preset amplitude, and L1<L2<L3.

[0021] Furthermore, when a pre-configured damage control model is called according to the etching intensity level and an energy correction factor for adjusting the ion beam energy and a gas correction factor for adjusting the reaction gas mixing ratio are output, the following steps are included:

[0022] Training a pre-configured damage control model, wherein the damage control model includes a multi-layer perceptron neural network model, a multi-layer convolutional neural network model, and a residual network model;

[0023] According to the etching intensity level, the trained damage control model is determined to be called;

[0024] A thickness deviation between the real-time quartz wafer thickness and the initial thickness of the quartz wafer is obtained, and an output energy correction factor and a gas correction factor are determined according to a relationship between the thickness deviation and a trained damage control model.

[0025] Furthermore, according to the etching intensity level, the call of the trained damage control model is determined, including:

[0026] When the etching intensity level is L1, the trained damage control model is determined to be a multi-layer perceptron neural network model;

[0027] When the etching intensity level is L2, it is determined that the trained damage control model is a multi-layer convolutional neural network model;

[0028] When the etching intensity level is L3, it is determined that the trained damage control model is called as the residual network model.

[0029] Furthermore, based on the relationship between the thickness deviation and the trained damage control model, the output energy correction factor and the gas correction factor are determined, including:

[0030] Determining a lattice damage weight coefficient within a preset value range of the lattice damage weight coefficient according to the trained damage control model;

[0031] Obtaining a ratio result between the thickness deviation and a configured preset target thickness, and determining a gas correction factor based on a relationship between the ratio result and a lattice damage weight coefficient;

[0032] The energy correction factor is determined based on the relationship between the complementary constraint relationship and the gas correction factor.

[0033] Furthermore, when training a pre-configured damage control model, it includes:

[0034] Obtain a historical etching data set including frequency offset, thickness error, and corresponding lattice damage indicators, and perform standardization on the historical etching data set;

[0035] The standardized frequency offset and thickness error are used as input, and the lattice damage index is used as the reference output to construct and train a multi-layer perceptron neural network model, a multi-layer convolutional neural network model, and a residual network model.

[0036] A grid search method is used to optimize the hyperparameters of the multilayer perceptron neural network model, the multilayer convolutional neural network model, and the residual network model. K-means clustering is performed on the training dataset to enhance the training sample structure.

[0037] Obtain the mean square error of the energy correction factor and gas correction factor of the predicted output in each model, and determine whether the training is complete based on the relationship between the mean square error and the configured preset mean square error, where:

[0038] If the mean square error is lower than the preset mean square error, the model training is determined to be complete.

[0039] Furthermore, when adjusting the etching parameters according to the output energy correction factor and the output gas correction factor, it includes:

[0040] Obtaining a proportional error between the frequency offset and a preset operating resonant frequency, and determining a preset ion beam energy based on a relationship between the proportional error and a correction factor for the output energy;

[0041] Obtaining a ratio between the thickness deviation and a preset target thickness, and correcting the oxygen volume fraction in the reaction gas based on a relationship between the ratio and an output gas correction factor;

[0042] determining an output voltage according to a relationship between a preset ion beam energy and a current ion beam energy;

[0043] The argon and oxygen mixing ratio is determined based on the adjusted oxygen volume ratio and the configured preset oxygen volume ratio.

[0044] Furthermore, determining the output voltage based on the relationship between the adjusted ion beam energy and the configured preset ion beam energy includes:

[0045] Obtain an energy difference between the adjusted ion beam energy and the configured preset ion beam energy, determine an adjustment coefficient based on a relationship between the energy difference and the configured first preset energy difference and the second preset energy difference, and determine an output voltage to be output based on the output voltage adjusted according to the adjustment coefficient:

[0046] When the energy difference is lower than the first preset energy difference, the adjustment coefficient is determined to be M1;

[0047] When the energy difference is higher than or equal to the first preset energy difference and lower than the second preset energy difference, the adjustment coefficient is determined to be M2;

[0048] When the energy difference is higher than or equal to the second preset energy difference, the adjustment coefficient is determined to be M3;

[0049] The first preset energy difference is smaller than the second preset energy difference, and M1<M2<M3<1.

[0050] Furthermore, the argon-oxygen mixing ratio is determined based on the adjusted oxygen volume ratio and the configured preset oxygen volume ratio, including:

[0051] Obtain a ratio deviation value between the adjusted oxygen volume ratio and the configured preset oxygen volume ratio, and determine a correction coefficient based on a relationship between the ratio deviation value and the configured first preset ratio deviation value and second preset ratio deviation value, and determine the argon and oxygen mixing ratio corrected according to the correction coefficient as the argon and oxygen mixing ratio to be output:

[0052] When the ratio deviation value is lower than the first preset ratio deviation value, the correction coefficient is determined to be N1;

[0053] When the ratio deviation value is higher than or equal to the first preset ratio deviation value and the ratio deviation value is lower than the second preset ratio deviation value, the correction coefficient is determined to be N2;

[0054] When the ratio deviation value is higher than or equal to the second preset ratio deviation value, the correction coefficient is determined to be N3;

[0055] The first preset ratio deviation value is smaller than the second preset ratio deviation value, and N1<N2<N3<1.

[0056] Compared with the prior art, the beneficial effect of the present invention is that by introducing a dual real-time monitoring mechanism of wafer thickness and resonant frequency during the ion beam etching process, the physical changes of the quartz crystal resonator during the processing process, especially the dynamic response characteristics of the frequency offset, can be accurately obtained. By comparing the real-time measured working resonant frequency with the preset target frequency, calculating the frequency offset, and judging the etching state based on the degree of offset, the controllability of the etching process is effectively improved, avoiding the frequency drift problem caused by traditional reliance on time setting or single thickness parameter control. Furthermore, the present invention combines the frequency offset to intelligently output energy correction factors and gas correction factors for dynamically adjusting ion beam etching parameters such as energy, current density, and oxygen content in the reaction gas, so that the etching process can be automatically compensated and optimized based on real-time feedback. This parameter control mechanism not only effectively reduces the physical damage and thermal stress accumulation of the ion beam on the wafer surface, but also ensures the balance between etching rate and etching uniformity, significantly improving the accuracy and stability of the frequency control of the quartz crystal resonator. Finally, by setting a preset frequency offset threshold, etching is automatically terminated when the frequency offset falls within an acceptable range, preventing performance degradation caused by over-etching and reducing crystal material waste and subsequent repair rates. Overall, this invention, by constructing a closed-loop control method for ion beam etching based on frequency feedback drive, achieves precise thinning, low-damage processing, and high-frequency stable output during the manufacturing process of quartz crystal resonators, demonstrating high practical value and industrial prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0058] Figure 1 A flowchart of a low-damage double-sided ion beam etching method for a high-frequency quartz crystal resonator provided by an embodiment of the present invention;

[0059] Figure 2 A schematic flow chart of a low-damage double-sided ion beam etching method for a high-frequency quartz crystal resonator provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0060] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, unless there is a conflict, the embodiments of the present disclosure and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0061] like Figure 1-Figure 2 As shown, in some embodiments of the present application, this embodiment provides a low-damage double-sided ion beam etching method for a high-frequency quartz crystal resonator, comprising:

[0062] Step S100: placing a quartz wafer on a rotating stage in a vacuum chamber, and starting symmetrically arranged upper and lower ion sources to form a double-sided etching field.

[0063] It's understandable that placing the quartz wafer in a vacuum chamber and mounting it on a rotating stage ensures high-precision ion beam etching in a highly pure, gas-free environment. The vacuum chamber effectively reduces scattering and attenuation of the ion beam path by air molecules, thereby improving the stability and controllability of the etching process. The rotating stage design ensures constant rotation of the wafer during etching. This structure helps even out the ion beam's impact on the wafer surface, thereby avoiding variations in etching depth caused by uneven beam distribution. The rotational motion continuously changes the beam angle at the wafer, further balancing energy distribution and significantly improving double-sided etching uniformity, reducing the occurrence of localized over- or under-etching. Furthermore, the key to this technology lies in the formation of a double-sided etching field through the symmetrical arrangement of ion sources. Conventional single-sided etching often leads to problems such as uneven wafer heating and stress accumulation on one side. Simultaneous double-sided etching not only improves processing efficiency but also significantly reduces crystal warpage and deformation caused by thickness variations or stress gradients. The upper and lower ion sources can regulate the beam parameters respectively to achieve symmetrical control of the etching behavior on the front and back sides, thereby ensuring the symmetry of the resonator morphology and the stability of the mechanical properties at the microscopic scale.

[0064] It can be seen that through the coordinated configuration of vacuum environment, rotating platform and symmetrical ion source, double-sided ion beam etching of quartz wafers with high uniformity, low stress and low damage is achieved, laying a physical foundation for subsequent precise frequency control and stable device performance.

[0065] Step S200: Acquire the real-time quartz crystal thickness and the operating resonant frequency of the quartz crystal resonator, and determine the frequency offset according to the relationship between the operating resonant frequency of the quartz crystal resonator and the configured preset operating resonant frequency.

[0066] Understandably, controlling the resonant frequency is crucial in the high-precision manufacturing of quartz crystal resonators. The thickness of a quartz crystal is inversely proportional to its resonant frequency: thinner wafers correspond to higher resonant frequencies. Therefore, precisely controlling the etching depth of the quartz wafer is crucial for ensuring frequency stability and consistent device performance. However, because the material removal rate during ion beam etching is susceptible to factors such as energy fluctuations and beam density variations, initial parameter settings alone cannot effectively guarantee etching consistency; dynamic adjustment requires a real-time feedback mechanism. A dual feedback monitoring mechanism is introduced to monitor both the thickness of the quartz wafer and its operating resonant frequency in real time. Specifically, thickness measurement can be accomplished using high-resolution, non-contact sensors such as laser interferometers or white light interferometry, typically achieving submicron accuracy. This information directly reflects the absolute amount of material removed, facilitating dynamic management of etching rates and total etching depth. Secondly, the operating resonant frequency is determined by exciting the quartz wafer with an external circuit and acquiring its response spectrum. Unlike thickness data, frequency information better reflects the actual impact of structural and material changes caused by etching on device performance. Especially in high-frequency devices (e.g., in the MHz range), even a slight frequency offset can cause the device to fail to meet design requirements. Therefore, by measuring the deviation between the current resonant frequency and the configured target frequency (i.e., the frequency offset), a dynamic assessment can be made of whether the current etching state is approaching the ideal state. Furthermore, the frequency offset serves as more than just a control signal; it also serves as a crucial criterion for determining whether to proceed with processing. A large offset indicates that the target thickness has not yet been reached, requiring further material removal. When the offset approaches the set threshold, the wafer is nearing the ideal state and can proceed to the refinement or termination phase to prevent performance degradation or structural damage caused by over-etching. Finally, through a coordinated damage control model, the frequency offset is used as an input feature, supporting intelligent algorithms in determining whether damage risks exist in the current etching process. This dual-modal, real-time sensing and feedback mechanism enables precise closed-loop control of the etching process, fundamentally improving the yield, stability, and consistency of high-frequency quartz crystal resonators.

[0067] It can be seen that by measuring the current operating resonant frequency in real time, comparing it with the preset frequency, and calculating the frequency offset between the two, it is possible to accurately reflect changes in wafer thickness and material state. This frequency offset serves as a key feedback signal, guiding the dynamic adjustment of the etching process, achieving precise control of etching depth and crystal performance, thereby ensuring that the resonator achieves the expected electrical performance and processing quality.

[0068] Step S300: Initialize etching parameters and determine the etching operation of the quartz wafer according to the relationship between the frequency offset and the configured preset frequency offset threshold.

[0069] Specifically, when determining the etching operation of the quartz wafer based on the relationship between the frequency offset and the configured preset frequency offset threshold, the process includes: if the frequency offset exceeds the preset frequency offset threshold, determining the output energy correction factor and the output gas correction factor based on the relationship between the frequency offset and the configured preset frequency offset threshold, and adjusting the etching parameters based on the output energy correction factor and the output gas correction factor; if the frequency offset is lower than the preset frequency offset threshold, terminating the etching operation based on the determination.

[0070] It is understood that by initializing key parameters during the etching process, including basic process conditions such as ion beam energy, beam current density, and gas composition ratio, these initial parameters provide a stable starting environment for etching, ensuring effective etching. Secondly, as etching progresses, the relationship between the frequency offset and the preset frequency offset threshold is monitored in real time to dynamically determine the next steps in the etching operation. If the frequency offset exceeds the preset threshold, it indicates that the wafer has not yet reached the ideal thickness or that the etching level still requires further adjustment. In this case, the output energy correction factor and the output gas correction factor are calculated based on the magnitude of the frequency offset. These correction factors map the quantitative relationship between the frequency offset and the etching parameters through a mathematical model, enabling fine-tuning of etching conditions to promote the etching process closer to the target state. Specifically, the output energy correction factor adjusts the ion beam energy, affecting the etching rate and uniformity; while the output gas correction factor regulates the proportion of the reactant gas (such as oxygen) in the gas mixture, affecting the chemical reactivity and surface damage during etching. This coordinated control mechanism effectively reduces wafer surface damage and internal stress accumulation, improving the device's quality factor and frequency stability. Conversely, when the frequency offset drops below a preset threshold, indicating that the etching has reached or is close to the design target, the etching operation is automatically terminated to prevent over-etching that could cause performance degradation or structural damage. This adaptive control strategy not only enhances the intelligence of the etching process but also significantly improves the production yield and product consistency of high-frequency quartz crystal resonators.

[0071] Specifically, when determining the output energy correction factor and the output gas correction factor based on the relationship between the frequency offset and the configured preset frequency offset threshold, it includes: determining the etching intensity level based on the amplitude of the frequency offset and the configured preset amplitude; calling the pre-configured damage control model according to the etching intensity level, and outputting the energy correction factor for adjusting the ion beam energy and the gas correction factor for adjusting the reaction gas mixing ratio.

[0072] Specifically, when determining the etching intensity level based on the amplitude of the frequency offset and the configured preset amplitude, it includes: determining the etching intensity level according to the relationship between the amplitude of the frequency offset and the configured first preset amplitude and second preset amplitude: when the amplitude of the frequency offset is lower than or equal to the first preset amplitude, the etching intensity level is determined to be L1; when the amplitude of the frequency offset is higher than the first preset amplitude and the amplitude of the frequency offset is lower than or equal to the second preset amplitude, the etching intensity level is determined to be L2; when the amplitude of the frequency offset is higher than the second preset amplitude, the etching intensity level is determined to be L3; wherein, 50ppm is less than the first preset amplitude and less than the second preset amplitude, and L1<L2<L3.

[0073] Specifically, when calling a pre-configured damage control model according to the etching intensity level and outputting an energy correction factor for adjusting the ion beam energy and a gas correction factor for adjusting the reaction gas mixing ratio, the method includes: training the pre-configured damage control model, wherein the damage control model includes a multi-layer perceptron neural network model, a multi-layer convolutional neural network model and a residual network model; determining to call the trained damage control model according to the etching intensity level; obtaining the thickness deviation between the real-time quartz wafer thickness and the initial thickness of the quartz wafer, and determining the output energy correction factor and the gas correction factor according to the relationship between the thickness deviation and the trained damage control model.

[0074] Specifically, according to the etching intensity level, the call of the trained damage control model is determined, including: when the etching intensity level is L1, the call of the trained damage control model is determined to be a multi-layer perceptron neural network model; when the etching intensity level is L2, the call of the trained damage control model is determined to be a multi-layer convolutional neural network model; when the etching intensity level is L3, the call of the trained damage control model is determined to be a residual network model.

[0075] Specifically, when determining the output energy correction factor and the gas correction factor based on the relationship between the thickness deviation and the trained damage control model, it includes: determining the lattice damage weight coefficient within the preset value range of the lattice damage weight coefficient based on the trained damage control model; obtaining the proportional result between the thickness deviation and the configured preset target thickness, and determining the gas correction factor based on the relationship between the proportional result and the lattice damage weight coefficient; determining the energy correction factor based on the relationship between the complementary constraint relationship and the gas correction factor.

[0076] As can be understood, the etching intensity grading system is established by real-time monitoring of the operating resonant frequency of the quartz crystal resonator, calculating the frequency offset, and comparing this offset with a pre-set threshold. Specifically, when the frequency offset is below a first preset amplitude, it indicates that the etching process is in a relatively mild stage, with minimal impact on the wafer, and is classified as the lowest etching intensity level, L1. When the frequency offset is between the first and second preset amplitudes, it is classified as the medium etching intensity level, L2, indicating that the etching process is gradually intensifying. When the frequency offset exceeds the second preset amplitude, it indicates that the etching process is more intense, with a greater impact on the wafer, and is classified as the highest etching intensity level, L3. This grading method makes the etching process more targeted, dynamically adjusting the etching intensity based on the wafer state to prevent over-etching and wafer damage. To achieve precise control, this technical solution incorporates deep learning technology to invoke different damage control models for different etching intensity levels. Specifically, these include three architectures: a multi-layer perceptron (MLP) neural network, a multi-layer convolutional neural network (CNN), and a residual network (ResNet). Each model is trained with extensive experimental data and is capable of capturing the complex, nonlinear relationship between wafer thickness variation and material damage under various etching conditions. By employing these models in a hierarchical manner, they more accurately match etching intensity with damage severity, improving model applicability and predictive accuracy. In practical applications, the model input data is the real-time measured wafer thickness deviation—the absolute value of the difference between the current thickness and the target thickness. This thickness deviation is a key indicator of etching effectiveness, directly reflecting the extent of material removal and etching uniformity. Based on this input data and the lattice damage characteristics learned during training, the model calculates two key adjustment parameters: an energy correction factor and a gas correction factor. The energy correction factor adjusts the ion beam energy intensity to ensure a moderate etching intensity. The gas correction factor adjusts the volume fraction of oxygen in the reaction gas, influencing the etching chemical reaction rate and surface conditions, thereby assisting in achieving low-damage etching. The gas correction factor is determined by the proportional relationship between the thickness deviation and the preset target thickness, combined with a lattice damage weighting factor. This weight coefficient reflects the material's sensitivity to damage under different etching intensities. It is usually obtained through feature layer extraction of a deep learning model and is constrained within a certain range (such as 0.2 to 0.5). Reasonable setting of the weight coefficient can make the gas ratio adjustment more flexible and scientific, avoid excessive or insufficient oxidation during the etching process, and maintain the stability of the surface structure. Finally, the energy correction factor and the gas correction factor satisfy a complementary constraint relationship, that is, the sum of the two is a constant, ensuring that the total etching adjustment energy is constant. This constraint design coordinates the adjustment parameters, which can not only ensure etching efficiency, but also effectively reduce thermal damage and mechanical stress. In addition, the adjustment mechanism realizes dynamic closed-loop control of etching parameters, responds to changes in chip status in real time, and greatly improves the stability of the etching process and the consistency of the product.

[0077] It can be seen that by establishing etching intensity levels based on the amplitude of the frequency offset and a preset threshold, hierarchical management of the etching process is effectively achieved, making the etching operation more precise and targeted. Different intensity levels correspond to different damage control models, and the most appropriate model can be dynamically selected and adjusted according to the etching intensity, improving the accuracy and applicability of model predictions, thereby better adapting to complex etching environments and diverse material conditions. Secondly, by training a multi-layer perceptron neural network, a multi-layer convolutional neural network, and a residual network model, the system can deeply explore the nonlinear relationships and material response characteristics during the etching process, achieving accurate prediction of wafer thickness variation and damage degree. Based on these predictions, dynamically calculated energy correction factors and gas correction factors can precisely control the ion beam energy and the reaction gas mixture ratio, effectively reducing thermal damage and mechanical stress during the etching process, and improving etching uniformity and quality stability. Finally, the gas correction factor is calculated using the proportional relationship between the lattice damage weight coefficient and thickness deviation, and the energy correction factor is determined through a complementary constraint relationship, making the etching parameter adjustment more scientific and reasonable, and ensuring the balance of overall etching energy. This closed-loop control mechanism not only improves the accuracy and repeatability of etching, but also significantly extends the service life and performance consistency of quartz crystal resonators, promoting the optimization and industrialization of high-frequency resonator manufacturing processes.

[0078] Specifically, when training a pre-configured damage control model, it includes: obtaining a historical etching data set including frequency offset, thickness error and corresponding lattice damage index, and standardizing the historical etching data set; using the standardized frequency offset and thickness error as input and the lattice damage index as reference output, constructing and training a multi-layer perceptron neural network model, a multi-layer convolutional neural network model and a residual network model; using a grid search method to optimize the hyperparameters in the multi-layer perceptron neural network model, the multi-layer convolutional neural network model and the residual network model, wherein K-means clustering is performed based on the training data set to enhance the training sample structure; obtaining the mean square error of the energy correction factor and the gas correction factor of the predicted output in each model, and determining whether the training is completed based on the relationship between the mean square error and the configured preset mean square error, wherein: if the mean square error is lower than the preset mean square error, then it is determined that the model training is completed.

[0079] As can be understood, a complete training dataset is formed by collecting historical etching data containing frequency offset, thickness error, and corresponding lattice damage indicators. The data is then normalized to eliminate dimensional differences and uneven data distribution, ensuring the stability and accuracy of the training process. Subsequently, three deep learning models with different structures—a multilayer perceptron neural network, a multilayer convolutional neural network, and a residual network—are used for modeling and training, using the standardized frequency offset and thickness error as input and the lattice damage indicators as reference output. These models are able to fully capture the complex nonlinear relationships in the etching process and the damage characteristics of crystal materials, improving prediction accuracy and generalization. To further optimize model performance, a grid search method is used to tune the model hyperparameters to ensure the optimal parameter combination. K-means clustering is also used to enhance the training data. Clustering and grouping enhances sample structure diversity, avoids model overfitting, and improves the model's adaptability to data under different etching conditions. During training, the mean squared error (MSE) of the model-predicted energy correction factor and gas correction factor is calculated and compared with a preset error threshold to determine the training completion status. Only when the mean square error of the model is lower than the preset threshold is the model training considered to have achieved the expected accuracy, ensuring that the trained damage control model has high accuracy and reliability, and can provide a scientific basis for the dynamic adjustment of etching parameters.

[0080] It can be seen that by constructing and training multiple deep learning models, the prediction accuracy and generalization ability of the damage control model were effectively improved. Standardizing historical etching data eliminated dimensional differences in the data, making model training more stable and reliable, and avoiding training bias caused by uneven data distribution. The multi-model fusion strategy of a multi-layer perceptron, a convolutional neural network, and a residual network enhanced the ability to characterize the complex nonlinear relationships in the etching process, more accurately reflecting the impact of frequency offset and thickness error on lattice damage, thus providing a scientific basis for precise adjustment of etching parameters. Secondly, grid search was used to optimize model hyperparameters, and K-means clustering was combined to enhance the training data structure. This not only improved model training efficiency but also enhanced the model's adaptability to diverse data under different etching conditions, avoiding overfitting and improving model robustness. Finally, a rigorous judgment mechanism based on mean squared error ensured the quality and accuracy of the trained model, ensuring that the damage control model can stably and efficiently support dynamic adjustment of the etching process in practical applications, effectively reducing wafer surface damage and improving the performance consistency and yield of quartz crystal resonators.

[0081] Specifically, when adjusting the etching parameters according to the output energy correction factor and the output gas correction factor, it includes: obtaining the proportional error between the frequency offset and the preset working resonant frequency, and determining the preset ion beam energy according to the relationship between the proportional error and the output energy correction factor; obtaining the proportional result between the thickness deviation and the preset target thickness, and correcting the oxygen volume proportion in the reaction gas according to the relationship between the proportional result and the output gas correction factor; determining the output voltage according to the relationship between the preset ion beam energy and the current ion beam energy; and determining the mixing ratio of argon and oxygen according to the adjusted oxygen volume proportion and the configured preset oxygen volume proportion.

[0082] Specifically, when determining the output voltage based on the relationship between the adjusted ion beam energy and the configured preset ion beam energy, it includes: obtaining the energy difference between the adjusted ion beam energy and the configured preset ion beam energy, and determining the adjustment coefficient based on the relationship between the energy difference and the configured first preset energy difference and second preset energy difference, and determining the output voltage to be output based on the output voltage adjusted by the adjustment coefficient: when the energy difference is lower than the first preset energy difference, the adjustment coefficient is determined to be M1; when the energy difference is higher than or equal to the first preset energy difference, and the energy difference is lower than the second preset energy difference, the adjustment coefficient is determined to be M2; when the energy difference is higher than or equal to the second preset energy difference, the adjustment coefficient is determined to be M3; wherein the first preset energy difference is less than the second preset energy difference, and M1<M2<M3<1.

[0083] Specifically, when determining the argon-oxygen mixing ratio based on the adjusted oxygen volume ratio and the configured preset oxygen volume ratio, the method includes: obtaining a ratio deviation value between the adjusted oxygen volume ratio and the configured preset oxygen volume ratio, and determining a correction coefficient based on a relationship between the ratio deviation value and the configured first preset ratio deviation value and second preset ratio deviation value, and determining the argon-oxygen mixing ratio corrected according to the correction coefficient as the argon-oxygen mixing ratio to be output: when the ratio deviation value is lower than the first preset ratio deviation value, the correction coefficient is determined to be N1; when the ratio deviation value is higher than or equal to the first preset ratio deviation value and the ratio deviation value is lower than the second preset ratio deviation value, the correction coefficient is determined to be N2; when the ratio deviation value is higher than or equal to the second preset ratio deviation value, the correction coefficient is determined to be N3; wherein the first preset ratio deviation value is less than the second preset ratio deviation value, and N1<N2<N3<1.

[0084] As can be understood, the proportional error is calculated by real-time acquisition of the frequency offset between the operating resonant frequency of the quartz crystal resonator and the preset target frequency. This error information is combined with a predetermined output energy correction factor to scientifically determine the preset ion beam energy. This feedback-based energy adjustment ensures that the ion beam energy level can respond promptly to actual changes in the resonant frequency, effectively controlling the etching rate and depth, thereby ensuring that the thickness of the quartz wafer closely matches the design requirements and maintaining the stability of the resonant frequency. Secondly, by utilizing the proportional relationship between the thickness deviation and the target thickness, combined with the output gas correction factor, the oxygen volume fraction in the reaction gas is dynamically adjusted. Precise regulation of oxygen concentration plays a key role in chemical reactivity and etch selectivity during the etching process. By controlling the oxygen volume fraction, the etching environment can be optimized, reducing the risk of thermal damage to the wafer surface, while also reducing the accumulation of residual stress caused by excessive chemical reactions, significantly improving wafer surface quality and etching uniformity. For the specific adjustment of ion beam energy, the technical solution has designed a multi-level energy difference threshold mechanism. According to the energy difference between the adjusted ion beam energy and the preset ion beam energy, the output voltage is corrected using graded adjustment coefficients M1, M2, and M3. Each adjustment coefficient corresponds to a different energy difference interval, ensuring that the ion beam energy can transition smoothly and adjust gradually, avoiding excessive energy changes that lead to instability or chip damage during the etching process. Through this meticulous graded adjustment mechanism, the balance between response speed and etching quality is taken into account. At the same time, the adjustment of the oxygen volume fraction also adopts graded correction coefficients N1, N2, and N3, and dynamically adjusts the mixing ratio of argon and oxygen according to the degree of deviation of the adjusted oxygen volume fraction from the preset value. This ensures the stability of the gas mixing ratio, ensures the continuity and consistency of the chemical reaction during the etching process, and effectively avoids uneven etching or performance degradation caused by fluctuations in gas composition.

[0085] It can be seen that through multi-dimensional dynamic feedback control, the use of graded correction factors and a segmented adjustment strategy achieves precise optimization of ion beam energy and reaction gas composition. This method not only improves etching accuracy and uniformity, but also effectively reduces thermal effects and the risk of microscopic damage, significantly improving the quality stability and yield rate of high-frequency quartz crystal resonators. Through these meticulous regulatory measures, this invention provides reliable technical support for the manufacturing process of high-precision quartz crystal resonators.

[0086] In the above embodiment, by introducing a dual real-time monitoring mechanism of wafer thickness and resonant frequency during the ion beam etching process, the physical changes of the quartz crystal resonator during the processing process, especially the dynamic response characteristics of the frequency offset, can be accurately obtained. By comparing the real-time measured working resonant frequency with the preset target frequency, calculating the frequency offset, and judging the etching state based on the degree of offset, the controllability of the etching process is effectively improved, avoiding the frequency drift problem caused by traditional reliance on time setting or single thickness parameter control. Furthermore, the present invention combines the frequency offset to intelligently output energy correction factors and gas correction factors for dynamically adjusting ion beam etching parameters such as energy, current density, and oxygen content in the reaction gas, so that the etching process can be automatically compensated and optimized based on real-time feedback. This parameter control mechanism not only effectively reduces the physical damage and thermal stress accumulation of the ion beam on the wafer surface, but also ensures the balance between etching rate and etching uniformity, significantly improving the accuracy and stability of the quartz crystal resonator frequency control. Finally, by setting a preset frequency offset threshold, etching is automatically terminated when the frequency offset falls within an acceptable range, preventing performance degradation caused by over-etching and reducing crystal material waste and subsequent repair rates. Overall, this invention, by constructing a closed-loop control method for ion beam etching based on frequency feedback drive, achieves precise thinning, low-damage processing, and high-frequency stable output during the manufacturing process of quartz crystal resonators, demonstrating high practical value and industrial prospects.

[0087] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0088] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0089] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0090] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A low-damage double-sided ion beam etching method for a high-frequency quartz crystal resonator, characterized in that: include: A quartz wafer is placed on a rotating stage in a vacuum chamber, and symmetrically arranged upper and lower ion sources are activated to form a double-sided etching field; Acquire the real-time quartz crystal thickness and the operating resonant frequency of the quartz crystal resonator, and determine the frequency offset based on the relationship between the operating resonant frequency of the quartz crystal resonator and the configured preset operating resonant frequency; Initialize the etching parameters and determine the etching operation of the quartz wafer based on the relationship between the frequency offset and the configured preset frequency offset threshold, where: If the frequency offset exceeds the preset frequency offset threshold, the output energy correction factor and the output gas correction factor are determined according to the relationship between the frequency offset and the configured preset frequency offset threshold, and the etching parameters are adjusted according to the output energy correction factor and the output gas correction factor; If the frequency offset is lower than a preset frequency offset threshold, the etching operation is terminated according to the determination; Based on the relationship between the frequency offset and the configured preset frequency offset threshold, the output energy correction factor and the output gas correction factor are determined, including: Determining an etching intensity level based on the amplitude of the frequency offset and a configured preset amplitude; The pre-configured damage control model is called according to the etching intensity level, and the energy correction factor for adjusting the ion beam energy and the gas correction factor for adjusting the reaction gas mixing ratio are output.

2. The low-damage double-sided ion beam etching method for a high-frequency quartz crystal resonator according to claim 1, wherein: When determining the etching intensity level based on the amplitude of the frequency offset and the configured preset amplitude, it includes: The etching intensity level is determined according to the relationship between the amplitude of the frequency offset and the configured first preset amplitude and second preset amplitude: When the amplitude of the frequency offset is lower than or equal to the first preset amplitude, the etching intensity level is determined to be L1; When the amplitude of the frequency offset is higher than the first preset amplitude and the amplitude of the frequency offset is lower than or equal to the second preset amplitude, the etching intensity level is determined to be L2; When the amplitude of the frequency offset is higher than the second preset amplitude, the etching intensity level is determined to be L3; Wherein, 50 ppm is smaller than the first preset amplitude and smaller than the second preset amplitude, and L1<L2<L3.

3. The low-damage double-sided ion beam etching method for a high-frequency quartz crystal resonator according to claim 2, wherein: When calling a pre-configured damage control model based on the etching intensity level, the output includes the energy correction factor for adjusting the ion beam energy and the gas correction factor for adjusting the reaction gas mixture ratio, including: Training a pre-configured damage control model, wherein the damage control model includes a multi-layer perceptron neural network model, a multi-layer convolutional neural network model, and a residual network model; According to the etching intensity level, the trained damage control model is called; A thickness deviation between the real-time quartz wafer thickness and the initial thickness of the quartz wafer is obtained, and an output energy correction factor and a gas correction factor are determined according to a relationship between the thickness deviation and a trained damage control model.

4. The low-damage double-sided ion beam etching method for a high-frequency quartz crystal resonator according to claim 3, wherein: According to the etching intensity level, the call of the trained damage control model is determined, including: When the etching intensity level is L1, the trained damage control model is determined to be a multi-layer perceptron neural network model; When the etching intensity level is L2, it is determined that the trained damage control model is a multi-layer convolutional neural network model; When the etching intensity level is L3, it is determined that the trained damage control model is called as the residual network model.

5. The low-damage double-sided ion beam etching method for a high-frequency quartz crystal resonator according to claim 4, characterized in that: Based on the relationship between thickness deviation and the trained damage control model, the output energy correction factor and gas correction factor are determined, including: Determining a lattice damage weight coefficient within a preset value range of the lattice damage weight coefficient according to the trained damage control model; Obtaining a ratio result between the thickness deviation and a configured preset target thickness, and determining a gas correction factor based on a relationship between the ratio result and a lattice damage weight coefficient; The energy correction factor is determined based on the relationship between the complementary constraint relationship and the gas correction factor.

6. The low-damage double-sided ion beam etching method for a high-frequency quartz crystal resonator according to claim 5, characterized in that: When training a pre-configured damage control model, this includes: Obtain a historical etching data set including frequency offset, thickness error, and corresponding lattice damage indicators, and perform standardization on the historical etching data set; The standardized frequency offset and thickness error are used as input, and the lattice damage index is used as the reference output to construct and train a multi-layer perceptron neural network model, a multi-layer convolutional neural network model, and a residual network model. A grid search method is used to optimize the hyperparameters of the multilayer perceptron neural network model, the multilayer convolutional neural network model, and the residual network model. K-means clustering is performed on the training dataset to enhance the training sample structure. Obtain the mean square error of the energy correction factor and gas correction factor of the predicted output in each model, and determine whether the training is complete based on the relationship between the mean square error and the configured preset mean square error, where: If the mean square error is lower than the preset mean square error, the model training is determined to be complete.

7. The low-damage double-sided ion beam etching method for a high-frequency quartz crystal resonator according to claim 6, wherein: When adjusting the etching parameters according to the output energy correction factor and the output gas correction factor, it includes: Obtaining a proportional error between the frequency offset and a preset operating resonant frequency, and determining a preset ion beam energy based on a relationship between the proportional error and a correction factor for the output energy; Obtaining a ratio between the thickness deviation and a preset target thickness, and correcting the oxygen volume fraction in the reaction gas based on a relationship between the ratio and an output gas correction factor; determining an output voltage according to a relationship between a preset ion beam energy and a current ion beam energy; The argon and oxygen mixing ratio is determined based on the adjusted oxygen volume ratio and the configured preset oxygen volume ratio.

8. The low-damage double-sided ion beam etching method for a high-frequency quartz crystal resonator according to claim 7, wherein: When determining the output voltage based on the relationship between the adjusted ion beam energy and the configured preset ion beam energy, the following are included: Obtain an energy difference between the adjusted ion beam energy and the configured preset ion beam energy, determine an adjustment coefficient based on a relationship between the energy difference and the configured first preset energy difference and the second preset energy difference, and determine an output voltage to be output based on the output voltage adjusted according to the adjustment coefficient: When the energy difference is lower than the first preset energy difference, the adjustment coefficient is determined to be M1; When the energy difference is higher than or equal to the first preset energy difference and lower than the second preset energy difference, the adjustment coefficient is determined to be M2; When the energy difference is higher than or equal to the second preset energy difference, the adjustment coefficient is determined to be M3; The first preset energy difference is smaller than the second preset energy difference, and M1<M2<M3<1.

9. The low-damage double-sided ion beam etching method for a high-frequency quartz crystal resonator according to claim 7, wherein: When determining the argon and oxygen mixing ratio based on the adjusted oxygen volume ratio and the configured preset oxygen volume ratio, the following steps are included: Obtain a ratio deviation value between the adjusted oxygen volume ratio and the configured preset oxygen volume ratio, and determine a correction coefficient based on a relationship between the ratio deviation value and the configured first preset ratio deviation value and second preset ratio deviation value, and determine the argon and oxygen mixing ratio corrected according to the correction coefficient as the argon and oxygen mixing ratio to be output: When the ratio deviation value is lower than the first preset ratio deviation value, the correction coefficient is determined to be N1; When the ratio deviation value is higher than or equal to the first preset ratio deviation value and the ratio deviation value is lower than the second preset ratio deviation value, the correction coefficient is determined to be N2; When the ratio deviation value is higher than or equal to the second preset ratio deviation value, the correction coefficient is determined to be N3; The first preset ratio deviation value is smaller than the second preset ratio deviation value, and N1<N2<N3<1.

Images