Resonator-based etch endpoint detection circuit

Abstract

The present application provides an etching endpoint detection circuit based on a resonator, which relates to the field of semiconductor manufacturing. The analyzer receives the resonance frequency of the resonator detected by the detector and determines the initial value of the etching depth. At the same time, it receives the ambient temperature of the etching chamber collected by the temperature collector, extracts the timing characteristics, and performs interactive modeling with the initial value of the etching depth to obtain a dynamically calculated temperature adaptive compensation factor for adaptively correcting the temperature drift error and outputting a highly robust etching depth correction value. Then, the comparator compares the etching depth correction value with the preset value to form a closed-loop feedback and trigger an endpoint determination signal. In this way, through the deep cooperation between the hardware perception layer and the algorithm analysis layer, the limitations of the traditional fixed compensation mode are broken through, and the decoupling optimization of environmental interference and device characteristics is realized.

Description

Technical Field

[0001] This application relates to the field of semiconductor manufacturing, and more particularly, in the embodiments of this application, it relates to an etch endpoint detection circuit based on a resonator. Background Art

[0002] In the field of semiconductor manufacturing, etch endpoint detection is a key technology for achieving high-precision nanomanufacturing. Currently, the mainstream detection methods include the optical emission spectrometer (OES) based on material spectra and the laser interferometer (LI) based on laser interference. However, OES has significant drawbacks: it relies on the spectral signals generated by material changes during etching, resulting in the ability to determine the endpoint only when etching different material interfaces, and is accompanied by a large amount of etch by-products, affecting the process cleanliness. In addition, OES has insufficient detection sensitivity for small etch areas (such as nanoscale structures), making it difficult to meet the requirements of advanced processes. On the other hand, the LI technology is limited by the laser wavelength and is only applicable to the detection of thin films with a thickness greater than 200 nm, and requires accurate refractive index parameters of known materials. In actual processes, measurement errors in the refractive index will directly lead to deviations in the calculation of the etch depth.

[0003] A more severe challenge comes from the harsh environment of the etch chamber: factors such as strong electromagnetic fields, high vacuum, ion bombardment, chemical corrosion, and dynamic temperature changes will seriously interfere with the signal stability of the detection equipment. Especially the temperature drift problem, as the core sensing unit, the resonance frequency of the resonator is not only affected by the mass change caused by etching, but also closely related to the ambient temperature of the chamber. Traditional temperature compensation methods mostly use fixed coefficient correction, which is difficult to adapt to the non-linear time-series changes of temperature during etching, resulting in the accumulation of etch depth calculation errors.

[0004] Therefore, an optimized etch endpoint detection circuit based on a resonator is desired. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. The embodiments of this application provide an etch endpoint detection circuit based on a resonator. First, an analyzer receives the resonance frequency of the resonator detected by a detector and determines the initial value of the etch depth. At the same time, it receives the ambient temperature of the etch chamber collected by a temperature collector, extracts the time-series characteristics, and performs interactive modeling with the initial value of the etch depth to obtain a dynamically calculated temperature adaptive compensation factor to adaptively correct the temperature drift error and output a highly robust etch depth correction value. Then, a comparator compares the etch depth correction value with a preset value to form a closed-loop feedback and trigger an endpoint determination signal. In this way, through the deep cooperation between the hardware sensing layer and the algorithm analysis layer, the limitations of the traditional fixed compensation mode are broken through, and the decoupling optimization of environmental interference and device characteristics is achieved.

[0006] According to one aspect of the present application, a resonator-based etch endpoint detection circuit is provided, which includes:

[0007] A detector, connected to the resonator, for detecting and outputting the resonance frequency of the resonator to an analyzer;

[0008] A temperature collector, for collecting and outputting the etch chamber ambient temperature to the analyzer;

[0009] An analyzer, for determining an initial etch depth value based on the resonance frequency of the resonator, and performing adaptive temperature drift correction on the initial etch depth value based on the etch chamber ambient temperature to obtain an etch depth correction value;

[0010] A comparator, for receiving the etch depth correction value and comparing the etch depth correction value with a preset value;

[0011] Wherein, the analyzer includes:

[0012] A temperature compensation factor calculation module, for performing temporal feature extraction on the etch chamber ambient temperature, and then performing adaptive depth collaborative analysis on the obtained temporal mode features of the etch chamber ambient temperature and the semantic description of the resonator's own characteristics to determine a temperature adaptive compensation factor;

[0013] A temperature drift adaptive correction module, for performing adaptive temperature drift correction on the initial etch depth value based on the temperature adaptive compensation factor to obtain the etch depth correction value.

[0014] Compared with the prior art, the resonator-based etch endpoint detection circuit provided by the present application first receives the resonance frequency of the resonator detected by the detector through the analyzer and determines the initial etch depth value, synchronously receives the etch chamber ambient temperature collected by the temperature collector, performs temporal feature extraction, and then performs interactive modeling with the initial etch depth value to obtain a dynamically calculated temperature adaptive compensation factor, so as to perform adaptive correction on the temperature drift error and output a highly robust etch depth correction value. Then, the comparator compares the etch depth correction value with the preset value to form a closed-loop feedback and trigger an endpoint determination signal. In this way, through the deep collaboration between the hardware perception layer and the algorithm analysis layer, the limitations of the traditional fixed compensation mode are broken through, and the decoupling optimization of environmental interference and device characteristics is realized. Description of the Drawings

[0015] The above and other objects, features, and advantages of the present application will become more apparent by describing the embodiments of the present application in more detail with reference to the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0016] Figure 1 It is a schematic diagram of a resonator-based etch endpoint detection circuit according to an embodiment of the present application.

[0017] Figure 2 It is a block diagram of an analyzer in a resonator-based etch endpoint detection circuit according to an embodiment of the present application.

[0018] Figure 3 It is a schematic diagram of data flow of an analyzer in a resonator-based etch endpoint detection circuit according to an embodiment of the present application.

[0019] Figure 4 It is a block diagram of a temperature compensation factor calculation module in a resonator-based etch endpoint detection circuit according to an embodiment of the present application.

[0020] Figure 5 It is a block diagram of a resonator characteristic-temperature time-domain depth collaborative analysis unit in a resonator-based etch endpoint detection circuit according to an embodiment of the present application. Detailed implementation manners

[0021] Various exemplary embodiments, features, and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the drawings denote elements having the same or similar functions. Although various aspects of the embodiments are shown in the drawings, the drawings do not have to be drawn to scale unless otherwise specified.

[0022] The special term "exemplary" here means "serving as an example, embodiment, or illustration". Any embodiment described as "exemplary" here does not have to be construed as superior to or better than other embodiments.

[0023] In addition, in order to better illustrate the present application, numerous specific details are given in the following detailed implementation manners. Those skilled in the art should understand that the present application can be implemented without some specific details. In some instances, methods, means, elements, and circuits well known to those skilled in the art are not described in detail so as to highlight the gist of the present application.

[0024] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of this application, "a plurality of" means two or more unless otherwise specifically defined.

[0025] In the field of semiconductor manufacturing, etch endpoint detection is crucial for high-precision nanomanufacturing. The mainstream methods include the optical emission spectrometer (OES) based on the spectrum of substances and the laser interferometer (LI) based on laser interference. However, OES can only determine the endpoint at the interface of different materials due to its dependence on the spectral signals generated by material changes, and its detection sensitivity for nanoscale structures is insufficient. The LI technology is limited by the laser wavelength and the requirement of accurately known material refractive index, and is applicable to the detection of thin films with a thickness exceeding 200 nm. In addition, harsh environmental factors in the etch chamber, such as strong electromagnetic fields, high vacuum, ion bombardment, chemical corrosion, and temperature changes, seriously affect the signal stability of the detection equipment. In particular, temperature drift affects the accuracy of the resonant frequency of the resonator. Traditional temperature compensation methods are difficult to adapt to this non-linear time series change, resulting in the accumulation of etch depth calculation errors. Coupled with the noise and non-linear characteristics in the detection circuit, the accuracy and reliability of the system in determining the nanoscale etch endpoint are further reduced.

[0026] It should be understood that although the resonator-based detection technology improves the detection consistency by co-locating the resonator and the sample in the etch environment and utilizing the same material characteristics, environmental interference (especially temperature drift) is still the key bottleneck restricting its practical application. Existing solutions lack in-depth analysis of the coupling effect between the characteristics of the resonator itself and the dynamic temperature field, and have not achieved the collaborative correction of multi-modal data. Therefore, there is an urgent need for a compensation mechanism that can adapt to environmental temperature changes, integrate the characteristics of the resonator and the real-time temperature time series characteristics, so as to break through the accuracy limit of nanoscale etch endpoint detection.

[0027] To address the above technical problems, this application proposes a resonator-based etch endpoint detection circuit. Figure 1 Schematic diagram of the resonator-based etch endpoint detection circuit according to an embodiment of this application. As Figure 1As shown, the resonator-based etch endpoint detection circuit 100 according to an embodiment of the present application includes: a detector 110 connected to the resonator for detecting and outputting the resonant frequency of the resonator to an analyzer; a temperature collector 120 for collecting and outputting the etch chamber ambient temperature to the analyzer; an analyzer 130 for determining an initial etch depth value based on the resonant frequency of the resonator and performing an adaptive temperature drift correction on the initial etch depth value based on the etch chamber ambient temperature to obtain an etch depth correction value; and a comparator 140 for receiving the etch depth correction value and comparing the etch depth correction value with a preset value.

[0028] In the above resonator-based etch endpoint detection circuit 100, the detector 110 is connected to the resonator and is used to detect and output the resonance frequency of the resonator to the analyzer. It should be understood that the resonator, as a highly sensitive sensor, can sense subtle mass changes during the etching process. As the etching process progresses, the material to be etched gradually decreases, which causes a change in the mass of the resonator and thus affects its resonance frequency. The detector can accurately capture this small but crucial frequency change, ensuring that every subtle change is recorded and transmitted to the analyzer. This requires the detector to have extremely high sensitivity and stability, and be able to maintain consistent performance in a complex and dynamically changing etching environment. Considering that the etching environment usually includes extreme conditions such as strong electromagnetic fields, ion bombardment, and chemical corrosion, the detector in this solution has sufficient anti-interference ability to avoid adverse effects of the external environment on signal acquisition. For example, a detector operating in a high-vacuum environment needs to use special materials and technologies to ensure its normal operation and prevent signal distortion caused by temperature fluctuations or electromagnetic interference. On the other hand, to achieve efficient data transmission, the detector also needs to support a high-speed data communication interface to quickly send the collected resonance frequency data to the analyzer for processing. This not only helps to improve the response speed of the entire system but also ensures that any important changes during the etching process can be detected and responded to in a timely manner. In addition, to improve the reliability and accuracy of the detector, its long-term stability and calibration mechanism also need to be considered. During long-term operation, due to the influence of various factors such as component aging and environmental condition changes, the performance of the detector may drift. Therefore, the design of this solution also includes a self-calibration function or a regular maintenance program to ensure that the detector is always in the best working state. In this way, errors caused by the equipment itself can be minimized to provide more accurate basic data for subsequent analysis. Specifically, since the resonator reflects the etching progress by sensing changes in its own mass. Therefore, the detector must be connected to it in a way that neither affects the normal operation of the resonator nor fails to accurately capture its resonance frequency change. Usually, this connection uses a high-precision electrical interface to ensure the stability and accuracy of signal transmission. At the same time, considering the extreme conditions in the etching environment, such as high temperature and strong electromagnetic interference, the connection between the detector and the resonator also needs to have good shielding performance to prevent external interference from affecting the measurement results. In addition, the selection of connection materials is also crucial, and they need to have excellent electrical conductivity, corrosion resistance, and thermal stability to adapt to various complex situations that may occur during the etching process. Next, during the process of detecting the resonance frequency, the detector utilizes the inherent physical characteristics of the resonator. When the resonator is affected by external factors (such as mass reduction caused by etching), its resonance frequency will change accordingly. The detector can sensitively capture these small frequency fluctuations through a precise electronic circuit design.Specifically, this involves a series of complex signal processing techniques, including but not limited to steps such as filtering, amplification, and analog-to-digital conversion. Filtering techniques are used to remove unnecessary noise interference, so that only signals related to the resonant frequency are retained; the amplifier circuit is responsible for enhancing the intensity of the original signal so that it is sufficient to be recognized by subsequent electronic components; and the analog-to-digital converter converts the processed analog signal into a digital form for further analysis and processing by a computer. These steps work together to ensure that even under extremely small frequency changes, the detector can provide reliable data output. Finally, the detector transmits the collected data to the analyzer losslessly and efficiently. Among them, in order to cope with possible data loss or error problems, redundant coding or error control techniques are usually adopted to ensure data integrity. On the other hand, considering the real-time requirements, a high-speed data transmission channel is also essential. Modern detectors often support multiple communication standards, such as USB, Ethernet, or dedicated industrial bus protocols, etc., in order to select the most suitable connection method according to different application scenarios. At the same time, with the development of wireless communication technology, some new detectors have also begun to explore the possibility of wireless data transmission, which provides more flexibility for future applications.

[0029] In the above resonator-based etch endpoint detection circuit 100, the temperature collector 120 is configured to collect and output the environmental temperature of the etch chamber to the analyzer. It should be understood that the environmental temperature inside the etch chamber is not constant but is affected by various factors, such as the release of plasma energy, the operating state of the cooling system, and changes in external environmental conditions. These changes will cause temperature fluctuations in the etch chamber, which in turn affect the performance of the resonator. Since the resonator is the core sensing unit, its resonance frequency is affected not only by the change in material quality during the etching process but also by temperature changes. Therefore, accurately obtaining the environmental temperature of the etch chamber is a key step in optimizing the measurement results of the resonator. The temperature collector continuously monitors the temperature changes in the etch chamber through highly sensitive sensors and outputs this information to the analyzer in the form of electrical signals. This ensures that detailed data on temperature changes can be obtained even in a complex and changing etching environment. Specifically, in the selection of sensors, considering the extreme working conditions inside the etch chamber, including high temperature, strong electromagnetic fields, chemical corrosion, etc., the temperature collector uses highly sensitive sensors that can adapt to these harsh environments. Such sensors usually have high temperature resistance, corrosion resistance, and good electromagnetic shielding performance to ensure stable operation in a complex environment. For example, certain temperature sensors based on the principle of platinum resistance or thermocouple are widely used in such scenarios due to their excellent stability and accuracy. These sensors measure temperature by converting temperature changes into electrical signal changes. Among them, platinum resistance sensors utilize the characteristic that the resistance of materials changes with temperature, while thermocouples measure temperature based on the thermoelectric emf generated at the contact points of two different metals. Moreover, the temperature collector has high-speed sampling capabilities to facilitate the capture of rapid temperature fluctuations that may occur during the etching process. Since the etching process is often accompanied by the release of plasma energy, the operation of the cooling system, and other factors that may cause temperature changes, the sensor not only needs to continuously monitor the temperature but also has a sufficient sampling frequency to record every subtle temperature change. This requires the temperature collector to be equipped with a high-performance analog front-end circuit to amplify the weak sensor signals and convert them into a form suitable for digital processing. In addition, to avoid signal loss or distortion, the analog front-end circuit should also integrate functional modules such as filters to remove unnecessary noise interference. In this way, even in a complex etching environment, the temperature collector can accurately obtain pure temperature signals, laying a solid foundation for subsequent data processing. In addition, to compensate for the nonlinear error of the sensor itself, the temperature collector usually has a built-in calibration algorithm. By fitting the sensor responses at a series of known temperature points, a calibration model is established. This model can automatically adjust the sensor output in actual applications, making the final temperature readings closer to the true value.

[0030] In the above-mentioned resonator-based etching endpoint detection circuit 100, the analyzer 130 is used to determine the initial value of the etching depth based on the resonant frequency of the resonator, and to perform adaptive temperature drift correction on the initial value of the etching depth based on the ambient temperature of the etching chamber to obtain the etching depth correction value. In particular, the analyzer 130 is used to: calculate the change value of the resonant frequency of the resonator; and convert the change value of the resonant frequency into the initial value of the etching depth.

[0031] Specifically, in the above-mentioned resonator-based etching endpoint detection circuit, the present application realizes the measurement of etching depth in harsh environments through multimodal data fusion and dynamic collaborative correction mechanism. Specifically, the resonant frequency change of the resonator is collected in real time by the detector, and is synchronously input into the analyzer with the ambient temperature time series data of the etching cavity. The analyzer first maps the resonant frequency change to the initial value of the etching depth, and then parses the material properties of the resonator (such as thermal expansion coefficient, thin film structure parameters, etc.) through semantic structured embedded coding technology, and converts it into a quantifiable semantic structured embedded representation of the resonator's own characteristics. At the same time, for the temperature time series data, a time series feature extraction algorithm is used to capture the nonlinear fluctuation pattern and periodic trend of the temperature during the etching process, and generate the time series pattern characteristics of the ambient temperature of the etching cavity.

[0032] On this basis, through deep collaborative analysis under multi-modal independent constraints, the resonator characteristic encoding and temperature timing characteristic encoding are interactively modeled to explore the fine-grained interactive response characteristics and correlations between the two (such as the coupling effect of temperature gradient on local deformation of the resonator). Finally, the resonator characteristic-temperature time domain fine-grained interactive response encoding representation is generated, and the temperature adaptive compensation factor is dynamically calculated based on this. This factor acts on the initial value of the etching depth in real time, and the temperature drift error is adaptively corrected through a nonlinear correction function (such as a compensation curve based on a physical model or data-driven), and a highly robust etching depth correction value is output. The whole process forms a closed-loop feedback to ensure that under strong electromagnetic interference, chemical corrosion and dynamic temperature changes, the system can still accurately compare the compensated etching depth with the preset threshold to trigger the endpoint judgment signal. This technical path breaks through the limitations of the traditional fixed compensation mode through the deep collaboration of the hardware perception layer and the algorithm analysis layer, and realizes the decoupling optimization of environmental interference and device characteristics.

[0033] Figure 2 4 is a block diagram of an analyzer in a resonator-based etching endpoint detection circuit according to an embodiment of the present application. Figure 3 FIG. 1 is a schematic diagram of data flow of an analyzer in a resonator-based etching endpoint detection circuit according to an embodiment of the present application. Figure 2 and Figure 3As shown, the analyzer 130 includes: a temperature compensation factor calculation module 131, configured to perform temporal feature extraction on the etching chamber ambient temperature, and then perform adaptive depth collaborative analysis on the obtained temporal pattern features of the etching chamber ambient temperature and the semantic description of the resonator's own characteristics to determine the temperature adaptive compensation factor; a temperature drift adaptive correction module 132, configured to perform adaptive temperature drift correction on the initial etching depth value based on the temperature adaptive compensation factor to obtain the corrected etching depth value.

[0034] Figure 4 It is a block diagram of the temperature compensation factor calculation module in the etching endpoint detection circuit based on a resonator according to an embodiment of the present application. As Figure 4 shown, in an embodiment of the present application, the temperature compensation factor calculation module 131 includes: a resonator own characteristic description acquisition unit 1311, configured to acquire the semantic description of the resonator's own characteristics; a resonator own characteristic semantic embedding encoding unit 1312, configured to perform semantic structured embedding encoding on the own characteristic description to obtain a resonator own characteristic semantic structured embedding encoding vector; an etching chamber ambient temperature temporal pattern feature extraction unit 1313, configured to perform temperature temporal feature extraction on the time series of the etching chamber ambient temperature to obtain an etching chamber ambient temperature temporal pattern feature encoding vector as the etching chamber ambient temperature temporal pattern feature; a resonator characteristic-temperature time domain depth collaborative analysis unit 1314, configured to perform adaptive depth collaborative analysis based on multi-modal independent constraints on the resonator own characteristic semantic structured embedding encoding vector and the etching chamber ambient temperature temporal pattern feature encoding vector to obtain a resonator characteristic-temperature time domain fine-grained interaction response encoding vector; a temperature adaptive compensation factor determination unit 1315, configured to determine the temperature adaptive compensation factor based on the resonator characteristic-temperature time domain fine-grained interaction response encoding vector.

[0035] Specifically, the resonator self - characteristic description acquisition unit 1311 is configured to acquire the self - characteristic description of the resonator. It should be understood that traditional fixed compensation models cannot fully address the frequency offset caused by environmental changes, especially the impact of temperature changes on the performance of the resonator. During the etching process in semiconductor manufacturing, the frequency change of the resonator is directly related to the accurate measurement of the etching depth, and temperature is an important factor affecting the frequency change of the resonator. Resonators with different materials and structures exhibit different thermal expansion characteristics and changes in elastic modulus under temperature changes, resulting in significant fluctuations in their resonant frequencies. Therefore, in order to more accurately compensate for the impact of temperature changes on the performance of the resonator, it is necessary to describe its self - characteristics in detail. These characteristics include the thermal expansion coefficient, material properties, geometry, and structural changes of the resonator, and the complex non - linear relationship between them and temperature needs to be revealed through modeling and analysis. These characteristics jointly determine the response mode and stability of the resonator under specific conditions, thus affecting the accuracy of the final measurement results.

[0036] Specifically, the resonator self - characteristic semantic embedding encoding unit 1312 is configured to perform semantic structured embedding encoding on the self - characteristic description to obtain a resonator self - characteristic semantic structured embedding encoding vector. It should be understood that considering the complex non - linear correlation between the self - characteristics of the resonator (such as thermal expansion coefficient, film thickness, lattice structure, etc.) and its temperature sensitivity, traditional fixed compensation models cannot effectively characterize the physical mechanism of the dynamic coupling of these characteristics with the ambient temperature. Therefore, in the technical solution of this application, first, semantic structured embedding encoding is performed on the self - characteristic description to obtain a resonator self - characteristic semantic structured embedding encoding vector. Through semantic structured embedding encoding, the unstructured characteristic data of the resonator (such as the material parameter table and film deposition conditions in the process document) can be transformed into a quantitative representation in a high - dimensional vector space, thereby extracting implicit physical laws (such as the gradient impact of material thermal expansion on the resonant frequency). The purpose of this process is to provide a basis for establishing a computable mapping relationship between the resonator characteristics and temperature drift in the subsequent process, enabling the subsequent multi - modal collaborative analysis to accurately decouple the mass change caused by etching and the frequency offset caused by temperature perturbation. Its implementation effect is reflected in two aspects: on the one hand, through embedding encoding, heterogeneous information such as process parameters and material properties is unified into a machine - interpretable mathematical expression, avoiding the limitations of manual experience - based modeling; on the other hand, the implicit physical semantic information in the structured vector (such as the correlation characteristics between film stress distribution and temperature sensitivity) can provide fine - grained prior knowledge for the deep collaborative analysis module, thereby achieving adaptive cancellation of temperature drift errors in a strong interference environment, ultimately improving the signal - to - noise ratio and robustness of the etching depth correction value, and solving the defect of the separate analysis of resonator characteristics and dynamic temperature field in the prior art.

[0037] In an embodiment of the present application, the etching chamber ambient temperature timing pattern feature extraction unit 1313 is used to: pass the time series of the etching chamber ambient temperature through a temperature timing feature extractor based on a forward LSTM to obtain the etching chamber ambient temperature timing pattern feature encoding vector. It should be understood that since the cavity temperature in the etching process is not steady-state, but presents dynamic nonlinear fluctuations (such as instantaneous temperature rise caused by plasma energy release and periodic cooling caused by cavity heat dissipation), these timing changes have a complex time domain coupling relationship with the thermal response characteristics of the resonator. The traditional fixed compensation model only focuses on the instantaneous value of temperature and cannot capture the gradient, trend and cumulative effect of temperature change, resulting in phase lag and amplitude deviation in the resonant frequency correction. Therefore, in the technical solution of the present application, the time series of the etching chamber ambient temperature is further subjected to temperature timing feature extraction to obtain the etching chamber ambient temperature timing pattern feature encoding vector. In particular, in a specific example of the present application, the time series of the ambient temperature of the etching chamber is passed through a temperature time series feature extractor based on a forward LSTM to obtain a time series pattern feature encoding vector of the ambient temperature of the etching chamber. Through temperature time series feature extraction, physical modes such as transient impact, thermal inertia delay, and periodic fluctuation in the original temperature data can be converted into high-dimensional feature vectors, such as using forward LSTM to extract dynamic modes such as temperature change rate, local extreme value distribution, and autocorrelation characteristics. By capturing the local mutation characteristics of the temperature time series (such as the instantaneous temperature rise caused by ion bombardment), the rapid change of the local thermal stress of the resonator can be predicted to avoid signal distortion caused by traditional mean filtering. In addition, the extracted temperature trend component (such as the overall temperature rise curve of the cavity during etching) can reflect the long-term drift effect of the heat accumulation effect on the fundamental frequency of the resonator, and realize advance compensation. In addition, the periodic fluctuation information implicit in the time series feature encoding (such as the temperature oscillation caused by the start and stop of the cooling system) provides a dynamic boundary condition for the resonator thermal-mechanical coupling model, so that the temperature adaptive compensation factor can accurately solve the superposition effect of transient and steady-state temperature interference. This method breaks through the bottleneck of confusion between temperature disturbance and etching signal in fixed compensation mode, and significantly improves the anti-interference ability of nano-level etching depth detection.

[0038] Figure 5 FIG. 1 is a block diagram of a resonator characteristic-temperature time domain depth collaborative analysis unit in a resonator-based etching endpoint detection circuit according to an embodiment of the present application. Figure 5In the embodiment of the present application, the resonator characteristic-temperature time-domain depth collaborative analysis unit 1314 includes: an etching cavity ambient temperature characteristic linear transformation subunit 1314-1, configured to perform linear transformation processing based on principal component analysis on the etching cavity ambient temperature time-series pattern feature encoding vector to obtain a set of etching cavity ambient temperature time-series principal component linear transformation feature encoding vectors; a resonator characteristic-temperature time-domain principal component inter-modal independence constraint analysis subunit 1314-2, configured to perform inter-modal independence constraint analysis on the set of etching cavity ambient temperature time-series principal component linear transformation feature encoding vectors and the resonator self-characteristic semantic structured embedding encoding vector to determine a set of resonator characteristic-temperature time-domain principal component inter-modal independence soft constraint factors; a resonator characteristic-temperature time-domain feature dynamic adaptive aggregation subunit 1314-3, configured to perform dynamic adaptive aggregation on the set of etching cavity ambient temperature time-series principal component linear transformation feature encoding vectors and the resonator self-characteristic semantic structured embedding encoding vector based on the set of resonator characteristic-temperature time-domain principal component inter-modal independence soft constraint factors to obtain a resonator characteristic-temperature time-domain fine-grained interaction response encoding vector. It should be understood that due to the highly non-linear and time-varying coupling effect between the material properties of the resonator (such as thermal expansion coefficient, thin film stress distribution) and the dynamic temperature field, it is difficult for traditional single-modal analysis or hard fusion strategies to decouple the redundant associations between the two (such as the local strong correlation and global weak correlation difference between temperature fluctuations and material thermal deformation), resulting in insufficient adaptability of the compensation model to complex working conditions. Therefore, in the technical solution of the present application, further perform adaptive depth collaborative analysis based on multi-modal independent constraints on the resonator self-characteristic semantic structured embedding encoding vector and the etching cavity ambient temperature time-series pattern feature encoding vector to obtain a resonator characteristic-temperature time-domain fine-grained interaction response encoding vector.

[0039] Specifically, the process of deep collaborative analysis under multi-modal independent constraints first uses principal component analysis to reduce the dimension and purify the information of the time series pattern characteristics of the etching chamber environment temperature, eliminate high-frequency noise and redundant fluctuation patterns in the temperature data, and achieve scale alignment of the resonator characteristic encoding and the temperature characteristic encoding through linear transformation (for example, unify the film thickness parameter and the temperature gradient dimension to the interactive space), avoiding the imbalance of feature contribution due to modal scale differences. Subsequently, the topological order matrix of the characteristic-temperature association is explicitly constructed through the inter-modal independence modeling unit to quantify the complementarity and independence of the two in the time domain (such as the long-range correlation between the instantaneous effect of thermal expansion on the resonant frequency and the temperature accumulation effect), and dynamically adjust the fusion weights of different modal features based on the soft constraint factor - when the temperature time series pattern and the resonator characteristics show strong independence in the local time domain, the interaction of complementary information is strengthened; when there is potential redundancy (such as the superposition of periodic temperature fluctuations and material thermal inertia), the interference of repeated features is suppressed. The purpose of this process is to extract physically interpretable time-domain interactive response features from multimodal data (such as the dynamic influence coefficient of temperature gradient changes on the local deformation of the resonator), and provide fine-grained time-varying correction parameters for temperature adaptive compensation. Specifically, the execution effect of deep collaborative analysis under multi-modal independent constraints is reflected in three aspects: first, through the optimization of inter-modal independence, the key modes directly related to the resonator characteristics in the temperature disturbance (such as the impact effect of the instantaneous temperature rise caused by ion bombardment on the film stress) are effectively distinguished, and irrelevant noise is stripped; second, the dynamic aggregation strategy under the soft constraint mechanism enables the compensation factor to adapt to the nonlinear evolution of the temperature field and the resonator characteristics during the etching process (such as the resonant frequency baseline drift caused by the overall temperature rise of the cavity in the later stage of etching); third, the physical correlation information implicit in the obtained resonator characteristic-temperature time domain fine-grained interactive response encoding vector (such as the dynamic phase relationship of thermal-mechanical coupling) provides a high-order correction basis for the subsequent compensation algorithm, significantly reducing the cumulative error introduced by the traditional method due to ignoring the temperature time series correlation, and finally achieving sub-nanometer temperature drift compensation accuracy in nano-etching endpoint detection. Through the fusion paradigm of independent constraints and dynamic collaboration, the accuracy bottleneck of the traditional fixed compensation model in complex environments has been broken through.

[0040] In an embodiment of the present application, the etching chamber ambient temperature feature linear transformation subunit 1314-1 is configured to: perform principal component analysis on the etching chamber ambient temperature time series pattern feature encoding vector to obtain a set of etching chamber ambient temperature time series principal component feature encoding vectors; perform linear transformation on each etching chamber ambient temperature time series principal component feature encoding vector in the set of etching chamber ambient temperature time series principal component feature encoding vectors to obtain a set of etching chamber ambient temperature time series principal component linear transformation feature encoding vectors, wherein each etching chamber ambient temperature time series principal component linear transformation feature encoding vector in the set of etching chamber ambient temperature time series principal component linear transformation feature encoding vectors has the same feature scale as the resonator self-characteristic semantic structured embedding encoding vector.

[0041] Specifically, performing principal component analysis on the etching chamber ambient temperature time series pattern feature encoding vector to obtain a set of etching chamber ambient temperature time series principal component feature encoding vectors, which is expressed by the etching chamber ambient temperature time series principal component feature encoding formula as: Wherein, is the etching chamber ambient temperature time series pattern feature encoding vector, is the principal component analysis, is the covariance matrix of the etching chamber ambient temperature time series pattern feature encoding vector, is the etching chamber ambient temperature time series pattern feature principal component orthogonal matrix, , and are respectively the 1st, 2nd and th etching chamber ambient temperature time series pattern feature principal component encoding vectors in the set of etching chamber ambient temperature time series principal component feature encoding vectors, is the etching chamber ambient temperature time series pattern feature diagonal matrix, , and are respectively , and corresponding eigenvalues, is 's transpose matrix, is the number of eigenvalues in the etching chamber ambient temperature time series pattern feature encoding vector, It represents extracting diagonal elements. It should be understood that during the semiconductor etching process, the chamber temperature is affected by multiple factors such as plasma energy release, cooling system operation, and external environmental disturbances, presenting non-linear time-series fluctuation characteristics. Due to relying on fixed coefficients or single-dimensional features, traditional temperature compensation methods are difficult to capture the gradients, trends, and periodic patterns of temperature changes, resulting in phase lags and amplitude deviations in resonator frequency correction. Therefore, it is necessary to perform multi-dimensional feature decoupling and information purification on the original temperature time-series data through principal component analysis. Principal component analysis can construct a feature subspace with high information entropy, and project the temperature fluctuation information scattered in the original high-dimensional feature space onto a set of ordered principal component axes through orthogonal transformation. Specifically, principal component analysis is used to reduce the dimension of the temperature time-series pattern feature coding vector, aiming to eliminate potential feature redundancy and multicollinearity problems within the modal data. For example, the spike temperature rise caused by the instantaneous energy release of the plasma and the periodic temperature drop caused by the chamber heat dissipation may show strong correlations in the original data. Through PCA, independent principal components with clear physical meanings such as the temperature change rate and local extreme value distribution can be separated, avoiding the interference of redundant features on subsequent interactions between modes. In this way, the computational complexity can be significantly reduced through dimensionality reduction, enabling efficient processing capabilities in etching scenarios with strict real-time requirements. Moreover, the ordered arrangement of the principal component axes enables a small number of head components to capture the main variation information of temperature changes, significantly improving the characterization accuracy of temperature drift features. In addition, noise interference and redundant fluctuation patterns in the original data can be eliminated, providing high-purity and low-coupling feature inputs for subsequent in-depth collaborative analysis of resonator characteristics and temperature in the time domain, thereby enhancing the discriminability and robustness of the calculation of temperature adaptive compensation factors. This process is essentially a decoupling and reconstruction of the physical properties and mathematical representations of temperature time-series data, laying a high-order feature foundation for resonator frequency correction in a dynamic environment.

[0042] Specifically, a linear transformation is performed on each etching chamber environment temperature time-series principal component feature coding vector in the set of etching chamber environment temperature time-series principal component feature coding vectors to obtain a set of etching chamber environment temperature time-series principal component linear transformation feature coding vectors, where each etching chamber environment temperature time-series principal component linear transformation feature coding vector in the set of etching chamber environment temperature time-series principal component linear transformation feature coding vectors has the same feature scale as the resonator's own characteristic semantic structured embedding coding vector, and is represented by the etching chamber environment temperature time-series principal component linear transformation formula: Where represents performing a linear transformation on each etching chamber environment temperature time-series principal component feature coding vector in , and are respectively the 1st, 2nd, and The principal component linear transformation feature coding vectors of the etching chamber ambient temperature time series, is the set of principal component linear transformation feature coding vectors of the etching chamber ambient temperature time series. It should be understood that considering the dynamic fluctuation characteristics of temperature time series data and the physical parameters of resonator material properties belong to different dimension systems. For example, the temperature gradient change is in the unit of °C / s, while the thin film stress parameter is in the unit of GPa. There are significant heterogeneities in the numerical range and statistical distribution between the two. If cross-modal interaction is directly carried out, the high-scale features will dominate the model attention mechanism, resulting in the key physical information carried by the low-scale features being submerged, causing misjudgment of temperature drift error compensation. By normalizing and aligning the feature scales, the interference of dimensional differences between modalities on subsequent in-depth collaborative analysis can be eliminated. Specifically, the linear transformation applies scale scaling and translation operations to the principal component feature coding vectors to make their distribution range match that of the resonator characteristic coding vectors. For example, the variance range of the temperature time series principal component is mapped from [0, 100] to [0, 1], while keeping the topological structure of its internal physical law unchanged. This alignment is not only a mathematical standardization process, but also a re-anchoring of physical meaning, ensuring that the gradient information of temperature change and the thermal expansion coefficient of the resonator are comparable in the feature space. In this way, the problem of unbalanced feature contributions caused by dimensional differences is eliminated, enabling the transient impact features of temperature time series and the cumulative effect of resonator thin film stress to obtain equal weights during the interaction process, avoiding the pseudo-correlation phenomenon caused by scale differences in traditional methods. Moreover, the unified feature scale reduces the convergence difficulty of the optimization process, improves the gradient stability of the temperature adaptive compensation factor calculation, and provides a high signal-to-noise ratio input basis for subsequent dynamic compensation algorithms.

[0043] In an embodiment of the present application, the resonator characteristic-temperature time-domain principal component inter-modal independence constraint analysis sub-unit 1314-2 is configured to: input each etching cavity ambient temperature time-series principal component linear transformation feature encoding vector in the set of the resonator self-characteristic semantic structured embedding encoding vector and the etching cavity ambient temperature time-series principal component linear transformation feature encoding vectors into the inter-modal independence modeling unit to obtain a set of resonator characteristic-temperature time-domain principal component inter-modal independence encoding matrices; perform canonical stability optimization on each resonator characteristic-temperature time-domain principal component inter-modal independence encoding matrix in the set of resonator characteristic-temperature time-domain principal component inter-modal independence encoding matrices to obtain a set of resonator characteristic-temperature time-domain principal component inter-modal independence canonical optimization encoding matrices, wherein the process of the canonical stability optimization is determined by the resonator characteristic-temperature time-domain eigenvectors composed of the respective eigenvalues of the resonator characteristic-temperature time-domain principal component inter-modal independence encoding matrix; and calculate a set of resonator characteristic-temperature time-domain principal component inter-modal independence soft constraint factors based on the set of resonator characteristic-temperature time-domain principal component inter-modal independence canonical optimization encoding matrices.

[0044] Specifically, inputting each etching cavity ambient temperature time-series principal component linear transformation feature encoding vector in the set of the resonator self-characteristic semantic structured embedding encoding vector and the etching cavity ambient temperature time-series principal component linear transformation feature encoding vectors into the inter-modal independence modeling unit to obtain a set of resonator characteristic-temperature time-domain principal component inter-modal independence encoding matrices, which is represented by the resonator characteristic-temperature time-domain feature inter-modal independence modeling formula as: Wherein, and are feature mapping functions, such as linear mapping or non-linear kernel functions, is the resonator self-characteristic semantic structured embedding encoding vector, is the length of the th etching cavity ambient temperature time-series principal component linear transformation feature encoding vector, Resonator characteristic - temperature time - domain principal component inter - modal independence coding matrix. It should be understood that there is a non - linear time - domain interaction between the material parameters of the resonator (such as the coefficient of thermal expansion, thin - film stress distribution) and the temperature time - series characteristics (such as the transient temperature rise caused by plasma energy release, the periodic fluctuation caused by cavity heat dissipation). Traditional implicit fusion methods cannot distinguish complementary information from redundant noise, resulting in insufficient capture accuracy of the temperature drift compensation model for key physical correlations. In this application, by explicitly constructing a learnable modal independence constraint, complementary features physically related to the resonator characteristics are screened from multi - dimensional temperature principal components. Specifically, the inter - modal independence modeling unit establishes an orthogonal projection relationship between the resonator characteristic coding vector and the characteristics of each temperature principal component. For example, the instantaneous influence mode of temperature gradient change on thin - film stress is extracted by maximizing mutual information, while suppressing the long - range redundant correlation between the overall cavity temperature rise and the material heat capacity. In this way, the physical fluctuation modes of the temperature time - series principal components (such as spike pulses, slow - varying trends) and the response characteristics of the resonator thermo - mechanical parameters (such as local deformation, fundamental frequency drift) are topologically correlated and modeled to form an interpretable resonator characteristic - temperature time - domain principal component inter - modal independence coding matrix. The set of obtained resonator characteristic - temperature time - domain principal component inter - modal independence coding matrices provides a physically interpretable weight basis for subsequent dynamic aggregation by quantifying the independence degree between the temperature principal components and the resonator characteristics.

[0045] Specifically, each resonator characteristic - temperature time - domain principal component inter - modal independence coding matrix in the set of resonator characteristic - temperature time - domain principal component inter - modal independence coding matrices is subjected to canonical stability optimization to obtain a set of resonator characteristic - temperature time - domain principal component inter - modal independence canonical optimization coding matrices, where the process of the canonical stability optimization is determined by the resonator characteristic - temperature time - domain eigenvector composed of the eigenvalues of each resonator characteristic - temperature time - domain principal component inter - modal independence coding matrix, and is expressed by the canonical stability optimization formula as: Where, is the eigenvalue of the th resonator characteristic - temperature time - domain principal component inter - modal independence coding matrix, is matrix multiplication, is the resonator characteristic - temperature time - domain canonical correlation coding vector, is the Resonator characteristics - temperature time domain principal component mode independence specification optimization coding matrix. It should be understood that considering the interaction relationship between the temperature time series principal components (such as the transient temperature rise gradient caused by plasma pulses) and the resonator material parameters (such as thin film stress anisotropy) presents a non-uniform topological structure with coexistence of local strong coupling and long-range weak correlation in the time domain, the matrix optimization method under the traditional symmetry assumption cannot maintain the stability of such asymmetric interaction topologies, resulting in global feature drift caused by local perturbations (such as cooling system vibration noise) during the mode fusion process. This application constructs a gauge potential field based on the resonator characteristics - temperature time domain eigenvectors through gauge stability optimization to achieve the stability control of the asymmetric topological structure. Specifically, the eigenvalue vector is used to characterize the topological order parameters of the temperature-characteristics interaction (such as the coupling strength between local temperature rise pulses and stress mutations, the phase correlation between long-range temperature rise trends and substrate thermal expansion), which is used as the mathematical basis of the gauge potential. The curvature of the independence coding matrix between the resonator characteristics - temperature time domain principal components is corrected through the connection 1-form in the gauge field theory. For example, for the strong coupling mode between the transient temperature rise spike caused by local ion bombardment at the initial stage of etching and the thin film edge stress gradient, the corresponding resonator characteristics - temperature time domain eigenvector adjusts the local curvature of the independence coding matrix between the resonator characteristics - temperature time domain principal components through the gauge potential to suppress the distortion effect of noise perturbations on the interaction topology; for the long-range weak correlation between the overall temperature rise of the cavity and the substrate thermal expansion in the later stage of etching, the global topology continuity is maintained through the gauge potential. The introduction of the gauge potential makes the independence coding matrix between the resonator characteristics - temperature time domain principal components remain invariant under the Lie group structure under local perturbations, ensuring the stability of the temperature-characteristics interaction topology. The optimized independence coding matrix between the resonator characteristics - temperature time domain principal components accurately characterizes the dynamic influence weight of the transient temperature rise gradient on the local deformation of the resonator (such as the instantaneous coupling coefficient of the microsecond-level temperature rise pulse to the thin film edge stress), while stabilizing the weak correlation between the long-range temperature accumulation effect and the fundamental frequency drift. Moreover, through the constraint of topological invariants, the interference of environmental noise (such as electromagnetic interference, mechanical vibration) on the mode fusion process during etching is significantly reduced, enabling the subsequent temperature drift compensation algorithm to achieve sub-angstrom-level dynamic correction accuracy under the asymmetric interaction topology, providing a stable feature space basis for the robust determination of the nano-level etching endpoint.

[0046] Specifically, based on the set of the resonator characteristics - temperature time domain principal component mode independence specification optimization coding matrix, calculate the set of the resonator characteristics - temperature time domain principal component mode independence soft constraint factors, which is expressed by the resonator characteristics - temperature time domain principal component mode independence soft constraint factor calculation formula as: Wherein, represents the matrix The square of the norm represents a set of soft constraint factors for the independence between the resonator characteristics - temperature time - domain principal component modes. It should be understood that the interaction relationship between the temperature time - series principal components (such as the transient temperature rise caused by plasma energy pulses and the hysteresis effect caused by the cavity thermal inertia) and the resonator characteristics (such as thin - film stress distribution, coefficient of thermal expansion) is not constant. The traditional hard - constraint method, which forces decoupling through a fixed threshold, will destroy the physical correlation between the two under specific working conditions (for example, the strong coupling relationship between the overall temperature rise of the cavity and the thermal deformation of the substrate in the later stage of etching), resulting in limited generalization ability of the temperature drift compensation model under complex working conditions. In this application, the physical correlation strength implicitly contained in the optimized encoding matrix for the independence between the resonator characteristics - temperature time - domain principal component modes is quantified as a dynamically adjustable soft constraint factor for the independence between the resonator characteristics - temperature time - domain principal component modes, realizing the adaptive allocation of modal fusion weights. Specifically, in this application, the square of the Frobenius norm of the optimized encoding matrix for the independence between the resonator characteristics - temperature time - domain principal component modes is extracted as the soft constraint factor for the independence between the resonator characteristics - temperature time - domain principal component modes, mapping the degree of independence between the resonator characteristics and the temperature principal components to a continuous weight value in the range of [0, 1]. For example, when there is a high degree of independence between the temperature gradient change and the thin - film stress (such as the instantaneous temperature rise caused by local ion bombardment in the initial stage of etching), the soft constraint factor for the independence between the resonator characteristics - temperature time - domain principal component modes approaches 1, strengthening the contribution of this temperature principal component to the compensation factor; when there is a physical coupling between the temperature accumulation effect and the thermal expansion of the material (such as the overall temperature rise of the cavity in the later stage of etching), the soft constraint factor for the independence between the resonator characteristics - temperature time - domain principal component modes approaches 0, allowing for a moderate fusion of their characteristics. The soft constraint mechanism for the independence between the resonator characteristics - temperature time - domain principal component modes enables the compensation model to adapt to the non - linear evolution of the temperature field and resonator characteristics during the etching process by dynamically balancing modal independence and correlation, avoiding the loss of physical information caused by hard constraints (such as ignoring the instantaneous impact of the temperature gradient on the local deformation of the resonator). Moreover, the continuously adjustable characteristic of the soft constraint factor for the independence between the resonator characteristics - temperature time - domain principal component modes enables the system to accurately distinguish the direct interference component (such as the temperature jitter caused by the vibration of the cooling system) and the effective correlation component (such as the phase - shift effect of the ion - bombardment temperature rise on the resonance frequency) in the temperature perturbation, thereby suppressing noise interference while retaining the key physical correlations, and finally achieving sub - angstrom - level dynamic precision control for temperature drift compensation in nanoscale etching depth detection.

[0047] In an embodiment of the present application, the resonator characteristic-temperature time-domain feature dynamic adaptive aggregation subunit 1314-3 is configured to: input each etching cavity environment temperature time-series principal component linear transformation feature encoding vector in the set of the resonator's own characteristic semantic structured embedding encoding vector and the etching cavity environment temperature time-series principal component linear transformation feature encoding vectors into a feature interaction response unit to obtain a set of resonator characteristic-temperature time-domain principal component mode fine-grained response interaction encoding vectors; and perform dynamic adaptive aggregation on the set of resonator characteristic-temperature time-domain principal component mode fine-grained response interaction encoding vectors based on the set of resonator characteristic-temperature time-domain principal component mode independence soft constraint factors to obtain the resonator characteristic-temperature time-domain fine-grained interaction response encoding vector.

[0048] Specifically, inputting each etching cavity environment temperature time-series principal component linear transformation feature encoding vector in the set of the resonator's own characteristic semantic structured embedding encoding vector and the etching cavity environment temperature time-series principal component linear transformation feature encoding vectors into a feature interaction response unit to obtain a set of resonator characteristic-temperature time-domain principal component mode fine-grained response interaction encoding vectors, which is represented by the resonator characteristic-temperature time-domain feature interaction response formula as: Wherein, represents cascade processing, 、 and represent element-wise multiplication, element-wise addition, and element-wise division respectively, and represent trainable weight matrices and trainable bias vectors, represents the Resonator characteristics - Temperature time-domain principal component modal fine-grained response interaction coding vector. It should be understood that the feature interaction response unit deconstructs the non-linear dependence relationship between the temperature principal component and the resonator characteristics through fine-grained interaction modeling at the feature dimension level. Among them, the feature interaction response unit constructs a high-order interaction space by using tensor product operations. For example, the temperature transient shock component (such as a microsecond-level temperature rise pulse) is multiplied element by element with the thin film stress gradient parameter to generate an interaction code representing the coupling strength between the two. At the same time, the gating mechanism is used to screen the phase-matching features of the temperature cumulative effect (such as a minute-level temperature rise trend) and the resonator thermal expansion coefficient. This design enables the system to simultaneously capture the non-linear correlation between the time-domain dynamic characteristics of the temperature time-series principal component (such as the sign reversal of the temperature change rate) and the spatial distribution characteristics of the resonator material response (such as the stress concentration phenomenon at the thin film edge). In this way, the set of resonator characteristics - temperature time-domain principal component modal fine-grained response interaction coding vectors forms a physically interpretable high-order feature representation by fusing the physical fluctuation modes of the temperature principal component (such as spike pulses, slow-changing trends) and the spatial distribution parameters of the resonator characteristics (such as stress gradient, thermal expansion anisotropy), such as quantifying the contribution weight of local temperature rise to the deformation of a specific area of the resonator. Moreover, the fine-grained interaction mechanism enables the temperature drift compensation model to distinguish different types of temperature-characteristic coupling modes during the etching process (such as phase shift caused by transient shock and baseline drift caused by long-term temperature rise), providing differential correction parameters for the dynamic compensation algorithm, and finally achieving sub-angstrom-level temperature drift correction accuracy for nanometer-level etching depth under strong noise interference.

[0049] Specifically, based on the set of independence soft constraint factors between the resonator characteristics - temperature time-domain principal component modes, the set of resonator characteristics - temperature time-domain principal component modal fine-grained response interaction coding vectors is dynamically adaptively aggregated to obtain the resonator characteristics - temperature time-domain fine-grained interaction response coding vector, which is expressed by the resonator characteristics - temperature time-domain feature dynamic adaptive aggregation formula as: Wherein, is function, is the number of vectors in the set of resonator characteristics - temperature time-domain principal component modal fine-grained response interaction coding vectors, It is the fine-grained interaction response coding vector of the resonator characteristic-temperature time domain. It should be understood that through the dynamic weight allocation mechanism guided by the soft constraint factor of the independence between the principal component modes of the resonator characteristic-temperature time domain, the time-domain adaptive optimization of multimodal feature fusion can be realized. Specifically, based on the soft constraint factor of the independence between the principal component modes of the resonator characteristic-temperature time domain, the fine-grained response interaction coding vector between the principal component modes of the resonator characteristic-temperature time domain is gated and weighted aggregated. For example, at the initial stage of etching, higher weights are given to the high-independence interaction features of the transient temperature rise peak and the stress mutation at the film edge to strengthen the instantaneous compensation of the resonance frequency by local impacts; while at the later stage of etching, the fusion ratio of the long-range correlation features of the cavity cumulative temperature rise and the substrate thermal expansion is increased to accurately correct the fundamental frequency drift. This dynamic aggregation strategy maps the independence strength of the soft constraint factor of the independence between the principal component modes of the resonator characteristic-temperature time domain into a probability distribution through the softmax function to realize the automatic switching of the dominant interaction modes in different process stages. In this way, the generated fine-grained interaction response coding vector of the resonator characteristic-temperature time domain forms a high-purity temperature-characteristic interaction representation by suppressing redundant noise (such as temperature jitter caused by the vibration of the cooling system) and strengthening physical correlation features (such as the coupling of the ion bombardment temperature rise gradient and local stress mutation). Moreover, the dynamic weight mechanism enables the compensation model to adapt to the non-linear evolution law of the temperature field and the resonator response during the etching process, avoiding the phase lag or over-compensation problems caused by the fixed fusion strategy in traditional methods.

[0050] In the embodiment of the present application, the temperature adaptive compensation factor determination unit 1315 is configured to: obtain a temperature adaptive compensation factor decoding value by passing the fine-grained interaction response coding vector of the resonator characteristic-temperature time domain through a temperature compensation factor calculator based on a decoder. In this way, the abstract fine-grained interaction response features of the resonator characteristic-temperature time domain can be converted into an operable compensation factor, and the precise calculation of the nanoscale etching depth can be realized by dynamically correcting the temperature drift component in the initial value of the etching depth.

[0051] Specifically, the temperature drift self - adaptive correction module 132 is used to perform self - adaptive temperature drift correction on the initial etch depth value based on the temperature self - adaptive compensation factor to obtain the corrected etch depth value. It should be understood that the temperature drift self - adaptive correction module can eliminate the influence of temperature drift on the measurement result through precise perception and dynamic compensation of temperature changes, ensuring the accuracy of the etch depth. Specifically, in a specific implementation of the present application, the initial etch depth value is adjusted through a non - linear correction function to better reflect the actual etching progress. Non - linear corrections (such as piece - wise linear fitting, polynomial regression) can accurately describe the influence of temperature drift on the resonance frequency. Among them, piece - wise linear fitting is used to handle the situation where the temperature change is relatively gentle, or when the relationship between temperature and etch depth can be reasonably divided into multiple linear intervals. In actual operation, the appropriate break points need to be determined first, which is usually decided based on experimental data or theoretical analysis. For example, in a typical etching process, if it is found that the temperature changes little within a certain range, while there are significant fluctuations in other ranges, then the entire temperature range can be divided into several sub - intervals, and the linear regression method is applied within each sub - interval to establish a model. Specifically, for each sub - interval, the slope and intercept of the best - fit straight line are calculated through statistical methods such as the least - squares method, so as to obtain the linear equation of this interval, for more accurate correction according to the characteristics of different temperature ranges, avoiding the large errors that may be brought by a single global model. In addition, piece - wise linear fitting also allows flexible adjustment of the boundaries of each sub - interval to adapt to the changing actual conditions, improving the adaptability and robustness of the model. Polynomial regression is used to handle scenarios where the temperature changes violently and complexly. It approximates the non - linear relationship between temperature and etch depth by constructing a high - order polynomial function. In the specific implementation process, an appropriate polynomial order needs to be selected, which depends on the complexity of the temperature change and the available data volume. Generally speaking, a higher order can capture the subtle changes in the data more accurately, but at the same time increases the risk of overfitting. Therefore, in practical applications, technical means such as cross - validation are often needed to find a balance point that can fully reflect the data characteristics without overfitting. Once the polynomial order is determined, the next step is to estimate the coefficients of the polynomial using the existing temperature - etch depth data set. This is usually achieved by solving an over - determined system of equations, that is, using the least - squares method or other optimization algorithms to minimize the difference between the predicted value and the true value. The finally obtained polynomial function can be used to predict the corrected etch depth value at any given temperature.

[0052] In the above resonator-based etch endpoint detection circuit 100, the comparator 140 is configured to receive the etch depth correction value and compare the etch depth correction value with a preset value. It should be understood that due to variations in etch rate, material properties, and environmental conditions, the actual etch progress often cannot fully meet expectations. By receiving the etch depth correction value and comparing it with the preset value, the comparator can detect any deviation from the expected situation in a timely manner and take necessary adjustment measures accordingly. For example, in some high-precision applications, even very small errors may lead to a significant decline in device performance, so a timely and accurate feedback mechanism is particularly important. Moreover, comparing with the preset target value not only helps to confirm whether the current process is proceeding as planned but also provides a basis for subsequent decision-making. For example, if it is found that the etch depth is close to the preset value, the etch process can be prepared to end; conversely, if it is found that the etch depth is far lower than expected, the parameter settings may need to be adjusted or the etch time extended to achieve the desired processing effect. In this way, the comparator effectively acts as the "eyes" in process control, continuously monitoring and evaluating the status of the etch progress. This dynamic adjustment mechanism can not only provide accurate feedback on the etch progress but also trigger corresponding control signals to ensure that the entire process flow proceeds smoothly according to the predetermined plan, thereby improving the system's response speed and stability, reducing the need for human intervention, and enhancing the overall automation level. Specifically, in a specific embodiment of the present application, the preset value is obtained through machine learning technology, and the comparison between the etch depth correction value and the preset value is calculated by taking the difference.

[0053] In summary, the resonator-based etch endpoint detection circuit 100 according to the embodiments of the present application is elucidated. It receives the resonance frequency of the resonator detected by the detector through an analyzer and determines the initial etch depth value. At the same time, it receives the etch chamber ambient temperature collected by the temperature collector, extracts the timing characteristics, and then interacts and models with the initial etch depth value to obtain a dynamically calculated temperature adaptive compensation factor to adaptively correct the temperature drift error and output a highly robust etch depth correction value. Then, the comparator compares the etch depth correction value with the preset value to form a closed-loop feedback and trigger the endpoint determination signal. In this way, through the deep cooperation between the hardware perception layer and the algorithm analysis layer, the limitations of the traditional fixed compensation mode are broken through, and the decoupling optimization of environmental interference and device characteristics is achieved.

[0054] As described above, the resonator-based etch endpoint detection circuit 100 according to the embodiments of the present application can be implemented in various terminal devices. In one example, the resonator-based etch endpoint detection circuit 100 can be integrated into the terminal device as a software module and / or a hardware module. For example, the resonator-based etch endpoint detection circuit 100 can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the resonator-based etch endpoint detection circuit 100 can also be one of the numerous hardware modules of the terminal device.

[0055] Alternatively, in another example, the resonator-based etch endpoint detection circuit 100 and the terminal device can also be separate devices, and the resonator-based etch endpoint detection circuit 100 can be connected to the terminal device through a wired and / or wireless network and transmit and interact information in accordance with a predefined data format.

[0056] In several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there can be other division methods in actual implementation.

[0057] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0058] In addition, in each embodiment of the present invention, the functional modules can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware, or in the form of a combination of hardware and software functional modules.

[0059] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-mentioned exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention.

[0060] In addition, it is obvious that the word "including" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units or apparatuses stated in the system can also be implemented by one unit or apparatus through software or hardware. The terms such as "second" are used to represent names and do not represent any specific order.

[0061] 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 them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit of the technical solutions of the present invention.

Claims

1. A resonator-based etch endpoint detection circuit, characterized in that, Including: A detector, connected to the resonator, for detecting and outputting the resonance frequency of the resonator to an analyzer; A temperature collector, for collecting and outputting the etching chamber ambient temperature to the analyzer; An analyzer, for determining an initial etching depth value based on the resonance frequency of the resonator, and performing an adaptive temperature drift correction on the initial etching depth value based on the etching chamber ambient temperature to obtain an etched depth correction value; A comparator, for receiving the etched depth correction value and comparing the etched depth correction value with a preset value; Wherein, the analyzer includes: A temperature compensation factor calculation module, including: a resonator self-characteristic description acquisition unit, for acquiring a description of the resonator's own characteristics; a resonator self-characteristic semantic embedding encoding unit, for performing semantic structured embedding encoding on the self-characteristic description to obtain a resonator self-characteristic semantic structured embedding encoding vector; an etching chamber ambient temperature time series pattern feature extraction unit, for performing temperature time series feature extraction on the time series of the etching chamber ambient temperature to obtain an etching chamber ambient temperature time series pattern feature encoding vector as the etching chamber ambient temperature time series pattern feature; a resonator characteristic-temperature time domain depth collaborative analysis unit, for performing an adaptive depth collaborative analysis based on multi-modal independent constraints on the resonator self-characteristic semantic structured embedding encoding vector and the etching chamber ambient temperature time series pattern feature encoding vector to obtain a resonator characteristic-temperature time domain fine-grained interaction response encoding vector; a temperature adaptive compensation factor determination unit, for determining a temperature adaptive compensation factor based on the resonator characteristic-temperature time domain fine-grained interaction response encoding vector; A temperature drift adaptive correction module, for performing an adaptive temperature drift correction on the initial etching depth value based on the temperature adaptive compensation factor to obtain the etched depth correction value; Wherein, the resonator characteristic-temperature time domain depth collaborative analysis unit includes: An etching chamber ambient temperature feature linear transformation sub-unit, for: performing principal component analysis on the etching chamber ambient temperature time series pattern feature encoding vector to obtain a set of etching chamber ambient temperature time series principal component feature encoding vectors; performing a linear transformation on each etching chamber ambient temperature time series principal component feature encoding vector in the set of etching chamber ambient temperature time series principal component feature encoding vectors to obtain a set of etching chamber ambient temperature time series principal component linear transformation feature encoding vectors, wherein each etching chamber ambient temperature time series principal component linear transformation feature encoding vector in the set of etching chamber ambient temperature time series principal component linear transformation feature encoding vectors has the same feature scale as the resonator self-characteristic semantic structured embedding encoding vector; A resonator characteristic-temperature time domain principal component inter-modal independence constraint analysis sub-unit, for performing an inter-modal independence constraint analysis on the set of etching chamber ambient temperature time series principal component linear transformation feature encoding vectors and the resonator self-characteristic semantic structured embedding encoding vector to determine a set of resonator characteristic-temperature time domain principal component inter-modal independence soft constraint factors; Resonator characteristic - temperature time - domain feature dynamic adaptive aggregation subunit, which is used to dynamically and adaptively aggregate the set of etched cavity environment temperature time - series principal component linear transformation feature encoding vectors and the semantic structured embedding encoding vector of the resonator's own characteristics based on the set of soft constraint factors for the independence between principal components of the resonator characteristic - temperature time - domain to obtain the resonator characteristic - temperature time - domain fine - grained interaction response encoding vector.

2. The resonator-based etch endpoint detection circuit according to claim 1, wherein The analyzer is used for: Calculating the change value of the resonant frequency of the resonator; Converting the change value of the resonant frequency into the initial value of the etching depth.

3. The resonator-based etch endpoint detection circuit according to claim 2, wherein The etched cavity environment temperature time - series pattern feature extraction unit is used for: passing the time - series of the etched cavity environment temperature through a temperature time - series feature extractor based on forward LSTM to obtain the etched cavity environment temperature time - series pattern feature encoding vector.

4. The resonator-based etch endpoint detection circuit according to claim 3, wherein The resonator characteristic - temperature time - domain principal component inter - modal independence constraint analysis subunit is used for: Inputting each etched cavity environment temperature time - series principal component linear transformation feature encoding vector in the set of the semantic structured embedding encoding vector of the resonator's own characteristics and the set of etched cavity environment temperature time - series principal component linear transformation feature encoding vectors into the inter - modal independence modeling unit to obtain a set of resonator characteristic - temperature time - domain principal component inter - modal independence encoding matrices; Performing canonical stability optimization on each resonator characteristic - temperature time - domain principal component inter - modal independence encoding matrix in the set of resonator characteristic - temperature time - domain principal component inter - modal independence encoding matrices to obtain a set of resonator characteristic - temperature time - domain principal component inter - modal independence canonical optimization encoding matrices, where the process of the canonical stability optimization is determined by the resonator characteristic - temperature time - domain eigenvectors composed of the eigenvalues of each resonator characteristic - temperature time - domain principal component inter - modal independence encoding matrix; Based on the set of resonator characteristic - temperature time - domain principal component inter - modal independence canonical optimization encoding matrices, calculating the set of soft constraint factors for the independence between principal components of the resonator characteristic - temperature time - domain.

5. The resonator-based etch endpoint detection circuit according to claim 4, wherein The resonator characteristic - temperature time - domain feature dynamic adaptive aggregation subunit is used for: Inputting each etched cavity environment temperature time - series principal component linear transformation feature encoding vector in the set of the semantic structured embedding encoding vector of the resonator's own characteristics and the set of etched cavity environment temperature time - series principal component linear transformation feature encoding vectors into the feature interaction response unit to obtain a set of resonator characteristic - temperature time - domain principal component inter - modal fine - grained response interaction encoding vectors; Based on the set of soft constraint factors for the independence between principal components of the resonator characteristic - temperature time - domain, dynamically and adaptively aggregating the set of resonator characteristic - temperature time - domain principal component inter - modal fine - grained response interaction encoding vectors to obtain the resonator characteristic - temperature time - domain fine - grained interaction response encoding vector.

6. The resonator-based etch endpoint detection circuit according to claim 5, characterized in that The temperature adaptive compensation factor determination unit is used for: passing the resonator characteristic - temperature time - domain fine - grained interaction response encoding vector through a temperature compensation factor calculator based on a decoder to obtain the temperature adaptive compensation factor decoding value.

Images