Deep silicon etching adaptive multi-modal end-point monitoring method, system, device and apparatus

Abstract

The application relates to the technical field of semiconductor detection, and discloses a deep silicon etching adaptive multi-modal endpoint monitoring method, system, device and equipment, which comprises the following steps: acquiring a plasma spectrum signal and a cumulative period number to determine an etching depth ratio; according to the etching depth ratio and a signal-to-noise ratio of the plasma spectrum signal, a current process stage is identified, and preset fusion weights corresponding to each mode are determined according to the process stage; the preset modes include a chaotic oscillator mode, a modulation depth mode and a spectral energy ratio mode; according to the plasma spectrum signal, endpoint confidence degrees corresponding to the preset modes are calculated in parallel; after weighted fusion according to the fusion weights, a comprehensive confidence degree is obtained, process state information is estimated, endpoint determination is carried out according to the comprehensive confidence degree and the process state information, and the application solves the problem that a single algorithm cannot adapt to non-monotonic changes of signal-to-noise ratios in a Bosch process whole process, thereby reducing endpoint detection reliability.

Description

Technical Field

[0001] This invention relates to the field of semiconductor testing technology, specifically to a method, system, device, and equipment for monitoring the endpoint of deep silicon etching adaptive multimodal processes. Background Technology

[0002] Through Silicon Via (TSV) deep silicon etching is a core process in 3D packaging, commonly employing the Bosch Deep Reactive Ion Etching process. This process achieves anisotropic etching by periodically switching between SF6 etching steps and C4F8 passivation steps at high frequency, with a typical cycle time of 50–120 ms. Due to the extremely low aperture ratio of TSVs (typically <0.5%, and as low as 0.1% in advanced packaging), the emission spectrum signal of the generated material in the plasma is extremely weak, resulting in a signal-to-noise ratio (SNR) typically below 5 dB or even close to 0 dB.

[0003] Existing endpoint detection technologies include intensity thresholding, modulation depth, spectral analysis, and chaotic oscillator methods. However, none of these methods consider the non-monotonic variation of signal-to-noise ratio with etching depth throughout the Bosch process. The signal-to-noise ratio is lowest in the initial stage due to the extremely low aperture ratio and the lack of stable etching trenches. The signal gradually increases in the middle stage and attenuates again in the final stage as it approaches the stop layer. A single algorithm cannot adapt to all three process stages simultaneously. Furthermore, existing technologies only output the endpoint signal and do not provide process information such as etching rate and remaining depth, making it impossible to form a closed-loop control with the etching equipment, leading to over-etching or under-etching. They also lack the ability to integrate and utilize multi-modal information, and existing technologies have never attempted to work multiple methods in tandem. Summary of the Invention

[0004] This invention provides an adaptive multimodal endpoint monitoring method, system, device, and equipment for deep silicon etching, to solve the problem that the existing technology lacks an adaptive multimodal collaborative monitoring scheme for the non-monotonic change of the signal-to-noise ratio throughout the Bosch process, resulting in low reliability of endpoint detection and inability to provide process status information for closed-loop adjustment of etching equipment.

[0005] In a first aspect, the present invention provides a deep silicon etching adaptive multimodal endpoint monitoring method for intelligent online semiconductor inspection, the method comprising: Acquire plasma spectral signals and cumulative cycle count, and determine the etching depth ratio based on the cumulative cycle count of the current etching process; Based on the etching depth ratio and the signal-to-noise ratio of the plasma spectral signal, the current process stage is identified, and the fusion weights corresponding to each preset mode are determined according to the process stage. The process stages include an initial stage, a middle stage, and a final stage. The preset modes include a chaotic oscillator mode, a modulation depth mode, and a spectral energy ratio mode. The initial stage is adapted to the chaotic oscillator mode, the middle stage is adapted to the modulation depth mode, and the final stage is adapted to the spectral energy ratio mode. Based on the plasma spectral signal, the endpoint confidence level corresponding to each preset mode is calculated in parallel. Based on the fusion weights, the endpoint confidence scores of each preset modality are weighted and fused to obtain the overall confidence score; The process status information of the current etching process is estimated in real time based on the plasma spectral signal. The endpoint is determined based on the overall confidence level and process status information; Output the endpoint signal and process status information for the etching equipment to perform closed-loop adjustments.

[0006] This invention provides an adaptive multimodal endpoint monitoring method for deep silicon etching. It determines the etching depth ratio by acquiring plasma spectral signals and cumulative cycle counts. After identifying the process stage based on the etching depth ratio and signal-to-noise ratio (SNR), it determines the fusion weights of the chaotic oscillator mode, modulation depth mode, and spectral energy ratio mode according to the stage. Parallel multimodal calculation and stage-adaptive weighted fusion solve the problem of low endpoint detection reliability caused by the inability of a single algorithm to adapt to the non-monotonic changes in SNR throughout the Bosch process. This method achieves full-process adaptive monitoring of non-monotonic SNR changes, significantly improving the accuracy and robustness of endpoint detection. Furthermore, by estimating the remaining etching depth in real time and jointly determining the endpoint with the comprehensive confidence level, it further reduces the risk of false positives and false negatives. By outputting the endpoint signal and process status information, the etching equipment can perform closed-loop adjustments based on this information, achieving linkage between monitoring and control. This solves the problem in existing technologies of lacking an adaptive multimodal collaborative monitoring scheme for non-monotonic changes in SNR throughout the Bosch process, which leads to low endpoint detection reliability and the inability to provide process status information for closed-loop adjustments of the etching equipment.

[0007] In one optional implementation, determining the etching depth ratio based on the cumulative number of cycles in the current etching process includes: Obtain the estimated total number of cycles; the estimated total number of cycles is determined based on the target etching depth and the theoretical etching rate. The ratio of the cumulative number of cycles to the estimated total number of cycles is calculated as the etching depth ratio.

[0008] In the above technical solution, by obtaining the estimated total number of cycles and calculating the ratio of the cumulative number of cycles to the estimated total number of cycles as the etching depth ratio, the progress of the etching process can be quantified. This provides an accurate and real-time quantitative basis for identifying the process stage based on the etching depth ratio. Moreover, the calculation method of this ratio is simple and efficient, without the need for additional hardware overhead.

[0009] In one alternative implementation, the method further includes: An online correction of the preset total number of cycles will be performed when all of the following conditions are met: The current etching process is in the mid-stage; The covariance of the current remaining depth estimate is less than a preset threshold; The estimated total number of cycles deviates from the current value by more than a preset percentage for several consecutive cycles.

[0010] In the above technical solution, by setting three conditions—the current etching process is in the middle stage, the covariance of the current remaining depth estimate is less than a preset threshold, and the estimated total number of cycles deviates from the current value by more than a preset percentage for several consecutive cycles—online correction of the estimated total number of cycles is only performed when all three conditions are met. This avoids erroneous corrections during unstable signal stages such as the early and late stages of etching. At the same time, the covariance threshold ensures that the depth estimate is reliable before triggering correction, and the continuous deviation judgment eliminates instantaneous fluctuation interference, thereby ensuring the accuracy and reliability of online correction.

[0011] In one optional implementation, performing an online correction of the preset total number of cycles includes: Based on the current estimated remaining etching depth and target etching depth, the estimated total number of cycles is calculated. The estimated values ​​are updated smoothly using an exponentially weighted moving average.

[0012] In the above technical solution, the estimated total number of cycles is calculated by back-calculating the remaining etching depth and the target etching depth, so that the correction of the estimated total number of cycles has a clear physical basis and avoids blind correction. At the same time, the estimated value is updated smoothly by using an exponentially weighted moving average, which eliminates the fluctuations that may exist in a single back-calculation value, makes the correction process converge smoothly, and improves the accuracy of the estimated total number of cycles.

[0013] In one alternative implementation, the signal-to-noise ratio of the plasma spectral signal is obtained in the following manner: The intensity of silicon fluoride spectral lines was extracted from the plasma spectral signal and normalized to obtain the normalized silicon fluoride spectral line intensity sequence. The noise floor is calculated based on the normalized spectral signal intensity sequence of silicon fluoride after the passivation step. The signal-to-noise ratio is determined by calculating the ratio of the standard deviation of the normalized silicon fluoride spectral signal intensity sequence to the noise floor.

[0014] In the above technical solution, by extracting and normalizing the fluoride spectral line signal intensity from the plasma spectral signal, the influence of optical window contamination and source power drift on the signal is eliminated, thus improving signal quality. By calculating the noise floor based on the normalized fluoride spectral line signal intensity sequence after the passivation step, the acquisition of the noise floor is synchronized with the current process state, accurately reflecting the actual noise level. The signal-to-noise ratio is determined by calculating the ratio of the standard deviation of the normalized fluoride spectral line signal intensity sequence to the noise floor, providing an accurate quantitative basis for subsequent stage identification. At the same time, the calculation method is simple and efficient.

[0015] In one alternative implementation, the current process stage is identified based on the etching depth ratio and the signal-to-noise ratio of the plasma spectral signal, including: When the etching depth ratio is less than a preset first ratio threshold and / or the signal-to-noise ratio is lower than a preset first threshold, it is identified as the initial stage; When the etching depth ratio is not less than a preset first ratio threshold and not greater than a preset second ratio threshold, and the signal-to-noise ratio is not less than a preset first threshold, it is identified as the intermediate stage; When the etching depth ratio is greater than the preset second ratio threshold, or when the signal-to-noise ratio continuously decreases, it is identified as the final stage.

[0016] In the above technical solution, the process stage is identified by combining two parameters: etching depth ratio and signal-to-noise ratio (SNR). This method is more accurate and reliable than using a single parameter. In the initial stage, the process is identified by the etching depth ratio being less than a threshold or the SNR being lower than a threshold, ensuring that the appropriate chaotic oscillator mode can be triggered in a timely manner under extremely low SNR conditions. In the final stage, in addition to the etching depth ratio exceeding the threshold, the process is also identified by the continuous decrease in SNR, covering the situation where the etching depth ratio has not yet exceeded the threshold but the signal has already begun to decay. The three stages of initial, middle, and final stages are completely divided, forming a seamless coverage of the entire process and providing a reliable basis for determining the appropriate main mode for each subsequent stage.

[0017] In one optional implementation, the fusion weights corresponding to each preset mode are determined according to the process stage, including: Based on the adaptation relationship between the process stage and the preset modes, the fusion weight of each preset mode under the process stage is determined.

[0018] In the above technical solution, by establishing an adaptation relationship that adapts to the chaotic oscillator mode in the initial stage, the modulation depth mode in the middle stage, and the spectral energy ratio mode in the final stage, and determining the fusion weight of each mode in the corresponding process stage based on the adaptation relationship, the adaptive switching of the main mode with the process stage is realized, ensuring that there is an adapted detection mode as the main mode in each stage, and avoiding the problem that a single algorithm fails due to changes in signal-to-noise ratio throughout the entire process.

[0019] In one optional implementation, the fusion weight of each mode under each process stage is determined based on the adaptation relationship between the process stage and each mode, including: Based on the adaptation relationship, the weight vector corresponding to the process stage is found from the preset stage weight mapping relationship; the preset stage weight mapping relationship stores the correspondence between each process stage and the preset fusion weight of each mode.

[0020] In the above technical solution, the correspondence between each process stage and the preset fusion weights of each modality is pre-stored as a stage weight mapping relationship. During actual operation, the weight vector corresponding to the current process stage can be directly found from the mapping relationship according to the adaptation relationship. There is no need to calculate the weight value of each modality online. The fusion weight of each modality in the corresponding process stage can be quickly obtained by searching, which simplifies the weight acquisition process and reduces the online calculation overhead.

[0021] In one optional implementation, based on the plasma spectral signal, the endpoint confidence levels corresponding to each preset mode are calculated in parallel, including: The spectral line signal of silicon fluoride is used as the external excitation of the nonlinear oscillator, and the endpoint confidence of the chaotic oscillator mode is calculated based on the state change of the nonlinear oscillator. The first average value of the silicon fluoride spectral line signal intensity in the etching step and the second average value of the silicon fluoride spectral line signal intensity in the passivation step are obtained. The ratio between the difference between the first average value and the second average value and the sum value is calculated as the modulation depth. The endpoint confidence of the modulation depth mode is calculated based on the modulation depth. The power spectrum is obtained by performing frequency domain transformation on the spectral line signal of silicon fluoride. The fundamental frequency power and noise power are extracted from the power spectrum. The ratio of the fundamental frequency power to the noise power is calculated, and the endpoint confidence of the spectral energy ratio mode is calculated based on the ratio.

[0022] In the above technical solution, the endpoint confidence of the chaotic oscillator mode is calculated by inputting the silicon fluoride spectral line signal into the nonlinear oscillator, the modulation depth is obtained by calculating the ratio of the difference and the sum of the average values ​​of the spectral line intensities of the etching step and the passivation step, and the endpoint confidence of the modulation depth mode is calculated accordingly. The endpoint confidence of the spectral energy ratio mode is calculated by performing frequency domain transformation on the silicon fluoride spectral line signal to extract the ratio of the fundamental frequency power to the noise power. The three modes based on different principles are calculated in parallel based on the same signal source, without dependence on or delay of each other. Compared with serial calculation, the overall calculation time is significantly shortened, ensuring the synchronization of the three mode outputs in time sequence, and providing timely and complete confidence input for subsequent weighted fusion.

[0023] In one optional implementation, the endpoint confidence scores of each preset modality are weighted and fused according to the fusion weights to obtain a comprehensive confidence score, including: The sum of the products of the endpoint confidence scores for each preset modality and their corresponding fusion weights is used as the overall confidence score.

[0024] In the above technical solution, the sum of the products of the endpoint confidence of each preset multimodal mode and its corresponding fusion weight is used as the comprehensive confidence. The detection results of three different principles, namely chaotic oscillator mode, modulation depth mode and spectral energy ratio mode, are integrated into a unified and quantitative comprehensive index. This weighted summation method is simple, efficient and computationally inefficient, and provides a clear and single decision basis for subsequent endpoint determination, avoiding the dilemma that multimodal outputs cannot be directly compared or comprehensively judged.

[0025] In one optional implementation, the process status information includes the remaining etching depth; the process status information of the current etching process is estimated in real time based on the plasma spectral signal, including: Based on the historical sequence of modulation depth, an empirical model between etching depth and modulation depth is fitted. Estimate the current etching depth based on the current modulation depth and the empirical model; The remaining etching depth is determined based on the difference between the target etching depth and the current etching depth.

[0026] In the above technical solution, by fitting an empirical model between the etching depth and the modulation depth based on the historical sequence of modulation depth, and estimating the current etching depth based on the current modulation depth and the empirical model, and then determining the remaining etching depth based on the difference between the target etching depth and the current etching depth, the remaining etching depth can be estimated online based solely on the spectral signal without relying on external measurement equipment. At the same time, the endpoint determination is expanded from a simple signal threshold judgment to a comprehensive judgment that combines physical depth estimation, providing an independent verification dimension for endpoint determination.

[0027] In one optional implementation, the endpoint determination is based on a comprehensive confidence level and process status information, including: When the overall confidence level exceeds the first confidence level threshold and the remaining etching depth is less than the preset depth threshold, and after a preset number of cycles, the current etching process is determined to have reached the etching endpoint.

[0028] In the above technical solution, the etching endpoint is determined only after two conditions are met simultaneously, namely, the comprehensive confidence level exceeds the first confidence level threshold and the remaining etching depth is less than the preset depth threshold, and this condition is maintained for a preset number of cycles. Compared with a single criterion, the dual-condition joint determination significantly improves the reliability of the endpoint determination. The requirement of maintaining a preset number of cycles effectively avoids false triggering caused by instantaneous signal fluctuations, reducing the risk of misjudgment and missed judgment.

[0029] In one optional implementation, the output endpoint signal and process status information include: During the etching process and / or when the etching endpoint is reached, an endpoint signal and process status information are output, including etching rate and etching uniformity index.

[0030] In the above technical solution, the endpoint signal provides a clear etching stop instruction to the etching equipment. At the same time, by outputting process status information including the remaining etching depth, etching rate and etching uniformity index, it provides multi-dimensional process status information to the operator or the host system, so that the changes in the entire etching process can be grasped while obtaining the endpoint determination result.

[0031] Secondly, this invention provides a deep silicon etching adaptive multimodal endpoint monitoring system for intelligent online semiconductor inspection, the system comprising: A spectrometer is used to receive plasma spectral signals; The synchronous signal isolation circuit is used to receive the gas switching trigger signal of the etching equipment and perform cycle counting based on the trigger signal to obtain the cumulative number of cycles; An industrial control computer, connected to a spectrometer and a synchronization signal isolation circuit respectively, is used to execute the deep silicon etching adaptive multimodal endpoint monitoring method of the first aspect or any corresponding embodiment described above.

[0032] In one alternative implementation, the industrial control computer includes a field-programmable gate array (FPGA) accelerator card, which is used to calculate in parallel the endpoint confidence scores corresponding to each preset mode based on the plasma spectral signal.

[0033] Thirdly, the present invention provides a deep silicon etching adaptive multimodal endpoint monitoring device for intelligent online semiconductor inspection, and is applied to the deep silicon etching adaptive multimodal endpoint monitoring method of the first aspect above or any corresponding embodiment thereof. The device includes: The signal acquisition and synchronization module is used to acquire plasma spectral signals and cumulative cycle count, and determine the etching depth ratio based on the cumulative cycle count of the current etching process. The stage identification and weight allocation module is used to identify the current process stage based on the etching depth ratio and the signal-to-noise ratio of the plasma spectral signal, and to determine the fusion weights corresponding to each preset mode according to the process stage. The process stages include an initial stage, a middle stage, and a final stage. The preset modes include a chaotic oscillator mode, a modulation depth mode, and a spectral energy ratio mode. The initial stage is adapted to the chaotic oscillator mode, the middle stage is adapted to the modulation depth mode, and the final stage is adapted to the spectral energy ratio mode. The multimodal calculation module is used to calculate the endpoint confidence level corresponding to each preset mode in parallel based on the plasma spectral signal. The fusion decision module is used to perform weighted fusion of the endpoint confidence scores of each preset modality according to the fusion weights to obtain the comprehensive confidence score; The state estimation module is used to estimate the process state information of the current etching process in real time based on the plasma spectral signal; The endpoint determination module is used to determine the endpoint based on the comprehensive confidence level and process status information. The output module is used to output the endpoint signal and process status information for the etching equipment to perform closed-loop adjustments.

[0034] Fourthly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the deep silicon etching adaptive multimodal endpoint monitoring method of the first aspect or any corresponding embodiment described above.

[0035] Fifthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the deep silicon etching adaptive multimodal endpoint monitoring method of the first aspect or any corresponding embodiment described above.

[0036] In a sixth aspect, the present invention provides a computer program product, including computer instructions for causing a computer to execute the deep silicon etching adaptive multimodal endpoint monitoring method of the first aspect or any corresponding embodiment described above.

[0037] This invention addresses the lack of an adaptive multimodal collaborative monitoring scheme for the non-monotonic change of the signal-to-noise ratio throughout the Bosch process in existing technologies, which leads to low reliability of endpoint detection and the inability to provide process status information for closed-loop adjustment of etching equipment. It provides a deep silicon etching adaptive multimodal endpoint monitoring method. This method acquires the plasma spectral signal and the cumulative cycle count, and determines the etching depth ratio based on the cumulative cycle count. Based on the etching depth ratio and the signal-to-noise ratio of the plasma spectral signal, it identifies the current process stage (early, middle, or late stage) and determines the fusion weights for each preset mode based on the identified process stage. The preset modes include chaotic oscillator mode, modulation depth mode, and spectral energy ratio mode. Based on the plasma spectral signal, the endpoint confidence of each preset mode is calculated in parallel. Based on the fusion weights, the endpoint confidence of each preset mode is weighted and fused to obtain a comprehensive endpoint confidence. The system employs a real-time estimation of remaining etching depth and endpoint determination based on comprehensive confidence level and remaining etching depth. This technology achieves a false positive rate of <0.5% and a detection delay of <5 cycles under conditions of aperture ratio ≤0.5% and signal-to-noise ratio as low as -3dB, while providing online estimation of etching rate and remaining depth. It utilizes chaotic oscillator modes to address the extremely low signal-to-noise ratio in the initial stage, modulation depth modes to address signal fluctuations in the middle stage, and spectral energy ratio modes to address signal attenuation in the final stage, enabling adaptive switching and smooth transition between the initial, middle, and final process stages. Through multi-modal parallel computing and dynamic weighted fusion, it overcomes the limitation of a single algorithm in adapting to changes in signal characteristics throughout the entire process. By estimating the remaining etching depth in real-time and jointly determining the endpoint with comprehensive confidence level, it avoids over-etching or under-etching. Finally, it outputs the endpoint signal and process monitoring data, achieving deep linkage and closed-loop feedback with the etching equipment. Attached Figure Description

[0038] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0039] Figure 1 This is a schematic diagram of the deep silicon etching adaptive multimodal endpoint monitoring system according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the first process of the deep silicon etching adaptive multimodal endpoint monitoring method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the second process of the deep silicon etching adaptive multimodal endpoint monitoring method according to an embodiment of the present invention; Figure 4This is a schematic diagram of the third process of the deep silicon etching adaptive multimodal endpoint monitoring method according to an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the running results of the deep silicon etching adaptive multimodal endpoint monitoring method according to an embodiment of the present invention; Figure 6 This is a structural block diagram of the deep silicon etching adaptive multimodal endpoint monitoring device according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0042] 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 number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0043] As an optional application scenario of this invention, such as Figure 1 As shown, this invention provides an adaptive multimodal endpoint monitoring system for deep silicon etching, connected to etching equipment 150. The system includes a spectrometer 110, a synchronization signal isolation circuit 120, and an industrial control computer 130. The spectrometer receives plasma spectral signals; the synchronization signal isolation circuit receives gas switching trigger signals from the etching equipment and performs period counting based on the trigger signals to obtain the cumulative number of periods; the industrial control computer, connected to both the spectrometer and the synchronization signal isolation circuit, executes the adaptive multimodal endpoint monitoring method for deep silicon etching. The system also includes an optical probe 140, connected to the spectrometer 110, for collecting plasma emission spectral signals and inputting them to the spectrometer.

[0044] Specifically, the spectrometer 110 is an acousto-optic tunable filter-optical emission spectrometer with a sampling frequency of not less than 1 kHz. The spectrometer 110 acquires the spectral intensities of SiF (Silicon Monofluoride, wavelength 440 nm) and Ar (Argon, wavelength 750 nm) at a sampling rate of ≥1 kHz. The synchronization signal isolation circuit is also used to receive the gas switching TTL signal (Transistor-Transistor Logic, synchronization trigger signal) from the etching device 150 for period alignment and stage counting.

[0045] The industrial control computer 130 includes a stage identification and signal-to-noise ratio (SNR) evaluation calculation module, which is used to normalize drift reduction (SiF / Ar ratio), calculate etching depth ratio (cumulative number of cycles / estimated total number of cycles), calculate real-time SNR (signal standard deviation / noise floor), and identify process stages (early / mid / late stage).

[0046] The multimodal algorithm engine module is implemented using a field-programmable gate array (FPGA) accelerator card. This FPGA accelerator card is used to calculate the endpoint confidence scores for each preset mode in parallel based on the plasma spectral signal. The preset modes include chaotic oscillator modes, modulation depth modes, and spectral energy ratio modes.

[0047] The industrial control computer 130 also includes: Process monitoring and state estimation module: Implemented by an embedded CPU. Based on the modulation depth history sequence, recursive least squares is used to fit the depth-modulation relationship, and Kalman filtering is used to estimate the etching rate and remaining etching depth in real time.

[0048] Fusion Decision and Communication Module: Dynamically allocates the weights of the three algorithms according to the current stage, calculates the comprehensive confidence level, jointly estimates the remaining etching depth, determines the endpoint, outputs the endpoint signal through digital I / O, and reports process monitoring parameters through EtherCAT (Ethernet for Control Automation Technology).

[0049] The system also includes an industrial bus communication module, which connects to an industrial control computer to send endpoint signals and process monitoring data to the host computer. The industrial bus communication module supports either the EtherCAT or Profinet protocol.

[0050] The workflow of a deep silicon etching adaptive multimodal endpoint monitoring system provided in this embodiment of the invention is as follows: Spectrometer 110 acquires plasma spectral signals in real time; synchronous signal isolation circuit 120 synchronously receives gas switching trigger signals from the etching equipment and performs cycle counting to obtain the cumulative cycle number; industrial control computer 130 identifies the current process stage based on the cumulative cycle number and the signal-to-noise ratio of the spectral signal, and determines the fusion weights of the chaotic oscillator mode, modulation depth mode, and spectral energy ratio mode based on the stage; field-programmable gate array (FPGA) accelerator card calculates the endpoint confidence of the three modes in parallel based on the spectral signal; industrial control computer performs weighted fusion of the confidence of each mode based on the fusion weights to obtain the comprehensive confidence; simultaneously, industrial control computer estimates the remaining etching depth in real time based on the historical modulation depth sequence, and jointly determines the endpoint based on the comprehensive confidence and the remaining etching depth; finally, when the endpoint is determined, industrial control computer outputs the endpoint signal through a digital input / output interface and reports process monitoring data through an industrial bus communication module.

[0051] According to an embodiment of the present invention, an embodiment of an adaptive multimodal endpoint monitoring method for deep silicon etching is provided. It should be noted that, in the appendix... Figure 2 To be continued Figure 4 The steps shown in the flowchart can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0052] This embodiment provides a deep silicon etching adaptive multimodal endpoint monitoring method, which can be used in the industrial control computer of the aforementioned deep silicon etching adaptive multimodal endpoint monitoring system. Figure 2 This is a flowchart of the deep silicon etching adaptive multimodal endpoint monitoring method according to an embodiment of the present invention, as follows: Figure 2 As shown, the process includes the following steps: Step S201: Obtain the plasma spectral signal and the cumulative number of cycles, and determine the etching depth ratio based on the cumulative number of cycles in the current etching process.

[0053] Plasma spectral signal refers to the light signal with characteristic wavelengths collected by the optical probe from the plasma etching chamber, which is released by excited-state particles (such as SiF, Ar, etc.) in the plasma when they transition back to the ground state. The signal is then output as an electrical signal after being split and converted by a spectrometer.

[0054] The cumulative cycle count refers to the number of times the gas switching TTL trigger signal has been received by the synchronous signal isolation circuit since the start of the current etching process, which is the total number of etching-passivation cycles that the Bosch process has performed.

[0055] The etching depth ratio refers to the degree of completion of the current etching process. It is calculated by the industrial control computer based on the cumulative number of cycles provided by the synchronous signal isolation circuit and is used to quantify the current etching progress.

[0056] Specifically, the spectrometer collects plasma spectral signals, the synchronous signal isolation circuit counts the TTL trigger signals of the etching equipment to obtain the cumulative number of cycles, and the industrial control computer determines the etching depth ratio based on the ratio of the cumulative number of cycles to the estimated total number of cycles. This ratio is used to quantify the progress of the current etching process and provide a quantitative basis for the progress dimension for subsequent stage identification.

[0057] For example, firstly, based on the TTL trigger signal, data is collected within each cycle. Moment of silicon fluoride spectral line signal intensity and Argon spectral line signal intensity at time 1 With a sampling rate of 1kHz, the industrial control computer receives the signal intensity values ​​at each of the above times and organizes them into a time-domain signal sequence with time as the horizontal axis and signal intensity as the vertical axis, thus forming a plasma spectral signal for subsequent processing.

[0058] The industrial control computer collects data from the spectrometer. real-time division The purpose is to eliminate the influence of common-mode interference such as optical window contamination and source power drift on the measurement results, and to ensure that the signal changes calculated in subsequent calculations truly reflect the changes in the etching reaction itself.

[0059] That is, through the dynamic baseline subtraction formula: This eliminates window contamination and source power drift, ensuring that subsequent signal changes accurately reflect the changes in the etching reaction itself. for The intensity of the silicon fluoride spectral line signal after time normalization.

[0060] Step S202: Based on the etching depth ratio and the signal-to-noise ratio of the plasma spectral signal, identify the current process stage and determine the fusion weights corresponding to each preset mode according to the process stage; the process stage includes an initial stage, a middle stage, and a final stage; the preset modes include chaotic oscillator mode, modulation depth mode, and spectral energy ratio mode; the initial stage is adapted to the chaotic oscillator mode, the middle stage is adapted to the modulation depth mode, and the final stage is adapted to the spectral energy ratio mode.

[0061] The signal-to-noise ratio (SNR) of the plasma spectral signal refers to the ratio of the standard deviation of the normalized silicon fluoride spectral signal intensity sequence to the noise floor, which is calculated by an industrial control computer and used to quantify the effective intensity of the current spectral signal relative to the background noise.

[0062] The current process stage refers to the stage of the current etching process as identified by the industrial control computer based on the etching depth ratio and signal-to-noise ratio, including the initial stage, intermediate stage, and final stage.

[0063] The preset fusion weights for each mode refer to the coefficients assigned by the industrial control computer to the chaotic oscillator mode, modulation depth mode, and spectral energy ratio mode according to the current process stage. These coefficients are used to perform a weighted summation of the endpoint confidence of each mode to reflect the importance of each mode at different stages.

[0064] Specifically, the industrial control computer comprehensively identifies the current process stage based on the etching depth ratio and signal-to-noise ratio, and determines the fusion weights of the chaotic oscillator mode, modulation depth mode, and spectral energy ratio mode from the preset weight mapping relationship according to the identified stage; among them, the chaotic oscillator mode has the highest weight in the initial stage, the modulation depth mode has the highest weight in the middle stage, and the spectral energy ratio mode has the highest weight in the final stage, so as to realize the adaptive switching of the main mode according to the process stage.

[0065] Step S203: Based on the plasma spectral signal, calculate the endpoint confidence level corresponding to each preset mode in parallel.

[0066] Specifically, this step is implemented by an FPGA accelerator in an industrial control computer. The current process stage (early / mid / late stage) is determined based on the cumulative number of cycles and the real-time signal-to-noise ratio. Three modes are run in parallel: chaotic oscillator, modulation depth, and spectral energy ratio. The endpoint confidence level corresponding to each mode is output.

[0067] Step S204: Based on the fusion weights, the endpoint confidence scores of each preset modality are weighted and fused to obtain the comprehensive confidence score.

[0068] Specifically, the industrial control computer multiplies the endpoint confidence scores of the chaotic oscillator mode, modulation depth mode, and spectral energy ratio mode by their respective weights according to the fusion weights corresponding to the current process stage, and then sums them to obtain a comprehensive confidence score, which serves as a unified quantitative basis for subsequent endpoint determination.

[0069] Step S205: Estimate the process status information of the current etching process in real time based on the plasma spectral signal.

[0070] Specifically, the industrial control computer 130 extracts characteristic parameters related to the etching process from the normalized silicon fluoride spectral signal, and estimates the process status information of the current etching process in real time based on the historical change trend of these characteristic parameters, providing an independent verification dimension for endpoint determination.

[0071] Step S206: Determine the endpoint based on the comprehensive confidence level and process status information.

[0072] Specifically, the industrial control computer fits an empirical model between the etching depth and the modulation depth based on the historical sequence of modulation depth, estimates the current etching depth based on the current modulation depth and the empirical model, and determines the remaining etching depth based on the difference between the target etching depth and the current etching depth. Then, it determines whether the overall confidence level exceeds the first confidence level threshold and whether the remaining etching depth is less than the depth threshold. When both conditions are met simultaneously and continue for a preset number of cycles, it is determined that the current etching process has reached the etching endpoint.

[0073] Step S207: Output the endpoint signal and process status information for the etching equipment to perform closed-loop adjustment.

[0074] Specifically, when the industrial control computer 130 determines that the etching endpoint has been reached, it outputs an endpoint signal to the etching equipment through a digital input / output interface, and at the same time outputs process status information to the etching equipment through an industrial bus communication module, so that the etching equipment can adjust process parameters or stop etching based on the information, thereby realizing the linkage between the monitoring system and the etching equipment.

[0075] The deep silicon etching adaptive multimodal endpoint monitoring method provided in this embodiment determines the etching depth ratio by acquiring plasma spectral signals and cumulative cycle counts. After identifying the process stage based on the etching depth ratio and signal-to-noise ratio (SNR), it determines the fusion weights of the chaotic oscillator mode, modulation depth mode, and spectral energy ratio mode according to the stage. Parallel multimodal calculation and stage-adaptive weighted fusion solve the problem that a single algorithm cannot adapt to the non-monotonic changes in SNR throughout the Bosch process, resulting in low endpoint detection reliability. It achieves full-process adaptive monitoring of non-monotonic SNR changes, significantly improving the accuracy and robustness of endpoint detection. At the same time, by estimating the remaining etching depth in real time and jointly determining the endpoint with the comprehensive confidence level, it further reduces the risk of false positives and false negatives. By outputting the endpoint signal and process status information, the etching equipment can make closed-loop adjustments based on this information, realizing the linkage between monitoring and control. This solves the problem in the prior art that there is a lack of adaptive multimodal collaborative monitoring schemes for non-monotonic changes in SNR throughout the Bosch process, resulting in low endpoint detection reliability and the inability to provide process status information for closed-loop adjustments of the etching equipment.

[0076] This embodiment provides a deep silicon etching adaptive multimodal endpoint monitoring method, which can be used in the industrial control computer of the aforementioned deep silicon etching adaptive multimodal endpoint monitoring system. Figure 3 This is a flowchart of the deep silicon etching adaptive multimodal endpoint monitoring method according to an embodiment of the present invention, as follows: Figure 3 As shown, the process includes the following steps: Step S301: Obtain the plasma spectral signal and the cumulative number of cycles, and determine the etching depth ratio based on the cumulative number of cycles in the current etching process.

[0077] Specifically, step S301 includes: Step a, obtain the estimated total number of cycles; the estimated total number of cycles is determined based on the target etching depth and the theoretical etching rate; calculate the ratio of the cumulative number of cycles to the estimated total number of cycles as the etching depth ratio.

[0078] The estimated total number of cycles refers to the total number of Bosch process cycles expected to be required from the start of the current etching process to the end of the etching process. It is estimated in advance by the industrial control computer based on the target etching depth and theoretical etching rate, and is continuously updated through online correction during the etching process.

[0079] Specifically, based on the cumulative number of cycles and preset total number of cycles The etching depth ratio is calculated using the following formula: (1); in, This indicates the percentage of etching depth.

[0080] In one alternative implementation, an online correction of the preset total number of cycles is performed when all of the following conditions are met: the current etching process is in the intermediate stage; the covariance of the current remaining depth estimate is less than a preset threshold; and the estimated total number of cycles deviates from the current value by more than a preset percentage for several consecutive cycles.

[0081] Specifically, the current etching process is in the middle stage, that is... Multiple consecutive cycles can be set to 5 or 10 consecutive cycles.

[0082] The estimated total number of cycles is set as follows: .

[0083] 5 or 10 consecutive cycles An update is triggered only when the value deviates from the current value by more than 5% (a preset percentage).

[0084] In one optional implementation, performing an online correction of the preset total number of cycles includes: Based on the current estimated remaining etching depth and target etching depth, the estimated total number of cycles is calculated; the estimated value is then updated smoothly using an exponentially weighted moving average.

[0085] Specifically, To estimate the total number of cycles, which is the total number of Bosch process cycles expected from the start to the end of etching. This is due to factors such as different wafer batches, variations in cavity conditions, and etching rate drift. Online adjustments are required to adapt to changes in the actual process. The specific adjustment method is as follows: (1) Initial value setting: During the process stabilization period, the average value is calculated using historical data (the number of etching endpoint cycles for the same product and the same machine) as... The initial value.

[0086] If historical data is unavailable, the target etching depth can be used as a reference. and theoretical etching rate (Based on the process formulation) Estimate: (2); in, This refers to the Bosch process cycle time. This is the initial value for estimating the total number of cycles.

[0087] (2) Online correction method (feedback correction based on endpoint prediction): At the end of each Bosch cycle, the remaining etching depth is estimated using the current estimate. (Given by Recursive Least Squares (RLS) + Kalman filter) The estimated total number of cycles is calculated by back-calculation, as shown in the following formula: (3); in, The target etching depth is known (process parameters are known).

[0088] The Exponentially Weighted Moving Average (EWMA) is used for smooth updates to avoid excessive fluctuations in a single estimate. (4); in, The learning rate is typically between 0.1 and 0.3. This is the updated estimated total number of cycles. This is the estimated total number of cycles before the update.

[0089] The parameter ranges for the above-mentioned exponentially weighted moving average learning rate, trigger threshold, and minimum update interval are shown in Table 1 below: Table 1 Parameter Range

[0090] Step S302: Based on the etching depth ratio and the signal-to-noise ratio of the plasma spectral signal, identify the current process stage and determine the fusion weights corresponding to each preset mode according to the process stage; the process stage includes an initial stage, a middle stage, and a final stage; the preset modes include chaotic oscillator mode, modulation depth mode, and spectral energy ratio mode; the initial stage is adapted to the chaotic oscillator mode, the middle stage is adapted to the modulation depth mode, and the final stage is adapted to the spectral energy ratio mode.

[0091] In one alternative implementation, the signal-to-noise ratio of the plasma spectral signal is obtained in the following manner: The intensity of silicon fluoride spectral lines is extracted from the plasma spectral signal and normalized to obtain a normalized silicon fluoride spectral line intensity sequence. Based on the normalized silicon fluoride spectral line intensity sequence after the passivation step, the noise floor is calculated. The ratio of the standard deviation of the normalized silicon fluoride spectral line intensity sequence to the noise floor is calculated to determine the signal-to-noise ratio.

[0092] Specifically, the formula for calculating the signal-to-noise ratio is as follows: (5); in, The standard deviation of the strength of the last quarter segment after the C4F8 passivation step. For signal-to-noise ratio, For normalization.

[0093] Specifically, the process stages include an initial stage, an intermediate stage, and a final stage; step S302 above includes: Step S3021: When the etching depth ratio is less than a preset first ratio threshold and / or the signal-to-noise ratio is lower than a preset first threshold, it is identified as the initial stage; when the etching depth ratio is not less than a preset first ratio threshold and not greater than a preset second ratio threshold, and the signal-to-noise ratio is not lower than a preset first threshold, it is identified as the intermediate stage; when the etching depth ratio is greater than a preset second ratio threshold, or when the signal-to-noise ratio continuously decreases, it is identified as the final stage.

[0094] Specifically, the preset first ratio threshold is 0.2, the preset second ratio threshold is 0.8, and the preset first threshold is... 。

[0095] The identification process includes: Initial stage: or The main mode is the chaotic oscillator mode, and the chaotic oscillator algorithm is adopted.

[0096] Mid-term stage: and The dominant mode is the modulation depth mode, and the modulation depth algorithm is used.

[0097] Final stage: or It begins to decline, and the dominant mode is the spectrum energy ratio mode, using the spectrum energy ratio algorithm.

[0098] Step S3022: Determine the fusion weight of each preset mode under the process stage based on the adaptation relationship between the process stage and each preset mode.

[0099] In some optional implementations, step S3022 above includes: Based on the adaptation relationship, the weight vector corresponding to the process stage is found from the preset stage weight mapping relationship; the preset stage weight mapping relationship stores the correspondence between each process stage and the preset fusion weight of each mode.

[0100] Specifically, the industrial control computer, based on the identified current process stage, searches for the weight vector corresponding to the current stage from the pre-stored stage weight mapping relationship, which contains the correspondence between each process stage and each mode fusion weight, thereby directly obtaining the fusion weights of the chaotic oscillator mode, modulation depth mode, and spectral energy ratio mode.

[0101] For example, the industrial control computer reads a preset stage weight mapping relationship from memory based on the identified current process stage (e.g., intermediate stage). In this mapping relationship, the initial stage corresponds to a weight vector of [0.6, 0.3, 0.1], the intermediate stage to [0.2, 0.6, 0.2], and the final stage to [0.1, 0.3, 0.6]. By matching the current stage with the stage identifier in the mapping relationship, the weight vector corresponding to that stage is found and extracted, thereby directly obtaining the fusion weights for the chaotic oscillator mode, modulation depth mode, and spectral energy ratio mode, which are then used for subsequent weighted fusion calculations. This mapping relationship is configured and stored before system operation, and does not need to be recalculated during operation; weight allocation can be completed simply by looking up the stage weight mapping relationship.

[0102] Step S303: Based on the plasma spectral signal, calculate the endpoint confidence level corresponding to each preset mode in parallel.

[0103] Specifically, step S303 includes: Step S3031: Use the silicon fluoride spectral line signal as the external excitation of the nonlinear oscillator, and calculate the endpoint confidence of the chaotic oscillator mode based on the state change of the nonlinear oscillator.

[0104] Specifically, the Duffing oscillator (nonlinear oscillator) equation is used to process the DC-free... The signal (Silicon Monofluoride emission spectrum intensity) is used as an external excitation to calculate the variance of the Poincaré section sliding window in real time, and outputs the period / chaotic state and confidence level. .

[0105] Among them, the endpoint confidence of the chaotic oscillator mode Calculated in the following way: Subtracting the arithmetic mean from the normalized silicon fluoride spectral signal intensity sequence yields the applied excitation signal after DC removal. ; External excitation signal Input the Duffing oscillator equations to a state of chaotic criticality; The Duffing oscillator equation is as follows: (6); Within each Bosch process cycle, record the phase space points of the Duffing oscillator equation on the Poincaré section, and calculate the sliding window variance of the Euclidean distance between consecutive points; When the sliding window variance jumps from greater than 0.02 to less than 0.01, it is determined that the Duffing oscillator has transitioned from a chaotic state to a large-scale periodic state. Based on the position of the sliding window variance relative to the interval from 0.02 to 0.01, the sigmoid function is used to map the sliding window variance to a continuous value between 0 and 1, which is then used as the endpoint confidence score for the chaotic oscillator mode. .

[0106] For example: The Duffing oscillator is a nonlinear dynamic system that is highly sensitive to weak periodic signals of a specific frequency. When no external signal is applied, its parameters are adjusted to make the oscillator operate in a chaotic critical state. At this time, the phase trajectory of the oscillator exhibits chaotic characteristics, and the landing points on the Poincaré section are dispersed (in this case, the variance of the sliding window is greater than 0.02).

[0107] In the early stages of etching, the plasma spectral signal contains a weak periodic component with the same frequency as the Bosch period. This component acts as an external excitation to drive the Duffing oscillator, causing it to transition from a chaotic state to a large-scale periodic state. The variance of the sliding window of the Poincaré section jumps from greater than 0.02 to less than 0.01. At this time, the confidence level of the chaotic oscillator mode output is 1 (representing "not reaching the endpoint").

[0108] As the actual etching depth increases, this confidence level is reflected as a continuous change from 1 to 0, representing the progress of the etching process.

[0109] When the etching reaches its endpoint, the ion flow disappears, the periodic component in the signal disappears, the Duffing oscillator jumps from a large-scale periodic state back to a chaotic state, and the variance of the Poincaré section sliding window jumps from less than 0.01 back to greater than 0.02. At this point, the endpoint confidence of the chaotic oscillator mode output... A value of 0 (representing "reaching the destination") is used by the system to determine the destination and output the destination signal.

[0110] In the above formula (6), The damping coefficient controls the energy dissipation rate of the oscillator, affecting the duration of the chaotic state and the sharpness of the transition between the periodic and chaotic states. The selection principle is as follows: The smaller the value, the higher the sensitivity of the oscillator to weak signals, but the stability of the chaotic state decreases. The higher the value, the more stable the system, but the lower the sensitivity. The value range is 0.3~0.6, typical value This value was calibrated through numerical simulation: under conditions of no external excitation. Under the conditions, adjust This causes the oscillator to exhibit clear chaotic / periodic transitions near the critical driving force. (And...) Signal frequency matching: It does not directly depend on the signal frequency, but it affects the response bandwidth of the oscillator. The effective response bandwidth of the time oscillator covers 5~50Hz, covering the fundamental frequency range of Bosch technology (8.3~20Hz).

[0111] The nonlinear restoring force term determines the double potential well structure of the oscillator. exist A stronghold was formed there, in A potential well is formed at the point where the oscillator can bounce between the two wells. The physical meaning is that the double potential well structure makes the oscillator produce a critical value effect for small signals. It does not respond below the critical value and transitions rapidly above the critical value. This is the principle basis for the chaotic oscillator to detect weak periodic signals. The nonlinear restoring force term is in the form of the standard Duffing oscillator equation and requires no adjustment.

[0112] The amplitude of the periodic driving force characterizes the magnitude of the built-in periodic driving force and directly determines whether the oscillator is in a chaotic critical state. The selection principle is: in the absence of an external signal... Adjustment Observe the Poincaré section of the oscillator. When... Reaching a certain critical value At this time, the oscillator transitions from a chaotic state to a large-scale periodic state. The operating point of this oscillator is set as... This makes the oscillator extremely sensitive to weak signals of the same frequency. The numerical range is: With damping coefficient Correlation, typical value The specific value needs to be obtained through numerical calibration. 1. Fixed , (Same as Bosch's process cycle). The driving force angular frequency in the Duffing oscillator equation is... The Bosch period is 50~120ms.

[0113] 2. From Start by gradually increasing the value by 0.001 each time, and calculate the variance of the Poincaré section.

[0114] 3. The moment when the variance of the Poincaré section first jumps from >0.02 to <0.01 The value is .

[0115] The angular frequency of the driving force, which is the built-in periodic driving force in the Duffing oscillator equations, must be matched with the fundamental frequency of the plasma spectral signal. The matching principle is as follows: Real-time acquisition via external trigger signal Dynamic adjustment .

[0116] The numerical ranges of the parameters related to the driving force angular frequency are shown in Table 2 below: Table 2. Numerical range of parameters related to the driving force angular frequency

[0117] For external excitation, i.e. after removing DC Signal, sampling rate 1kHz. Represented as: (7); in, This is the arithmetic mean of the normalized spectral line intensity sequence of silicon fluoride. The amplitude is usually much smaller (Weak signal), when superimposed with the built-in driving force, can induce the oscillator to transition from a chaotic state to a periodic state, or vice versa, under critical conditions.

[0118] The parameters of the Duffing oscillator equations are summarized in Table 3 below: Table 3 Summary of parameters for the Duffing oscillator equations

[0119] Step S3032: Obtain the first average value of the silicon fluoride spectral line signal intensity in the etching step and the second average value of the silicon fluoride spectral line signal intensity in the passivation step. Calculate the ratio between the difference between the first average value and the second average value and the sum value as the modulation depth, and calculate the endpoint confidence of the modulation depth mode based on the modulation depth.

[0120] Specifically, calculate each Bosch process cycle. Output And its moving average, and mapped to confidence level. .

[0121] in, For modulation depth, The intensity of the silicon fluoride spectral line signal during the etching step. To passivate the intensity of the silicon fluoride spectral line signal within the step, To modulate the endpoint confidence of the deep mode, It is a sigmoid activation function. This is the sensitivity coefficient. This is the modulation depth threshold.

[0122] Step S3033: Perform frequency domain transformation on the silicon fluoride spectral line signal to obtain the power spectrum, extract the fundamental frequency power and noise power from the power spectrum, calculate the ratio of the fundamental frequency power to the noise power, and calculate the endpoint confidence of the spectral energy ratio mode based on the ratio.

[0123] Specifically, an FFT (Fast Fourier Transform) is performed on the data from the most recent 8 periods to calculate the spectral energy ratio of the fundamental frequency power to the noise power (out-of-band high-frequency power). Output endpoint confidence level .

[0124] in, The endpoint confidence level, The threshold for the spectral energy ratio. This represents the spectral energy ratio.

[0125] (1) Fundamental frequency : Fundamental frequency Equal to the reciprocal of the Bosch process cycle: The range of values ​​for the fundamental frequency and the Bosch process cycle is shown in Table 4 below: Table 4. Range of fundamental frequency and Bosch process cycle time

[0126] In this embodiment, the fundamental frequency It is acquired in real time via an external trigger signal, requiring no user configuration. Specifically, the period of the trigger signal is... Therefore, the fundamental frequency is known precisely.

[0127] (2) Calculation method of spectral energy ratio mode: For the most recent 8 cycles Signal data (total) Perform a Fast Fourier Transform (FFT) on each sampling point (fs=1000Hz) to obtain the power spectrum. .

[0128] Fundamental frequency power : base frequency and its second harmonic The sum of the powers (because Bosch modulated square waves contain odd and even harmonics, but the fundamental frequency and second harmonic already contain the vast majority of the energy), the formula for calculating the fundamental frequency power is as follows: (8); Out-of-band high-frequency power (i.e., noise power): defined as the sum of power above the third harmonic of the fundamental frequency (excluding DC components). (and fundamental frequency / second harmonic), out-of-band high-frequency power) The calculation formula is as follows: (9); The formula for calculating the spectral energy ratio is as follows: (10); in, Small constants (e.g.) Prevent division by zero.

[0129] The out-of-band high-frequency power definition range is shown in Table 5 below: Table 5 Out-of-band high-frequency power definition range

[0130] It should be noted that the modulation signal of the Bosch process is approximately a square wave, and its energy is mainly concentrated at the fundamental frequency and low harmonics. The amplitude of harmonics above the third is already small (less than 1 / 3 of the fundamental frequency) and is greatly affected by sampling noise, so it is included in the noise power for calculating the signal-to-noise ratio.

[0131] Step S304: Based on the fusion weights, the endpoint confidence scores of each preset modality are weighted and fused to obtain the comprehensive confidence score.

[0132] Specifically, step S304 includes: Step b: Calculate the sum of the products of the endpoint confidence scores of each preset modality and their corresponding fusion weights, and use this as the overall confidence score.

[0133] For example, dynamically assigning modal weights : Initial stage: [0.6, 0.3, 0.1].

[0134] Mid-term stage: [0.2, 0.6, 0.2].

[0135] Final stage: [0.1, 0.3, 0.6].

[0136] The formula for calculating the overall confidence level is: (11); in, For the mode weights of the chaotic oscillator, To modulate the depth mode weights, The spectral energy ratio is the mode weight. To assess the overall confidence level, For the confidence level of the chaotic oscillator mode, To modulate the confidence level of the deep mode, is the spectral energy ratio modal confidence level.

[0137] Step S305: Estimate the process status information of the current etching process in real time based on the plasma spectral signal.

[0138] Specifically, the process status information includes the remaining etching depth; step S305 above includes: Step c: Fit an empirical model between the etching depth and the modulation depth based on the historical sequence of modulation depth; estimate the current etching depth based on the current modulation depth and the empirical model; determine the remaining etching depth based on the difference between the target etching depth and the current etching depth.

[0139] Specifically, based on historical modulation depth The value is obtained by recursive least squares fitting of the current etching depth and modulation depth. Empirical models: (12); in, At the current etching depth, The scale factor controls the current etching depth. The overall magnitude of the change, in μm. The physical meaning of: The total range of changes in current etching depth as the value changes from 1 to 0. Typical range: 50~200μm. The sensitivity coefficient controls the effect of the current etching depth on... Sensitivity to change, dimensionless. The physical meaning of: The larger the depth, the greater The steeper the change. Typical range: -5 to -15 (negative values, because...) The depth increases as the etching process progresses. For offset constant: when The depth offset when it approaches 0, in μm. Physical meaning: Etching complete ( Total etching depth when ≈0, typical range: 0~10μm (close to 0).

[0140] By using a Kalman filter to fuse the current etching depth estimate and etching rate (the derivative of depth with respect to time), the remaining etching depth is output in real time. And its 95% confidence interval. Calculate the modulation depth. The standard deviation of the sliding window value is used as an index of etching uniformity.

[0141] Step S306: Determine the endpoint based on the overall confidence level and process status information.

[0142] Specifically, step S306 includes: Step d: When the overall confidence level exceeds the first confidence level threshold and the remaining etching depth is less than the preset depth threshold, and after a preset number of cycles, it is determined that the current etching process has reached the etching endpoint.

[0143] Specifically, the etching endpoint conditions are as follows: and And it lasts for two cycles.

[0144] For example, assuming the current process stage is the intermediate stage, the industrial control computer obtains the fusion weight vector for the intermediate stage as 0.2, 0.6, 0.2 according to the preset stage weight mapping relationship, i.e., the chaotic oscillator mode weight is 0.2, the modulation depth mode weight is 0.6, and the spectral energy ratio mode weight is 0.2. Simultaneously, based on the silicon fluoride spectral line signal acquired by the spectrometer, the industrial control computer calculates the current endpoint confidence levels for the three modes in parallel: the chaotic oscillator mode confidence level... Modulation depth mode confidence Spectral energy ratio modal confidence The industrial control computer calculates according to the following weighted fusion formula: The overall confidence level was obtained as 0.53, which served as the unified quantitative basis for subsequent endpoint determination. Since 0.53 did not exceed the endpoint determination threshold of 0.8, the industrial control computer continued monitoring for the next cycle without triggering endpoint determination.

[0145] Step S307: Output the endpoint signal and process status information for the etching equipment to perform closed-loop adjustment.

[0146] Specifically, step S307 includes: Step e: During the etching process and / or when the etching endpoint is reached, output the endpoint signal and process status information, which also includes the etching rate and etching uniformity index.

[0147] Specifically, the endpoint signal (falling edge of the trigger signal) is output through the digital I / O interface.

[0148] The remaining etching depth, etching rate, and uniformity index are sent to the host computer via Ethernet control automation technology.

[0149] In one alternative implementation, if the etching uniformity index exceeds the standard, a warning signal is issued, and the etching equipment can automatically pause or adjust the RF power (Radio Frequency) of the etching equipment.

[0150] The deep silicon etching adaptive multimodal endpoint monitoring method provided in this embodiment identifies the initial, middle, and final stages based on the etching depth ratio and signal-to-noise ratio, and accordingly assigns fusion weights to three modes: chaotic oscillator, modulation depth, and spectral energy ratio. This ensures that each stage has the most suitable dominant mode playing a role, overcoming the problem that a single algorithm cannot adapt to changes in signal characteristics throughout the entire process. By parallel calculation of the endpoint confidence of the three modes and weighted fusion according to stage weights, smooth transition and collaborative decision-making between algorithms are achieved, solving the result jump problem caused by traditional hard switching and significantly reducing the false positive rate. By estimating the remaining etching depth in real time and jointly deciding with the comprehensive confidence, an independent physical verification dimension is provided for endpoint determination, effectively avoiding over-etching or under-etching caused by relying solely on intensity signals. At the same time, by outputting endpoint signals and process monitoring data, the monitoring system can be linked with the etching equipment. Ultimately, under the conditions of aperture ratio ≤0.5% and signal-to-noise ratio as low as -3dB, a false positive rate of less than 0.5% and a detection delay of less than 5 cycles are achieved, while simultaneously providing process information such as etching rate and uniformity index.

[0151] As one or more specific application embodiments of the present invention, combined with Figure 4 The adaptive multimodal endpoint monitoring method for deep silicon etching provided by this invention will be further described in detail, such as... Figure 4 As shown, the specific process is as follows: Step 1: Acquisition and trigger signal acquisition: The industrial control computer 130 acquires the intensity of silicon fluoride spectral lines and argon spectral lines in real time through the spectrometer 110, and receives the gas switching trigger signal of the etching equipment through the synchronous signal isolation circuit 120.

[0152] Step 2: Signal normalization: The industrial control computer 130 normalizes the intensity of the collected silicon fluoride spectral lines by calculating the ratio of the intensity of the silicon fluoride spectral lines to the intensity of the argon spectral lines, in order to eliminate common-mode interference such as window contamination and source power drift.

[0153] Step 3: Calculate the etching depth ratio: The industrial control computer 130 calculates the etching depth ratio based on the cumulative number of cycles and the estimated total number of cycles provided by the synchronization signal isolation circuit 120.

[0154] Step 4: Calculate the real-time signal-to-noise ratio: The industrial control computer 130 calculates the real-time signal-to-noise ratio based on the normalized silicon fluoride spectral signal intensity sequence.

[0155] Step 5: Determine the process stage: The industrial control computer 130 comprehensively determines the current process stage based on the etching depth ratio and the real-time signal-to-noise ratio: when the depth ratio is less than the first ratio threshold or the signal-to-noise ratio is lower than the first signal-to-noise ratio threshold, it is identified as the initial stage; when the depth ratio is not less than the first ratio threshold and not greater than the second ratio threshold, and the signal-to-noise ratio is not lower than the first signal-to-noise ratio threshold, it is identified as the intermediate stage; when the depth ratio is greater than the second ratio threshold or the signal-to-noise ratio continuously decreases, it is identified as the final stage.

[0156] Step 6: Determine the dominant mode: The industrial control computer 130 determines the dominant mode based on the identified process stage: the initial stage uses the chaotic oscillator method as the dominant mode, the middle stage uses the modulation depth method as the dominant mode, and the final stage uses the spectrum energy ratio method as the dominant mode.

[0157] Step 7: Parallel computing and process monitoring: The industrial control computer 130 calculates the endpoint confidence of three different principles—chaotic oscillator mode, modulation depth mode, and spectral energy ratio mode—in parallel based on the normalized silicon fluoride spectral line signal. At the same time, based on the modulation depth historical sequence, it uses recursive least squares to fit an empirical model between the etching depth and the modulation depth, and uses Kalman filtering to update the remaining etching depth and etching uniformity index in real time.

[0158] Step 8: Output remaining depth and uniformity: The industrial control computer 130 outputs the currently estimated remaining etching depth and etching uniformity index.

[0159] Step 9: Dynamic weighted fusion: The industrial control computer 130 calculates the weighted sum of the endpoint confidence scores of the three modes according to the fusion weights corresponding to the current stage, and obtains the comprehensive confidence score.

[0160] Step 10: Endpoint determination: The industrial control computer 130 determines whether the overall confidence level is greater than or equal to 0.8 and whether the remaining etching depth is less than 5μm. If both conditions are met, the etching endpoint is determined to have been reached, and step 11 is executed. If either condition is not met, the process returns to step 1 to continue the next cycle.

[0161] Step 11: Output the endpoint signal and report the monitoring data: The industrial control computer 130 outputs the endpoint signal through the digital input / output interface and reports the process monitoring data through the industrial bus communication module.

[0162] The quantization performance of the deep silicon etching adaptive multimodal endpoint monitoring method provided in this embodiment is compared with that of the single modulation depth method and the single chaotic oscillator method, as shown in Table 6 below. Table 6 Comparison of Quantitative Effects

[0163] This embodiment also provides a display diagram of the running effect of the deep silicon etching adaptive multimodal endpoint monitoring method. In this embodiment, when the actual etching endpoint is the 250th Bosch cycle, the industrial control computer determines that the etching endpoint has been reached in the 252nd cycle, with a detection delay of 2 cycles (corresponding to 0.24 seconds, based on a 120ms cycle length). Figure 5 As shown, the first subplot is the Bosch process input signal (SiF raw signal amplitude) curve, with the vertical axis representing the SiF raw signal amplitude (in dB). This curve illustrates the trend of SiF signal intensity variation throughout the etching process. The SiF signal amplitude remains relatively stable throughout the etching process, reflecting the basic level of the emission spectrum intensity of the etching reaction products in the plasma under normal etching conditions. It should be noted that this curve shows the absolute amplitude of the SiF raw signal, while the signal-to-noise ratio (SNR) used for stage identification is a derived parameter calculated based on this and the noise floor. The SNR exhibits a non-monotonic change characteristic of first decreasing, then increasing, and then decreasing again, while the SiF raw amplitude remains relatively stable during the etching process.

[0164] The second sub-figure is the modulation depth. The curve shows the output of the modulation depth mode. The changing trend throughout the entire process. In the initial stage of etching, due to the extremely low signal-to-noise ratio, the modulation depth... The value fluctuates drastically and shows no stable trend, rendering the modulation depth method ineffective; in the medium term, the signal-to-noise ratio improves, and the modulation depth... The value shows a monotonically decreasing trend, which can effectively reflect the etching process; in the final stage, the modulation depth The value drops to a low level but the change tends to level off, making it difficult to modulate the depth of the modulation. The value indicates the precise endpoint. This curve shows that the modulation depth mode is most effective in the mid-term.

[0165] The third subplot is the spectral energy ratio curve, which shows the output of the spectral energy ratio mode. The variation pattern throughout the entire etching process is as follows: In the early and middle stages of etching, the fundamental frequency energy accounts for a relatively high and stable proportion of the total energy, and the spectral energy ratio R value does not change significantly; in the final stage of etching, the etching step signal begins to attenuate, the fundamental frequency power decreases, resulting in a significant decrease in the spectral energy ratio. This curve shows that the spectral energy ratio mode can sensitively capture changes in frequency domain characteristics in the final stage.

[0166] The fourth subplot shows the confidence curve of the chaotic oscillator, illustrating the trend of the final confidence level of the chaotic oscillator mode throughout the entire etching process. In the early stages of etching, the chaotic oscillator is extremely sensitive to weak periodic signals at very low signal-to-noise ratios, capable of detecting weak etching step signals with the same frequency as the Bosch process cycle, and the confidence level begins to rise. In the middle stages of etching, as signal quality improves, the chaotic oscillator continues to respond to periodic signals, and the confidence level remains at a high level. In the final stages of etching, the signal attenuates again, but the chaotic oscillator still retains a certain detection capability for weak signals. This curve indicates that the chaotic oscillator mode is most effective in the initial stage when the signal-to-noise ratio is extremely low.

[0167] The fifth sub-figure shows the overall confidence curve and endpoint determination, displaying both the overall confidence curve and the 0.8 threshold line. The overall confidence curve illustrates the trend of the overall confidence after weighted fusion of the three modes according to stage weights throughout the entire process. In the early stage of etching, the overall confidence is mainly contributed by the confidence of the chaotic oscillator, with a low value but starting to rise. In the middle stage of etching, the modulation depth confidence gradually increases, and the overall confidence continues to climb. In the late stage of etching, the overall confidence reaches the threshold of 0.8 or higher (around 55s), and the remaining etching depth is less than 5μm. After both conditions are met simultaneously and maintained for a preset number of cycles, the etching endpoint is determined to have been reached. This overall confidence curve is the final decision-making basis for endpoint determination: when the overall confidence curve crosses the 0.8 threshold line, it indicates that both criteria have been met.

[0168] The deep silicon etching adaptive multimodal endpoint monitoring method provided in this embodiment upgrades the traditional single-point endpoint detection to an intelligent monitoring system of "multimodal perception-weighted fusion-state estimation-closed-loop feedback" for the entire Bosch process. Under the conditions of detectable aperture ratio ≤0.5% and signal-to-noise ratio (SNR) as low as -3dB, it achieves a false positive rate of <0.5% and a detection delay of <5 cycles, while providing online estimation of etching rate and remaining etching depth.

[0169] This embodiment also provides a deep silicon etching adaptive multimodal endpoint monitoring device, which is used to implement the above embodiments and preferred embodiments, and will not be repeated as described herein. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0170] This embodiment provides a deep silicon etching adaptive multimodal endpoint monitoring device, applied to the above-mentioned... Figures 2 to 4 The deep silicon etching adaptive multimodal endpoint monitoring method shown is as follows: Figure 6 As shown, it includes: The signal acquisition and synchronization module 601 is used to acquire plasma spectral signals and cumulative cycle count, and determine the etching depth ratio based on the cumulative cycle count of the current etching process. The stage identification and weight allocation module 602 is used to identify the current process stage based on the etching depth ratio and the signal-to-noise ratio of the plasma spectral signal, and to determine the fusion weights corresponding to each preset mode according to the process stage; the process stages include an initial stage, a middle stage, and a final stage; the preset modes include a chaotic oscillator mode, a modulation depth mode, and a spectral energy ratio mode; the initial stage is adapted to the chaotic oscillator mode, the middle stage is adapted to the modulation depth mode, and the final stage is adapted to the spectral energy ratio mode; The multimodal calculation module 603 is used to calculate the endpoint confidence level corresponding to each preset mode in parallel based on the plasma spectral signal. The fusion decision module 604 is used to perform weighted fusion of the endpoint confidence scores of each preset modality according to the fusion weights to obtain the comprehensive confidence score. The state estimation module 605 is used to estimate the process state information of the current etching process in real time based on the plasma spectral signal. The endpoint determination module 606 is used to estimate the remaining etching depth in real time and determine the endpoint based on the comprehensive confidence level and process status information. Output module 607 is used to output the endpoint signal and process status information for the etching equipment to perform closed-loop adjustment.

[0171] In some alternative implementations, the signal acquisition and synchronization module 601 includes: The etching depth ratio calculation unit is used to obtain the estimated total number of cycles. The estimated total number of cycles is determined based on the target etching depth and the theoretical etching rate. The ratio of the cumulative number of cycles to the estimated total number of cycles is calculated as the etching depth ratio.

[0172] In one alternative embodiment, the device further includes: The online correction module is used to perform online correction of the preset total number of cycles when all of the following conditions are met: the current etching process is in the intermediate stage; the covariance of the current remaining depth estimate is less than a preset threshold; and the estimated value of the total number of cycles deviates from the current value by more than a preset percentage for several consecutive cycles.

[0173] In one alternative implementation, the correction module includes: The estimation back-calculation and update unit is used to back-calculate the estimated total number of cycles based on the currently estimated remaining etching depth and target etching depth; and to smooth and update the estimated value using an exponentially weighted moving average.

[0174] In one alternative implementation, the signal-to-noise ratio of the plasma spectral signal is obtained in the following manner: The intensity of silicon fluoride spectral lines was extracted from the plasma spectral signal and normalized to obtain the normalized silicon fluoride spectral line intensity sequence. The noise floor is calculated based on the normalized spectral signal intensity sequence of silicon fluoride after the passivation step. The signal-to-noise ratio is determined by calculating the ratio of the standard deviation of the normalized silicon fluoride spectral signal intensity sequence to the noise floor.

[0175] In some optional implementations, the process stages include an initial stage, an intermediate stage, and a final stage; the stage identification and weight allocation module 602 includes: The three-stage identification unit is used to identify the initial stage when the etching depth ratio is less than a preset first ratio threshold and / or the signal-to-noise ratio is lower than a preset first threshold; to identify the intermediate stage when the etching depth ratio is not less than a preset first ratio threshold and not greater than a preset second ratio threshold, and the signal-to-noise ratio is not lower than a preset first threshold; and to identify the final stage when the etching depth ratio is greater than a preset second ratio threshold, or when the signal-to-noise ratio continuously decreases.

[0176] In one optional implementation, the stage identification and weight allocation module 602 includes: The fusion weight determination unit is used to determine the fusion weight of each preset mode under the process stage according to the adaptation relationship between the process stage and the preset modes. The adaptation relationship includes: the initial stage is adapted to the chaotic oscillator mode, the middle stage is adapted to the modulation depth mode, and the final stage is adapted to the spectral energy ratio mode.

[0177] In one optional implementation, the fusion weight determination unit includes: The sub-unit lookup function is used to find the weight vector corresponding to the process stage from the preset stage weight mapping relationship based on the adaptation relationship. The preset stage weight mapping relationship stores the correspondence between each process stage and the preset fusion weight of each mode.

[0178] In one alternative implementation, the multimodal computing module 603 includes: The first confidence calculation unit is used to use the silicon fluoride spectral line signal as the external excitation of the nonlinear oscillator and calculate the endpoint confidence of the chaotic oscillator mode based on the state change of the nonlinear oscillator. The second confidence calculation unit is used to obtain the first average value of the silicon fluoride spectral line signal intensity in the etching step and the second average value of the silicon fluoride spectral line signal intensity in the passivation step, calculate the ratio between the difference between the first average value and the second average value and the sum value, and use it as the modulation depth, and calculate the endpoint confidence of the modulation depth mode based on the modulation depth. The third confidence calculation unit is used to perform frequency domain transformation on the silicon fluoride spectral line signal to obtain the power spectrum, extract the fundamental frequency power and noise power from the power spectrum, calculate the ratio of the fundamental frequency power to the noise power, and calculate the endpoint confidence of the spectral energy ratio mode based on the ratio.

[0179] In one alternative implementation, the fusion decision module 604 includes: The weighted fusion unit is used to calculate the sum of the products of the endpoint confidence of each preset mode and its corresponding fusion weight, which is used as the overall confidence.

[0180] In one optional implementation, the process status information includes the remaining etching depth; the status estimation module 605 includes: The real-time estimation unit is used to fit an empirical model between the etching depth and the modulation depth based on the historical sequence of modulation depth; estimate the current etching depth based on the current modulation depth and the empirical model; and determine the remaining etching depth based on the difference between the target etching depth and the current etching depth.

[0181] In one optional implementation, the endpoint determination module 606 includes: The etching endpoint determination unit is used to determine that the current etching process has reached the etching endpoint when the overall confidence level exceeds the first confidence level threshold and the remaining etching depth is less than the preset depth threshold, and after a preset number of cycles.

[0182] In one alternative implementation, the output module 607 includes: The output unit is used to output an endpoint signal and process status information during the etching process and / or when the etching endpoint is reached. The process status information also includes etching rate and etching uniformity index.

[0183] The deep silicon etching adaptive multimodal endpoint monitoring device provided in this embodiment of the invention can execute the deep silicon etching adaptive multimodal endpoint monitoring method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0184] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0185] This diagram illustrates a suitable structural design for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processing unit, etc.) 701, which can perform various appropriate actions and processes based on a program stored in a read-only memory 702 or a program loaded from a memory 708 into a random access memory 703. The read-only memory may be a ROM, and the random access memory may be RAM. The random access memory 703 also stores various programs and data required for the operation of the electronic device. The processor 701, the read-only memory 702, and the random access memory 703 are interconnected via a bus 704. An input / output interface 705 is also connected to the bus 704.

[0186] Typically, the following devices can be connected to the input / output interface 705: input devices 706 including, for example, a touchscreen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; output devices 707 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; memory devices 708 including, for example, magnetic tape, hard disk, etc.; and communication devices 709. Communication device 709 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 7 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0187] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 709, or installed from a memory 708 (the memory may be magnetic tape, hard disk, etc.), or installed from a read-only memory 702. When the computer program is executed by the processor 701, it performs the functions defined in the cooperative control method of the etching apparatus and the spectral imager according to embodiments of the present invention.

[0188] Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0189] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the deep silicon etching adaptive multimodal endpoint monitoring method shown in the above embodiments is implemented.

[0190] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0191] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A deep silicon etching adaptive multimodal endpoint monitoring method for intelligent online semiconductor inspection, characterized in that, The method includes: Acquire plasma spectral signals and cumulative cycle count, and determine the etching depth ratio based on the cumulative cycle count of the current etching process; Based on the etching depth ratio and the signal-to-noise ratio of the plasma spectral signal, the current process stage is identified, and the fusion weights corresponding to each preset mode are determined according to the process stage; the process stage includes an initial stage, a middle stage, and a final stage; the preset modes include a chaotic oscillator mode, a modulation depth mode, and a spectral energy ratio mode; the initial stage is adapted to the chaotic oscillator mode, the middle stage is adapted to the modulation depth mode, and the final stage is adapted to the spectral energy ratio mode; Based on the plasma spectral signal, the endpoint confidence level corresponding to each preset mode is calculated in parallel. Based on the fusion weights, the endpoint confidence scores of each preset modality are weighted and fused to obtain a comprehensive confidence score. The process status information of the current etching process is estimated in real time based on the plasma spectral signal. The endpoint is determined based on the overall confidence level and the process status information. The endpoint signal and the process status information are output for the etching equipment to perform closed-loop adjustments.

2. The method according to claim 1, characterized in that, The step of determining the etching depth ratio based on the cumulative number of cycles in the current etching process includes: Obtain the estimated total number of cycles; the estimated total number of cycles is determined based on the target etching depth and the theoretical etching rate; The ratio of the cumulative number of cycles to the estimated total number of cycles is calculated as the etching depth ratio.

3. The method according to claim 2, characterized in that, The method further includes: An online correction of the preset total number of cycles will be performed when all of the following conditions are met: The current etching process is in the mid-stage; The covariance of the current remaining depth estimate is less than a preset threshold; The estimated total number of cycles deviates from the current value by more than a preset percentage for several consecutive cycles.

4. The method according to claim 3, characterized in that, The online correction of the preset total number of cycles includes: Based on the current estimated remaining etching depth and target etching depth, the estimated total number of cycles is calculated. The estimated value is then updated smoothly using an exponentially weighted moving average.

5. The method according to claim 1, characterized in that, The signal-to-noise ratio of the plasma spectral signal is obtained in the following manner: The intensity of silicon fluoride spectral lines is extracted from the plasma spectral signal, and the intensity of silicon fluoride spectral lines is normalized to obtain a normalized sequence of silicon fluoride spectral line intensity. The noise floor is calculated based on the normalized silicon fluoride spectral signal intensity sequence after the passivation step. The signal-to-noise ratio is determined by calculating the ratio of the standard deviation of the normalized silicon fluoride spectral signal intensity sequence to the noise floor.

6. The method according to claim 1, characterized in that, Based on the etching depth ratio and the signal-to-noise ratio of the plasma spectral signal, the current process stage is identified, including: When the etching depth ratio is less than a preset first ratio threshold and / or the signal-to-noise ratio is lower than a preset first threshold, it is identified as the initial stage. When the etching depth ratio is not less than the preset first ratio threshold and not greater than the preset second ratio threshold, and the signal-to-noise ratio is not lower than the preset first threshold, it is identified as the intermediate stage; When the etching depth ratio is greater than the preset second ratio threshold, or when the etching depth ratio is not greater than the preset second ratio threshold but the signal-to-noise ratio continuously decreases, it is identified as the final stage.

7. The method according to claim 1, characterized in that, The step of determining the preset fusion weights corresponding to each modality based on the process stage includes: Based on the adaptation relationship between the process stage and the preset modes, the fusion weight of the preset modes under the process stage is determined.

8. The method according to claim 7, characterized in that, Based on the adaptation relationship between the process stage and each mode, the fusion weight of each mode under the process stage is determined, including: According to the adaptation relationship, the weight vector corresponding to the process stage is found from the preset stage weight mapping relationship; the preset stage weight mapping relationship stores the correspondence between each process stage and the preset fusion weight of each mode.

9. The method according to claim 5, characterized in that, Based on the plasma spectral signal, the endpoint confidence levels corresponding to each preset mode are calculated in parallel, including: The silicon fluoride spectral line signal is used as an external excitation for the nonlinear oscillator, and the endpoint confidence of the chaotic oscillator mode is calculated based on the state changes of the nonlinear oscillator. The first average value of the silicon fluoride spectral line signal intensity in the etching step and the second average value of the silicon fluoride spectral line signal intensity in the passivation step are obtained. The ratio between the difference between the first average value and the second average value and the sum value is calculated as the modulation depth. The endpoint confidence of the modulation depth mode is calculated based on the modulation depth. The power spectrum is obtained by performing a frequency domain transformation on the silicon fluoride spectral line signal. The fundamental frequency power and noise power are extracted from the power spectrum. The ratio of the fundamental frequency power to the noise power is calculated, and the endpoint confidence of the spectral energy ratio mode is calculated based on the ratio.

10. The method according to claim 1, characterized in that, The step of weighted fusion of the endpoint confidence scores of each preset modality according to the fusion weights to obtain a comprehensive confidence score includes: The sum of the products of the endpoint confidence scores of each preset modality and their corresponding fusion weights is used as the overall confidence score.

11. The method according to claim 9, characterized in that, The process status information includes the remaining etching depth; The step of estimating the process status information of the current etching process in real time based on the plasma spectral signal includes: Based on the historical sequence of modulation depth, an empirical model between etching depth and modulation depth is fitted. Estimate the current etching depth based on the current modulation depth and the empirical model; The remaining etching depth is determined based on the difference between the target etching depth and the current etching depth.

12. The method according to claim 11, characterized in that, The endpoint determination is based on the comprehensive confidence level and the process status information, including: When the overall confidence level exceeds the first confidence level threshold and the remaining etching depth is less than the preset depth threshold, and after a preset number of cycles, it is determined that the current etching process has reached the etching endpoint.

13. The method according to claim 12, characterized in that, The output endpoint signal and the process status information include: During the etching process and / or when the etching endpoint is reached, an endpoint signal and process status information are output, including etching rate and etching uniformity index.

14. A deep silicon etching adaptive multimodal endpoint monitoring system for intelligent online semiconductor inspection, characterized in that, The system includes: A spectrometer is used to receive plasma spectral signals; A synchronous signal isolation circuit is used to receive the gas switching trigger signal of the etching equipment and perform cycle counting based on the trigger signal to obtain the cumulative number of cycles; An industrial control computer is connected to the spectrometer and the synchronization signal isolation circuit, respectively, for executing the deep silicon etching adaptive multimodal endpoint monitoring method according to any one of claims 1 to 13.

15. The system according to claim 14, characterized in that, The industrial control computer includes a field-programmable gate array (FPGA) accelerator card, which is used to calculate the endpoint confidence level corresponding to each preset mode in parallel based on the plasma spectral signal.

16. A deep silicon etching adaptive multimodal endpoint monitoring device for intelligent online semiconductor inspection, characterized in that, The apparatus, used for performing the deep silicon etching adaptive multimodal endpoint monitoring method according to any one of claims 1 to 13, comprises: The signal acquisition and synchronization module is used to acquire plasma spectral signals and cumulative cycle count, and determine the etching depth ratio based on the cumulative cycle count of the current etching process. The stage identification and weight allocation module is used to identify the current process stage based on the etching depth ratio and the signal-to-noise ratio of the plasma spectral signal, and to determine the fusion weights corresponding to each preset mode according to the process stage. The process stages include an initial stage, a middle stage, and a final stage. The preset modes include a chaotic oscillator mode, a modulation depth mode, and a spectral energy ratio mode. The initial stage is adapted to the chaotic oscillator mode, the middle stage is adapted to the modulation depth mode, and the final stage is adapted to the spectral energy ratio mode. The multimodal calculation module is used to calculate the endpoint confidence level corresponding to each preset mode in parallel based on the plasma spectral signal. The fusion decision module is used to perform weighted fusion of the endpoint confidence scores of each preset mode according to the fusion weights to obtain a comprehensive confidence score; The state estimation module is used to estimate the process state information of the current etching process in real time based on the plasma spectral signal; The endpoint determination module is used to determine the endpoint based on the comprehensive confidence level and the process status information. The output module is used to output the endpoint signal and the process status information for the etching equipment to perform closed-loop adjustment.

17. An electronic device, characterized in that, include: A memory and a processor are communicatively connected, the memory storing computer instructions, and the processor executing the computer instructions to perform the deep silicon etching adaptive multimodal endpoint monitoring method according to any one of claims 1 to 13.

18. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to execute the deep silicon etching adaptive multimodal endpoint monitoring method according to any one of claims 1 to 13.

19. A computer program product, characterized in that, Includes computer instructions for causing a computer to execute the deep silicon etching adaptive multimodal endpoint monitoring method according to any one of claims 1 to 13.

Images