A semiconductor device, and a method and apparatus for loading a radio frequency monitoring threshold value thereof

Abstract

The application discloses a semiconductor device and a loading method and device of a radio frequency monitoring threshold thereof, relates to the technical field of semiconductors, and is applied to a semiconductor device. The semiconductor device comprises a process chamber, a radio frequency generating device, a gas supply device and a monitoring device, and comprises the following steps: in the process of processing a wafer, a first radio frequency monitoring threshold is loaded to the monitoring device, and a signal representing a radio frequency parameter is obtained; the signal representing the radio frequency parameter is processed to obtain a second radio frequency monitoring threshold; the current radio frequency monitoring threshold of the monitoring device is determined based on the second radio frequency monitoring threshold; and the current radio frequency monitoring threshold is loaded to the monitoring device. Therefore, the current radio frequency monitoring threshold adapted to the monitoring device can be determined by combining the first radio frequency monitoring threshold and the second radio frequency monitoring threshold and loaded to the monitoring device, the dynamic loading of the radio frequency monitoring threshold in the wafer processing process is realized, false positives and false negatives can be reduced, and the radio frequency monitoring accuracy is ensured.

Description

Technical Field

[0001] This application relates to the field of semiconductor technology, and in particular to a semiconductor device and a method and apparatus for loading radio frequency monitoring thresholds thereon. Background Technology

[0002] In semiconductor wafer fabrication, the radio frequency (RF) power within the process chamber is a key parameter affecting the quality of processes such as etching and deposition. To ensure process stability and product yield, semiconductor equipment is typically equipped with monitoring devices to monitor RF parameters (such as RF power, current, and spectrum) within the process chamber in real time, and to identify and issue alarms for abnormal states based on preset RF monitoring thresholds.

[0003] In existing technologies, RF monitoring thresholds are mostly fixed preset values, typically set once during equipment manufacturing or process formulation development, and remain unchanged during subsequent wafer processing. However, as the equipment operates for a long time, components within the chamber gradually age, the RF matching network status drifts, and the process conditions of different batches of wafers vary. Fixed monitoring thresholds are difficult to adapt to these changes, lack the ability to dynamically adjust thresholds based on real-time process data, and cannot achieve adaptive matching between the threshold and the current process state.

[0004] Therefore, how to achieve dynamic updating and real-time adjustment of radio frequency monitoring thresholds during wafer fabrication has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] This application discloses a semiconductor device and a method and apparatus for loading the radio frequency (RF) monitoring threshold, which solves the technical problem of the inability to dynamically update and adjust the RF monitoring threshold in real time during wafer fabrication. The specific solution is as follows: In a first aspect, this application provides a method for loading a radio frequency (RF) monitoring threshold, applied to a semiconductor device used for processing wafers. The semiconductor device includes a process chamber, an RF generation device, and a monitoring device. The RF generation device generates RF signals within the process chamber, and the monitoring device monitors RF parameters within the process chamber. The method includes: during wafer processing, loading a first RF monitoring threshold onto the monitoring device and acquiring a signal characterizing the RF parameters using the monitoring device; processing the signal characterizing the RF parameters to obtain a second RF monitoring threshold; determining a current RF monitoring threshold for the monitoring device based on the second RF monitoring threshold; and loading the current RF monitoring threshold onto the monitoring device.

[0006] Therefore, in the wafer processing process, this application can load a first radio frequency (RF) monitoring threshold onto the monitoring device and use the monitoring device to acquire a signal characterizing the RF parameters. Then, the signal characterizing the RF parameters is processed to obtain a second RF monitoring threshold. Next, based on the second RF monitoring threshold, the current RF monitoring threshold of the monitoring device is determined. Finally, the current RF monitoring threshold is loaded onto the monitoring device. In this way, during wafer processing, this application directly acquires the signal characterizing the RF parameters of the monitoring device based on the first RF monitoring threshold, processes the signal to obtain the second RF monitoring threshold, and then combines the first and second RF monitoring thresholds to load the determined current RF monitoring threshold adapted to the monitoring device. This achieves dynamic updating and real-time adjustment of the RF monitoring threshold, enabling the monitoring device to always dynamically adjust based on a threshold matching the current process state, significantly improving the accuracy of RF monitoring.

[0007] In one possible embodiment, determining the current radio frequency monitoring threshold of the monitoring device based on the second radio frequency monitoring threshold includes: acquiring wafer information and processing parameters in the process chamber; determining whether the second radio frequency monitoring threshold matches the wafer information and the processing parameters; if they match, determining the second radio frequency monitoring threshold as the current radio frequency monitoring threshold of the monitoring device; if they do not match, determining the first radio frequency monitoring threshold as the current radio frequency monitoring threshold of the monitoring device.

[0008] By acquiring wafer information and processing parameters, it is determined whether the second RF monitoring threshold matches the wafer information and processing parameters. If they match, the second RF monitoring threshold is set as the current RF monitoring threshold. If they do not match, the first RF monitoring threshold is set as the current RF monitoring threshold. Wafer identity and process parameters can be introduced as matching conditions, so that the threshold can be intelligently selected according to specific batches and process conditions, avoiding false alarms or missed alarms due to process deviations or differences between different wafers, and improving the applicability of the threshold.

[0009] In one possible embodiment, the signals characterizing the radio frequency parameters include spectral signals during the processing of the wafer, the time consumed by the pin movement, and current signals associated with the radio frequency parameters.

[0010] Since the signals characterizing radio frequency parameters include spectral signals, pin movement time, and current signals related to radio frequency parameters, they can provide multi-dimensional radio frequency state characterization information, thereby comprehensively reflecting the changes in intracavity radio frequency parameters from multiple perspectives such as optics, mechanics, and electricity, and improving the accuracy of threshold adjustment.

[0011] In one possible embodiment, processing the signal characterizing the radio frequency parameters to obtain a second radio frequency monitoring threshold includes: performing data statistics on the signal characterizing the radio frequency parameters to obtain corresponding current data distribution characteristics; and updating the first radio frequency monitoring threshold based on the current data distribution characteristics to obtain a second radio frequency monitoring threshold.

[0012] In this embodiment, the second radio frequency monitoring threshold can be determined efficiently by using the signal characterizing the radio frequency parameters through data statistics. It can be adaptively adjusted based on the statistical characteristics of real-time data, thereby enabling the threshold to change with process fluctuations and improving the robustness of monitoring.

[0013] In one possible embodiment, updating the first radio frequency monitoring threshold based on the current data distribution characteristics to obtain a second radio frequency monitoring threshold includes: calculating the current data distribution characteristics, and performing a weighted calculation on the corresponding calculation results and the first radio frequency monitoring threshold based on a preset weight coefficient to obtain the second radio frequency monitoring threshold.

[0014] In this embodiment, when updating the radio frequency monitoring threshold, relevant weighting coefficients are set. The weighting adjustment can be combined with historical benchmarks and current statistical results, which can introduce real-time correction while maintaining threshold stability, thus taking into account both threshold smoothness and fast response capability.

[0015] In one possible embodiment, processing the signal characterizing the radio frequency parameters to obtain a second radio frequency monitoring threshold includes: adaptively adjusting the first radio frequency monitoring threshold based on the signal characterizing the radio frequency parameters through online learning to obtain a second radio frequency monitoring threshold.

[0016] In this embodiment, the first radio frequency monitoring threshold can be adaptively updated using the signal representing the radio frequency parameter through online learning. This allows for real-time optimization of the monitoring threshold, thereby automatically capturing non-stationary changes and long-term trends in the process and enabling self-updating of the threshold without manual intervention.

[0017] In one possible embodiment, the process of determining the first radio frequency monitoring threshold includes: determining target process data in historical process data that matches the processing parameters of the wafer, and determining the radio frequency monitoring threshold corresponding to the target process data as the first radio frequency monitoring threshold.

[0018] In this embodiment, the first radio frequency monitoring threshold is initialized based on the target process data that matches the processing parameters of the wafer in the historical process data, which can provide a reasonable starting monitoring benchmark for the new batch of wafers and improve the effectiveness of initial monitoring.

[0019] In one possible embodiment, after loading the current radio frequency monitoring threshold into the monitoring device, the method further includes: acquiring real-time processing data of the wafer during the processing; and generating radio frequency monitoring results based on the matching relationship between the real-time processing data and the current radio frequency monitoring threshold.

[0020] In this embodiment, during the radio frequency monitoring process, the corresponding radio frequency monitoring results are generated by matching the monitored real-time processing data with the current radio frequency monitoring threshold. This enables the comparison and judgment between real-time data and dynamic thresholds, allowing for timely identification of abnormal radio frequency states and improving the real-time monitoring and fault alarm capabilities of the processing process.

[0021] In one possible embodiment, generating an RF monitoring result based on the matching relationship between the real-time processing data and the current RF monitoring threshold includes: generating an RF monitoring result based on the matching relationship between the real-time processing data and the current RF monitoring threshold, and the matching relationship between the wafer processing parameters and the processing parameters corresponding to the current RF monitoring threshold.

[0022] In this embodiment, the process of generating radio frequency monitoring results not only considers the matching relationship between real-time processing data and the current radio frequency monitoring threshold, but also the matching relationship between processing parameters. This allows for dual verification of data matching and process matching, thereby distinguishing between normal process fluctuations and real anomalies, reducing false alarm rates, and improving the reliability of monitoring results.

[0023] Secondly, this application provides a device for loading a radio frequency monitoring threshold, comprising: The signal acquisition module is used to load a first radio frequency monitoring threshold onto the monitoring device during the processing of the wafer, and to acquire a signal characterizing the radio frequency parameters using the monitoring device. The information processing module is used to process the signal characterizing the radio frequency parameters to obtain a second radio frequency monitoring threshold. A threshold determination module is used to determine the current radio frequency monitoring threshold of the monitoring device based on the second radio frequency monitoring threshold; A threshold loading module is used to load the current radio frequency monitoring threshold into the monitoring device.

[0024] In this embodiment, during the wafer processing, the signal acquisition module can load a first radio frequency (RF) monitoring threshold onto the monitoring device and use the monitoring device to acquire signals characterizing the RF parameters. The information processing module can process the signals characterizing the RF parameters to obtain a second RF monitoring threshold. The threshold determination module can determine the current RF monitoring threshold of the monitoring device based on the second RF monitoring threshold. The threshold loading module can load the current RF monitoring threshold onto the monitoring device. Therefore, the dynamic acquisition, processing, decision-making, and loading of the RF monitoring threshold are modularly implemented, enabling the monitoring device to dynamically adjust based on a threshold that matches the current process state, significantly improving the accuracy of RF monitoring.

[0025] Thirdly, this application provides a semiconductor device for processing wafers, comprising: A process chamber for generating plasma by exciting the process gas with the radio frequency, and using the plasma to process the wafer surface; A radio frequency generating device is connected to the process chamber, and the radio frequency generating device is used to generate radio frequency within the process chamber; A monitoring device is installed in or communicates with the process chamber. The monitoring device is used to monitor the radio frequency parameters in the process chamber and output a signal characterizing the radio frequency parameters. The processor, which is communicatively connected to the monitoring device, is used to perform the following steps: The signal characterizing the radio frequency parameter is processed to obtain a second radio frequency monitoring threshold. Based on the second radio frequency monitoring threshold, the current radio frequency monitoring threshold of the monitoring device is determined; The current radio frequency monitoring threshold is loaded into the monitoring device.

[0026] In this embodiment, the semiconductor device includes a process chamber, a radio frequency (RF) generation device, a monitoring device, and a processor. The process chamber generates plasma by exciting the process gas with the RF, and uses the plasma to process the wafer surface. The RF generation device is connected to the process chamber and can generate RF within the process chamber. The monitoring device is located in or communicatively connected to the process chamber and can monitor the RF parameters within the process chamber and output signals characterizing the RF parameters. The processor processes the signals characterizing the RF parameters to obtain a second RF monitoring threshold, determines the current RF monitoring threshold based on the second RF monitoring threshold, and loads the current RF monitoring threshold into the monitoring device. This enables the semiconductor device to have the ability to self-optimize the RF monitoring threshold online, thereby adaptively adjusting the monitoring parameters according to the actual process conditions. This allows the monitoring device to dynamically adjust based on a threshold that matches the current process state, significantly improving the accuracy of RF monitoring.

[0027] In one possible embodiment, the processor is further configured to: Receive real-time sensor data, which includes optical sensor signals, radio frequency output power values, and current values; The data is processed using outlier filtering conditions and stored in a database.

[0028] In this embodiment, the processor can receive real-time sensor data (including optical sensor signals, radio frequency output power values, and current values), process the data using outlier filtering conditions, and store it in a database. It can preprocess and filter multi-source sensor data, thereby eliminating instantaneous noise or outliers, improving the purity of input data, and providing a reliable data foundation for subsequent threshold calculations.

[0029] In one possible embodiment, the optical sensor signal includes a reflection spectrum signal and a diffraction spectrum signal; the processor is used to perform multidimensional cross-validation of the optical sensor signal and the radio frequency current signal to generate a radio frequency health judgment.

[0030] In this embodiment, since the processor is used to perform multi-dimensional cross-verification of optical signals and radio frequency current signals to generate radio frequency health judgment, it can realize the joint analysis of optical and electrical signals, thereby enabling mutual verification from different physical dimensions and improving the accuracy of radio frequency health assessment.

[0031] In one possible embodiment, the processor is located in a server connected to the semiconductor device.

[0032] In this embodiment, since the processor is located in a server connected to the semiconductor device, the computing tasks can be deployed on a remote server. This allows for complex threshold calculations to be performed by leveraging the server's stronger computing power and storage resources, reducing the load on the device and facilitating centralized management and algorithm upgrades.

[0033] In one possible embodiment, the semiconductor device further includes a host computer that is communicatively connected to the processor; the monitoring device is integrated into the host computer software of the semiconductor device, and the host computer software configures monitoring parameters through an interactive interface.

[0034] In this application, since the monitoring device is integrated into the host computer software of the semiconductor device, the host computer is connected to the processor, and the host computer software configures the monitoring parameters through the interactive interface. The monitoring function can be software-based and configured on the user interface, which makes it convenient for operators to view and adjust the monitoring parameters intuitively, thereby improving the human-computer interaction and the flexibility of equipment configuration. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is an architectural diagram of a semiconductor device disclosed in an embodiment of this application; Figure 2 This application discloses a method for loading a radio frequency monitoring threshold. Figure 3 This application discloses a specific method for loading a radio frequency monitoring threshold. Figure 4 This application discloses a radio frequency monitoring threshold loading system. Figure 5 This application discloses a radio frequency monitoring threshold loading device. Figure 6 This application discloses a specific radio frequency monitoring threshold loading device. Figure 7 This is another specific radio frequency monitoring threshold loading device disclosed in the embodiments of this application.

[0037] Explanation of reference numerals in the attached figures: 10-Semiconductor equipment; 11-Gas supply device; 12-RF generation device; 13-Process chamber; 14-Monitoring device; 15-Processor; 16-Wafer; 151 - Data acquisition unit; 152 - Calculation unit; 153 - Matching unit; 21-Signal acquisition module; 22-Information processing module; 23-Threshold determination module; 24-Threshold loading module; 30 - Loading device for radio frequency monitoring threshold; 301 - Memory; 302 - Transceiver. Detailed Implementation

[0038] This application's technical solution can be applied to semiconductor equipment, such as etching equipment and deposition equipment. The etching equipment generates radio frequency (RF) power to produce plasma containing electrons, ions, and free radicals in a vacuum environment. The direction of ion bombardment is controlled by an electric field, enabling the etching operation on the wafer within the process chamber. During this process, sensors are needed to monitor fluctuations in RF parameters in real time. When RF parameters exceed dynamic thresholds, timely alarms or automatic adjustments are required to prevent uneven etching or wafer scrap due to RF instability. Furthermore, as the process chamber is used for longer periods (e.g., chamber aging, component wear), the sensor signals will slowly drift. This solution, however, directly acquires the signal representing the RF parameters from the monitoring device based on a first RF monitoring threshold during wafer processing. This signal is then processed to obtain a second RF monitoring threshold. Combining the first and second RF monitoring thresholds, a determined current RF monitoring threshold adapted to the monitoring device is loaded, enabling dynamic updating and real-time adjustment of the RF monitoring threshold. This allows the monitoring device to continuously adjust based on thresholds matching the current process state, significantly improving the accuracy of RF monitoring.

[0039] Example 1

[0040] like Figure 1As shown, this application discloses an architecture diagram of a semiconductor device 10, which includes a gas supply device 11, a radio frequency (RF) generation device 12, a process chamber 13 (containing process gas generated by the gas supply device 11 and a wafer 16 for which processing is to be performed), a monitoring device 14, and a processor 15. The process chamber 13 is used to generate plasma by RF excitation of the process gas, and to process the surface of the wafer 16 using the plasma. The RF generation device 12 is connected to the process chamber 13 and is used to generate RF within the process chamber 13. The gas supply device 11 is connected to the process chamber 13 and is used to supply process gas to the process chamber 13. The monitoring device 14 is disposed in or communicatively connected to the process chamber 13 and is used to monitor RF parameters within the process chamber 13 and output signals characterizing the RF parameters. The processor 15 is communicatively connected to the monitoring device 14. The processor 15 processes the signals characterizing the RF parameters to obtain a second RF monitoring threshold; and based on the second RF monitoring threshold, determines the current RF monitoring threshold of the monitoring device 14; finally, the current RF monitoring threshold is loaded into the monitoring device 14. In one possible implementation, the processor 15 is further configured to: receive real-time sensor data, including optical sensor signals, radio frequency output power values, and current values; process the data using outlier filtering conditions and store it in a database. Furthermore, the optical sensor signals in the semiconductor device 10 may include reflection spectrum signals and diffraction spectrum signals; the processor 15 is configured to perform multi-dimensional cross-validation between the optical sensor signals and the radio frequency current signals to generate a radio frequency health assessment. In a possible implementation, the processor 15 may be located in a server connected to the semiconductor device 10. In a possible implementation, the semiconductor device 10 may also include a host computer, which is communicatively connected to the processor 15; and the monitoring device 14 is integrated into the host computer software of the semiconductor device 10, which configures monitoring parameters through an interactive interface.

[0041] Furthermore, the wafer 16 to be processed is placed in the process chamber 13. The radio frequency (RF) generation device 12 provides RF parameters to the process chamber 13 to excite the process gas introduced by the gas supply device 11 to form plasma, which is used to etch the surface of the wafer 16. Before or at the beginning of wafer 16 processing, the processor 15 first loads a first RF monitoring threshold (e.g., a threshold matched from a historical process database, or a preset safety default value) into the monitoring device 14. Subsequently, the data acquisition unit 151 in the processor 15 starts working: during wafer 16 processing, the data acquisition unit 151 acquires signals characterizing the RF parameters in real time using a high-frequency sampling method. Specifically, the signals include spectral signals (reflection spectrum and diffraction spectrum) from optical sensors, the ejector pin movement time from the ejector pin position encoder, and the RF current signal from the current sensor on the RF transmission path. This approach comprehensively captures the transmission efficiency, coupling state, and plasma stability information of radio frequency (RF) parameters from three different physical dimensions: optical, mechanical, and electrical. This overcomes the limitations of traditional solutions that rely on a single data source (such as optical signals or conductive current) and cannot detect faults in specific scenarios. Correspondingly, the data acquisition unit 151 transmits the acquired signals representing RF parameters to the computing unit 152. The computing unit 152 uses online learning algorithms (such as an adaptive adjustment strategy based on reinforcement learning) to process these signals in real time. Specifically, the computing unit 152 uses a first RF monitoring threshold as the initial threshold, and takes the current real-time signal value, the signal statistical characteristics (mean, standard deviation) within the recent time window, and the current process step identifier as state inputs. It drives the threshold to iteratively fine-tune towards the optimal boundary through a preset reward function (e.g., a positive reward if the signal is within the threshold range and the process is normal; a negative reward if the signal exceeds the threshold and triggers a false alarm). As the wafer 16 processing continues, the computing unit 152 continuously outputs updated thresholds. When the threshold change is less than a preset convergence threshold in multiple consecutive updates, the convergence value is determined as the second RF monitoring threshold. This allows monitoring thresholds to overcome the limitations of static presets, enabling real-time tracking of chamber aging, component drift, and process fluctuations. Therefore, it avoids false alarms or missed alarms caused by fixed thresholds, significantly improving the adaptability and long-term accuracy of RF health monitoring. Next, the matching unit 153 intervenes in the decision-making process, first acquiring the wafer information of the current wafer 16 (such as batch number and material type) and the currently executing processing parameters (such as recipe name, RF power setting, and chamber pressure). Then, the matching unit 153 matches and verifies the second RF monitoring threshold generated by the calculation unit 152 against these process conditions: determining whether the historical process conditions corresponding to the second threshold (i.e., the wafer information and processing parameters at the time the threshold was generated) are consistent with or highly similar to the current process conditions.If the matching is successful, the matching unit 153 determines the second RF monitoring threshold as the current RF monitoring threshold; if the matching fails (e.g., the current process formula is completely different from the formula used to generate the second threshold), the matching unit 153 reverts to using the first RF monitoring threshold as the current RF monitoring threshold. This avoids incorrectly applying a high-precision threshold generated under a specific process condition to a completely different process scenario, preventing monitoring failures caused by threshold-process mismatch, ensuring dynamic optimization while guaranteeing the safety and reliability of RF monitoring. Finally, the monitoring device 14 formally loads the current RF monitoring threshold determined by the matching unit 153 into the real-time monitoring. During the subsequent processing of the wafer 16, the monitoring device 14 continuously acquires real-time processing data and compares it with the corresponding threshold; if the real-time processing data falls within the corresponding threshold range, the RF health is determined to be normal; if it exceeds the threshold range, an alarm can be triggered immediately or a report can be sent to the host computer. Simultaneously, the monitoring device 14 can also perform dual matching verification, that is, not only comparing the numerical matching relationship between the real-time data and the threshold, but also verifying whether the current process parameters are consistent with the historical process parameters associated with the threshold, thereby further reducing the false alarm rate.

[0042] Therefore, this embodiment, through the division of labor and cooperation of the data acquisition unit 151, the calculation unit 152, and the matching unit 153, organically combines multi-source sensor acquisition, online learning and dynamic optimization, and adaptive matching of process conditions to form a complete intelligent loading and updating mechanism for radio frequency monitoring thresholds. This enables semiconductor equipment to have the ability to self-optimize radio frequency monitoring thresholds online, thereby adaptively adjusting monitoring parameters according to actual process conditions. This allows the monitoring device to always dynamically adjust based on thresholds that match the current process state, significantly improving the accuracy of radio frequency monitoring.

[0043] Example 2

[0044] like Figure 2 As shown in the figure, this application discloses a method for loading a radio frequency monitoring threshold, applied to the aforementioned semiconductor device. The loading method specifically includes: Step S11: During the wafer processing, the first radio frequency monitoring threshold is applied to the monitoring device, and the monitoring device is used to acquire signals characterizing radio frequency parameters.

[0045] In this application, a first RF monitoring threshold needs to be pre-determined before wafer processing. This first RF monitoring threshold refers to an initial monitoring threshold generated before wafer processing, used as a benchmark for judging RF parameters at the current moment. Specifically, it can be determined through historical process data matching or equipment default values, ensuring both security and process specificity. Specifically, when a process engineer or host computer system needs to configure an initial monitoring threshold for a wafer to be processed, it can first obtain the wafer information (such as wafer batch number, material type, film structure) and the processing parameters to be executed in the process chamber (such as process recipe name, RF power setting, chamber pressure, gas flow rate, etc.). Then, it retrieves historical process data records from the historical process database that match the aforementioned wafer information and processing parameters. If one or more matching historical process data are retrieved (i.e., similar wafers have been processed under the same or highly similar process conditions in the past), the corresponding RF monitoring thresholds, which have been previously verified as effective through dynamic algorithms, are extracted and used as the RF monitoring threshold for the target process data. This threshold is then determined as the first RF monitoring threshold for the current wafer processing task. This provides a reasonable initial monitoring benchmark for the new batch of wafers, improving the effectiveness of initial monitoring. If no matching historical process data is found, a safe default threshold preset by the equipment manufacturer and verified through extensive experiments can be directly used as the first RF monitoring threshold. It should be noted that when multiple monitoring thresholds correspond to the same process conditions, the most suitable monitoring threshold can be output according to a preset strategy (e.g., prioritizing versions generated within the most recent week without alarm records). This approach leverages historical success experience to provide a reasonable initial monitoring benchmark for the current processing, avoiding the use of overly lenient or overly stringent fixed thresholds due to a lack of reference. This achieves the technical effect of ensuring that the first RF monitoring threshold is both safe and process-specific, laying a good foundation for subsequent dynamic adjustments.

[0046] Furthermore, after determining the first RF monitoring threshold, the formal wafer fabrication can begin. During fabrication, the determined first RF monitoring threshold is loaded into the monitoring device as the initial judgment benchmark for the current moment. Simultaneously, the monitoring device begins to acquire signals characterizing RF parameters in real time. Specifically, during process runs where RF power has been output and plasma has been ignited or stabilized, the monitoring device can monitor relevant sensor data based on the first RF monitoring threshold. It should be noted that signals characterizing RF parameters refer to one or more measurable physical quantities that are directly or indirectly related to the RF energy state output by the RF generation device within the process chamber of the semiconductor equipment, and that can be acquired, processed, and used by the monitoring device to characterize RF performance or process stability. RF parameters include, but are not limited to: RF current signals (such as RMS value, peak value, waveform variation coefficient), RF voltage signals, spectral signals (such as average light intensity at characteristic wavelengths, fluctuation amplitude, peak-valley characteristics), pin movement time (such as the average time and fluctuation of multiple movement cycles), forward power, reflected power, impedance matching state parameters, etc. The time taken for a pusher to move refers to the time it takes for a pusher to move from its initial position to its target position during the process of lifting or lowering a wafer within the process chamber. This allows for the comprehensive acquisition of radio frequency (RF) parameters from three dimensions: optical (spectral), mechanical (pusher motion), and electrical (current). Therefore, it enables cross-validation of multi-source heterogeneous data, improves the accuracy of monitoring threshold adjustments, and effectively overcomes the shortcomings of traditional solutions that rely on only a single signal (such as optical or current signals) and cannot detect faults in specific scenarios. This significantly improves the comprehensiveness, reliability, and robustness of RF health status monitoring.

[0047] Step S12: Process the signal characterizing the radio frequency parameters to obtain the second radio frequency monitoring threshold.

[0048] The above steps load the first RF monitoring threshold into the monitoring device, which then acquires signals characterizing RF parameters. These signals, after filtering, can be stored in a database. Through dynamic processing, a new monitoring threshold, denoted as the second RF monitoring threshold, can be generated that better reflects the current chamber status and process conditions. The second RF monitoring threshold is dynamically generated based on real-time signals acquired during the current wafer fabrication process, reflecting the actual state of the current chamber. It should be noted that the database can store each monitoring threshold along with its corresponding generation time, wafer batch, process recipe, sample size, and verification results (whether there were false alarms / missed alarms).

[0049] Furthermore, the processing can be performed in real-time during the process run, or in batches after a batch of wafers or a single wafer has been processed. The processor, which houses the computing module, reads all signals representing RF parameters acquired during the current wafer processing from a database. This processor can be a standalone server connected to the semiconductor equipment or a computing unit within a host computer. The processor first verifies the validity of these signals. Specifically, it checks whether each signal value is within a preset reasonable physical range (e.g., current is not negative, light intensity does not exceed the sensor's range) and uses pre-configured outlier filtering conditions (e.g., based on local outlier factors) to remove invalid data caused by momentary sensor jitter, electromagnetic interference, or abnormal data acquisition. The signals that have passed validity verification form a valid sample set. Then, the various signals in the valid sample set are comprehensively analyzed to extract characteristic parameters that reflect the performance of the RF parameters. For example, for spectral signals (including reflectance and diffraction spectra), the processor extracts the average light intensity at characteristic wavelengths, the amplitude of light intensity fluctuations, or the peak and valley characteristics of the spectral curve. These characteristics can characterize the concentration changes of active particles in the plasma and the efficiency of RF parameters coupling to the plasma. For the ejector pin movement time, the processor can calculate the average time and fluctuation of multiple movement cycles. This parameter can indirectly reflect the consistency of mechanical movement within the cavity, thus relating to the stability of the cavity's physical state during wafer transfer. For current signals related to RF parameters, the processor extracts features such as the effective value, peak value, and coefficient of variation of the current waveform. These features directly reflect the impedance matching and power transmission stability in the RF transmission path. Finally, combining the current signal characteristics obtained from processing various signals, a preset dynamic threshold generation strategy can be used to calculate the second RF monitoring threshold. This threshold typically includes an upper limit and a lower limit, forming a monitoring interval. It is important to note that the second RF monitoring threshold is a dynamically generated value, different from the static first RF monitoring threshold loaded in step S1. The first threshold may come from historical matching data or device default values, while the second threshold is calculated entirely based on signals acquired in real time during the current wafer processing, thus better reflecting the actual state of the current cavity.

[0050] Step S13: Determine the current radio frequency monitoring threshold of the monitoring device based on the second radio frequency monitoring threshold.

[0051] In this embodiment, a second radio frequency (RF) monitoring threshold is dynamically generated through the above steps. A first RF monitoring threshold (from historical matching or a default value) is also present. A reasonable choice can then be made between these two thresholds to determine the current RF monitoring threshold that should actually be loaded into the monitoring device for real-time monitoring. It should be noted that the current RF monitoring threshold refers to the threshold that is ultimately loaded into the monitoring device for real-time monitoring; it may be the dynamically generated second RF monitoring threshold or a reverted first RF monitoring threshold.

[0052] Specifically, after the second RF monitoring threshold is generated, it can be combined with the wafer information of the wafer currently being processed and the processing parameters used in this processing. The wafer information mentioned here includes, but is not limited to: wafer batch number, wafer serial number, wafer material type (e.g., silicon, germanium, compound semiconductor), wafer size (e.g., 200mm, 300mm), film structure and thickness, etc. Processing parameters include, but are not limited to: recipe name, RF power setting, process chamber pressure setting, process gas type and flow rate, electrostatic chuck adsorption voltage, process temperature, process duration, etc. Furthermore, the wafer information and processing parameters can be directly read from the equipment's host computer or parsed from the processing instructions issued by the manufacturing execution system. It can be understood that the wafer information and processing parameters provide a clear process condition index for subsequent threshold matching judgments, ensuring that the selection of the monitoring threshold strictly corresponds to the current actual process scenario.

[0053] Furthermore, the acquired current wafer information and processing parameters are matched with a generated second RF monitoring threshold to determine whether the second RF monitoring threshold is suitable for the current wafer processing. Specifically, when storing the second RF monitoring threshold, historical wafer information and historical processing parameters that the threshold is compatible with can be recorded simultaneously. To improve robustness, a "fuzzy matching" level can be set: if the current wafer information and processing parameters are highly similar to the corresponding information corresponding to the second RF monitoring threshold (e.g., similarity greater than 90%), it can be considered a match. It is understood that information matching can verify the applicability of the dynamically generated second threshold, avoiding the erroneous application of a high-precision threshold generated under a specific process condition to completely different process scenarios. Furthermore, when the judgment result shows that the current wafer information and processing parameters match the historical conditions when the second RF monitoring threshold was generated, it can be considered that the second RF monitoring threshold is also applicable to the current processing scenario. Therefore, the processor uses this second RF monitoring threshold as the current RF monitoring threshold. When the judgment result indicates that the current wafer information and processing parameters do not match the historical conditions when the second RF monitoring threshold was generated, it can be considered that the second RF monitoring threshold may not be applicable to the current process scenario (for example, the second threshold was generated from a completely different process recipe or different batches of wafers). To avoid misuse, the processor abandons the use of the second threshold and instead uses the first RF monitoring threshold as the current RF monitoring threshold. Although the first RF monitoring threshold is not as instantly adaptive as the dynamically generated second threshold, it has a wider range of applicability and can provide a safe and reliable monitoring benchmark when a suitable dynamic threshold is lacking. It is understandable that, by combining specific process batches and process conditions, a suitable monitoring threshold can be intelligently selected from the first and second RF monitoring thresholds to avoid false alarms or missed alarms due to process offsets or differences between different wafers, thereby improving the applicability of the threshold.

[0054] Step S14: Load the current radio frequency monitoring threshold into the monitoring device.

[0055] In this embodiment, the current radio frequency (RF) monitoring threshold has been determined through the above steps (this threshold may be a dynamically generated second RF monitoring threshold or a first RF monitoring threshold). Then, the current RF monitoring threshold is applied to the monitoring device, and this threshold is used to judge the real-time data of the wafer fabrication process, outputting meaningful RF monitoring results. Specifically, after the current RF monitoring threshold is loaded into the monitoring device, it is used as the benchmark for all subsequent real-time judgments. This threshold is persistently stored in the configuration parameters until the next threshold update (a new threshold is reloaded before the next wafer is processed).

[0056] In one specific embodiment, after the current RF monitoring threshold is loaded into the monitoring device, real-time signals from optical sensors, RF current sensors, and ejector pin position sensors can be continuously read during the wafer processing, i.e., real-time processing data during the wafer processing can be acquired in real time. It should be noted that this real-time processing data can be the same type of data as the "signals characterizing RF parameters" acquired in the above steps (such as spectral signals, current signals, ejector pin movement time), but the difference is that the signals characterizing RF parameters are mainly used to dynamically generate thresholds, while the real-time processing data acquired here is used to compare with the currently loaded threshold in real time to determine whether the RF parameters are normal at the current moment. The sampling rate of the real-time processing data is usually high to ensure that instantaneous abnormal fluctuations can be captured. Furthermore, if the real-time processing data is within the current RF monitoring threshold range, it is determined as a "match," indicating that the current RF parameters are performing normally. The monitoring device can generate a "normal" or "pass" monitoring result and continue monitoring the next sampling point. If the real-time processing data exceeds the current RF monitoring threshold range, it is determined as a "mismatch," indicating that the current RF parameters are abnormal (such as overshoot, undershoot, or excessive fluctuation). The monitoring device immediately generates an "abnormal" or "alarm" monitoring result. In this way, by comparing real-time data with dynamic thresholds, abnormal RF states can be identified in a timely manner, improving the real-time monitoring and fault alarm capabilities of the processing process.

[0057] It should be noted that, to further improve the reliability of monitoring results and avoid false alarms caused by accidental fluctuations in single data points, while comparing the matching relationship between real-time data and corresponding thresholds, the matching relationship between the current processing parameters and the processing parameters corresponding to the current RF monitoring threshold can be verified. This involves employing a dual-matching strategy to generate the final RF monitoring results. Dual-matching refers to, based on a preliminary judgment of whether the real-time data matches or does not match the threshold, especially in edge cases (where real-time data is just near the threshold boundary), determining the matching relationship between the current processing parameters and the processing parameters corresponding to the current RF monitoring threshold. This combines data-level matching with process condition-level matching, effectively distinguishing between "genuine RF parameter anomalies" and "threshold mismatches caused by process parameter changes," significantly reducing false alarm rates and improving the interpretability and reliability of monitoring results.

[0058] Therefore, this application determines the first RF monitoring threshold based on historical process data matching or safety default values ​​before processing, and uses historical success experience to provide a reasonable initial monitoring benchmark for the current processing. This avoids using overly lenient or stringent fixed thresholds due to a complete lack of reference, ensuring that the first threshold is both safe and process-specific. Furthermore, it simultaneously acquires spectral signals (optical), pin movement time (mechanical), and RF current signals (electrical), comprehensively collecting RF parameter characterization parameters from three dimensions. This enables cross-validation of multi-source heterogeneous data, effectively overcoming the shortcomings of traditional solutions that rely on a single signal and cannot detect failures in specific fault scenarios. This significantly improves the comprehensiveness, reliability, and robustness of RF health status monitoring. Subsequently, a second RF monitoring threshold is dynamically generated based on the monitored real-time signals. This threshold is calculated entirely based on current wafer processing data, accurately reflecting the current chamber state and solving the problem that fixed thresholds cannot adapt to chamber aging and component drift. Furthermore, by acquiring wafer information and processing parameters, it is determined whether the second RF monitoring threshold matches the current process conditions: if they match, a dynamically optimized second RF monitoring threshold is used; if they do not match, it reverts to the first RF monitoring threshold, which has a wider applicability. This avoids misusing high-precision thresholds generated under specific process conditions in different scenarios and ensures reliable system operation even when a suitable dynamic threshold is unavailable. Finally, the current threshold is loaded into the monitoring device for real-time comparison of processing data at a high sampling rate. A dual matching strategy can be optionally employed (simultaneously verifying the matching relationship between data and thresholds, and the matching relationship between process parameters and the source process of the thresholds), effectively distinguishing between genuine RF parameter anomalies and threshold mismatches caused by process parameter changes, significantly reducing false alarm rates and improving the interpretability and reliability of monitoring results.

[0059] Example 3

[0060] like Figure 3 As shown in the figure, this application discloses a method for loading a radio frequency monitoring threshold, applied to a semiconductor device. The semiconductor device includes a process chamber, a radio frequency generation device, a gas supply device, and a monitoring device, specifically including: Step S21: During the wafer processing, the first radio frequency monitoring threshold is applied to the monitoring device, and the monitoring device is used to acquire signals characterizing radio frequency parameters.

[0061] Step S22: Perform data statistics on the signals characterizing the radio frequency parameters to obtain the corresponding current data distribution characteristics.

[0062] In this embodiment, the above steps yield various signals characterizing radio frequency (RF) parameters during the current wafer fabrication process (e.g., spectral signals, RF current signals, ejector pin movement time, etc.). These data can then be statistically analyzed to transform the raw sensor data into probability distribution characteristics (e.g., central tendency and dispersion) that reflect the typical performance and fluctuation range of RF parameters under current process conditions, providing a basis for subsequent updates to monitoring thresholds. Specifically, the signals to be statistically analyzed include at least: the effective value of the RF current (reflecting RF power transmission efficiency), the average light intensity at spectral characteristic wavelengths (reflecting plasma density and RF coupling efficiency), and the average ejector pin movement time (reflecting mechanical consistency). For each signal, the size of the statistical window also needs to be determined, i.e., how many recent sampling points or how much recent data are used for statistical analysis. The statistical window can be a fixed length (e.g., the most recent 1000 sampling points, the most recent 10 seconds of data) or a dynamically adjusted length depending on the process steps (e.g., a wider window in the steady-state process stage and a narrower window in the transient startup stage). Correspondingly, data statistics can include central tendency statistics and dispersion statistics. Central tendency describes the typical value or central location of data, with common indicators such as the mean (arithmetic mean) and median. The mean or median can be freely chosen based on signal characteristics, or both can be calculated and cross-checked. For example, when the difference between the mean and median exceeds a preset threshold, it indicates a significant skew in the signal distribution, requiring further investigation of data quality or process status. In the dispersion statistics process, key indicators include standard deviation, interquartile range, and range. In practical applications, the processor can select either standard deviation or interquartile range as the primary dispersion indicator based on signal characteristics and noise levels. For signals with low noise (such as RF current in a stable phase), standard deviation is preferred; for signals potentially experiencing spike interference (such as light intensity signals), interquartile range is preferred. Furthermore, statistical analysis yields a complete picture of the current data distribution characteristics.

[0063] Step S23: Update the first radio frequency monitoring threshold based on the current data distribution characteristics to obtain the second radio frequency monitoring threshold.

[0064] In this embodiment, the current data distribution characteristics corresponding to the signal characterizing the radio frequency parameters have been obtained through the above steps. Then, a recommended threshold can be calculated based on the current data distribution characteristics. This recommended threshold reflects the reasonable monitoring boundary considered only from the current wafer processing data. Furthermore, during the calculation of the recommended threshold, it is necessary to ensure that the recommended threshold does not exceed preset physical limits (e.g., the radio frequency current cannot be negative, and the light intensity cannot exceed the sensor range) to guarantee the rationality of the recommended value. This transforms real-time statistics into specific monitoring boundary values, thus objectively reflecting the maximum allowable deviation of the current chamber within the normal fluctuation range, thereby achieving the effect of real-time matching between the recommended threshold and the current process state. Further, a weighted fusion method can be used to combine the first radio frequency monitoring threshold and the recommended threshold according to a certain weight ratio to balance the historical benchmark and the current real-time state, obtaining the final second radio frequency monitoring threshold. The weight ratio used here can be a weight ratio set in advance based on empirical values, or a weight ratio dynamically determined considering the service life of the process chamber. After the weighted calculation is completed, the rationality of the obtained second radio frequency monitoring threshold can be verified. If a problem is detected during verification, restrictive measures can be taken, such as truncating the threshold beyond a reasonable range to a boundary value, or abandoning the update (keeping the first threshold unchanged), and logging a warning. This introduces real-time correction while maintaining threshold stability, balancing threshold smoothness and rapid response capabilities. Accordingly, the second RF monitoring threshold, which has been verified to be correct, can be determined as the current RF monitoring threshold corresponding to the current wafer processing stage.

[0065] Step S24: Through online learning, the first radio frequency monitoring threshold is adaptively adjusted based on the signal representing the radio frequency parameters to obtain the second radio frequency monitoring threshold.

[0066] In this embodiment, the current data distribution characteristics corresponding to the signal representing the radio frequency parameters have been obtained through the above steps. Subsequently, online learning technology (an adaptive adjustment strategy within a reinforcement learning framework) can be used to iteratively optimize the first radio frequency monitoring threshold in real time during wafer fabrication, continuously bringing it closer to the optimal judgment boundary under the current chamber and process conditions. It is understood that online learning allows the monitoring threshold to be continuously adjusted based on the processing data of each wafer or even each sampling cycle, thereby achieving immediate response to changes in the chamber state. Reinforcement learning is used to adaptively adjust the monitoring threshold: the agent executes actions such as "raising the threshold," "lowering the threshold," or "maintaining the threshold" based on the current threshold and real-time signal value, and receives reward feedback based on whether false alarms or missed alarms occur subsequently, thus gradually learning the optimal threshold setting strategy. Furthermore, online learning initialization uses the first radio frequency monitoring threshold as the starting point of the initial strategy, i.e., the initial threshold before the agent begins learning. Simultaneously, hyperparameters such as the learning rate (e.g., 0.1) and exploration rate (e.g., 0.2) are set to balance the relationship between "utilizing existing experience" and "exploring new actions." This approach can transform the threshold adjustment problem into an online learning decision-making process, thus overcoming the limitations of traditional fixed thresholds or offline training models. It allows for adaptive adjustment based on the statistical characteristics of real-time data, enabling the threshold to change with process fluctuations. This achieves the technical effect of automatically capturing non-stationary changes and long-term trends in the process, and realizing self-updating of the threshold without manual intervention.

[0067] Online learning refers to the use of algorithms such as reinforcement learning during wafer fabrication to iteratively adjust monitoring thresholds based on real-time signals and reward feedback, gradually approaching the optimal judgment boundary. Specifically, during wafer fabrication, signals representing radio frequency parameters can be acquired in real-time at intervals of each sampling period or each fixed time window (e.g., every 100 milliseconds or after each wafer is processed). An action is then selected and executed according to the current strategy. Specifically, the current state is observed, and the latest real-time signal value (e.g., the current effective value of the radio frequency current) and statistical characteristics within the most recent sliding window (e.g., the average and standard deviation of the signal over the past second) are read from the monitoring device or data buffer. This information is combined to form the current state. Then, based on the current state and the current strategy (e.g., an ε-greedy strategy, where the optimal action is selected with a probability of 1-ε, and other actions are randomly explored with a probability of ε; ε represents the exploration probability of the greedy strategy, a decimal between 0 and 1, which typically decays over time), the online learning agent selects an action. For example, if the average signal value of the recent sampling points is significantly close to the current upper threshold and the fluctuation shows an increasing trend, the "increase upper threshold" action might be selected to avoid frequent false alarms. If the average signal value is far below the upper threshold and the fluctuation is small, the "decrease upper threshold" action might be selected to tighten the monitoring window and improve sensitivity. If the average signal value is stable in the middle of the threshold range, the "keep threshold unchanged" action can be selected. The selected action is then applied to the current monitoring threshold to obtain an updated temporary threshold. The process continues for a period of time (e.g., one decision cycle, or the completion of one wafer processing cycle), recording whether any alarms occur under this temporary threshold and whether the alarms are correct. The reward value is calculated based on the actual alarm results and subsequent process verification (e.g., judging wafer quality through online measurement equipment or operator feedback). Then, reinforcement learning update rules (such as policy gradient methods) are used to update the action value function or policy network, increasing the probability of selecting actions that generate higher rewards in similar states. Understandably, online learning can form a closed-loop adaptive optimization process, which allows the monitoring threshold to be finely adjusted towards a better direction at every moment, and can quickly approach the optimal monitoring boundary under non-stationary and non-linear changing chamber conditions.

[0068] Step S25: Determine the current radio frequency monitoring threshold of the monitoring device based on the second radio frequency monitoring threshold.

[0069] Step S26: Load the current radio frequency monitoring threshold into the monitoring device.

[0070] For more detailed processing procedures of steps S21, S25 and S26, please refer to the corresponding content disclosed in the foregoing embodiments, which will not be repeated here.

[0071] Therefore, statistical analysis of signals characterizing RF parameters reveals their distribution characteristics, objectively reflecting the chamber's true performance under current technological conditions. Based on these distribution characteristics, a weighted fusion and rationality verification of the first RF monitoring threshold generates a second RF monitoring threshold. This allows the monitoring threshold to break free from static presets, smoothly adjusting to changes such as chamber aging and component drift, avoiding false alarms or missed alarms caused by fixed thresholds, and ensuring the effectiveness and rationality of the monitoring threshold. Furthermore, employing a reinforcement learning framework for online learning enables the threshold to rapidly approach the optimal judgment boundary under non-stationary and non-linear chamber conditions, possessing continuous self-evolution capabilities and significantly reducing manual intervention.

[0072] like Figure 4 As shown, this embodiment discloses a loading system for radio frequency monitoring thresholds, and... Figure 1 Unlike the corresponding embodiment, which uses computing unit 152 to perform real-time online reinforcement learning and iteratively fine-tune RF monitoring thresholds during wafer 16 processing, this embodiment focuses on historical data mining across wafers and batches. It pre-determines high-confidence personalized monitoring thresholds before wafer 16 processing begins, achieving proactive and accurate monitoring of the RF health of process chamber 13. Furthermore, the loading system in this embodiment is based on… Figure 1 The semiconductor device 10 shown implements dynamic updates and intelligent matching loading of radio frequency monitoring thresholds. This loading system deploys a computing module (corresponding to...). Figure 1 The computing unit 152), monitoring module (corresponding to monitoring device 14 and processor 15, matching unit 153), and process module (corresponding to Figure 1 The gas supply device 11, radio frequency generation device 12, process chamber 13 and wafer 16 are included, along with the real-time data module (corresponding to...). Figure 1 The data acquisition unit 151 in the middle works in concert with other modules to achieve dynamic updating and intelligent matching of monitoring thresholds. The specific content of this embodiment is as follows.

[0073] First, during wafer fabrication, the real-time data module collects various sensor signals using high-frequency sampling, including optical sensors (reflection spectrum, diffraction spectrum), radio frequency current sensors (output current, reflected power), and pin movement time signals. These signals are periodically read by the computation module (e.g., after each wafer is processed or after accumulating a fixed number of samples). The computation module internally deploys a machine learning algorithm, which can employ a dynamic threshold adjustment method based on statistical indicators or an online adaptive adjustment strategy using reinforcement learning. The statistical algorithm calculates the current data distribution characteristics (mean, standard deviation) and, based on preset weighting coefficients, weights and fuses the first radio frequency monitoring threshold (a threshold matched from the historical process database or the equipment's preset safety default value) with the real-time recommended threshold (a threshold statistically obtained based on the distribution characteristics) to generate an updated monitoring threshold, i.e., the second radio frequency monitoring threshold. The reinforcement learning algorithm uses the first radio frequency monitoring threshold as the initial state and iteratively adjusts the threshold based on the reward signals from real-time signal feedback (positive reward for correct alarms, negative reward for false alarms or missed alarms), gradually approaching the optimal judgment boundary.

[0074] Then, after generating a new RF monitoring threshold, the calculation module can determine whether to save the new threshold to the process database. The criteria for this determination include: whether the difference between the new second RF monitoring threshold and the current first RF monitoring threshold exceeds a preset threshold (e.g., a change rate > 5%), and whether the process conditions corresponding to the second RF monitoring threshold are representative (e.g., sufficient sample size, normal process results). If the saving conditions are met, the calculation module can save the second RF monitoring threshold along with the corresponding wafer information (batch, material type) and processing parameters (recipe name, RF power, chamber pressure, etc.) to the process database. If the saving conditions are not met, the new threshold is discarded, awaiting the next calculation. It is understandable that the above steps allow the monitoring threshold to be continuously optimized as the actual state of the chamber changes, avoiding the failure of fixed thresholds due to chamber aging or component drift.

[0075] Furthermore, when a new wafer is ready for processing, the monitoring module first acquires the wafer's information (such as wafer batch, material type, and film structure) and the machine's real-time processing information (i.e., the process recipe to be executed, RF power setting, chamber pressure, gas flow rate, etc.). Using this information as an index, the monitoring module performs a matching query in the process database. Specifically, it compares the current wafer information and processing parameters with the process conditions corresponding to each historical monitoring threshold saved in the database. If one or more highly matching records are found (e.g., identical recipe names and key parameter deviations <5%), the monitoring module outputs this historical monitoring threshold as the dynamic threshold for successful matching, i.e., the first RF monitoring threshold. If no matching record is found, the monitoring module calls a preset monitoring threshold (i.e., the equipment's default safety threshold or a conservative threshold based on extensive process verification) as the output. Furthermore, before outputting the monitoring threshold, the monitoring module can perform a verification: it performs a secondary verification by comparing the matched historical monitoring threshold with the current real-time status of the machine (such as the cumulative RF power-on time of the chamber and the number of wafers since the last maintenance) to ensure the applicability of the threshold. Understandably, the above steps enable "on-demand matching" of thresholds, using high-precision dynamic thresholds when reliable historical data is available and safe, universal monitoring thresholds when data is lacking, thus ensuring both monitoring accuracy and the robustness of the process system.

[0076] Furthermore, the first RF monitoring threshold output by the monitoring module is transmitted to the process module and the real-time data module respectively. The process module is responsible for executing the wafer fabrication recipe and loading the received threshold into the monitoring device as the current monitoring threshold before process start. It should be noted that the operation of updating the monitoring threshold can be performed when processing different wafers, or at different process steps or different time points on the same wafer. The real-time data module continuously collects real-time sensor data during wafer fabrication and compares the value of each sampling point with the loaded current threshold in real time: if the real-time data falls within the threshold range, the RF health is determined to be normal; if it exceeds the threshold range, an alarm signal is immediately generated, which can trigger the process module to perform protective actions (such as reducing RF power, stopping the process, or raising the ejector pin). In addition, after completing the monitoring judgment, the real-time data collected by the real-time data module is also fed back to the process database as input for the next round of calculation modules. It can be understood that by dynamically updating the RF monitoring threshold, the entire process system can maintain its adaptive capability during long-term operation.

[0077] It is understood that the RF monitoring threshold loading system in this embodiment achieves offline batch learning and periodic updating of the chamber health monitoring threshold through the calculation module, intelligent selection of general and non-general thresholds based on historical conditions through the monitoring module, and continuously expands and optimizes the process database through real-time data feedback closed loop. It should be noted that this embodiment is similar to... Figure 1 The different embodiments have different focuses. Figure 1 The corresponding embodiment uses computing unit 152 to perform online reinforcement learning in real time during the wafer 16 processing to iteratively fine-tune the threshold, focusing on dynamic fine adjustment of the threshold within a single wafer processing cycle; this embodiment adopts a combination strategy of offline batch historical data mining and online intelligent threshold matching, focusing on pre-determining a high-confidence personalized threshold based on cross-batch historical experience before processing begins, to achieve forward-looking and accurate threshold pre-setting.

[0078] In summary, this embodiment constructs a complete RF monitoring threshold loading system through machine learning dynamic updates in the computing module, intelligent matching output in the monitoring module, and closed-loop application of the process and real-time data modules, effectively solving the problem of poor monitoring reliability caused by fixed thresholds and missing matching.

[0079] like Figure 5 As shown, this embodiment discloses a device for loading a radio frequency monitoring threshold, specifically including: The signal acquisition module 21 is used to load a first radio frequency monitoring threshold onto the monitoring device during the wafer processing process, and to acquire a signal characterizing radio frequency parameters using the monitoring device. Information processing module 22 is used to process the signal characterizing the radio frequency parameters to obtain the second radio frequency monitoring threshold; The threshold determination module 23 is used to determine the current radio frequency monitoring threshold of the monitoring device based on the second radio frequency monitoring threshold; The threshold loading module 24 is used to load the current radio frequency monitoring threshold into the monitoring device.

[0080] Therefore, the technical solution of this application provides a baseline monitoring threshold at the initial stage of process operation and collects multi-dimensional radio frequency signals (such as spectral signals, ejector pin movement time, and radio frequency current signals) in real time, thus providing an initial reference and a real data basis for subsequent dynamic adjustments. Then, the real-time signals are converted into a dynamically optimized new threshold, and the resulting second radio frequency monitoring threshold reflects the true state of the current chamber and process. Subsequently, the current radio frequency monitoring threshold is determined from the first and second radio frequency monitoring thresholds and applied to real-time monitoring. This enables radio frequency monitoring to possess adaptive and highly robust threshold management capabilities, effectively solving the problem of fixed threshold mismatch with the process, and significantly improving the accuracy and real-time performance of radio frequency health monitoring.

[0081] Figure 6 This is a schematic structural diagram of the radio frequency monitoring threshold loading device 30 provided in this application. The device 30 includes a processor 15, which is coupled to a memory 301. The memory 301 is used to store computer programs or instructions and / or data. The processor 15 is used to execute the computer programs or instructions stored in the memory 301, or to read the data stored in the memory 301, in order to perform the methods in the above method embodiments.

[0082] Optionally, there may be one or more processors 15.

[0083] Optionally, the memory 301 may be one or more.

[0084] Alternatively, the memory 301 can be integrated with the processor 15, or it can be set separately.

[0085] Optionally, such as Figure 7 As shown, the device 30 also includes a transceiver 302 for receiving and / or transmitting signals. For example, the processor 15 controls the transceiver 301 to receive and / or transmit signals.

[0086] In addition, this application also provides a computer-readable storage medium storing computer instructions that, when executed on a computer, cause the methods in the various method embodiments of this application to be performed.

[0087] This application also provides a computer program product, which includes computer program code or instructions, such that when the computer program code or instructions are run on a computer, the methods in the various method embodiments of this application are executed.

[0088] Furthermore, this application also provides a chip including a processor. A memory for storing a computer program is provided independently of the chip, and the processor is used to execute the computer program stored in the memory to perform the methods in the various method embodiments of this application.

[0089] Furthermore, the chip may also include a communication interface. The communication interface may be an input / output interface or an interface circuit, etc. Furthermore, the chip may also include a memory.

[0090] It should be understood that the processor in the embodiments of this application can be an integrated circuit chip with the ability to process signals. In implementation, each step of the above method embodiments can be completed by the integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware encoding processor, or implemented by a combination of hardware and software modules in the encoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above method.

[0091] The memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), which serves as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), and synchronous link dynamic memory (SLDRAM).

[0092] It should be noted that when the processor is a general-purpose processor, DSP, ASIC, FPGA, or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, the memory can be integrated into the processor.

[0093] It should also be noted that the memory described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0094] Each of the above modules or units can be implemented through software, hardware, or a combination of both. For example, the real-time data module, monitoring module, and computing module can all be implemented based on software.

[0095] In this application, "implemented through software" means that the processor reads and executes program instructions stored in memory to implement the functions corresponding to the aforementioned modules or units. Here, the processor refers to a processing circuit capable of executing program instructions, including but not limited to at least one of the following: a central processing unit (CPU), a microprocessor, a digital signal processor (DSP), a microcontroller unit (MCU), or an artificial intelligence processor, etc., and other processing circuits capable of running program instructions. In other embodiments, the processor may also include circuits with other processing functions (such as hardware circuits for hardware acceleration, bus and interface circuits, etc.). The processor can be presented as an integrated chip, for example, as an integrated chip whose processing function only includes executing software instructions, or it can also be presented as a SoC (system on a chip), that is, on a single chip, in addition to the processing circuit capable of running program instructions (usually referred to as the "core"), it also includes other hardware circuits for implementing specific functions (of course, these hardware circuits can also be implemented separately based on ASIC or FPGA). Correspondingly, the processing functions, in addition to executing software instructions, may also include various hardware acceleration functions (such as AI calculation, encoding / decoding, compression / decompression, etc.).

[0096] In this application, "implemented in hardware" means that the functions of the above-mentioned modules or units are implemented through hardware processing circuits that do not have program instruction processing capabilities. These hardware processing circuits can be composed of discrete hardware components or integrated circuits. To reduce power consumption and size, integrated circuits are typically used. Hardware processing circuits can include ASICs (application-specific integrated circuits) or PLDs (programmable logic devices); PLDs can include FPGAs (field-programmable gate arrays), CPLDs (complex programmable logic devices), and so on. These hardware processing circuits can be a single packaged semiconductor chip (e.g., packaged as an ASIC); or they can be integrated with other circuits (e.g., CPUs, DSPs) and packaged into a single semiconductor chip. For example, multiple hardware circuits and a CPU can be formed on a silicon substrate and packaged into a single chip; this type of chip is also called a SoC. Alternatively, circuits for implementing FPGA functions and a CPU can be formed on a silicon substrate and encapsulated into a single chip; this type of chip is also called a SoPC (system on a programmable chip).

[0097] It should be noted that when this application is implemented through software, hardware, or a combination of both, different software or hardware can be used, and it is not limited to using only one type of software or hardware. For example, one module or unit can be implemented using a CPU, while another module or unit can be implemented using a DSP. Similarly, when implemented using hardware, one module or unit can be implemented using an ASIC, while another module or unit can be implemented using an FPGA. Of course, it is not limited to using the same software (e.g., all through a CPU) or the same hardware (e.g., all through an ASIC) to implement some or all modules or units. Furthermore, those skilled in the art will understand that software is generally more flexible but less performant than hardware, while hardware is the opposite. Therefore, those skilled in the art can choose software, hardware, or a combination of both based on actual needs.

[0098] The foregoing preferred embodiments have further illustrated the objectives, technical solutions, and advantages of the present invention. It should be understood that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for loading a radio frequency monitoring threshold, applied to a semiconductor device for processing wafers, the semiconductor device comprising a process chamber, a radio frequency generation device, and a monitoring device, wherein the radio frequency generation device is used to generate radio frequency within the process chamber; The monitoring device is used to monitor radio frequency parameters within the process chamber, characterized in that it includes: During the processing of the wafer, a first radio frequency monitoring threshold is applied to the monitoring device, and the monitoring device is used to acquire signals characterizing the radio frequency parameters. The signal characterizing the radio frequency parameter is processed to obtain a second radio frequency monitoring threshold. Based on the second radio frequency monitoring threshold, the current radio frequency monitoring threshold of the monitoring device is determined; The current radio frequency monitoring threshold is loaded into the monitoring device.

2. The method for loading the radio frequency monitoring threshold according to claim 1, characterized in that, Determining the current radio frequency monitoring threshold of the monitoring device based on the second radio frequency monitoring threshold includes: Obtain the wafer information and the processing parameters in the process chamber; Determine whether the second radio frequency monitoring threshold matches the wafer information and the processing parameters; If a match is found, the second radio frequency monitoring threshold is determined as the current radio frequency monitoring threshold of the monitoring device; If they do not match, the first radio frequency monitoring threshold will be determined as the current radio frequency monitoring threshold of the monitoring device.

3. The method for loading the radio frequency monitoring threshold according to claim 1, characterized in that, The signals characterizing the radio frequency parameters include spectral signals during wafer fabrication, pin movement time, and current signals related to the radio frequency parameters.

4. The method for loading a radio frequency monitoring threshold according to any one of claims 1 to 3, characterized in that, The step of processing the signal characterizing the radio frequency parameters to obtain the second radio frequency monitoring threshold includes: Data statistics are performed on the signals characterizing the radio frequency parameters to obtain the corresponding current data distribution characteristics; The first radio frequency monitoring threshold is updated based on the current data distribution characteristics to obtain the second radio frequency monitoring threshold.

5. The method for loading the radio frequency monitoring threshold according to claim 4, characterized in that, The step of updating the first radio frequency monitoring threshold based on the current data distribution characteristics to obtain the second radio frequency monitoring threshold includes: The current data distribution characteristics are calculated, and the corresponding calculation results and the first radio frequency monitoring threshold are weighted based on preset weight coefficients to obtain the second radio frequency monitoring threshold.

6. The method for loading a radio frequency monitoring threshold according to any one of claims 1 to 3, characterized in that, The step of processing the signal characterizing the radio frequency parameters to obtain the second radio frequency monitoring threshold includes: By learning online, the first radio frequency monitoring threshold is adaptively adjusted based on the signal representing the radio frequency parameters to obtain the second radio frequency monitoring threshold.

7. The method for loading a radio frequency monitoring threshold according to any one of claims 1 to 6, characterized in that, The process of determining the first radio frequency monitoring threshold includes: Identify the target process data in the historical process data that matches the processing parameters of the wafer, and determine the radio frequency monitoring threshold corresponding to the target process data as the first radio frequency monitoring threshold.

8. The method for loading a radio frequency monitoring threshold according to any one of claims 1 to 7, characterized in that, After loading the current radio frequency monitoring threshold into the monitoring device, the method further includes: Acquire real-time processing data of the wafer during the processing; Radio frequency monitoring results are generated based on the matching relationship between the real-time processed data and the current radio frequency monitoring threshold.

9. The method for loading radio frequency monitoring thresholds according to claim 8, characterized in that, The step of generating radio frequency monitoring results based on the matching relationship between the real-time processed data and the current radio frequency monitoring threshold includes: Based on the matching relationship between the real-time processing data and the current RF monitoring threshold, and the matching relationship between the wafer processing parameters and the processing parameters corresponding to the current RF monitoring threshold, RF monitoring results are generated.

10. A device for loading a radio frequency monitoring threshold, characterized in that, include: The signal acquisition module is used to load a first radio frequency monitoring threshold onto the monitoring device during the processing of the wafer, and to acquire a signal characterizing the radio frequency parameters using the monitoring device. The information processing module is used to process the signal characterizing the radio frequency parameters to obtain a second radio frequency monitoring threshold. A threshold determination module is used to determine the current radio frequency monitoring threshold of the monitoring device based on the second radio frequency monitoring threshold; A threshold loading module is used to load the current radio frequency monitoring threshold into the monitoring device.

11. A semiconductor device for processing wafers, characterized in that, include: A process chamber for generating plasma by exciting the process gas with the radio frequency, and using the plasma to process the wafer surface; A radio frequency generating device is connected to the process chamber, and the radio frequency generating device is used to generate radio frequency within the process chamber; A monitoring device is installed in or communicates with the process chamber. The monitoring device is used to monitor the radio frequency parameters in the process chamber and output a signal characterizing the radio frequency parameters. The processor, which is communicatively connected to the monitoring device, is used to perform the following steps: The signal characterizing the radio frequency parameter is processed to obtain a second radio frequency monitoring threshold. Based on the second radio frequency monitoring threshold, the current radio frequency monitoring threshold of the monitoring device is determined; The current radio frequency monitoring threshold is loaded into the monitoring device.

12. The semiconductor device according to claim 11, characterized in that, The processor is also used for: Receive real-time sensor data, which includes optical sensor signals, radio frequency output power values, and current values; The data is processed using outlier filtering conditions and stored in a database.

13. The semiconductor device according to claim 11 or 12, characterized in that, The optical sensor signal includes a reflection spectrum signal and a diffraction spectrum signal; the processor is used to perform multidimensional cross-validation of the optical sensor signal and the radio frequency current signal to generate a radio frequency health judgment.

14. The semiconductor device according to any one of claims 11 to 13, characterized in that, The processor is located in a server connected to the semiconductor device.

15. The semiconductor device according to any one of claims 11 to 14, characterized in that, The semiconductor device also includes a host computer, which is communicatively connected to the processor; the monitoring device is integrated into the host computer software of the semiconductor device, and the host computer software configures monitoring parameters through an interactive interface.

Images