A method for controlling backside deposition of AuGe alloy coatings

Abstract

This invention relates to the field of semiconductor thin film deposition and vacuum coating process control, specifically to a method for controlling the backside deposition of AuGe alloy. The method is applied to a deposition chamber equipped with an edge airflow adjustment device and an electromagnetic confinement coil. The method includes: acquiring real-time gas pressure data and real-time plasma impedance data of the edge region; extracting gas pressure fluctuation signals and plasma impedance signals to generate an initial state dataset; extracting high-frequency noise features based on the initial state dataset to generate an edge flow field instability index characterizing the degree of AuGe atom free cluster accumulation; calculating the critical time to reach a preset failure threshold based on this index and determining a target control strategy; and dynamically adjusting the edge protection airflow parameters and confinement magnetic field parameters according to the target control strategy to reduce the accumulation degree and prevent abnormal deposition of AuGe alloy on the backside.

Description

Technical Field

[0001] This invention relates to the field of semiconductor thin film deposition and vacuum coating process control, specifically to a method for controlling the backside deposition of AuGe alloy films. Background Technology

[0002] In existing AuGe alloy coating processes, the vacuum deposition chamber used for wafer deposition typically includes a main deposition area and an edge area near the outer periphery of the wafer. When AuGe alloy is deposited in the main deposition area, it is usually protected by a pressure ring, edge protection gas, or confinement structure to reduce the diffraction of deposited particles to the back side of the wafer. During the deposition process, as the wafer warps due to heat, the target material undergoes asymmetric degradation after use, and the local flow field and plasma distribution at the edge of the chamber fluctuate, AuGe atom free clusters are prone to accumulate in the edge area and migrate along the outer periphery of the wafer to the back side, thereby causing abnormal deposition, arcing, or contamination risks on the back side. However, most existing methods use fixed gas flow parameters, fixed magnetic field parameters, or rely on passive handling after back side anomalies occur, making it difficult to identify risks in advance and make graded adjustments based on the real-time status of the edge area, resulting in an inability to effectively control the problem of abnormal AuGe alloy deposition on the back side of the wafer. Summary of the Invention

[0003] To solve the above-mentioned technical problems, the present invention provides a method for controlling the back-side deposition of AuGe alloy coatings. Specifically, the technical solution of the present invention includes:

[0004] S1. The edge state sensing device acquires real-time air pressure data and real-time plasma impedance data of the edge region, and extracts air pressure fluctuation signals and plasma impedance signals to generate an initial state dataset.

[0005] S2. Based on the initial state dataset, high-frequency noise features are extracted to generate an edge flow field instability index, which is used to characterize the degree of accumulation of AuGe atomic free clusters in the edge region;

[0006] S3. Based on the edge flow field instability index, calculate the critical time when the edge flow field instability index of the current deposition state reaches the preset failure threshold, and determine the corresponding target control strategy;

[0007] S4. Based on the target control strategy, the edge protection airflow parameters are dynamically adjusted by the edge airflow adjustment device, and the constraint magnetic field parameters are dynamically adjusted by the electromagnetic constraint coil to reduce the accumulation degree and prevent abnormal deposition of AuGe alloy on the back side.

[0008] Optionally, step S1 includes the following sub-steps:

[0009] S11. The real-time air pressure data and real-time plasma impedance data are collected through the edge state sensing device;

[0010] S12. The real-time air pressure data and real-time plasma impedance data are filtered to separate the high-frequency disturbance component as the air pressure fluctuation signal and plasma impedance signal.

[0011] S13. The pressure fluctuation signal and the plasma impedance signal are combined to generate the initial state dataset.

[0012] Optionally, step S2 includes the following sub-steps:

[0013] S21. Analyze the initial state dataset to extract the amplitude variance of the air pressure fluctuation signal and the phase shift of the plasma impedance signal as the high-frequency noise features;

[0014] S22. Input the amplitude variance and phase offset into a preset state observer model;

[0015] S23. Calculate using the state observer model to output the edge flow field instability index.

[0016] Optionally, the preset state observer model is constructed in the following way:

[0017] Acquire the sample pressure fluctuation signal, sample plasma impedance signal and corresponding back-side arcing record during the historical deposition process;

[0018] Based on the sample air pressure fluctuation signal, sample plasma impedance signal, and corresponding sample back arcing record, an initial neural network is trained to generate the state observer model.

[0019] Optionally, step S3 includes the following sub-steps:

[0020] S31. Calculate the rate of change of the edge flow field instability index within a preset time window containing a predetermined number of continuous control cycles;

[0021] S32. Based on the edge flow field instability index and rate of change, and in conjunction with the preset failure threshold, estimate the remaining time to reach the preset failure threshold;

[0022] S33. The remaining time is determined as the critical time.

[0023] Optionally, step S3 further includes the step of determining the target control strategy based on the critical time, specifically including:

[0024] A preset first time threshold and a second time threshold are obtained, wherein the first time threshold is greater than the second time threshold; the first time threshold and the second time threshold are obtained by joint calibration based on the historical average of the free cluster accumulation rate in the edge flow field region and the mechanical response delay time of the exhaust valve execution command in the edge airflow regulating device.

[0025] If the critical time is greater than the first time threshold, a first control strategy to maintain the current deposition rate is generated as the target control strategy; if the critical time is less than or equal to the first time threshold and greater than the second time threshold, a local adjustment strategy is generated as the target control strategy; if the critical time is less than or equal to the second time threshold, an active degradation reset strategy is generated as the target control strategy.

[0026] Optionally, when the target control strategy is the proactive degradation reset strategy, step S4 includes the following sub-steps:

[0027] S41. In response to the active degradation reset strategy, generate an airflow surge command and a magnetic field trajectory change command;

[0028] S42. Based on the airflow surge command, adjust the protective gas intake flow rate in the edge protection airflow parameters to a preset maximum intake flow rate value to disperse the AuGe atom free clusters in the accumulation state;

[0029] S43. Based on the magnetic field trajectory change command, adjust the parameters of the confinement magnetic field to change the plasma distribution pattern and simultaneously reduce the deposition rate of the main deposition region.

[0030] Optionally, when the target control strategy is the local adjustment strategy, step S4 includes the following sub-steps:

[0031] The local adjustment strategy is analyzed, and the corresponding flow fine-tuning step size is calculated based on the difference between the critical time and the second time threshold.

[0032] Based on the flow rate fine-tuning step size, the edge protection airflow parameter is increased while the constraint magnetic field parameter remains unchanged.

[0033] Optionally, the process may further include the following before step S1:

[0034] Obtain target usage cycle data and target discharge voltage data;

[0035] Calculate the offset of the target discharge voltage data relative to the preset initial discharge voltage;

[0036] When the offset is greater than zero, an asymmetric degradation compensation factor for the target material is generated based on the ratio of the offset to the target material usage cycle data; when the offset is less than or equal to zero, the asymmetric degradation compensation factor for the target material is set to zero or a preset lower limit safety value.

[0037] S2 specifically includes: inputting the target material asymmetric degradation compensation factor as a gain correction term into a preset state observer model to correct the output edge flow field instability index.

[0038] Optionally, the edge protection airflow parameters include the intake flow rate of the protective gas and the opening degree of the exhaust valve, and the constraint magnetic field parameters include the magnitude of the excitation current of the electromagnetic constraint coil.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] By collecting real-time gas pressure and plasma impedance data from the edge region of the deposition chamber, filtering and separating high-frequency disturbance components, and packaging them into an initial state dataset by time slices, the risk of abnormal deposition on the back side can be indirectly characterized by the real-time state of the edge. During the index generation stage, the variance of gas pressure fluctuation amplitude and impedance phase shift are extracted and input into the state observer model. Combined with the corresponding relationships obtained from training historical back-side arcing records, the accuracy of identifying the degree of free cluster accumulation in gold-germanium alloys is improved. In the risk extrapolation stage, the critical time to reach the failure threshold is calculated based on the edge flow field instability index and its rate of change, shifting risk assessment from post-event handling to pre-event intervention. Prediction: During the control decision-making stage, strategies for maintaining the current deposition rate, local adjustment, or active degradation reset are generated based on the critical time level to avoid response lag or excessive disturbance caused by fixed airflow parameters and fixed magnetic field parameters. At the same time, an asymmetric degradation compensation factor is generated by combining the target material usage cycle and discharge voltage offset to conservatively correct the instability index. The edge protection airflow is implemented as the protective gas inlet flow rate and exhaust valve opening, and the constraint magnetic field is implemented as the coil excitation current magnitude. This achieves closed-loop coordinated control of airflow scouring, magnetic field constraint, and deposition rate reduction, thereby reducing the risk of abnormal deposition, discharge arcing, and contamination of gold-germanium alloys on the back side of the wafer. Attached Figure Description

[0041] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0042] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0044] like Figure 1 As shown, a method for controlling the backside deposition of AuGe alloy coatings is applied to a deposition chamber comprising a main deposition region and an edge region, and equipped with an edge airflow adjustment device, an electromagnetic confinement coil, and an edge state sensing device. The method includes the following steps:

[0045] S1. Real-time air pressure data and real-time plasma impedance data of the edge region are obtained through the edge state sensing device, and the air pressure fluctuation signal and plasma impedance signal are extracted to generate the initial state dataset.

[0046] S2. Based on the initial state dataset, high-frequency noise features are extracted to generate the edge flow field instability index, which is used to characterize the degree of accumulation of AuGe atomic free clusters in the edge region;

[0047] S3. Based on the edge flow field instability index, calculate the critical time when the edge flow field instability index of the current deposition state reaches the preset failure threshold, and determine the corresponding target control strategy;

[0048] S4. Based on the target control strategy, the edge protection airflow parameters are dynamically adjusted by the edge airflow adjustment device, and the constraint magnetic field parameters are dynamically adjusted by the electromagnetic constraint coil to reduce the accumulation degree and prevent abnormal deposition of AuGe alloy on the back side.

[0049] This embodiment provides an overall execution mechanism for a method to control the backside deposition of AuGe alloy coating. Specifically, the method is applied to a vacuum deposition chamber for co-sputtering deposition of AuGe alloy. The chamber is equipped with a main deposition area for supporting the front side of the wafer and an edge area near the outer periphery of the wafer. The edge area is equipped with a protective gas nozzle, an exhaust channel, a local pressure sensor, a plasma impedance detection terminal, and an electromagnetic confinement coil arranged around the outer periphery of the wafer. In this implementation environment, the wafer deforms due to heat and clamping. Traditional fixed pressure rings cannot completely prevent AuGe atoms from diffracting to the backside. Therefore, an adjustable edge barrier needs to be formed through the dynamic coordination of airflow and magnetic field. During execution, the process controller acquires real-time pressure data and real-time plasma impedance data of the edge area, and extracts the pressure fluctuation signal and plasma impedance signal related to local disturbances to form an initial state dataset.

[0050] The process controller extracts high-frequency noise features from the initial state dataset, such as the amplitude dispersion of the air pressure signal in a short time and the phase shift trend of the impedance signal, and maps these features to the edge flow field instability index. This index does not directly represent the back deposition thickness, but is used to characterize the degree of accumulation of AuGe atomic free clusters in the eddy stagnation zone within the edge region.

[0051] For ease of understanding, a basic example data segment can be set: at three consecutive sampling times, the normalized perturbation values ​​corresponding to the pressure fluctuation signal are 0.10, 0.18, and 0.26, and the plasma impedance phase shift values ​​are 0.03, 0.08, and 0.15. If the preset model converts the pressure perturbation dispersion and impedance phase shift together into an edge flow field instability index between 0 and 1, then this segment can yield a current index of 0.62. If the preset failure threshold is 0.85, and the index increases by an average of 0.04 per control cycle in the most recent time window, then the process controller can estimate that it has approximately 6 control cycles remaining before the failure threshold. Based on this critical time, the process controller further determines whether to maintain the current deposition state, perform local airflow adjustment, or execute an active degradation reset.

[0052] In boundary cases, if pressure data is missing but impedance data is still valid within a certain sampling period, the process controller can use the pressure fluctuation trend of the previous valid period as a short-term substitute input and reduce the confidence weight of the output index for that period. If both pressure and impedance data are abnormal, such as a sensor disconnection or the detection value exceeding the hardware range, the controller will not directly output a strategy to maintain the deposition rate, but will instead enter a safety protection branch, limiting the deposition power in the main deposition area and adjusting the edge protection airflow to a preset safe flow rate until valid data is continuously obtained.

[0053] In a series of high-throughput AuGe alloy deposition processes, the edge protection gas flow was stable in the early stages of the process, and the edge flow field instability index remained between 0.30 and 0.45. As the local consumption of the target surface intensified, the atomic plume gradually deflected to the outer periphery of the wafer, and the edge gas pressure signal began to show dense high-frequency fluctuations, and the impedance phase also showed continuous shifts. Without directly measuring the back side of the wafer, the process controller inferred the risk of abnormal deposition on the back side based on the edge sensing signal, and changed the edge gas flow and confinement magnetic field in advance before the index approached the threshold, thereby avoiding the abnormal deposition of AuGe free clusters on the back side of the wafer.

[0054] The purpose of this step is to transform the risk of back-side contamination, which is difficult to measure directly in real time, into a process state that can be indirectly characterized by the chamber edge pressure and plasma impedance, and to keep the edge flow field under control under high deposition rate conditions through dynamic control driven by critical time.

[0055] In this embodiment, S1 includes the following sub-steps:

[0056] S11. Real-time air pressure data and real-time plasma impedance data are collected through the edge state sensing device;

[0057] S12. Filter the real-time air pressure data and real-time plasma impedance data to separate the high-frequency disturbance component as the air pressure fluctuation signal and the plasma impedance signal.

[0058] S13. Combine the air pressure fluctuation signal with the plasma impedance signal to generate an initial state dataset.

[0059] This embodiment provides an edge state data acquisition and packaging mechanism. Specifically, in the aforementioned overall scheme, if only the raw readings output by the barometer and impedance detection terminal are used, the raw readings will simultaneously include low-frequency adjustment changes of the chamber pumping system, trend terms caused by the target power increasing at a rate lower than the preset rate of change, and high-frequency disturbance terms that truly reflect local edge disturbances. If these are not separated, the process controller may easily misjudge normal pumping valve adjustments as AuGe cluster accumulation, thereby causing unnecessary airflow surges or magnetic field adjustments.

[0060] Therefore, in this embodiment, the real-time sampling signal is filtered and the separated high-frequency disturbance components are constructed as the initial state dataset. The filtering process specifically includes: using a moving average detrending algorithm to obtain the low-frequency mean of the real-time air pressure data and the real-time plasma impedance data as a baseline, and calculating the deviation value by subtracting the data of each sampling point from the low-frequency mean value, and defining the deviation value as the high-frequency disturbance component.

[0061] In practice, the sensors at the edge of the deposition chamber output two types of data with the same or time-synchronized sampling period: one is local gas pressure data at the edge, and the other is plasma impedance data; the filtering process can be carried out by combining moving average detrending and high-frequency component extraction.

[0062] For example, the air pressure readings at five sampling points before and after a certain moment are 10.00, 10.02, 10.06, 10.12, and 10.20, with the units being the internal calibration units of the process system. If the low-frequency mean of this set of data is 10.08, then the deviation of 0.12 from the low-frequency mean at the current sampling point of 10.20 can be taken as a component of the air pressure fluctuation signal. The impedance signal can also be processed in a similar way, separating the original impedance phase from the low-frequency drift trend and retaining only the component that shifts rapidly in a short period of time. When packaging the data, the air pressure fluctuation signal and the plasma impedance signal within the same time window can be aligned according to the timestamp.

[0063] For example, the initial state dataset can be represented as a binary sequence of three time slices: the first time slice contains a pressure perturbation of 0.02 and an impedance perturbation of 0.01, the second time slice contains a pressure perturbation of 0.07 and an impedance perturbation of 0.04, and the third time slice contains a pressure perturbation of 0.12 and an impedance perturbation of 0.09; this binary sequence can provide input for subsequent instability index calculations without recording complete chamber images or back-side deposition data.

[0064] The specific steps for packaging and generating the initial state dataset include: strictly aligning the air pressure fluctuation signal and plasma impedance signal within the same preset time window according to the sampling timestamp, constructing a multivariate time series matrix with time slice as the dimension, and using it as the initial state dataset;

[0065] In boundary cases, if there is a discrepancy between the sampling timestamps of the barometric pressure sensor and the impedance detection end, adjacent data with a time difference of less than one sampling period can be paired; if the time difference exceeds the preset synchronization range, the time slice is marked as asynchronous and does not participate in the current exponent calculation; if the high-frequency disturbance component obtained after filtering is close to zero for a long time, but the original signal shows obvious drift, it is determined that the current change is more likely to come from slow variables of the equipment, such as pump speed changes or temperature drift. In this case, the equipment status verification can be triggered instead of directly performing edge airflow compensation.

[0066] During the deposition of the same batch of AuGe alloy wafers, the chamber exhaust valves undergo slow opening changes to maintain total pressure, causing the initial gas pressure reading to gradually increase. The process controller first removes this low-frequency trend, then retains the short-term fluctuations near the wafer edge caused by atomic plumes impacting the gas flow barrier, and packages these fluctuations and impedance phase perturbations into time slices. The inputs used for subsequent calculations thus focus on reflecting local anomalies at the edge, rather than being masked by global chamber pressure regulation.

[0067] The purpose of this step is to improve the targeting of input data to local flow field anomalies at the edge, reduce the probability of false triggering caused by slow variables of the device, and provide structurally consistent and time-aligned state data for subsequent exponential generation.

[0068] In this embodiment, S2 includes the following sub-steps:

[0069] S21. Analyze the initial state dataset to extract the amplitude variance of the air pressure fluctuation signal and the phase shift of the plasma impedance signal as high-frequency noise features;

[0070] S22. Input the amplitude variance and phase offset into the preset state observer model;

[0071] S23. Calculate using the state observer model to output the edge flow field instability index.

[0072] This embodiment provides a mechanism for generating an edge flow field instability index from high-frequency noise characteristics. Specifically, obtaining only high-frequency disturbance signals is insufficient to directly determine whether AuGe free clusters accumulate in the edge region, because the instantaneous peak of local disturbances may come from brief pumping pulsations, while the continuously increasing amplitude dispersion and impedance phase shift better reflect the stagnation of the edge flow field and changes in plasma distribution. Therefore, this embodiment extracts the amplitude variance of the pressure fluctuation signal and the phase shift of the plasma impedance signal from the initial state dataset, and then inputs them into a preset state observer model to output the edge flow field instability index.

[0073] In the microscopic simulation, assuming there are four pressure fluctuation values ​​within a time window: 0.02, 0.04, 0.10, and 0.16, with a mean of 0.08, the deviation from the mean is greater than a preset deviation threshold, resulting in an amplitude variance greater than a preset variance threshold. Further assuming the plasma impedance phase offset relative to the reference phase within the same time window is 0.01, 0.03, 0.07, and 0.11, respectively, indicating a continuous amplification trend in phase offset, the state observer model can normalize these two features to form a set of inputs. For example, with an amplitude variance feature of 0.55 and a phase offset feature of 0.60, the model output edge flow field instability index is 0.58. If the amplitude variance increases to 0.75 and the phase offset increases to 0.80 in the next time window, the output index can increase to 0.78, indicating an increase in the accumulation of AuGe free clusters in the edge stagnation zone.

[0074] In practical deployments, the state observer model can be a lookup table model, a regression model, or a neural network model trained with historical sedimentation data, as long as its input is amplitude variance and phase offset, and its output is a value used to characterize the degree of anomaly in the edge flow field. To prevent misjudgment caused by a single feature, the model can be set with feature balance constraints: when the amplitude variance increases but the phase offset is still low, the increase in the output exponent is limited; when both increase simultaneously and change in the same direction for multiple consecutive time windows, the output exponent increases rapidly; in boundary cases, if the amplitude variance is zero or close to zero but the phase offset suddenly increases significantly, the process controller can first determine whether there is an impedance sensing terminal contact abnormality or an RF matching device switching event, and temporarily not directly interpret this single-point anomaly as cluster accumulation.

[0075] If the phase shift is zero but the gas pressure amplitude variance continues to increase, the risk level can be set to an intermediate state and wait for the next window verification. If the input features exceed the model training or calibration range, the input is truncated at the upper limit and enters a conservative control branch to avoid the model extrapolation leading to an excessively low output. In this AuGe alloy deposition mainline scenario, the wafer edge forms local gaps due to warping, and the AuGe atomic plume is deflected outward due to the local degradation of the target material. At this time, the amplitude variance of the edge gas pressure fluctuation first increases, and the plasma impedance phase shifts synchronously. The state observer model converts this dual-feature unidirectional deterioration into an increased edge flow field instability index, enabling the controller to identify risks before obvious anomalies occur on the back side.

[0076] The purpose of this step is to further compress the original high-frequency disturbances into process state quantities that can be used for control decisions, reduce reliance on human experience judgment, and enable data from different batches and different time windows to be compared on a uniform scale.

[0077] In this embodiment, the preset state observer model is constructed in the following way:

[0078] Acquire the sample pressure fluctuation signal, sample plasma impedance signal and corresponding back-side arcing record during the historical deposition process;

[0079] Based on the sample air pressure fluctuation signal, sample plasma impedance signal and corresponding sample back arcing record, an initial neural network is trained to generate a state observer model.

[0080] This embodiment provides a mechanism for constructing a state observer model. Specifically, in the aforementioned scheme, if the state observer model relies solely on manually set weights, such as simply assigning equal weights to the gas pressure amplitude variance and impedance phase shift, it may not accurately reflect the real risks at different stages of target material use or under different chamber evacuation conditions. In particular, after the AuGe alloy target material experiences asymmetric consumption, the atomic plume deflection will cause the impedance phase to change beyond the preset fluctuation threshold before the gas pressure, and the fixed-weight model may respond with lag.

[0081] Therefore, this embodiment uses sample signals and back-side arcing records from the historical deposition process to train the initial neural network to obtain a state observer model that can adapt to the characteristics of the process history. When building the model, the sample pressure fluctuation signal, sample plasma impedance signal and corresponding sample back-side arcing records are first extracted from the historical batches. The sample back-side arcing records can come from the equipment event log, the re-inspection record after wafer back-side contamination, the chamber abnormal discharge record or the process interruption record.

[0082] To facilitate training, sample back-side arcing records can be converted into risk labels. For example, samples that experience back-side arcing within a preset first time period after a certain time window are labeled as 1, and samples that do not experience back-side arcing and whose subsequent processes are stable are labeled as 0. They can also be labeled with risk levels such as 0, 0.5, and 1. To avoid confusion between the training label source and the concept of sample back-side arcing records in the embodiments, the aforementioned re-inspection records, abnormal discharge records in the chamber, or process interruption records after wafer back-side contamination are only used to restore or supplement the sample back-side arcing records when it is possible to trace and confirm a corresponding relationship with the back-side arcing event. If an abnormal record only reflects an equipment maintenance event, a general shutdown, or a discharge disturbance that has no causal relationship with back-side arcing, then the abnormal record is not used as a separate training label input.

[0083] In other words, the focus of training labels remains on the back-end arcing itself. Other historical information is only used as a basis for time positioning, level verification, or missing information completion of back-end arcing records, thereby ensuring that the instability index output by the model is consistent with the actual back-end failure boundary. In the simplified extrapolation example, assume there are three sets of historical samples: the first set has a pressure amplitude variance of 0.20 and an impedance phase shift of 0.18, and no back-end arcing occurred afterward, so the label is 0; the second set has a pressure amplitude variance of 0.50 and an impedance phase shift of 0.62, and a slight abnormal discharge occurred afterward, so the label is 0.5; the third set has a pressure amplitude variance of 0.82 and an impedance phase shift of 0.90, and a back-end arcing interruption occurred afterward, so the label is 1. The initial neural network adjusts its internal connection parameters by repeatedly comparing the difference between the output value and the label, so that it outputs a high-risk index close to 1 when inputting similar features from the third set, and a low-risk index close to 0 when inputting similar features from the first set.

[0084] In boundary cases, if the number of back-end arcing events in the historical samples is small, the model training can use oversampling or risk sample weighting to ensure that a small number of high-risk samples are not overwhelmed by a large number of normal samples, causing the feature weight ratio to fall below the preset recognition lower limit. If some historical records lack complete air pressure or impedance signals, the entire batch of data is not directly deleted. Instead, complete segments are extracted as training samples, and missing segments are marked as invalid. If the new chamber does not yet have enough historical records, a general model of the same type of chamber can be used first, and local samples are gradually added for recalibration during the operation of this chamber.

[0085] In the high-throughput AuGe deposition line, no abnormalities were found in the first ten batches of wafers, and the model obtained a large number of normal samples. When a certain target material entered the later stage of use, the edge impedance phase continued to shift and triggered back-side arcing logs several minutes later. This event and the signal segments before it were added to the training set, enabling the model to learn that a rapid shift in impedance phase accompanied by an increase in gas pressure amplitude variance corresponds to a higher failure risk. In subsequent batches, even if back-side arcing has not yet occurred, the model can output an increased instability index in advance.

[0086] The purpose of this mechanism is to establish a correspondence between edge sensing signals and back-side abnormal events through historical process data, improve the fit of the instability index to the actual failure boundary, and enable the controller to adapt to signal differences caused by target degradation and changes in chamber state.

[0087] In this embodiment, S3 includes the following sub-steps:

[0088] S31. Calculate the rate of change of the edge flow field instability index within a preset time window containing a predetermined number of continuous control cycles;

[0089] S32. Based on the edge flow field instability index and rate of change, combined with the preset failure threshold, the remaining time to reach the preset failure threshold is estimated;

[0090] S33. Determine the remaining time as the critical time.

[0091] This embodiment provides a critical time calculation mechanism. Specifically, in the aforementioned scheme, if only the current edge flow field instability index is judged to exceed the preset failure threshold, the control action often has a response lag. Since AuGe free clusters may cause back-side arcing to occur in a short time once they enter the rapid accumulation stage, this embodiment not only focuses on the absolute value of the current index, but also calculates its rate of change within a preset time window, and uses this to extrapolate the remaining time to the preset failure threshold. The preset time window contains a predetermined number of continuous control cycles. The rate of change is obtained by calculating the difference between the instability index of the latest control cycle and the oldest control cycle within the preset time window, and dividing it by the number of continuous control cycles. During execution, the process controller saves the edge flow field instability index of the most recent control cycles.

[0092] For example, in four consecutive control cycles, the indices are 0.54, 0.60, 0.67, and 0.75, respectively, and the preset failure threshold is 0.90. If the average rate of change of the most recent three intervals is used, then each control cycle increases by approximately 0.07. The difference between the current index and the preset failure threshold is 0.15. Based on this, the remaining time can be estimated to be approximately 0.15 divided by 0.07, which is about 2 to 3 control cycles. This remaining time is used as the critical time for subsequent selection of control strategies. To avoid distortion of the rate of change caused by single-point fluctuations, the index sequence within the preset time window can be smoothed.

[0093] For example, if the exponential sequence is 0.54, 0.60, 0.88, 0.62, and 0.88 appears only once and corresponds to a sensor synchronization anomaly, then this point can be marked as an anomaly, and the rate of change can be estimated using adjacent effective values; if the exponential sequence shows a continuous increase, then the critical time can be deduced based on the positive rate of change; if the exponential decreases or remains stable, then the critical time can be set to a value greater than the preset safety time, indicating that it is not necessary to enter the strong intervention branch for the time being.

[0094] In boundary cases, if the current index has reached or exceeded the preset failure threshold, the critical time is set to zero and a safety control strategy is triggered. If the rate of change is zero but the current index is close to the threshold (e.g., the current index is 0.86 and the threshold is 0.90), safety cannot be determined simply because the rate of change is zero. Instead, the critical time can be set to a preset lower limit and the risk can be eliminated only after several consecutive windows of decline. If the rate of change is negative and the difference between the current index and the threshold is greater than the preset safety margin, the risk can be determined to be mitigated, but the minimum monitoring frequency is still maintained.

[0095] In the AuGe deposition process, the asymmetric degradation of the target material caused the edge flow field instability index to rise slowly from 0.52 to 0.60, and then rapidly to 0.76 due to the intensified deflection of the atomic plume. Although 0.76 has not yet reached the preset failure threshold of 0.90, the rate of change has increased significantly. Based on this, the process controller can estimate the remaining time to be shortened and prepare for airflow and magnetic field adjustments in advance, rather than waiting for the actual arcing to occur.

[0096] The purpose of this step is to shift the risk assessment from whether a dangerous state has been reached to how much time is left before a dangerous state, so that the controller can select intervention actions of different intensities based on the speed at which the risk approaches.

[0097] In this embodiment, S3 further includes the step of determining the target control strategy based on the critical time, specifically including:

[0098] A preset first time threshold and a second time threshold are obtained, wherein the first time threshold is greater than the second time threshold; the first time threshold and the second time threshold are obtained by joint calibration based on the historical average of the accumulation rate of free clusters in the edge flow field region and the mechanical response delay time of the exhaust valve execution command in the edge airflow regulating device.

[0099] If the critical time is greater than the first time threshold, a first control strategy to maintain the current deposition rate is generated as the target control strategy; if the critical time is less than or equal to the first time threshold and greater than the second time threshold, a local adjustment strategy is generated as the target control strategy; if the critical time is less than or equal to the second time threshold, an active degradation reset strategy is generated as the target control strategy.

[0100] This embodiment provides a mechanism for determining the target control strategy based on the critical time classification. Specifically, based on the aforementioned critical time calculation, if all risk states adopt the same control action, such as increasing the protective airflow and changing the magnetic field according to a preset surge ratio, it is easy to cause unnecessary gas path load increase, increase in chamber pressure regulation amplitude, and disturbance of film formation state in the main deposition area. Conversely, if a high deposition rate is always maintained, abnormal deposition on the back side cannot be effectively prevented when the risk approaches rapidly.

[0101] Therefore, this embodiment sets a first time threshold and a second time threshold, and generates different control strategies based on the interval into which the critical time falls; in specific execution, the first time threshold is greater than the second time threshold; the first time threshold is used to distinguish between a safe observation state and a state requiring local intervention, and the second time threshold is used to distinguish between a state that can be slowly adjusted and an emergency state that requires active degradation and reset;

[0102] For example, the first time threshold can be set to 10 control cycles, and the second time threshold can be set to 3 control cycles. If the critical time is 15 control cycles, it means that although the edge flow field fluctuates, the distance from the failure threshold is greater than the preset safe time difference, and the process controller generates a first control strategy to maintain the current deposition rate. If the critical time is 6 control cycles, a local adjustment strategy is generated, such as increasing the edge protection airflow by a preset flow step. If the critical time is 2 control cycles, an active degradation reset strategy is generated to rebuild the edge flow field order in a preset high control gain mode.

[0103] In boundary cases, when the critical time is exactly equal to the first time threshold, a local adjustment strategy is entered instead of maintaining the current deposition rate to avoid risk exceeding the limit due to control delay; when the critical time is exactly equal to the second time threshold, an active degradation reset strategy is entered instead of the local adjustment strategy to ensure the priority of backside contamination protection; if the critical time calculation result is negative, it means that the current index has exceeded the preset failure threshold or the model extrapolation has exceeded the limit, so it is directly treated as zero and active degradation reset is triggered; if the critical time frequently crosses the threshold in adjacent cycles, a hysteresis interval can be set, for example, requiring strong degradation to be performed after two consecutive cycles meet the degradation conditions, or requiring multiple consecutive cycles to restore safety before exiting after entering strong degradation;

[0104] In the same batch of AuGe wafer deposition, the critical time in the front-end process is longer than the first time threshold for a long time, and the system maintains a high deposition rate. When uneven consumption occurs at the edge of the target material, the critical time drops below the first time threshold. The controller first adopts a local adjustment strategy to increase the edge protection airflow. If the impedance phase continues to deteriorate, the critical time is further shortened to within the second time threshold. The controller no longer insists on the highest deposition rate, but switches to an active degradation reset strategy.

[0105] Furthermore, in order to keep the execution actions in this embodiment consistent with those in S4, when the target control strategy is the first control strategy, the process controller does not disconnect from the edge airflow adjustment device and the electromagnetic constraint coil, but outputs a holding command to the corresponding execution component to maintain the set values ​​of the edge protection airflow parameters and constraint magnetic field parameters in the previous stable cycle, and continues to update the critical time according to the current monitoring frequency.

[0106] In other words, the target control strategy corresponds to executable control outputs at all three levels: the first control strategy corresponds to controlled maintenance, the local adjustment strategy corresponds to limited amplitude adjustment, and the active degradation reset strategy corresponds to strong intervention adjustment. Through this hierarchical definition, the misunderstanding of maintaining the current deposition rate as not performing edge control at all can be avoided, thus ensuring that the strategy determination of S3 and the execution object of S4 are consistent with the meaning of the entire text.

[0107] The purpose of this mechanism is to convert critical time into executable graded control actions, to achieve a dynamic balance between deposition rate stability, gas path operating load, cavity pressure regulation amplitude and back-side contamination protection, and to ensure that strong intervention is triggered only when necessary.

[0108] In this embodiment, when the target control policy is an active degradation reset policy, S4 includes the following sub-steps:

[0109] S41. In response to the active degradation reset strategy, generate airflow surge command and magnetic field trajectory change command;

[0110] S42. Based on the airflow surge command, adjust the protective gas intake flow rate in the edge protection airflow parameters to the preset maximum intake flow rate value to disperse the AuGe atom free clusters in the accumulation state;

[0111] S43. Based on the magnetic field orbit change command, adjust the constraint magnetic field parameters to change the plasma distribution pattern and simultaneously reduce the deposition rate of the main deposition region.

[0112] This embodiment provides an active degradation reset control mechanism; specifically, in the aforementioned graded strategy, when the critical time is less than or equal to the second time threshold, the edge region may have formed a vortex stagnation zone of AuGe free clusters; if a small airflow fine adjustment is still used at this time, it may not be able to disperse the clusters in time, and simply increasing the airflow may also change the plasma edge morphology, causing a new local stagnation.

[0113] Therefore, in this embodiment, in an emergency, a gas flow surge command and a magnetic field trajectory change command are generated simultaneously, and the deposition rate of the main deposition region is reduced in sync to quickly destroy the accumulation state and reduce the intensity of new atomic supply. In specific execution, the process controller responds to the active degradation reset strategy and sends a gas flow surge command to the edge gas flow adjustment device to adjust the inlet flow rate of the edge protection gas to the preset maximum flow rate value. If necessary, the opening of the exhaust valve is adjusted to enable the fluid channel to have the flushing capability to meet the preset cleaning standard. At the same time, the process controller sends a magnetic field trajectory change command to the electromagnetic confinement coil, and changes the edge plasma distribution pattern by adjusting the excitation current, so that the high-density plasma region that was originally close to the wafer back entrance is shifted to the controllable region of the chamber.

[0114] At the same time, the deposition rate of the main deposition region is reduced synchronously, for example by reducing the sputtering power or reducing the power of at least one of the Au and Ge targets, thereby reducing the flux of new AuGe atoms. Specifically, the synchronous reduction of the deposition rate of the main deposition region is manifested by controlling the sputtering power supply of the process chamber and reducing the sputtering power applied to the AuGe target to physically reduce the atomic flux to the front side of the wafer, thereby limiting the source of new free clusters.

[0115] In the simplified simulation example, it is assumed that the normal flow rate of the edge protection gas is 40, the preset maximum flow rate is 80, the normal excitation current of the electromagnetic confinement coil is 5, the excitation current in the orbital change state is 7, and the deposition rate level corresponding to the main deposition power is 100%. When the critical time is 2 control cycles and the second time threshold is 3 control cycles, the controller increases the airflow from 40 to 80, adjusts the excitation current from 5 to 7, and reduces the main deposition rate level to 70%. This combined action is not to increase the front deposition rate, but to prioritize reducing the degree of edge cluster accumulation.

[0116] In boundary conditions, if the edge airflow regulating device cannot reach the preset maximum flow rate, for example, if the valve feedback only reaches 80% of the maximum value, the process controller can further reduce the deposition rate in the main deposition area and extend the reset duration. If the temperature or current of the electromagnetic confinement coil reaches the hardware protection limit, the excitation current will not be increased. Instead, the maximum safe current within the allowable range will be used, and the insufficient magnetic field reset will be compensated by reducing the deposition power. If the edge flow field instability index continues to rise after the active degradation reset is performed, the process suspension or chamber safety shutdown procedure will be triggered to prevent backside contamination from expanding into batch anomalies.

[0117] Furthermore, in conjunction with the parameter definitions in the embodiments, adjusting the edge protection airflow parameter to the preset maximum flow value means that the main control action is to make the protective gas intake flow reach the maximum setting under the corresponding strategy, and to make the exhaust valve opening enter the cooperative opening range that matches the maximum intake flow, so as to ensure that the total pressure of the chamber is still within the allowable range when the edge flushing capability is improved; accordingly, the airflow surge command in this embodiment preferably includes a main command for the intake flow and a matching command for the exhaust valve opening, so that the edge protection airflow parameter always corresponds to the same type of execution object throughout the text, avoiding its understanding as only a single intake flow parameter;

[0118] When the AuGe target material enters the later stage of use, after a batch of wafers has been deposited to the middle section, the edge flow field instability index rapidly approaches the failure threshold, and the critical time drops to within the second time threshold. The process controller immediately increases the edge protection airflow to the maximum flow rate and changes the distribution of the confining magnetic field, while reducing the deposition rate in the main deposition area. This action changes the front deposition rate and gas path operating parameters in a short time, but it can disperse the AuGe free clusters at the wafer edge and block their jetting path to the back of the wafer.

[0119] The purpose of this mechanism is to prioritize the protection of the system's safety boundary when the risk of anomalous deposition on the back side is rapidly approaching, and to reconstruct the flow field and plasma order in the edge region through the combined effects of airflow scouring, magnetic field reconstruction, and reduction of deposition flux.

[0120] In this embodiment, when the target control strategy is a local adjustment strategy, S4 includes the following sub-steps:

[0121] The local adjustment strategy is analyzed, and the corresponding flow fine-tuning step size is calculated based on the difference between the critical time and the second time threshold.

[0122] Based on the flow fine-tuning step size, the edge protection airflow parameters are increased while keeping the constraint magnetic field parameters unchanged.

[0123] This embodiment provides a local adjustment and control mechanism. Specifically, when the critical time is less than or equal to the first time threshold and greater than the second time threshold, the edge region has already shown an increasing risk trend, but has not yet reached the point where it is necessary to change the magnetic field and reduce the main deposition rate. If active degradation and reset are directly performed at this stage, it is easy to cause the process state of the main deposition region to fluctuate beyond the preset steady-state allowable range. If no intervention is performed at all, the risk may continue to approach the second time threshold.

[0124] Therefore, this embodiment slightly increases the edge protection airflow by calculating the flow rate fine-tuning step size while keeping the constraint magnetic field parameters unchanged, thereby alleviating edge cluster accumulation while reducing process disturbances. In specific execution, the process controller analyzes the local adjustment strategy and calculates the difference between the critical time and the second time threshold. The smaller the difference, the closer the risk is to the active degradation reset interval, so the flow rate fine-tuning step size is larger. The larger the difference, the more sufficient the adjustment time is, so the fine-tuning step size can be smaller.

[0125] For example, if the first time threshold is 10 control cycles, the second time threshold is 3 control cycles, and the current critical time is 8 control cycles, then the difference between the first and second time thresholds is 5, and the flow fine-tuning step size can be set to the first preset step size; if the current critical time is 4 control cycles, then the difference is 1, and the flow fine-tuning step size can be set to the second preset step size of 15; in this way, the airflow adjustment intensity increases progressively with the approach of risk.

[0126] Furthermore, keeping the constraint magnetic field parameters unchanged here does not mean that the electromagnetic constraint coil has left the control loop, but rather that the process controller maintains the excitation current output command or the set value of the previous cycle in the current control cycle, so that the magnetic field parameters enter a controlled holding state. In other words, the dynamic adjustment of the constraint magnetic field parameters in S4 includes both changing the excitation current and actively maintaining its current setting after strategy determination. The reason for adopting this controlled holding method is that the main cause of the medium-risk stage is the weakening of the edge gas flow barrier, rather than the instability of the plasma distribution. If the magnetic field is changed too early at this time, it may cause incidental disturbances to the uniformity of the main deposition area.

[0127] In boundary cases, if the calculated flow fine-tuning step size is less than the minimum adjustable resolvable step size that the device can execute, then the minimum executable step size is used; if the calculated step size causes the flow to exceed the allowable upper limit, then the flow is limited to the allowable upper limit, and the need to switch to active degradation reset is reassessed in the next cycle; if the instability index decreases after local adjustment, then the new airflow is maintained or gradually rolled back; if it continues to rise for several consecutive control cycles, then the airflow is not further increased to areas that may cause turbulence, but instead, a higher intensity control strategy is switched according to the critical time.

[0128] Furthermore, considering the limitations on the edge protection airflow parameters in the embodiments, the increase in edge protection airflow parameters in this embodiment is preferably achieved by gradually increasing the intake airflow of the protective gas as the main adjustment variable. The exhaust valve opening can be simultaneously fine-tuned based on changes in the total chamber pressure, the exhaust capacity of the edge region, and the current exhaust valve opening margin, so that the increased intake airflow matches the exhaust capacity. In other words, the flow rate fine-tuning step size directly corresponds to the increase in the intake airflow, while the exhaust valve opening participates as a supporting maintenance variable or a small compensation variable. During the local adjustment phase, the control focus remains on the gradual enhancement of the edge protection gas, rather than changing the distribution of the constraint magnetic field.

[0129] In AuGe alloy deposition, the edge sensing signal indicates that the gas pressure amplitude variance is starting to increase, but the impedance phase shift has not yet reached an emergency level, with a critical time of 6 control cycles. Based on this, the process controller determines that it is still in the local adjustment stage, calculates the flow rate fine-tuning step size, increases the edge protection gas inlet flow rate, and keeps the electromagnetic confinement coil excitation current unchanged. If the total chamber pressure is close to the upper limit at this time, the exhaust valve opening is slightly adjusted to release the pressure burden brought by the increased gas volume. If the index falls back in the next window, the system continues to maintain a high deposition rate. If the index continues to rise, further adjustments are made or active degradation reset is initiated.

[0130] The purpose of this mechanism is to prioritize low-disturbance adjustment measures during the medium-risk stage to avoid premature changes in the magnetic field distribution and main deposition rate, while preventing the edge flow field instability index from continuing to approach the failure threshold.

[0131] In this embodiment, the following is included before S1:

[0132] Obtain target usage cycle data and target discharge voltage data;

[0133] Calculate the offset of the target discharge voltage data relative to the preset initial discharge voltage;

[0134] When the offset is greater than zero, a target asymmetric degradation compensation factor is generated based on the ratio of the offset to the target usage cycle data; when the offset is less than or equal to zero, the target asymmetric degradation compensation factor is set to zero or a preset lower limit safety value.

[0135] S2 specifically includes: inputting the target asymmetric degradation compensation factor as a gain correction term into the preset state observer model to correct the output edge flow field instability index.

[0136] This embodiment provides a target asymmetric degradation compensation mechanism. Specifically, in the aforementioned scheme, the edge flow field instability index is mainly generated based on edge gas pressure fluctuations and plasma impedance changes. However, in long-term production, the AuGe alloy target surface may experience uneven consumption, local nodule formation, or changes in discharge state, causing the atomic plume angle to deflect. Such disturbances may not significantly affect the overall deposition rate of the main deposition area in the early stages, but they will have a continuous impact on the gas flow barrier at the wafer edge. If the target state is not included in the correction, the same edge gas pressure disturbance represents different risks in the new target and degraded target stages.

[0137] Therefore, in this embodiment, before acquiring edge state data, the asymmetric degradation compensation factor of the target material is calculated and used to correct the edge flow field instability index. In specific execution, the process controller acquires the target material usage cycle data and the target material discharge voltage data. The target material usage cycle data can represent the number of deposition batches the target material has undergone, the cumulative discharge time, or the cumulative process cycle. The target material discharge voltage data can come from the sputtering power supply or the discharge monitoring module.

[0138] The offset of the current target discharge voltage relative to the preset initial discharge voltage is calculated, and the ratio of this offset to the target usage cycle data is obtained to obtain the target asymmetric degradation compensation factor. In order to make the compensation process have a clear data flow relationship, the target asymmetric degradation compensation factor can be treated as a monotonic gain correction term for the S2 output result. That is, the original instability index is first obtained from the edge pressure fluctuation signal and the plasma impedance signal, and then the compensation factor is used to adjust and correct the original instability index upward. When the compensation factor is greater than the preset factor threshold, the upward adjustment magnitude increases accordingly; when the compensation factor is less than or equal to the preset factor threshold, the upward adjustment magnitude decreases accordingly.

[0139] Can be adjusted according to the formula The calculation is performed by using the original index obtained without incorporating target degradation information. Compensation factor for asymmetric degradation of target material and the compensation weights obtained based on chamber calibration The products are added together to obtain the corrected edge flow field instability index. ;

[0140] To avoid the correction results losing a unified comparative scale, the corrected index can be further limited within a preset risk dimension range, such as between 0 and 1. In the simplified extrapolation example, assuming the preset initial discharge voltage is 500, the current target discharge voltage is 530, and the offset is 30; the target's used cycle data is 10, then the compensation factor is 3; if another target's current voltage is also offset by 30, but its used cycle is 30, then the compensation factor is 1; the former indicates that a voltage offset greater than the preset offset threshold occurs within the first preset used cycle, and there is a greater possibility of degradation exceeding the preset wear standard. Therefore, a stronger correction can be applied when calculating the input edge flow field instability index.

[0141] For example, the original instability index is 0.60, which is corrected to 0.63 when the compensation factor is low and to 0.70 when the compensation factor is high. Furthermore, the correction preferably adopts a conservative strategy of only increasing or maintaining the original value. That is, when the discharge voltage deviation reflects the increased risk of asymmetric degradation of the target material, the instability index is increased; when the discharge voltage deviation does not reflect a clear degradation trend, the original risk result obtained directly from the edge sensing signal is not artificially suppressed by this compensation term. This ensures that the real-time edge state is always the main basis for judgment, while the target material compensation factor participates in the decision-making as a risk amplification term on the source side, avoiding the situation where the true edge anomaly is masked by the fluctuation of target material information.

[0142] In boundary cases, if the target usage cycle data is zero or close to zero, the denominator can be set to be no less than the preset minimum cycle value to avoid division by zero; if the discharge voltage data shows a momentary spike, a short-term smoothing can be performed first to prevent the power switching transient from being misjudged as target degradation; if the voltage offset is negative, it means that the current discharge voltage is lower than the initial reference, and the lower limit of the compensation factor can be truncated to zero based on process experience, or it can be recorded as a low-risk correction item without directly reducing the risk index generated by the edge signal; if the target usage data is missing, a default compensation value that is not conducive to safety should be adopted, and the equipment maintenance system should be prompted to verify the target history.

[0143] During the high-throughput deposition of AuGe alloy, after multiple batches of the same target were run consecutively, the discharge voltage showed a significant shift relative to the initial state, but the rate monitoring of the main deposition area still did not show any abnormalities. Based on this, the process controller generated a higher target asymmetric degradation compensation factor and adjusted the risk result upward when calculating the edge flow field instability index. In this way, when the edge gas pressure and impedance only showed moderate disturbances, the system could also identify the high-risk state that might be caused by the deflection of the target plume.

[0144] The purpose of this mechanism is to introduce the latent perturbation of long-term target degradation into edge risk assessment, so that the instability index can not only reflect the current edge sensing state, but also reflect the asymmetric changes on the AuGe atom source side, thereby improving the adaptability to long-term production scenarios.

[0145] In this embodiment, the edge protection airflow parameters include the intake flow rate of the protective gas and the opening degree of the exhaust valve, and the constraint magnetic field parameters include the magnitude of the excitation current of the electromagnetic constraint coil.

[0146] This embodiment provides a mechanism for limiting executable control parameters; specifically, in the aforementioned scheme, if only the adjustment of edge protection airflow and constraint magnetic field are described in general terms, it may be difficult to determine which execution components the controller actually outputs to during engineering implementation, and it is also difficult to perform closed-loop feedback.

[0147] Therefore, in this embodiment, the edge protection airflow parameters are specified as the inlet flow rate of the protective gas and the opening degree of the exhaust valve, and the confinement magnetic field parameters are specified as the magnitude of the excitation current of the electromagnetic confinement coil, so that the control strategy can be directly converted into control commands that can be executed by the chamber equipment. During execution, the protective gas can be an inert protective gas that is permissible by the process, the inlet flow rate is executed by the mass flow controller, and the exhaust valve opening is executed by the chamber exhaust control valve. Increasing the inlet flow rate can enhance the airflow scouring ability of the wafer edge, and adjusting the exhaust valve opening can change the discharge capacity and local flow direction of the edge region. The combination of the two can avoid abnormal local pressure in the chamber caused by simply increasing the inlet flow. The excitation current of the electromagnetic confinement coil is executed by the coil power supply, and the edge plasma distribution pattern is adjusted by changing the magnitude of the excitation current.

[0148] In the simplified simulation example, if the local adjustment strategy requires increased edge protection capability, the controller can adjust the intake flow rate from 40 to 48, while simultaneously adjusting the exhaust valve opening from 30% to 34% to maintain the matching of edge scouring and discharge capabilities, and keep the excitation current constant at 5. If the active degradation reset strategy is entered, the controller can increase the intake flow rate to 80, adjust the exhaust valve opening to match the maximum flow rate, and adjust the excitation current from 5 to 7, while reducing the deposition rate in the main deposition area. The above values ​​are only used to illustrate the coordination relationship between control quantities; the actual values ​​can be pre-calibrated based on the chamber size, gas path capacity, coil rated current, and process window.

[0149] In boundary conditions, if the total pressure in the chamber exceeds the allowable range after the intake flow rate is increased, the exhaust valve opening is increased first; if the exhaust valve has reached the upper limit of the allowable range but the total pressure is still abnormal, the intake flow rate is limited to a further increase and the deposition rate is reduced; if the coil temperature approaches the protection threshold after the excitation current is adjusted, the excitation current is stopped from being increased further and compensation is made by adjusting the airflow parameters and deposition rate; if the feedback values ​​of intake flow rate, exhaust valve opening and excitation current deviate from the set values ​​beyond the allowable range, the controller marks the execution action as incomplete and reassesses whether to enter the safe shutdown or process pause branch.

[0150] In the AuGe alloy deposition mainline scenario, when the process controller selects a local adjustment strategy based on the edge flow field instability index, it only increases the protective gas inlet flow rate and fine-tunes the exhaust valve opening, without changing the excitation current, in order to maintain the stability of the plasma distribution in the main deposition area; when the risk further increases and triggers active degradation reset, the controller simultaneously adjusts the inlet flow rate, exhaust valve opening and excitation current, so that the airflow scouring and magnetic field confinement work together on the wafer edge area.

[0151] The purpose of this mechanism is to translate the abstract airflow and magnetic field control into specific, executable, feedback-enabled, and amplitude-limitable equipment parameters, thereby ensuring that the back-side deposition control method can be stably implemented in the actual AuGe alloy deposition chamber.

[0152] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for controlling the back-side deposition of AuGe alloy coatings, applied to a deposition chamber comprising a main deposition region and an edge region, and equipped with an edge airflow adjustment device, an electromagnetic confinement coil, and an edge state sensing device, characterized in that, Includes the following steps: S1. The edge state sensing device acquires real-time air pressure data and real-time plasma impedance data of the edge region, and extracts air pressure fluctuation signals and plasma impedance signals to generate an initial state dataset. S2. Based on the initial state dataset, high-frequency noise features are extracted to generate an edge flow field instability index, which is used to characterize the degree of accumulation of AuGe atomic free clusters in the edge region; S3. Based on the edge flow field instability index, calculate the critical time when the edge flow field instability index of the current deposition state reaches the preset failure threshold, and determine the corresponding target control strategy; S4. Based on the target control strategy, the edge protection airflow parameters are dynamically adjusted by the edge airflow adjustment device, and the constraint magnetic field parameters are dynamically adjusted by the electromagnetic constraint coil to reduce the accumulation degree and prevent abnormal deposition of AuGe alloy on the back side.

2. The method for controlling backside deposition of AuGe alloy coating according to claim 1, characterized in that, S1 includes the following sub-steps: S11. The real-time air pressure data and real-time plasma impedance data are collected through the edge state sensing device; S12. The real-time air pressure data and real-time plasma impedance data are filtered to separate the high-frequency disturbance component as the air pressure fluctuation signal and plasma impedance signal. S13. The pressure fluctuation signal and the plasma impedance signal are combined to generate the initial state dataset.

3. The method for controlling back-side deposition of AuGe alloy coating according to claim 1, characterized in that, S2 includes the following sub-steps: S21. Analyze the initial state dataset to extract the amplitude variance of the air pressure fluctuation signal and the phase shift of the plasma impedance signal as the high-frequency noise features; S22. Input the amplitude variance and phase offset into a preset state observer model; S23. Calculate using the state observer model to output the edge flow field instability index.

4. The method for controlling back-side deposition of AuGe alloy coating according to claim 3, characterized in that, The preset state observer model is constructed in the following way: Acquire the sample pressure fluctuation signal, sample plasma impedance signal and corresponding back-side arcing record during the historical deposition process; Based on the sample air pressure fluctuation signal, sample plasma impedance signal, and corresponding sample back arcing record, an initial neural network is trained to generate the state observer model.

5. The method for controlling backside deposition of AuGe alloy coating according to claim 1, characterized in that, S3 includes the following sub-steps: S31. Calculate the rate of change of the edge flow field instability index within a preset time window containing a predetermined number of continuous control cycles; S32. Based on the edge flow field instability index and rate of change, and in conjunction with the preset failure threshold, estimate the remaining time to reach the preset failure threshold; S33. The remaining time is determined as the critical time.

6. The method for controlling back-side deposition of AuGe alloy coating according to claim 5, characterized in that, S3 further includes the step of determining the target control strategy based on the critical time, specifically including: A preset first time threshold and a second time threshold are obtained, wherein the first time threshold is greater than the second time threshold; the first time threshold and the second time threshold are obtained by joint calibration based on the historical average of the free cluster accumulation rate in the edge flow field region and the mechanical response delay time of the exhaust valve execution command in the edge airflow regulating device. If the critical time is greater than the first time threshold, a first control strategy to maintain the current deposition rate is generated as the target control strategy; if the critical time is less than or equal to the first time threshold and greater than the second time threshold, a local adjustment strategy is generated as the target control strategy; if the critical time is less than or equal to the second time threshold, an active degradation reset strategy is generated as the target control strategy.

7. The method for controlling back-side deposition of AuGe alloy coating according to claim 6, characterized in that, When the target control strategy is the proactive degradation reset strategy, step S4 includes the following sub-steps: S41. In response to the active degradation reset strategy, generate an airflow surge command and a magnetic field trajectory change command; S42. Based on the airflow surge command, adjust the protective gas intake flow rate in the edge protection airflow parameters to a preset maximum intake flow rate value to disperse the AuGe atom free clusters in the accumulation state; S43. Based on the magnetic field trajectory change command, adjust the parameters of the confinement magnetic field to change the plasma distribution pattern and simultaneously reduce the deposition rate of the main deposition region.

8. The method for controlling back-side deposition of AuGe alloy coating according to claim 6, characterized in that, When the target control strategy is the local adjustment strategy, step S4 includes the following sub-steps: The local adjustment strategy is analyzed, and the corresponding flow fine-tuning step size is calculated based on the difference between the critical time and the second time threshold. Based on the flow rate fine-tuning step size, the edge protection airflow parameter is increased while the constraint magnetic field parameter remains unchanged.

9. The method for controlling back-side deposition of AuGe alloy coating according to claim 1, characterized in that, Before S1, the following also applies: Obtain target usage cycle data and target discharge voltage data; Calculate the offset of the target discharge voltage data relative to the preset initial discharge voltage; When the offset is greater than zero, an asymmetric degradation compensation factor for the target material is generated based on the ratio of the offset to the target material usage cycle data; when the offset is less than or equal to zero, the asymmetric degradation compensation factor for the target material is set to zero or a preset lower limit safety value. S2 specifically includes: inputting the target material asymmetric degradation compensation factor as a gain correction term into a preset state observer model to correct the output edge flow field instability index.

10. The method for controlling back-side deposition of AuGe alloy coating according to claim 1, characterized in that, The edge protection airflow parameters include the intake flow rate of the protective gas and the opening degree of the exhaust valve, and the constraint magnetic field parameters include the magnitude of the excitation current of the electromagnetic constraint coil.

Images