Semiconductor equipment temperature control system fault prediction method based on digital twinning

Abstract

The invention discloses a semiconductor equipment temperature control system fault prediction method based on digital twinning, which relates to the field of temperature control fault prediction, and comprises the following steps: configuring a structural topological graph of a temperature control system, defining a health parameter vector of the temperature control system, configuring an online grey box digital twinning model, and obtaining temperature setting parameters and process stage state parameters. Temperature control response segments are divided, response feature vectors are configured based on the temperature control response segments, and event labeling is carried out. A temperature control system grey box digital twinborn model with on-line updating capability is constructed on the basis of a temperature control system structure topological graph, a corresponding state space relation is established, and a health parameter vector formed by a circulating pump efficiency parameter, a heat exchanger heat transfer capability parameter, a fluid regulating valve opening offset parameter and a temperature sensor zero offset parameter is matched. And the operation state of the temperature control system has good physical interpretability.

Description

Technical Field

[0001] This invention relates to the field of temperature control fault prediction, and in particular to a method for predicting faults in a semiconductor device temperature control system based on digital twins. Background Technology

[0002] Temperature control systems in semiconductor manufacturing equipment are used to maintain the temperature stability of components such as process chambers, pipelines, and heat exchangers. Their control effect directly affects the uniformity of film thickness, the consistency of etching morphology, and the overall line yield. A typical semiconductor equipment temperature control system consists of a heat source unit, a heat exchanger unit, a circulating pump control unit, a fluid regulating valve control unit, a pipeline unit, and temperature sensor measurement units, flow sensor measurement units, and pressure sensor measurement units. It achieves heat exchange between the process heat load and the environment through a closed-loop circulating heat carrier. Under advanced process nodes and high-density integration process conditions, temperature set parameters change frequently and process stage switching is complex. Temperature control systems need to maintain long-term operational stability and high availability while ensuring dynamic response performance. This has driven the industry to gradually introduce digital modeling, process monitoring, and predictive maintenance to improve the observability and manageability of temperature control systems.

[0003] Existing temperature control system fault monitoring relies heavily on alarm thresholds from temperature sensor measurements, simple statistical process control charts, or single data-driven models. These systems lack physically interpretable health parameter vector descriptions of performance degradation in key functional units within the temperature control system's topology, such as heat source units, heat exchanger units, circulating pump control units, and fluid regulating valve control units. Furthermore, they cannot differentiate between different performance degradation modes, such as heat exchanger fouling, circulating pump performance decline, temperature sensor zero-point drift, and fluid regulating valve sticking, based on the dynamic characteristics of temperature control response segments. This results in coarse-grained fault identification and reliance on manual experience for root cause location.

[0004] On the other hand, existing maintenance strategies mostly adopt fixed-cycle maintenance or simple cumulative operating time threshold methods, without combining the evolution trend of health parameter vectors with the prediction results of future process thermal load. It is difficult to calculate the remaining time for key functional units to reach the warning threshold, planned maintenance threshold and forced shutdown threshold based on the health index and multi-level health thresholds. It is even more difficult to match the maintenance window with the production task scheduling of the manufacturing execution system, which can easily lead to premature maintenance that wastes the available life of equipment or maintenance delays that cause unplanned downtime. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a fault prediction method for semiconductor equipment temperature control systems based on digital twins. This method involves configuring the temperature control system's structural topology and constructing a gray-box digital twin model of the temperature control system. It defines a health parameter vector for the temperature control system, consisting of parameters such as circulating pump efficiency, heat exchanger heat transfer capacity, fluid regulating valve opening offset, and temperature sensor zero-point offset. The method then divides the temperature control response into segments based on temperature setpoint steps, wafer actions, and process stage switching events, extracting response feature vectors. Finally, it performs online identification of the health parameter vector within a perturbation excitation calibration window.

[0006] Furthermore, multi-timescale residual decomposition and fault mode fingerprint database are used to achieve online identification of typical performance degradation modes. On this basis, a health index and multi-level health thresholds are introduced. The rate of change of the health index is coupled with the predicted heat load time series to deduce the fault time window of key functional units and output maintenance decision suggestions linked with the equipment management platform. Thus, without changing the existing proportional-integral-derivative control parameters and fuzzy control parameters, early fault trend prediction, fine fault mode identification and predictive maintenance scheduling that matches the production schedule of the temperature control system can be achieved.

[0007] Therefore, this application provides a fault prediction method for a semiconductor device temperature control system based on digital twins, comprising the following steps:

[0008] Step S100: Configure the structural topology diagram of the temperature control system, define the health parameter vector of the temperature control system, and configure an online gray box digital twin model.

[0009] Step S200: Obtain temperature setting parameters and process stage status parameters, divide temperature control response segments, configure response feature vectors based on temperature control response segments, and label events.

[0010] Step S300: Configure the perturbation excitation conditions and temperature setpoint perturbation signal, configure the online identification error term of health parameters, and configure the health parameter update algorithm.

[0011] Step S400: Configure residuals and perform multi-timescale decomposition operations, configure the fault mode fingerprint database, and configure online fault mode identification and fault trend extrapolation.

[0012] Step S500: Obtain the production task schedule, predict the process heat load, configure the health index of key functional units, set multi-level thresholds, set the fault time window simulation and maintenance time point estimation, output the fault prediction results and link them with the maintenance decision system.

[0013] In some specific embodiments, step S100 specifically includes:

[0014] Step S100.1: Configure the structural topology diagram of the temperature control system and define the health parameter vector of the temperature control system.

[0015] Step S100.2: Configure a gray-box digital twin model with online functionality.

[0016] In some specific embodiments, step S200 specifically includes:

[0017] Step S200.1: Obtain the temperature setting parameters and process stage status parameters, and divide the temperature control response segments.

[0018] Step S200.2: Configure response feature vectors and label events based on temperature control response segments.

[0019] For each segment of the temperature control response, dynamic response characteristic indicators closely related to the temperature control response performance are extracted. These dynamic response characteristic indicators include rise time parameter, overshoot parameter, settling time parameter, initial slope parameter, steady-state deviation parameter, and flow response deviation parameter.

[0020] The rise time parameter, overshoot parameter, settling time parameter, initial slope parameter, steady-state deviation parameter, and flow response deviation parameter are constructed into a response feature vector in a fixed order.

[0021] Each response feature vector is associated with the process event type corresponding to the temperature control response segment to which the response feature vector belongs, and an event tag is configured for the temperature control response segment. The event tags include a set value step response tag, a wafer thermal shock response tag, and a power on response tag.

[0022] In some specific embodiments, step S300 specifically includes:

[0023] Step S300.1: Configure the perturbation excitation conditions and temperature setpoint perturbation signal.

[0024] A temperature control calibration window is configured for each process formula. The temperature control calibration window is selected during the time period when the thermal stability requirements are low in the process. The time period includes the standby phase between two adjacent batches and the time interval of 5 to 10 seconds before the preheating phase before the cold start of the equipment.

[0025] Within the temperature control calibration window, a small-amplitude temperature setpoint perturbation signal is superimposed on the temperature setpoint signal. The temperature control system is subjected to a perturbation provided that the temperature setpoint perturbation signal does not exceed the process temperature tolerance range.

[0026] Step S300.2: Configure the online identification of error items for health parameters and configure the health parameter update algorithm.

[0027] In some specific embodiments, step S400 specifically includes:

[0028] Step S400.1: Configure residuals and perform multi-timescale decomposition operation, and configure the fault mode fingerprint database.

[0029] Step 400.2: Configure online fault mode identification and fault trend extrapolation.

[0030] In some specific embodiments, step S500 specifically includes:

[0031] Step S500.1: Obtain the production task schedule, predict the process heat load, configure the health index of key functional units, and set multi-level thresholds.

[0032] Step S500.2: Set the fault time window for simulation and maintenance time point estimation, output the fault prediction results and link them with the maintenance decision system.

[0033] At the current time At any given time, calculate the health index for each key functional unit. The rate of change of the health index was estimated based on its recent changes. Assuming that the rate of change of the health index is proportional to the average heat load power, the [further details are needed for the next step]. The rate of change of the health index of each key functional unit is approximated as follows:

[0034]

[0035] Based on the rate of change of the health index and the health threshold, the time interval required for the current health index to decay to different thresholds is calculated.

[0036] Combining the available maintenance windows of semiconductor manufacturing equipment, Matching with equipment idle time periods or process batch intervals yields the first... Maintenance time intervals for key functional units, including when the corresponding health index approaches the health warning threshold. The earliest recommended maintenance time corresponds to the health index reaching the mandatory shutdown threshold. The previous latest permitted maintenance time.

[0037] Based on the configured fault mode fingerprint database, the fault mode type of the current critical functional unit is queried. The queried fault modes are associated with the health index and maintenance time interval of the critical functional unit to generate a list of recommended maintenance items for the critical functional unit.

[0038] When the obtained fault time window data, the time series of health index of key functional units, and the fault mode identifier obtained by matching the fault mode fingerprint database are transmitted to the equipment management platform and the predictive maintenance platform through the data interface protocol, the structured data fields include equipment identifier, key functional unit number, current value of health index, predicted trend of health index change, suggested maintenance time range, fault mode name, and fault mode confidence score.

[0039] In summary, the fault prediction method for semiconductor equipment temperature control systems based on digital twins provided in this application constructs a gray-box digital twin model of the temperature control system with online update capability based on the structural topology diagram of the temperature control system. It establishes a one-to-one state-space relationship between the controlled hot cavity temperature state quantity, the liquid supply pipe temperature state quantity, the liquid return pipe temperature state quantity, and the liquid storage unit temperature state quantity and the heating power adjustment parameter, the cooling power adjustment parameter, the circulating pump speed control parameter, and the fluid regulating valve opening adjustment parameter. Combined with the health parameter vector composed of the circulating pump efficiency parameter, the heat exchanger heat transfer capacity parameter, the fluid regulating valve opening offset parameter, and the temperature sensor zero-point offset parameter, the operating state of the temperature control system has good physical interpretability.

[0040] By configuring response feature vectors for temperature control response segments such as temperature setpoint step response, thermal shock response of the carrier plate, and power activation response, and using recursive least squares algorithm or extended Kalman filter algorithm to identify health parameter vectors online under the action of temperature setpoint perturbation signal, and combining multi-timescale residual decomposition and fault mode fingerprint database, the system can distinguish and identify heat exchanger scaling mode, circulating pump performance degradation mode, temperature sensor zero drift mode, and fluid control valve jamming or opening offset mode. This effectively makes up for the shortcomings of existing technologies that rely on a single sensor alarm and cannot identify the performance degradation of key functional units in the early stage.

[0041] Furthermore, by obtaining production task schedules from the manufacturing execution system and constructing predicted thermal load time series, defining health indices and multi-level health thresholds for key functional units, and extrapolating the time intervals for reaching early warning thresholds, planned maintenance thresholds, and forced shutdown thresholds based on the rate of change of the health index, and matching this with the available maintenance window of the equipment, maintenance decision data including fault mode names, health index trends, and suggested maintenance time ranges are output to the equipment management platform and predictive maintenance platform. Without interfering with the existing fuzzy PID+Smith control structure and process digital twin system, visualized health management of the temperature control system, load-aware fault time window prediction, and maintenance strategy optimization coordinated with the production schedule are achieved, thereby reducing the risk of unplanned downtime, reducing over-maintenance, and improving the overall uptime and process stability of semiconductor manufacturing equipment. Attached Figure Description

[0042] Figure 1 This is an overall flowchart of a fault prediction method for a semiconductor device temperature control system based on digital twin, provided in an embodiment of this application. Detailed Implementation

[0043] Please refer to Figure 1 The diagram illustrates a flow chart of an embodiment of a fault prediction method for a semiconductor device temperature control system based on digital twins, according to the present disclosure.

[0044] like Figure 1 As shown, a fault prediction method for a semiconductor equipment temperature control system based on digital twins includes the following steps:

[0045] Step S100: Configure the structural topology diagram of the temperature control system, define the health parameter vector of the temperature control system, and configure an online gray box digital twin model.

[0046] Step S200: Obtain temperature setting parameters and process stage status parameters, divide temperature control response segments, configure response feature vectors based on temperature control response segments, and label events.

[0047] Step S300: Configure the perturbation excitation conditions and temperature setpoint perturbation signal, configure the online identification error term of health parameters, and configure the health parameter update algorithm.

[0048] Step S400: Configure residuals and perform multi-timescale decomposition operations, configure the fault mode fingerprint database, and configure online fault mode identification and fault trend extrapolation.

[0049] Step S500: Obtain the production task schedule, predict the process heat load, configure the health index of key functional units, set multi-level thresholds, set the fault time window simulation and maintenance time point estimation, output the fault prediction results and link them with the maintenance decision system.

[0050] In some specific embodiments, step S100 specifically includes:

[0051] Step S100.1: Configure the structural topology diagram of the temperature control system and define the health parameter vector of the temperature control system.

[0052] For the temperature control system in semiconductor manufacturing equipment, the various physical components inside the semiconductor manufacturing equipment used to maintain the temperature stability of the process cavity are identified, and a temperature control system structure topology diagram covering all heat exchange units and fluid circulation units is constructed based on the identification results.

[0053] The temperature control system topology includes: a heat source unit for providing heat energy to the process chamber, a heat exchanger unit for exchanging heat with the environment or cooling medium, a pipeline unit for transporting the heat carrier, a circulating pump control unit for driving the heat carrier to circulate in the pipeline unit, a fluid regulating valve control unit for regulating the flow rate of the heat carrier, a temperature sensor measurement unit for measuring temperature, a flow sensor measurement unit for measuring flow rate, and a pressure sensor measurement unit for measuring pressure.

[0054] Each structural unit has a corresponding mapping relationship between physical entities and digital modeling parameters, and in the process of digital twin modeling of the temperature control system, a parameterized component corresponding to the structural unit is established.

[0055] The temperature control system topology is constructed in the form of a closed loop. The heat source unit and the heat exchanger unit are connected through the pipeline unit. The heat carrier circulates along the pipeline unit under the driving force provided by the circulating pump control unit, and the flow rate of the heat carrier is regulated by the fluid regulating valve control unit. The heat carrier flows through the controlled heat cavity that bears the process heat load and the heat exchanger unit that performs heat exchange in sequence, and then returns to the liquid storage unit for storing the heat carrier, thus forming a complete heat flow circulation path.

[0056] Each physical unit in the temperature control loop is mapped to a low-order thermofluid dynamic model component. The heat exchanger unit, the controlled thermal cavity, and the pipeline unit are divided into several thermal capacity nodes and thermal resistance nodes. The node-level thermofluid state space model is established through the heat conduction behavior, heat exchange behavior, and heat convection behavior between the thermal capacity nodes and thermal resistance nodes, forming a thermal energy state vector.

[0057] The thermal state vector includes at least the temperature state variables of the controlled thermal cavity, the liquid supply pipe, the liquid return pipe, and the liquid storage unit. The control input variables include heating power adjustment parameters, cooling power adjustment parameters, circulating pump speed control parameters, and fluid regulating valve opening adjustment parameters. Each control input variable is mapped to the thermal state vector through a dynamic function relationship, so that the dynamic model of the temperature control system can reflect the influence of changes in control input variables on the evolution process of the thermal state vector.

[0058] A health parameter vector for the temperature control system is constructed. The health parameter vector consists of several physically interpretable scalar parameters, which are used to characterize the performance degradation of key functional units in the temperature control system. The health parameter vector includes circulating pump efficiency parameters, heat exchanger heat transfer capacity parameters, fluid regulating valve opening offset parameters, and temperature sensor zero point offset parameters.

[0059] The circulating pump efficiency parameter is a real number ranging from 0 to 1, used to describe the driving efficiency of the circulating pump control unit relative to the standard reference state under the current operating condition. The heat exchanger heat transfer capacity parameter is a real number ranging from 0 to 1, used to describe the effective heat transfer capacity of the heat exchanger unit relative to the new equipment state under actual working conditions. The fluid regulating valve opening offset parameter is ranging from −5% to +5%, used to describe the static opening offset of the fluid regulating valve control unit due to mechanical wear or response lag. The temperature sensor zero point offset parameter is ranging from −0.5℃ to +0.5℃, used to describe the degree of reference measurement offset of the temperature sensor measurement unit due to drift or aging.

[0060] The health parameter vector of the temperature control system is coupled with the thermodynamic behavior parameters of the temperature control system. The coupling methods between the health parameter vector of the temperature control system and the thermodynamic model of the temperature control system include: defining the heat transfer coefficient of the heat exchanger unit as the product of the reference heat transfer coefficient and the heat transfer capacity parameter of the heat exchanger; and defining the actual output flow of the circulating pump control unit as the product of the standard output flow and the efficiency parameter of the circulating pump. This is used to reflect the direct impact of the degradation of key performance on the thermal response behavior of the temperature control system in the thermodynamic model of the temperature control system.

[0061] Step S100.2: Configure a gray-box digital twin model with online functionality.

[0062] By combining the thermal fluid dynamic model of the temperature control system based on the structural topology diagram of the temperature control system with the health parameter vector of the temperature control system, a gray box digital twin model of the temperature control system is constructed. The gray box digital twin model of the temperature control system expresses the dynamic process of the thermal energy state vector of the temperature control system changing over time in the form of state space.

[0063] The state-space form of the gray box digital twin model of the temperature control system includes state update equations and output observation equations. The state update equations are used to describe the evolution process of the current thermal energy state vector under the combined action of control input variables and temperature control system health parameter vectors. The output observation equations are used to describe the mapping relationship between the temperature data, flow data and pressure data collected by the temperature sensor measurement unit, flow sensor measurement unit and pressure sensor measurement unit and the internal thermal energy state vector.

[0064] Each state variable in the state-space form of the gray box digital twin model of the temperature control system corresponds to a specific physical meaning. The observed variables in the output observation equation include at least the temperature of the controlled hot chamber, the supply liquid temperature, the return liquid temperature, the flow sensor measurement value, and the pressure sensor measurement value. Process noise terms and measurement noise terms are introduced into the gray box digital twin model structure of the temperature control system to simulate external disturbances and sensor measurement errors during the operation of the temperature control system. The health parameter vector of the temperature control system is set according to the factory calibration value of the new equipment in the initial state, and is updated by an online identification mechanism during subsequent operation. The online identification mechanism adjusts the health parameter vector of the temperature control system according to the deviation between the sensor measurement data and the output of the gray box digital twin model of the temperature control system.

[0065] In some specific embodiments, step S200 specifically includes:

[0066] Step S200.1: Obtain the temperature setting parameters and process stage status parameters, and divide the temperature control response segments.

[0067] During the execution of process recipes in semiconductor manufacturing equipment, temperature control-related setting parameters and operating status parameters are obtained from the equipment control system. The setting parameters include temperature setpoint parameters, cavity pressure parameters, and radio frequency power parameters. The operating status parameters include wafer action identifier parameters and process stage number parameters.

[0068] The temperature setpoint parameter is used to indicate the target temperature value of each heating unit and each cooling unit of the semiconductor manufacturing equipment. The cavity pressure parameter is used to indicate the real-time pressure level in the reaction cavity. The radio frequency power parameter is used to indicate the power level in the plasma process. The wafer action identifier parameter is used to mark whether the silicon wafer has been fed into the cavity. The process stage number parameter is used to identify the current formulation step.

[0069] Based on the changing characteristics of temperature setpoint parameters, cavity pressure parameters, RF power parameters, wafer action identifier parameters, and process stage number parameters, multiple event triggering rules are defined to mark the boundary moments of temperature control system response changes. The event triggering rules include: when the absolute difference of the temperature setpoint parameter between two adjacent sampling moments is greater than or equal to 0.3℃, it is determined as a temperature setpoint step event; when the wafer action identifier parameter changes from 0 to 1, it is determined as a wafer event; and when the process stage number parameter changes, it is determined as a recipe stage switching event.

[0070] The event time determined by the event triggering rules is used as the boundary condition for dividing the temperature control response segments. Any two adjacent event triggering times constitute the time interval of a temperature control response segment. The measurement data collected by the temperature control system in each time interval is used to construct an independent temperature control response segment data set. Each temperature control response segment contains the data of temperature measurement sequence, flow measurement sequence and pressure measurement sequence in that time interval, forming multiple temperature control response segments with clear boundaries and independent response behavior.

[0071] Step S200.2: Configure response feature vectors and label events based on temperature control response segments.

[0072] For each segment of the temperature control response, dynamic response characteristic indicators closely related to the temperature control response performance are extracted. These dynamic response characteristic indicators include rise time parameter, overshoot parameter, settling time parameter, initial slope parameter, steady-state deviation parameter, and flow response deviation parameter.

[0073] The rise time parameter characterizes the time required for the temperature response to reach 90% of the target value. It is calculated as the length of the time interval from the start of the temperature response curve's change in the temperature measurement sequence to reaching 90% of the setpoint. The overshoot parameter characterizes the positive deviation of the maximum temperature value relative to the setpoint during the temperature response. It is calculated as the difference between the maximum temperature value and the setpoint in the temperature measurement sequence. The settling time parameter characterizes the response time required for the temperature response to reach the setpoint ±0.1℃ range and remain within that range. The initial slope parameter characterizes the rate of change of the temperature response curve in the initial stage of the response. It is taken as the first-order differential slope of the temperature measurement sequence at the beginning of the temperature control response segment. The steady-state deviation parameter characterizes the deviation between the final stable value of the temperature response and the setpoint. The flow response deviation parameter characterizes the degree of deviation between the actual measured flow rate and the theoretical flow rate predicted by the gray-box digital twin model of the temperature control system.

[0074] The rise time parameter, overshoot parameter, settling time parameter, initial slope parameter, steady-state deviation parameter, and flow response deviation parameter are constructed into a response feature vector in a fixed order. This response feature vector is expressed by the formula... The response feature vector of each temperature control response segment is as follows:

[0075]

[0076] In the formula: The rise time parameter, This is the overshoot parameter. To adjust the time parameters, The initial slope parameter, For steady-state deviation parameters, This is the flow response deviation parameter.

[0077] Each response feature vector is associated with the process event type corresponding to the temperature control response segment to which the response feature vector belongs, and an event tag is configured for the temperature control response segment. The event tags include a set value step response tag, a wafer thermal shock response tag, and a power on response tag.

[0078] Event tags are used to distinguish different types of temperature control response segments in subsequent temperature control health status determination and fault trend identification tasks.

[0079] In some specific embodiments, step S300 specifically includes:

[0080] Step S300.1: Configure the perturbation excitation conditions and temperature setpoint perturbation signal.

[0081] A temperature control calibration window is configured for each process formula. The temperature control calibration window is selected during the time period when the thermal stability requirements are low in the process. The time period includes the standby phase between two adjacent batches and the time interval of 5 to 10 seconds before the preheating phase before the cold start of the equipment.

[0082] Within the temperature control calibration window, a small-amplitude temperature setpoint perturbation signal is superimposed on the temperature setpoint signal. This perturbation is applied to the temperature control system provided the temperature setpoint perturbation signal does not exceed the process temperature tolerance range. The temperature setpoint perturbation signal is defined as follows:

[0083]

[0084] In the formula: Indicates time The temperature setpoint disturbance value superimposed over time. Indicates the disturbance amplitude, disturbance amplitude The value range is from −0.2℃≤A≤+0.2℃ to −0.2℃≤A≤+0.2℃. Indicates the perturbation frequency, perturbation frequency The value is not higher than This is used to reduce the impact of temperature setpoint perturbation excitation on the stability of silicon wafer temperature control within the cavity.

[0085] The temperature setpoint perturbation signal is independently recorded as a perturbation calibration segment marker in the equipment control system and managed separately from the normal process segment. This ensures that the temperature control calibration window segment data and the normal process segment data are distinguishable from each other, thereby improving the accuracy and stability of process data analysis results.

[0086] Step S300.2: Configure the online identification of error items for health parameters and configure the health parameter update algorithm.

[0087] During the perturbation excitation calibration window, a sequence of temperature measurements is acquired from the temperature sensor measurement unit. Obtain the flow measurement value sequence from the flow sensor measurement unit. , Number the discrete sampling time points, and link the current control input variable with the current health parameter vector. Input the gray box digital twin model of the temperature control system to obtain the temperature prediction value sequence. With the sequence of predicted flow values .

[0088] Based on the deviations between measured and predicted temperature values, and between measured and predicted flow rates, an error term is constructed for online identification of health parameters. The error term is defined as follows:

[0089]

[0090]

[0091] The temperature error term and the flow rate error term are combined column-wise to form the overall error vector. The overall error vector is defined as:

[0092]

[0093] In the formula: This indicates that the sampling sequence number in discrete time is Temperature error term at time, This indicates that the sampling sequence number in discrete time is The flow error term at any given time. This indicates that the sampling sequence number in discrete time is The temperature measurement value obtained by the temperature sensor measurement unit at all times. This indicates that the sampling sequence number in discrete time is Time, in the health parameter vector The predicted temperature value calculated by the gray box digital twin model of the temperature control system under the given conditions. This indicates that the sampling sequence number in discrete time is The flow measurement value obtained by the flow sensor measurement unit at all times. This indicates that the sampling sequence number in discrete time is Time, in the health parameter vector The predicted flow rate calculated by the digital twin model of the ash box of the temperature control system under the given conditions. This indicates the time sequence number during the discrete sampling process of the temperature control system.

[0094] In the gray box digital twin model of the temperature control system constructed based on state-space form, the health parameter vector is updated online using either the recursive least squares algorithm or the extended Kalman filter algorithm. The update relation is defined as follows:

[0095]

[0096] In the formula: For the first A vector of health parameters at each sampling time. It is the adaptive gain matrix. Output the partial derivative matrix of the gray box digital twin model of the temperature control system with respect to the health parameter vector. For the first The error vector at each sampling time point.

[0097] To maintain the physical interpretability of the health parameter vector during online updates, reasonable constraint intervals are set for each component of the health parameter vector. The constraint intervals are defined as follows:

[0098]

[0099]

[0100]

[0101]

[0102] The aforementioned health parameter constraint range is used to prevent the health parameters from generating solutions that do not conform to the physical characteristics of the temperature control system when process drift or abnormal disturbances occur during the online identification process.

[0103] After each perturbation excitation calibration window ends and the online identification and update of the health parameter vector is completed, the updated health parameter vector is correlated with the response feature vector of the temperature control response segment corresponding to the process event. A time-series correlation is then performed between the health parameter vector and the response feature vector to construct a joint evolution time series data structure for subsequent analysis. The joint evolution time series is represented as follows:

[0104]

[0105] in, For the first The health parameter vector corresponding to each update time. For process response segments The response feature vector, Number the process formula. This refers to the cavity numbering of semiconductor manufacturing equipment. To record timestamps, the joint time series is used in subsequent steps for fault evolution trend identification, multi-timescale residual analysis, and cluster analysis of temperature control anomaly patterns.

[0106] In some specific embodiments, step S400 specifically includes:

[0107] Step S400.1: Configure residuals and perform multi-timescale decomposition operation, and configure the fault mode fingerprint database.

[0108] During the operation of the temperature control system, the current control input variables and the current health parameter vector are used as inputs. By calling the updated gray box digital twin model of the temperature control system, the current sampling time is generated. The predicted output vector, denoted as The predicted output vector includes predicted temperature, predicted flow rate, and predicted pressure.

[0109] The temperature sensor measurement unit, flow sensor measurement unit, and pressure sensor measurement unit acquire the data at the sampling time. The temperature measurement value, flow measurement value, and pressure measurement value are collected, and the three types of measurement values ​​are combined to form the measured output vector. Sampling time is constructed based on the measured output vector and the predicted output vector. residual vector The residual vector is defined as:

[0110]

[0111] in, It is a three-dimensional residual vector, which includes temperature residual components, flow rate residual components and pressure residual components. To be at the sampling time The measured output vector obtained by combining the measurement units of each sensor To be at the sampling time The predicted output vector is calculated from the gray box digital twin model of the temperature control system.

[0112] The residual vector time series obtained by arranging in chronological order Decoupling is achieved using a multi-time-scale decomposition method, which includes at least one of wavelet decomposition algorithm and multi-rate filter structure. This method decomposes the residual vector into fast perturbation components and slow drift components, with the following decomposition relationship:

[0113]

[0114] in, This is a fast perturbation residual vector, used to characterize the rapidly changing components caused by high-frequency signal perturbations and sensor noise. The slow-drift residual vector is used to characterize the structural deviation caused by long-term slow changes in health parameters. The residual decoupling process is used to separate the time-domain features of the fault dynamic response, providing a stable residual criterion for subsequent fault mode identification.

[0115] During the equipment trial operation phase, historical maintenance record phase, and manually set accelerated aging test phase, data on different types of performance degradation modes that may occur in the temperature control system of semiconductor manufacturing equipment are collected and analyzed. Data on the changes of health parameter vectors and response feature vectors over time corresponding to each performance degradation mode are obtained, and combined with the slow drift residual vector obtained by multi-timescale decomposition. Construct a failure mode sample set.

[0116] The failure mode sample set includes typical temperature control system performance degradation modes such as heat exchanger scaling mode, circulating pump performance degradation mode, temperature sensor zero drift mode, and fluid control valve sticking or opening offset mode.

[0117] Heat exchanger fouling mode: heat exchanger heat transfer capacity parameters The temperature residual component in the slow-drift residual vector exhibits a monotonically decreasing trend over time. A persistent positive deviation occurs, affecting the rise time parameter in the response eigenvector. With adjustment time parameter Compared to a significantly increased state of health, among which, Represents the slow drift residual vector The component in the temperature channel.

[0118] Circulating pump performance degradation mode: circulating pump efficiency parameters The components of the slow-drift residual vector in the flow channel during descent. The deviation is negative, and the rise time of the temperature response is significantly longer compared to the healthy state. Represents the slow drift residual vector The component on the flow channel.

[0119] Temperature sensor zero-point drift mode: Temperature sensor zero-point offset parameters The absolute value of increases continuously over time, while the changes in flow residual and pressure residual are not significant.

[0120] Fluid control valve stuck or opening offset mode: Fluid control valve opening offset parameter A significant bias is observed compared to zero, with the overshoot parameter in the response feature vector. The magnitude of the response increases, and the response amplitude varies significantly to different process events.

[0121] The range of health parameter vector changes, slow drift residual vector features, and response feature vector features corresponding to the performance degradation modes of each typical temperature control system are summarized by combining structured rules and a lightweight pattern recognition model to construct a fault mode fingerprint database. This database is established when the following conditions are met: And the duration of the deviation is greater than If so, the current operating status of the temperature control system will be identified as the heat exchanger scaling development stage mode in the fault mode fingerprint database.

[0122] Step 400.2: Configure online fault mode identification and fault trend extrapolation.

[0123] During the continuous operation of semiconductor manufacturing equipment, based on the time-series recorded health parameter vector sequences Slow drift residual vector sequence With response feature vector sequence A fixed-length sliding time window is used to select the health parameter vector, slow drift residual vector, and response feature vector within a continuous sampling time period, and the data within the sliding time window are combined to form a joint analysis sample.

[0124] For each health parameter component in the joint analysis sample Calculate the rate of change of time within the sliding time window, denoted as . The rate of change over time is used to characterize the components of health parameters. The degradation trend is analyzed by calculating the mean and trend slope of each component of the slow drift residual vector within the sliding time window, which is used to quantify the cumulative deviation and direction of change of the slow drift residual within the current time window.

[0125] The current operating status characteristics obtained based on the time change rate of health parameters and the slow drift residual statistics are compared with the fault mode characteristics in the fault mode fingerprint database to identify that the current operating status of the temperature control system belongs to one of the following states: normal operation state, early performance degradation state, fault critical state, and recommended immediate shutdown state.

[0126] After obtaining the current operating status of the temperature control system, the system outputs the fault mode identifier that matches the current operating status and the corresponding fault mode confidence value. The fault mode confidence value is used to characterize the degree of matching between the current operating status and the characteristics of each fault mode in the fault mode fingerprint database, providing a basis for subsequent predictive maintenance scheduling strategies and temperature control system operation strategy optimization.

[0127] In some specific embodiments, step S500 specifically includes:

[0128] Step S500.1: Obtain the production task schedule, predict the process heat load, configure the health index of key functional units, and set multi-level thresholds.

[0129] Obtain production task scheduling information for a preset future time period from the manufacturing execution system or scheduling system of semiconductor manufacturing equipment. The production task scheduling information includes the target execution process recipe sequence, process cavity usage frequency, batch processing interval time and cavity, and parallel operation occupancy rate.

[0130] Based on production task scheduling information and combined with the statistical mapping relationship between historical process formula types and heat load levels, a heat load prediction model for temperature control system load estimation is constructed. The heat load prediction model outputs a predicted heat load time series corresponding to future time intervals, denoted as […]. , Indicates at a point in time The predicted unit heat load power is obtained. The heat load prediction model predicts the heat load based on the mapping relationship between process formulation type and average heat load power range. The prediction results satisfy... .

[0131] in, This refers to the minimum heat load power obtained from historical operations for the corresponding process formulation type. This refers to the maximum heat load power obtained from historical operations for the corresponding process formulation type.

[0132] For the circulating pump control unit, heat exchanger unit, temperature sensor measurement unit, and fluid regulating valve control unit in the temperature control system, define corresponding health parameter components and health indexes respectively. Used for comprehensive evaluation of the The health index is calculated as follows: This reflects the performance deviation of each key functional unit's current operating state from its reference calibration state.

[0133]

[0134] in, For the first Key functional units in time Current health parameter values, For the first The reference health parameter values ​​for each key functional unit during the equipment's factory calibration phase. For the first The maximum allowable variation in health parameters for each key functional unit within physically acceptable limits.

[0135] Multiple health thresholds are set for the health index to support different levels of fault warning and maintenance decisions. The health thresholds include: health warning threshold, planned maintenance threshold, and forced shutdown threshold.

[0136] The health warning threshold is: .

[0137] The planned maintenance threshold is: .

[0138] The forced shutdown threshold is: .

[0139] Step S500.2: Set the fault time window for simulation and maintenance time point estimation, output the fault prediction results and link them with the maintenance decision system.

[0140] At the current time At any given time, calculate the health index for each key functional unit. The rate of change of the health index was estimated based on its recent changes. Assuming that the rate of change of the health index is proportional to the average heat load power, the [further details are needed for the next step]. The rate of change of the health index of each key functional unit is approximated as follows:

[0141]

[0142] in, For the first Thermal degradation sensitivity coefficients for each key functional unit This is the average predicted heat load power calculated within a preset sliding time window.

[0143] Based on the rate of change of the health index and the health threshold, the time interval required for the current health index to decay to different thresholds is calculated.

[0144] The time interval required to reach the health warning threshold is:

[0145]

[0146] The time interval required to reach the planned maintenance threshold is:

[0147]

[0148] Combining the available maintenance windows of semiconductor manufacturing equipment, Matching with equipment idle time periods or process batch intervals yields the first... Maintenance time intervals for key functional units, including when the corresponding health index approaches the health warning threshold. The earliest recommended maintenance time corresponds to the health index reaching the mandatory shutdown threshold. The previous latest permitted maintenance time.

[0149] Based on the configured fault mode fingerprint database, the fault mode type of the current critical functional unit is queried. The queried fault modes are associated with the health index and maintenance time interval of the critical functional unit to generate a list of recommended maintenance items for the critical functional unit.

[0150] When the obtained fault time window data, the time series of health index of key functional units, and the fault mode identifier obtained by matching the fault mode fingerprint database are transmitted to the equipment management platform and the predictive maintenance platform through the data interface protocol, the structured data fields include equipment identifier, key functional unit number, current value of health index, predicted trend of health index change, suggested maintenance time range, fault mode name, and fault mode confidence score.

[0151] The generated fault prediction results and maintenance decision suggestions are only used as the basis for operation and maintenance decisions. They do not adjust the existing proportional, integral, derivative and fuzzy control parameters in the temperature control system. They are clearly distinguished from the existing fuzzy PID+Smith control structure and ProficyCSense process digital twin system in terms of control layer implementation path.

[0152] In practical applications, when modeling the temperature control system of a process chamber on a mass-production wafer production line, engineers establish physical mapping relationships for each of the following units: heat source unit, heat exchanger unit, circulating pump control unit, fluid regulating valve control unit, pipeline unit, liquid storage unit, and temperature sensor measurement unit, flow sensor measurement unit, and pressure sensor measurement unit. The controlled heat chamber temperature, supply pipe temperature, return pipe temperature, and liquid storage unit temperature are combined to form a thermal energy state vector. Values ​​from 0 to 1 are set for the circulating pump efficiency parameter and the heat exchanger heat transfer capacity parameter; a value range of −5% to +5% is set for the fluid regulating valve opening offset parameter; and a value range of −0.5℃ to +0.5℃ is set for the temperature sensor zero-point offset parameter. This constitutes a temperature control system health parameter vector. Based on this, a gray-box digital twin model of the temperature control system is configured, enabling the gray-box digital twin model to output predicted values ​​of temperature, flow rate, and pressure under given control input variables and the temperature control system health parameter vector.

[0153] During the process of formula operation, the temperature setpoint parameters, cavity pressure parameters, RF power parameters, wafer action identifier parameters, and process stage number parameters are continuously read from the equipment control system. When the absolute difference of the temperature setpoint parameters at adjacent sampling times reaches 0.3℃, it is marked as a temperature setpoint step event. When the wafer action identifier parameter switches from a wafer-free state to a wafer-on state, it is marked as a wafer-on event. When the process stage number parameter changes, it is marked as a formula stage switching event. The temperature control response segment is divided with any two adjacent event times as the boundary. For each temperature control response segment, the rise time, overshoot, settling time, initial slope, steady-state deviation, and flow response deviation are extracted from the temperature measurement sequence and flow measurement sequence. These are used to form a response feature vector in a fixed order. Setpoint step response label, wafer thermal shock response label, and power on response label are configured according to the event type for subsequent health analysis and fault identification.

[0154] A temperature control calibration window is set in each process formula. The temperature control calibration window is placed in the standby stage between adjacent batches or the preheating stage before cold start of the equipment for a continuous time interval of 5 to 10 seconds. Within the temperature control calibration window, a temperature setpoint perturbation signal with an amplitude in the range of -0.2℃ to +0.2℃ and a frequency not higher than 0.1 Hz is superimposed on the temperature setpoint. The temperature setpoint perturbation signal is marked as the perturbation calibration segment in the equipment control system. The measurement values ​​of the temperature sensor measurement unit and the flow sensor measurement unit within the temperature control calibration window are collected and compared with the temperature prediction value and flow prediction value output by the gray box digital twin model of the temperature control system. The temperature error and flow error are calculated, and the health parameter vector of the temperature control system is updated using the recursive least squares algorithm or the extended Kalman filter algorithm. At the same time, it is ensured that the circulating pump efficiency parameter is not lower than 0.6, the heat exchanger heat transfer capacity parameter is not lower than 0.5, the fluid regulating valve opening offset parameter does not exceed -5% to +5%, and the temperature sensor zero point offset parameter does not exceed -0.5℃ to +0.5℃.

[0155] During the continuous operation of the temperature control system, the current control input variables and the current health parameter vector of the temperature control system are input into the gray box digital twin model of the temperature control system to obtain the predicted values ​​of temperature, flow, and pressure. These predicted values ​​are then compared with the measured values ​​obtained by the temperature sensor measurement unit, flow sensor measurement unit, and pressure sensor measurement unit to form a residual time series. Multi-timescale decomposition is performed on the residual time series to decompose the residuals into fast disturbance residual vectors and slow drift residual vectors. During the equipment trial operation phase, historical maintenance phase, and accelerated aging test phase, heat exchanger fouling mode, circulating pump performance degradation mode, temperature sensor zero-point drift mode, and fluid regulating valve sticking or opening deviation mode are collected. The corresponding temperature control system health parameter vector trajectory, slow drift residual vector features, and response feature vector features are used to construct a fault mode fingerprint database. The judgment condition for the heat exchanger fouling mode is set as follows: the heat exchanger heat transfer capacity parameter is lower than 0.75 and the component of the slow drift residual vector in the temperature channel is continuously higher than 0.3℃ for more than 30 minutes. The rate of change of each temperature control system health parameter component and the statistical features of the slow drift residual vector are calculated using a fixed-length sliding time window. The current operating status features are compared with the fault mode fingerprint database to output the normal operating status, early performance degradation status, fault critical status, and suggested shutdown status, as well as the corresponding fault mode confidence.

[0156] Based on online fault mode identification, production task scheduling information for future time periods is retrieved from the Manufacturing Execution System (MES) and Production Scheduling System. This includes process recipe sequences, process chamber usage frequency, batch processing intervals, and chamber parallel occupancy rates. Based on the mapping relationship between process recipe types and heat load power in historical operating data, a heat load prediction model is constructed, outputting a predicted heat load time series limited to the historical minimum and maximum heat load power. For the circulating pump control unit, heat exchanger unit, temperature sensor measurement unit, and fluid regulating valve control unit, a health index is calculated based on the temperature control system's health parameter vector, and a unified setting is implemented. The system sets a health warning threshold of 0.8, a planned maintenance threshold of 0.6, and a forced shutdown threshold of 0.4. It then calculates the time interval for the health index to decay to each threshold based on the recent rate of change of the health index and the predicted heat load time series. This information is matched with idle time periods in the production task schedule to determine the earliest recommended maintenance time and the latest allowed maintenance time. The system writes the equipment identifier, functional unit number, health index, recommended maintenance time range, fault mode name, and fault mode confidence level to the equipment management platform and predictive maintenance platform via an interface. This information guides maintenance decisions for the semiconductor equipment temperature control system while maintaining the existing proportional-integral-derivative control and fuzzy control logic unchanged.

Claims

1. A fault prediction method for a semiconductor equipment temperature control system based on digital twins, characterized in that, Includes the following steps: S100: Configure the structural topology diagram of the temperature control system, define the health parameter vector of the temperature control system, and configure an online gray box digital twin model; S200: Obtain temperature setting parameters and process stage status parameters, divide temperature control response segments, configure response feature vectors based on temperature control response segments, and label events. S300, configure perturbation excitation conditions and temperature setpoint perturbation signals, configure online identification of health parameters error terms, and configure health parameter update algorithm; S400: Configure residuals and perform multi-timescale decomposition operations; configure fault mode fingerprint database; configure online fault mode identification and fault trend extrapolation. S500 acquires production task schedules, predicts process heat load, configures health indices for key functional units, sets multi-level thresholds, sets fault time windows for simulation and maintenance time point estimation, outputs fault prediction results, and links with the maintenance decision system.

2. The fault prediction method for a semiconductor equipment temperature control system based on digital twin according to claim 1, characterized in that, S100 specifically includes: S100.1 Configure the structural topology diagram of the temperature control system and define the health parameter vector of the temperature control system; For the temperature control system in semiconductor manufacturing equipment, the various physical components inside the semiconductor manufacturing equipment used to maintain the temperature stability of the process cavity are identified, and a temperature control system structure topology diagram covering all heat exchange units and fluid circulation units is constructed based on the identification results. The temperature control system topology diagram includes: a heat source unit for providing heat energy to the process chamber, a heat exchanger unit for exchanging heat with the environment or cooling medium, a pipeline unit for transporting the heat carrier, a circulating pump control unit for driving the heat carrier to circulate in the pipeline unit, a fluid regulating valve control unit for regulating the flow rate of the heat carrier, a temperature sensor measurement unit for measuring temperature, a flow sensor measurement unit for measuring flow rate, and a pressure sensor measurement unit for measuring pressure. The temperature control system structure topology is constructed in the form of a closed loop. The heat source unit and the heat exchanger unit are connected through the pipeline unit. The heat carrier circulates along the pipeline unit under the driving force provided by the circulating pump control unit, and the flow rate of the heat carrier is regulated by the fluid regulating valve control unit. The heat carrier flows through the controlled heat cavity that bears the process heat load and the heat exchanger unit that performs heat exchange in sequence, and then returns to the liquid storage unit for storing the heat carrier, thus forming a complete heat flow circulation path. Each physical unit in the temperature control loop is mapped to a low-order thermal fluid dynamic model component. The heat exchanger unit, the controlled thermal cavity and the pipeline unit are divided into several thermal capacity nodes and thermal resistance nodes. The node-level thermal fluid state space model is established by the heat conduction behavior, heat exchange behavior and heat convection behavior between the thermal capacity nodes and thermal resistance nodes, forming a thermal energy state vector. The thermal state vector includes at least the temperature state variables of the controlled thermal cavity, the liquid supply pipe, the liquid return pipe, and the liquid storage unit. The control input variables include heating power adjustment parameters, cooling power adjustment parameters, circulating pump speed control parameters, and fluid regulating valve opening adjustment parameters. Each control input variable is mapped to the thermal state vector through a dynamic function relationship, so that the dynamic model of the temperature control system can reflect the influence of changes in control input variables on the evolution process of the thermal state vector. A health parameter vector for the temperature control system is constructed. The health parameter vector for the temperature control system consists of several scalar parameters with physical interpretability, which are used to characterize the performance degradation of key functional units in the temperature control system. The health parameter vector for the temperature control system includes circulating pump efficiency parameters, heat exchanger heat transfer capacity parameters, fluid regulating valve opening offset parameters, and temperature sensor zero point offset parameters. The health parameter vector of the temperature control system is coupled with the thermodynamic behavior parameters of the temperature control system. The coupling methods between the health parameter vector of the temperature control system and the thermodynamic model of the temperature control system include: defining the heat transfer coefficient of the heat exchanger unit as the product of the reference heat transfer coefficient and the heat transfer capacity parameter of the heat exchanger; and defining the actual output flow of the circulating pump control unit as the product of the standard output flow and the efficiency parameter of the circulating pump. This is used to reflect the direct impact of the degradation of key performance on the thermal response behavior of the temperature control system in the thermodynamic model of the temperature control system. S100.2, Configured with an online gray-box digital twin model; By combining the dynamic model of the thermal fluid of the temperature control system based on the structural topology diagram of the temperature control system with the health parameter vector of the temperature control system, a gray box digital twin model of the temperature control system is constructed. The gray box digital twin model of the temperature control system expresses the dynamic process of the thermal energy state vector of the temperature control system changing with time in the form of state space. The state space form of the gray box digital twin model of the temperature control system includes state update equations and output observation equations. The state update equations are used to describe the evolution process of the current thermal energy state vector under the combined action of control input variables and temperature control system health parameter vectors. The output observation equations are used to describe the mapping relationship between the temperature data, flow data and pressure data collected by the temperature sensor measurement unit, flow sensor measurement unit and pressure sensor measurement unit and the internal thermal energy state vector. Each state variable in the state-space form of the gray box digital twin model of the temperature control system corresponds to a specific physical meaning. The observed variables in the output observation equation include at least the temperature of the controlled hot chamber, the supply liquid temperature, the return liquid temperature, the flow sensor measurement value, and the pressure sensor measurement value. Process noise terms and measurement noise terms are introduced into the gray box digital twin model structure of the temperature control system to simulate external disturbances and sensor measurement errors during the operation of the temperature control system. The health parameter vector of the temperature control system is set according to the factory calibration value of the new equipment in the initial state, and is updated by an online identification mechanism during subsequent operation. The online identification mechanism adjusts the health parameter vector of the temperature control system according to the deviation between the sensor measurement data and the output of the gray box digital twin model of the temperature control system.

3. The fault prediction method for a semiconductor equipment temperature control system based on digital twin as described in claim 1, characterized in that, S200 specifically includes: S200.1 Obtain temperature setting parameters and process stage status parameters, and divide the temperature control response segments; During the execution of process recipes in semiconductor manufacturing equipment, temperature control-related setting parameters and operating status parameters are obtained from the equipment control system. The setting parameters include temperature setpoint parameters, cavity pressure parameters, and radio frequency power parameters. The operating status parameters include wafer action identifier parameters and process stage number parameters. The temperature setpoint parameter is used to indicate the target temperature value of each heating unit and each cooling unit of the semiconductor manufacturing equipment. The cavity pressure parameter is used to indicate the real-time pressure level in the reaction cavity. The radio frequency power parameter is used to indicate the power level in the plasma process. The wafer action identifier parameter is used to mark whether the silicon wafer has been fed into the cavity. The process stage number parameter is used to identify the current formulation step. Based on the changing characteristics of temperature setpoint parameters, cavity pressure parameters, RF power parameters, wafer action identifier parameters, and process stage number parameters, multiple event triggering rules are defined to mark the boundary moments of temperature control system response changes. The event triggering rules include: when the absolute difference of the temperature setpoint parameter between two adjacent sampling moments is greater than or equal to 0.3℃, it is determined as a temperature setpoint step event; when the wafer action identifier parameter changes from 0 to 1, it is determined as a wafer event; and when the process stage number parameter changes, it is determined as a recipe stage switching event. The event time determined by the event triggering rules is used as the boundary condition for dividing the temperature control response segment. Any two adjacent event triggering times constitute the time interval of a temperature control response segment. The measurement data collected by the temperature control system in each time interval is used to construct an independent temperature control response segment data set. Each temperature control response segment contains the data of temperature measurement sequence, flow measurement sequence and pressure measurement sequence in that time interval, forming multiple temperature control response segments with clear boundaries and independent response behavior. S200.2 Configuring response feature vectors and labeling events based on temperature control response segments; For each segment of the temperature control response, dynamic response characteristic indicators closely related to the temperature control response performance are extracted. These dynamic response characteristic indicators include rise time parameter, overshoot parameter, settling time parameter, initial slope parameter, steady-state deviation parameter, and flow response deviation parameter. The rise time parameter, overshoot parameter, settling time parameter, initial slope parameter, steady-state deviation parameter, and flow response deviation parameter are constructed into a response feature vector in a fixed order; Each response feature vector is associated with the process event type corresponding to the temperature control response segment to which the response feature vector belongs, and an event tag is configured for the temperature control response segment. The event tags include a set value step response tag, a wafer thermal shock response tag, and a power on response tag.

4. The fault prediction method for a semiconductor equipment temperature control system based on digital twin according to claim 1, characterized in that, The S300 specifically includes: S300.1, Configure perturbation excitation conditions and temperature setpoint perturbation signals; A temperature control calibration window is configured for each process formula. The temperature control calibration window is selected during the time period when the thermal stability requirements are low in the process. The time period includes the standby phase between two adjacent batches and the time interval of 5 to 10 seconds before the preheating phase before the cold start of the equipment. Within the temperature control calibration window, a small-amplitude temperature setpoint perturbation signal is superimposed on the temperature setpoint signal. The temperature control system is subjected to a perturbation provided that the temperature setpoint perturbation signal does not exceed the process temperature tolerance range. S300.2 Configure online identification of error items for health parameters and configure the health parameter update algorithm; During the perturbation excitation calibration window, a sequence of temperature measurements is acquired from the temperature sensor measurement unit. Obtain the flow measurement value sequence from the flow sensor measurement unit. , Number the discrete sampling time points, and link the current control input variable with the current health parameter vector. Input the gray box digital twin model of the temperature control system to obtain the temperature prediction value sequence. With the sequence of predicted flow values ; Based on the deviations between measured and predicted temperature values, and between measured and predicted flow rates, an error term is constructed for online identification of health parameters. The error term is defined as follows: The temperature error term and the flow rate error term are combined column-wise to form the overall error vector. The overall error vector is defined as: In the gray box digital twin model of the temperature control system constructed based on state space form, the health parameter vector is updated online using the recursive least squares algorithm or the extended Kalman filter algorithm. To ensure the physical interpretability of the health parameter vector during online updates, reasonable constraint ranges are set for each component in the health parameter vector. After each perturbation excitation calibration window ends and the online identification and update of the health parameter vector is completed, the updated health parameter vector is associated with the response feature vector of the temperature control response segment corresponding to the process event. The health parameter vector and the response feature vector are then correlated in time to construct a joint evolution time series data structure for subsequent analysis.

5. The fault prediction method for a semiconductor equipment temperature control system based on digital twin according to claim 1, characterized in that, The S400 specifically includes: S400.1 Configure residuals and perform multi-time-scale decomposition operations, and configure the fault mode fingerprint database; During the operation of the temperature control system, the current control input variables and the current health parameter vector are used as inputs. By calling the updated gray box digital twin model of the temperature control system, the current sampling time is generated. The predicted output vector, denoted as The predicted output vector includes predicted temperature, predicted flow rate, and predicted pressure. The temperature sensor measurement unit, flow sensor measurement unit, and pressure sensor measurement unit acquire the data at the sampling time. The temperature measurement value, flow measurement value, and pressure measurement value are collected, and the three types of measurement values ​​are combined to form the measured output vector. Sampling time is constructed based on the measured output vector and the predicted output vector. residual vector The residual vector is defined as: The residual vector time series obtained by arranging in chronological order Decoupling is performed using a multi-time-scale decomposition method, which includes at least one of wavelet decomposition algorithm and multi-rate filter structure. The residual vector is decomposed into fast perturbation component and slow drift component by the multi-time-scale decomposition method. During the equipment trial operation phase, historical maintenance record phase, and manually set accelerated aging test phase, data on different types of performance degradation modes that may occur in the temperature control system of semiconductor manufacturing equipment are collected and analyzed. Data on the changes of health parameter vectors and response feature vectors over time corresponding to each performance degradation mode are obtained, and combined with the slow drift residual vector obtained by multi-timescale decomposition. Construct a failure mode sample set; The failure mode sample set includes typical temperature control system performance degradation modes such as heat exchanger scaling mode, circulating pump performance degradation mode, temperature sensor zero drift mode, and fluid control valve sticking or opening offset mode. The range of health parameter vector changes, slow drift residual vector features, and response feature vector features corresponding to the performance degradation modes of each typical temperature control system are summarized by combining structured rules and a lightweight pattern recognition model to construct a fault mode fingerprint database. This database is established when the following conditions are met: And the duration of the deviation is greater than In this case, the current operating status of the temperature control system will be identified as the heat exchanger fouling development stage mode in the fault mode fingerprint database. S400.2, Configure online fault mode identification and fault trend extrapolation; During the continuous operation of semiconductor manufacturing equipment, based on the time-series recorded health parameter vector sequences Slow drift residual vector sequence With response feature vector sequence A fixed-length sliding time window is used to select the health parameter vector, slow drift residual vector and response feature vector within a continuous sampling time period, and the data within the sliding time window are combined to form a joint analysis sample. For each health parameter component in the joint analysis sample Calculate the rate of change of time within the sliding time window, denoted as . The rate of change over time is used to characterize the components of health parameters. The degradation trend is analyzed by calculating the mean and trend slope of each component of the slow drift residual vector within the sliding time window, which is used to quantify the cumulative deviation and direction of change of the slow drift residual within the current time window. The current operating status characteristics obtained based on the time change rate of health parameters and the slow drift residual statistics are compared with the fault mode characteristics in the fault mode fingerprint database to identify that the current operating status of the temperature control system belongs to one of the following states: normal operation state, early performance degradation state, fault critical state, and recommended immediate shutdown state.

6. The fault prediction method for a semiconductor equipment temperature control system based on digital twin according to claim 1, characterized in that, The S500 specifically includes: S500.1: Obtain production task schedule, predict process heat load, configure health index of key functional units, and set multi-level thresholds. Obtain production task scheduling information for a preset future time period from the manufacturing execution system or scheduling system of semiconductor manufacturing equipment. The production task scheduling information includes the process recipe sequence to be executed, the frequency of use of process chambers, batch processing interval time and chambers, and parallel operation occupancy rate. Based on production task scheduling information and combined with the statistical mapping relationship between historical process formula types and heat load levels, a heat load prediction model for temperature control system load estimation is constructed. The heat load prediction model outputs a predicted heat load time series corresponding to future time intervals, denoted as […]. , Indicates a point in time The predicted unit heat load power is obtained. The heat load prediction model predicts the heat load based on the mapping relationship between process formulation type and average heat load power range. The prediction results satisfy... ; For the circulating pump control unit, heat exchanger unit, temperature sensor measurement unit, and fluid regulating valve control unit in the temperature control system, define corresponding health parameter components and health indexes respectively. Used for comprehensive evaluation of the The degree of performance deviation of the current operating state of each key functional unit relative to the reference calibration state; Multiple health thresholds are set for the health index to support different levels of fault warning and maintenance decisions. The health thresholds include: health warning threshold, planned maintenance threshold, and forced shutdown threshold. S500.2: Set up fault time window simulation and maintenance time point estimation, output fault prediction results and link with the maintenance decision system; At the current time At any given time, calculate the health index for each key functional unit. And based on the changes in the health index over a recent period, the rate of change of the health index is estimated; under the assumption that the rate of change of the health index is proportional to the average heat load power, the first... The rate of change of the health index of each key functional unit is approximated as follows: Based on the rate of change of the health index and the health threshold, calculate the time interval required for the current health index to decay to different thresholds; Combining the available maintenance windows of semiconductor manufacturing equipment, Matching with equipment idle time periods or process batch intervals yields the first... Maintenance time intervals for key functional units, including when the corresponding health index approaches the health warning threshold. The earliest recommended maintenance time corresponds to the health index reaching the mandatory shutdown threshold. The previous latest permitted maintenance time; Based on the configured fault mode fingerprint database, the fault mode type of the current critical functional unit is queried, and the queried fault modes are associated with the health index and maintenance time interval of the critical functional unit to generate a list of recommended maintenance items for the critical functional unit. When the obtained fault time window data, the time series of health index of key functional units, and the fault mode identifier obtained by matching the fault mode fingerprint database are transmitted to the equipment management platform and the predictive maintenance platform through the data interface protocol, the structured data fields include equipment identifier, key functional unit number, current value of health index, predicted trend of health index change, suggested maintenance time range, fault mode name, and fault mode confidence score.

Images