Semiconductor processing equipment and control method thereof

Abstract

The present invention discloses a semiconductor processing equipment and a control method thereof. The semiconductor processing equipment comprises a power supply, an impedance matcher and a semiconductor processing module. At least one current / voltage sensor is arranged between the power supply and the impedance matcher, and the real-time sampling signal of the current / voltage sensor is synchronized with the power signal output by the power supply in real time; at least one working state sensing and collecting device is arranged between the impedance matcher and the semiconductor processing module, and the working state sensing and collecting device collects the state quantity of the semiconductor processing module in real time; and also comprises a control system, which determines whether the impedance is matched based on the real-time sampling signal and the state quantity of the current / voltage sensor, performs root cause analysis when the impedance is not matched, and selectively regulates the working state of the impedance matcher and / or the semiconductor processing module based on the result of the root cause analysis, so as to eliminate the abnormal working state of the semiconductor processing module and / or make the impedance matched.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor equipment, and in particular to semiconductor processing equipment and a control method thereof. Background Art

[0002] With the rapid development of the advanced logic and storage device market, the iteration speed and technical performance requirements of semiconductor devices are increasing, and more stringent requirements are being placed on the processing technology. In order to achieve more precise critical dimension control and process higher aspect ratios, high-speed pulses and ultra-high voltage radio frequency or square wave signals are widely used in the semiconductor etching process for process implementation. These process conditions cause more drastic changes in the processing environment inside the cavity and more stringent reflection control requirements for the power circuit, which also puts higher standards on the control capabilities of the etching equipment itself.

[0003] The functional modules of the current mainstream etching equipment are independent of each other in terms of control logic, which makes it difficult for the equipment to make the most accurate strategy for the abnormal situations that need to be handled when controlling the etching process. For example, when a high reflection appears in the power circuit, the existing machine will think that it is caused by impedance mismatch, and will reduce the reflection energy by adjusting the overall impedance matching value. However, the cause of the abnormal reflection value may also be a fault in the power circuit, or even the plasma may be turned off due to poor process conditions during the process. Although the current control strategy of the equipment will shut down the power for protection after the abnormal time reaches a certain set threshold. However, such a protection strategy cannot start from the root cause of the abnormal situation and deal with the problem in a timely manner at the early stage of the problem, which leads to instability of the process and damage to the internal components of the equipment.

[0004] Therefore, a new approach is needed with more intelligent control capabilities that can accurately and efficiently handle abnormal situations in the etching process, while improving process effects and protecting the equipment so that it works under healthier conditions. Summary of the invention

[0005] To solve the above problems, the present invention proposes a method for determining the working state of semiconductor equipment and intelligently controlling the working state of semiconductor equipment based on the power signal matching situation, combined with global perception data and machine learning algorithms. In the case where the power signal is mismatched due to matching parameter factors, the intelligent matcher is controlled to quickly and accurately complete the impedance matching adjustment in a manner that best suits the process settings. In the case of mismatch due to equipment abnormalities, the functional module corresponding to the control device is adjusted or processed urgently, thereby ensuring stable operation of the equipment and reducing damage to the equipment caused by abnormalities.

[0006] Specifically, according to one aspect of the present invention, a semiconductor processing equipment is proposed, comprising a power supply, an impedance matcher and a semiconductor processing module, wherein at least one current / voltage sensor is provided between the power supply and the impedance matcher, and the real-time sampling signal of the current / voltage sensor is synchronized with the power signal output by the power supply in real time; at least one working state sensing and collecting device is provided between the impedance matcher and the semiconductor processing module, and the working state sensing and collecting device collects the state quantity of the semiconductor processing module in real time; and further comprising a control system, wherein the control system determines whether the impedance is matched based on the real-time sampling signal of the current / voltage sensor and the state quantity, and performs root cause analysis when the impedance is not matched, and selectively regulates the working state of the impedance matcher and / or the semiconductor processing module based on the result of the root cause analysis to eliminate the abnormal working state of the semiconductor processing module and / or make the impedance matched.

[0007] Optionally, the power supply is a radio frequency power supply or a high-voltage pulse power supply.

[0008] Optionally, the semiconductor processing module includes a reaction chamber and / or an equipment base, and the reaction chamber is a load terminal of the power signal.

[0009] Optionally, the sampling signal of the current / voltage sensor and the state quantity collected in real time by the working state sensing and collecting device are updated with the impedance matcher and the control system through a high-speed communication protocol at a frequency of at least 500 Hz.

[0010] Optionally, the root cause analysis is implemented to determine the cause of the impedance mismatch, and the cause includes abnormal working state of the semiconductor processing module and matching mode error of the impedance matcher.

[0011] Optionally, the power signal includes amplitude, phase, forward / backward propagation characteristics, and harmonics.

[0012] Optionally, the state quantity includes at least one of the electrical, temperature, air pressure, and stress states of the semiconductor processing module.

[0013] Optionally, a data training system is also included to provide the storage space and computing power required for encoding and inferring training data required for machine learning.

[0014] Optionally, the data training system is electrically connected to the control system via a high-speed communication bus to achieve training data transmission and inference model update between the control system and the data training system.

[0015] Optionally, the control system includes a machine learning processing unit for inferring the working state of the semiconductor processing equipment through a machine learning algorithm model.

[0016] Optionally, the algorithm model includes a feedforward neural network inference model, a feedback neural network or a graph network.

[0017] Optionally, the algorithm model includes an input layer, a network layer and an output layer, wherein the input layer includes one or more intermediate layers or hidden layers, each intermediate layer contains multiple nodes, each node is interconnected with each existing node in the previous layer or the next layer, each node has different calculation weights, and constitutes a complete weighted neural network.

[0018] Optionally, the weight of the node is obtained by training the data training system and is dynamically updated according to the power signal and the state quantity.

[0019] Optionally, the input layer is composed of the power signal and state quantity obtained by sampling, and the output layer is a data set composed of equipment working status result labels obtained by neural network inference.

[0020] Optionally, the equipment working status result label corresponds to execution measures for eliminating abnormal working status of the semiconductor processing module and / or achieving impedance matching.

[0021] Optionally, the power supply and the current / voltage sensor package constitute a power module, and the working status sensing and collecting device is arranged on the semiconductor processing module to constitute a load end.

[0022] Optionally, the impedance matcher includes an impedance matching network and a communication module, and the communication module included in at least one of the impedance matchers is integrated with an industrial Ethernet control master station, which is used to build a self-determining communication closed loop based on the impedance matching network, power module, load end and control system.

[0023] Optionally, the control system / impedance matcher and data training system integrated installation topology is one-to-one or many-to-one.

[0024] According to another aspect of the present invention, a semiconductor processing equipment control method is provided, comprising:

[0025] S10: obtaining a power signal of a power source for plasma excitation, and obtaining a state quantity of a working state of a semiconductor processing module,

[0026] S20: performing impedance matching judgment based on the power signal and the state quantity, and performing root cause analysis when the impedance does not match,

[0027] S30 Selectively regulating the working state of the impedance matcher and / or the semiconductor processing module based on the result of the root cause analysis to eliminate the abnormal working state of the semiconductor processing module and / or achieve impedance matching.

[0028] Optionally, the power signal includes at least one of amplitude, phase, forward / backward propagation characteristics, and harmonics.

[0029] Optionally, the state quantity includes at least one of the electrical, temperature, air pressure, and stress states of the semiconductor processing module.

[0030] Optionally, the root cause analysis is implemented as follows: determining the cause of the impedance mismatch, wherein the cause includes an abnormal working state of the semiconductor processing module and an error in a matching mode of the impedance matcher.

[0031] Optionally, the root cause analysis includes:

[0032] S21 collects the power signal and the state quantity into a data set, and converts the current data set and the historical data set into two dimensions to form a state original signal that conforms to the input layer of the neural network model;

[0033] S22 extracts the features of the state parameters after the original state signal is processed by the hidden layer convolution of the network layer.

[0034] S23 When the original state signal passes through the last hidden layer of the network layer, the current working state label is output to the output layer.

[0035] S24 compares the current working state label with the device working state result label in the output layer to determine the root cause of the device impedance mismatch.

[0036] Optionally, after the root cause analysis is completed, the historical data set is updated.

[0037] Optionally, the abnormal working state of the semiconductor processing module includes plasma leakage, ignition inside the reaction chamber, and plasma extinction.

[0038] Optionally, when the working state of the semiconductor processing module is abnormal and sparks are about to occur in the reaction chamber, the power supply is controlled to reduce or shut down the power output, and the concentration of the reaction gas in the reaction chamber is adjusted to avoid the occurrence of sparks.

[0039] Optionally, when the semiconductor processing module is in an abnormal working state and the plasma in the reaction chamber is extinguished, the control power supply increases the output power of the high-frequency RF signal, adjusts the impedance matcher to the impedance matching point, and re-ignites the plasma inside the chamber.

[0040] Optionally, when the root cause analysis determines that the impedance parameters are incorrect, the impedance matching network is adjusted through matching mode algorithm processing, and the impedance adjustment state is monitored in real time. When the impedance is adjusted to the set target, one monitoring control is ended and preparations are made for the next monitoring control.

[0041] Unlike traditional semiconductor processing equipment, where each functional module is independent of each other in terms of control logic, the present invention proposes to judge and control the working state of the equipment based on the power signal matching situation and combined with global perception data. In addition, the working state of the equipment is further judged and controlled in combination with a machine learning algorithm. When the power signal is mismatched due to matching parameter factors, the matcher is controlled to quickly and accurately complete the impedance matching adjustment in a manner that best suits the process settings. In the event of mismatch due to abnormal equipment conditions, the functional module corresponding to the control equipment is adjusted or processed urgently, thereby ensuring stable operation of the equipment and reducing damage to the equipment caused by abnormal conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A schematic diagram of the hardware structure of a semiconductor processing device according to the present invention;

[0043] Figure 2 A schematic diagram of a module structure of an impedance matcher of the present invention;

[0044] Figure 3 It is a schematic diagram of the functional modules of a control system of the present invention;

[0045] Figure 4 It is a schematic diagram of a feedforward neural network inference model carried by a machine learning processing unit of the present invention;

[0046] Figure 5 It is a schematic diagram of the execution of a control algorithm of a semiconductor processing equipment of the present invention;

[0047] Figure 6 It is a topological diagram of the integrated installation of a control system / impedance matcher and a data training system of the present invention. DETAILED DESCRIPTION

[0048] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Although the existing control schemes for semiconductor processing equipment can meet the requirements of automatic operation and monitoring under many complex process conditions, due to the independence of functional modules in control logic, it is inevitable that some abnormal working conditions will be handled with lags. In this case, the processing may fail at best, or even damage the equipment's own parts, affecting process stability and equipment life.

[0050] To solve the above problems, according to one aspect of the present invention, a semiconductor processing equipment is provided, comprising a power supply 101, an impedance matcher 102 and a semiconductor processing module 103 (Process Module), at least one current / voltage sensor 104 is provided between the power supply 101 and the impedance matcher 102, and the real-time sampling signal of the current / voltage sensor 104 is synchronized with the power signal output by the power supply 101 in real time; at least one working state sensing and collecting device 105 is provided between the impedance matcher 102 and the semiconductor processing module 103, and the working state sensing and collecting device 105 collects the state quantity of the semiconductor processing module 103 in real time; and also includes a control system 202, the control system 202 determines whether the impedance is matched based on the real-time sampling signal and the state quantity of the current / voltage sensor 104, and performs root cause analysis when the impedance is not matched, and selectively adjusts the working state of the impedance matcher 102 and / or the semiconductor processing module 103 based on the result of the root cause analysis, so as to eliminate the abnormal working state of the semiconductor processing module 103 and / or make the impedance match.

[0051] Different from the traditional semiconductor processing equipment where each functional module is independent of each other in control logic, the present invention proposes to judge and control the working state of the equipment based on the power signal matching situation and combined with the global perception data. Therefore, when the power signal is mismatched due to matching parameter factors, the matcher can be controlled to quickly and accurately complete the impedance matching adjustment in a way that is most suitable for the process setting. In the case of mismatch due to abnormal conditions of the equipment, the functional module corresponding to the control equipment is adjusted or processed urgently, thereby ensuring the stable operation of the equipment and reducing the damage caused to the equipment by abnormal conditions.

[0052] Specifically, Figure 1It is a schematic diagram of the hardware structure of the semiconductor processing equipment control device, wherein the power supply 101 is a radio frequency or high-voltage pulse power supply, covering all the working frequency bands required by the current semiconductor process, and supporting a variety of energy output modes such as continuous wave and pulse waveform. The impedance matcher 102 includes an impedance matching network circuit suitable for radio frequency signals, a matching modulation control system, and related sensors. The semiconductor processing module 103 (Process Module) includes key parts such as a reaction chamber and a device platform, wherein the reaction chamber is the load terminal of the power signal. There needs to be at least one set of high-sampling rate current / voltage sensors 104 between the impedance matcher 102 and the power supply 101. The sampling signal of the sensor is synchronized with the output signal of the radio frequency or high-voltage pulse power supply in real time, and then collects and provides power signal information including but not limited to amplitude, phase, forward / backward propagation characteristics, harmonic composition, etc. The signal is updated with the impedance matcher 102 through a high-speed communication protocol (including but not limited to industrial Ethernet control automation technology (EtherCAT), gigabit transceiver (GTH), etc.) at a frequency of at least 500 Hz. At least one working state sensing and collecting device 105 is required between the impedance matching device 102 and the semiconductor processing module 103. The device can collect real-time state quantities including but not limited to electrical, temperature, and air pressure in the semiconductor processing module 103, and also transmit and update data with the impedance matching device 102 at a frequency of at least 500 Hz through a high-speed communication protocol. "High sampling rate" generally refers to a sampling rate of 400 kSa / s or more.

[0053] The working state sensing and collecting device 105 is used to collect working state parameters in the semiconductor processing module 103 in real time. The working state parameters include but are not limited to electrical information, cavity temperature, cavity pressure, and stress state quantity.

[0054] When the working state sensing and collecting device 105 is arranged at the lower electrode, the electrical information of the lower electrode, such as the current, voltage and its changes, can be collected in real time, and the temperature of the lower electrode and the stress state of the lower electrode can also be collected. When the working state sensing and collecting device 105 is arranged at the shower head, the electrical information of the shower head, such as the current, voltage and its changes, can be collected in real time, and the temperature, air pressure and stress state of the shower head can also be collected. When the working state sensing and collecting device 105 is located in the gaseous space of the reaction chamber, the discharge state, the temperature in the chamber and the air pressure in the chamber can be collected in real time in the semiconductor processing module 103.

[0055] Therefore, the working status sensing and collecting device 105 is a kit composed of one or more types of sensors such as current and voltage sensors, temperature sensors, pressure sensors, stress sensors, optical sensors (such as optical fiber sensors), chemical sensors (such as free radical sensors), etc., and one or more sets of them can be set in different areas of the semiconductor processing module 103 to collect various parameters in the semiconductor processing module 103 in real time.

[0056] Preferably, the working status sensing and collecting device 105 is provided with an EtherCAT or Gigabit transceiver port, and can also perform data transmission and update with the impedance matcher 102 at a frequency of at least 500 Hz through a high-speed communication protocol.

[0057] By real-time acquisition of various parameters in the semiconductor processing module 103 through the working status sensing and acquisition device 105 and real-time feedback to the control system 202, the control system 202 can not only adjust the impedance based on the current / voltage sensor 104 signal, but also make a comprehensive judgment and adjust the impedance according to various parameters in the semiconductor processing module 103. Compared with the prior art that only uses the RF end signal as the basis for impedance adjustment, more complete impedance matching judgment and regulation can be performed, thereby further shortening the regulation and matching process and reducing the impedance matching delay.

[0058] Typically, the power supply 101 and the current / voltage sensor 104 are in the same closed frame power module 106 in the physical space. When the current / voltage sensor 104 meets the requirements of the invention, it can also be placed independently or in the same closed frame with the impedance matcher 102. The semiconductor processing module 103 and the working state sensing and collecting device 105 are in the same closed frame load end 107 in the physical space, and based on the characteristics of the collected physical quantity, the working state sensing and collecting device 105 can be placed in the load end 107 in a centralized or distributed manner. The hardware system also needs to include a data training system 108 to provide the large-capacity storage space and large-scale computing power required for encoding and inferring training data required for machine learning.

[0059] Figure 2 It is a schematic diagram of the module composition of an impedance matcher 102, wherein the impedance matching network 201 is an impedance matching network composed of electronic components through appropriate settings, and the matching network includes at least one element that can be adjusted by external physical quantities (mechanical, electrical, etc.) to complete impedance matching adjustment.

[0060] In some embodiments, the impedance matching network 201 includes at least one variable capacitor, and the variable capacitor is connected to a motor. The variable capacitor is adjusted by the motor to adjust the impedance.

[0061] The impedance matching parameters and the automatic control of the whole equipment are realized by the control system 202, and the control system 202 includes at least one processing unit that can collect real-time information of the process and optimize the impedance matching through the corresponding algorithm. Usually, the impedance matching network 201 and the control system 202 are integrated and packaged in a metal shell to facilitate installation and maintenance and reduce the interference of the external environment on the matching system during the process.

[0062] In some cases, the control system 202 can also be independently packaged and exist in a separate constrained space in the physical space with the impedance matching network 201. A power transmission link 203 is reserved in the impedance matcher 102 to realize the transmission of power signals between the power module 106 and the load end 107. A communication bus 204 is reserved for high-speed transmission of network communication signals, sensor sensing signals and actuator action signals. An Ethernet communication bus 205 is reserved to access the control system 202 to realize the detection and control of key information of the working status of the matcher by the master control unit. A high-speed communication bus 206 is reserved to connect the control system 202 with the data training system 108 to complete the transmission of a large amount of training data between the control system and the data training system and to regularly update the inference model.

[0063] Figure 3301 is a diagram of the functional modules of the control system. The processor 301 is the core processor of the control system, which can be ARM, FPGA, etc., and is used to realize data processing and logic algorithm deduction and execution. The storage module 302 is a system storage module, including volatile and non-volatile memories, providing the cache resources required for sampling and calculation process execution during the process, and recording the operation data, historical logs and other information during the working process of the matcher. The communication module 303 is a system communication module, which enables the control system to support but not limited to interconnection with other subsystems of the equipment through protocols such as Ethernet, Industrial Ethernet Control Automation Technology (EtherCAT), Gigabit Transceiver (GTH), etc., and sends control instructions, sampling results, execution information, etc. to the corresponding subsystem or module according to the characteristics such as the amount of information data to be transmitted and the response time. The sensor input module 304 is a system sensor input module, which includes analog-to-digital conversion, digital signal processing and other related functions, and can sample the sensor sensing information directly related to the control system 202 and transmit it to the processor 301 for algorithm processing. The execution output module 305 is a system execution output module, which includes related functions such as digital-to-analog conversion and digital signal generation, and can pass the system operation instructions obtained by running the matching algorithm in the processor 301 to the actuator directly related to the control system 202. The machine learning processing unit 306 is a processing system that can infer the working state of the device through the machine learning algorithm model. The system is isolated from the processor 301 in logical space or physical space, and calls the limited computing resources in the storage module 302 through related algorithms. In some examples, the resource allocation of the storage module 302 will be automatically optimized and allocated according to the computational complexity of the processor 301 and the internal model scale of the machine learning processing unit 306.

[0064] In particular, in some embodiments, the communication module 303 integrates an industrial Ethernet control master station (master) for forming a self-decision-making communication closed loop between the impedance matcher 102, the power module 106, the load end 107 and the control system 202, without the need to communicate with the host computer in real time and wait for the control instructions of the host computer. Thus, by combining the high-speed transmission of the Ethernet control automation technology port and the parallel calculation of the FPGA, various sensor data can be transmitted to the FPGA almost without delay, and the parallel calculation of the FPGA rather than the serial calculation can simultaneously process multiple data information without calculating the queuing delay, thereby significantly reducing the mismatch and oscillation problems caused by the current impedance control time delay dislocation, and facilitating the timely response of distributed control.

[0065] Figure 4It is a schematic diagram of a feedforward neural network inference model carried by the machine learning processing unit 306. The model mainly includes an input layer 402, a network layer 401, and an output layer 403. The network layer 401 includes one or more intermediate layers or hidden layers. Each intermediate layer includes multiple nodes 405, each node 405 is interconnected with each node 405 in the previous layer or the next layer (if any), each node 405 has a different calculation weight, and constitutes a complete weighted neural network. The weight of the node 405 is obtained by training the data training system 108, and is dynamically updated according to external input and the data collected by sampling in the actual process, including but not limited to current, voltage, temperature, pressure, and stress.

[0066] The input layer 402 is composed of the device working state sampling data 404 transmitted through the communication module 303 and the sensor input module 304 and saved in the storage module 302, and the data refresh frequency is above 500 Hz. The output layer 403 is a set of device working state result labels 406 inferred by the neural network, and the corresponding execution measures in the device working state result labels 406 are transmitted to the corresponding calculation and execution units. In other examples, the machine learning processing unit 306 can carry a feedback neural network or a graph network, or other types of machine learning algorithm models.

[0067] According to another aspect of the present invention, a semiconductor processing equipment control method is provided, comprising:

[0068] S10: obtaining a power signal of a power source for plasma excitation, and obtaining a state quantity of a working state of a semiconductor processing module,

[0069] S20 determines impedance matching based on power signals and state quantities, and performs root cause analysis when impedance mismatch occurs.

[0070] S30 selectively regulating the working state of the impedance matcher and / or the semiconductor processing module based on the result of the root cause analysis to eliminate the abnormal working state of the semiconductor processing module and / or achieve impedance matching.

[0071] Different from the traditional semiconductor processing equipment where each functional module is independent of each other in control logic, the present invention proposes to judge and control the working state of the equipment based on the power signal matching situation and combined with the global perception data. Therefore, when the power signal is mismatched due to matching parameter factors, the matcher can be controlled to quickly and accurately complete the impedance matching adjustment in a way that is most suitable for the process setting. In the case of mismatch due to abnormal conditions of the equipment, the functional module corresponding to the control equipment is adjusted or processed urgently, thereby ensuring the stable operation of the equipment and reducing the damage caused to the equipment by abnormal conditions.

[0072] In some embodiments, the root cause analysis is implemented as follows: determining the cause of the impedance mismatch, where the cause includes abnormal working state of the semiconductor processing module and matching mode error of the impedance matcher.

[0073] In some embodiments, root cause analysis includes:

[0074] S21 collects the power signal and the state quantity into a data set, and converts the current data set and the historical data set into two dimensions to form a state original signal that conforms to the input layer of the neural network model;

[0075] The original signal of the S22 state is processed by the hidden layer convolution of the network layer to extract the features of the state parameters.

[0076] S23 When the original state signal passes through the last hidden layer of the network layer, the current working state label is output to the output layer.

[0077] S24 compares the current working status label with the device working status result label in the output layer to determine the root cause of the device impedance mismatch.

[0078] Combine global perception data with machine learning algorithms to control the working state of semiconductor equipment. With the flexibility and efficiency of the matcher, it can fully cover the timely matching and adjustment of impedance parameters under different process conditions. At the same time, based on a more comprehensive perception architecture and algorithm, the power signal mismatch problem that occurs during equipment operation is handled from the perspective of root cause analysis, so that equipment operating abnormalities can be discovered as early as possible, and the conditions that induce abnormalities can be adjusted in time, reducing the probability of failure from the source, improving production yield and extending equipment life. In addition, according to the scale of machine use, the algorithm training architecture can be flexibly configured and installed to achieve a balance between efficiency and cost.

[0079] Specifically, Figure 5 shown. Figure 5 It is a schematic diagram of the execution of a control algorithm of the present invention. When the algorithm is started, the electrical, temperature, and mechanical real-time state sampling data of the equipment are obtained from the storage module 302 as the input layer 402, and the sampling data is represented as xi. If a total of n types of sampling data are used, the data set obtained at one time is represented as (x1, x2, x3, ... xn). The acquired data is transmitted to the processor 301, and the processor 301 determines whether impedance mismatch behavior occurs 501. If the impedance matching state obtained at this time is consistent with the target setting, the monitoring control is ended once, and the next monitoring control is prepared. If the impedance matching state obtained at this time is inconsistent with the target setting, the impedance mismatch root cause inference is started through the network layer 401.

[0080] In an example of impedance mismatch root cause inference, the current operating state parameters S0 (x1, x2, x3, … xn) of the device represented by the data set and the historical operating state parameter set G{St (x1, x2, x3, … xn)} of the device, (t = 1, 2, … T) are two-dimensionalized to form the state original signal O (S0, S1, …, ST) that conforms to the input layer 402 of the neural network model. The input quantity is processed by the hidden layer convolution of the network layer 401 to extract the features of the state parameters, and the output result of the qth hidden layer is represented by Fq (S0, S1, …, ST). After the data passes through the last hidden layer of the network layer 401, the current working state label Y is output to the output layer 403, and the current working state label Y is compared with the device working state result label 406 in the output layer 403 to make a judgment on the root cause of the device impedance mismatch, and the result is transmitted to the processor 301 for subsequent processing.

[0081] After one inference is completed, the equipment historical operation status parameter set G{St(x1, x2, x3, … xn)} is updated, (t= 0, 1, … T - 1), and a new equipment operation status monitoring process is entered.

[0082] Example description:

[0083] During the process, the equipment experienced a ignition failure inside the cavity.

[0084] At this time, the control algorithm finds that the device is in an impedance mismatch state based on the current device operating state parameters, and performs root cause reasoning based on S0 and G{St(x1, x2, x3, … xn)} according to the example method. After the reasoning process of the network layer 401, the current working state label Y representing the ignition fault inside the cavity is output, and the inference result is cached in the storage module 302 for subsequent abnormal processing. At the same time, the historical operating state parameter set G is updated using S0, and the subsequent operating state monitoring continues.

[0085] If the impedance matching state is consistent with the target setting at this time, the monitoring control is ended and the next monitoring control is prepared. If the impedance matching state is inconsistent with the target setting at this time, the impedance mismatch root cause inference is started through the network layer 401, and the inference result output layer 403 is cached in the storage module 302. The processor 301 determines whether the impedance mismatch at this time is caused by the abnormal situation of the equipment through the output layer 403 through the cache status data in the storage module 302, which is recorded as step 502, wherein the abnormal situation includes but is not limited to plasma leakage, ignition inside the reaction chamber, plasma extinction, etc. If the result of step 502 is that an abnormality occurs, the processor 301 starts to handle the abnormality according to the corresponding algorithm strategy, which is recorded as step 506. After the processing is completed, the detection control is ended and the next monitoring control is prepared.

[0086] In an example of step 506 , if the state quantity indicates that arcing will occur in the reaction chamber, the control system 202 controls the power module 106 through the communication bus 204 to reduce or shut down the power output and adjust the concentration of the reaction gas at the load end 107 to avoid arcing.

[0087] In an example of step 506, if the state quantity indicates that the plasma in the reaction chamber is extinguished, the control system 202 controls the power module 106 through the communication bus 204 to increase the output power of the high-frequency RF signal, adjusts the impedance matcher 102 to the impedance matching point, and re-ignites the plasma inside the chamber.

[0088] If the result of step 502 is that the impedance parameter is incorrect, the processor 301 processes the matching mode algorithm (recorded as step 504), and outputs a control signal through the control system 202, so that the impedance matching network 201 adjusts the impedance in an automatic matching, fixed matching or intelligent matching manner (recorded as step 505), and monitors the impedance adjustment state in real time (recorded as step 503). When the impedance is adjusted to the set target, one monitoring control ends and prepares to enter the next monitoring control.

[0089] Figure 66(a) is a topological diagram of the integrated installation of the control system 202 / impedance matcher 102 and the data training system 108. 6(a) is a one-to-one topological form, where each set of data training systems 108 is connected to a corresponding set of control systems 202 / impedance matchers 102. The computing resources consumed are relatively small, and the inference network is mainly adapted to a single PM. 6(b) is a one-to-many topological form for a set of machines. For example, a set of machines includes 4 sets of PMs. The data training system 108 trains the model with the control data of all control systems 202 / impedance matchers 102 on the integrated machine. The computing resources consumed are large, and the inference network model obtained by training has better compatibility. 6(c) is a one-to-many topological form for multiple sets of machines. At this time, the data training system 108 formally serves as the data collection center of multiple sets of machines. It can obtain the control data of all control systems 202 / impedance matchers 102 in the topology and train the model on this basis. This situation consumes a lot of computing resources. The inference network model obtained by training has extremely high compatibility and has higher support for long-term upgrades and flexible configurations of machines.

[0090] The present invention controls the working state of semiconductor equipment based on the power signal matching situation, combined with global perception data and machine learning algorithms. With the flexibility and efficiency of the matcher, it can fully cover the timely matching and adjustment of impedance parameters under different process conditions. At the same time, based on a more comprehensive perception architecture and algorithm, the power signal mismatch problem that occurs during the operation of the equipment is handled from the perspective of root cause analysis, so that the equipment can be found to be abnormal as early as possible, and the conditions that induce the abnormality can be adjusted in time, thereby reducing the probability of failure from the source, improving production yield and extending the service life of the equipment. In addition, according to the scale of machine use, the algorithm training architecture can be flexibly configured and installed to achieve a balance between efficiency and cost.

[0091] It should be noted that, in the present invention, "semiconductor processing equipment" and its abbreviation "semiconductor equipment" and "equipment" have the same meaning; "impedance matcher" and its abbreviation "matcher" have the same meaning. "High-speed communication" refers to industrial Ethernet control automation technology (EtherCAT), gigabit transceiver (GTH) or other technologies with a communication rate not lower than these two, and data is updated at a frequency of at least 500 Hz.

[0092] Although the content of the present invention has been described in detail through the above preferred embodiments, it should be appreciated that the above description should not be considered as a limitation of the present invention. After reading the above content, it will be apparent to those skilled in the art that various modifications and substitutions of the present invention will occur. Therefore, the protection scope of the present invention should be limited by the appended claims.

Claims

1. A semiconductor processing equipment, characterized in that: include Power supplies, impedance matching devices and semiconductor processing modules, At least one current / voltage sensor is provided between the power supply and the impedance matcher. The real-time sampling signal of the current / voltage sensor is synchronized with the power signal output by the power supply in real time; At least one working state sensing and collecting device is provided between the impedance matcher and the semiconductor processing module. The working status sensing and collecting device collects the status quantity of the semiconductor processing module in real time; The system further includes a control system, wherein the control system determines whether the impedance is matched based on the real-time sampling signal of the current / voltage sensor and the state quantity, and performs a root cause analysis when the impedance is not matched, wherein the root cause analysis is implemented to determine the cause of the impedance mismatch, wherein the cause includes an abnormal working state of the semiconductor processing module and an error in the matching mode of the impedance matcher. Based on the result of the root cause analysis, the working state of the impedance matcher and / or the semiconductor processing module is selectively regulated to eliminate the abnormal working state of the semiconductor processing module and / or achieve impedance matching.

2. The semiconductor processing equipment according to claim 1, characterized in that The power supply is a radio frequency power supply or a high voltage pulse power supply.

3. The semiconductor processing equipment according to claim 1, characterized in that The semiconductor processing module includes a reaction chamber and / or an equipment base, and the reaction chamber is a load terminal of the power signal.

4. The semiconductor processing equipment according to claim 1, wherein: The sampling signal of the current / voltage sensor and the state quantity collected in real time by the working state sensing and collecting device are updated with the impedance matcher and the control system through a high-speed communication protocol at a frequency of at least 500 Hz.

5. The semiconductor processing equipment according to claim 1, characterized in that The power signal includes amplitude, phase, forward / backward propagation characteristics, and harmonics.

6. The semiconductor processing equipment according to claim 1, wherein: The state quantity includes at least one of the electrical, temperature, pressure, and stress states of the semiconductor processing module.

7. The semiconductor processing equipment according to any one of claims 1 to 6, characterized in that: It also includes a data training system that provides the storage space and computing power required to encode and infer training data required for machine learning.

8. The semiconductor processing equipment according to claim 7, characterized in that The data training system is electrically connected to the control system via a high-speed communication bus to achieve training data transmission and inference model update between the control system and the data training system.

9. The semiconductor processing equipment according to claim 7, characterized in that The control system includes a machine learning processing unit for inferring the working state of the semiconductor processing equipment through a machine learning algorithm model.

10. The semiconductor processing equipment according to claim 9, characterized in that The algorithm model includes a feedforward neural network inference model, a feedback neural network or a graph network.

11. The semiconductor processing equipment according to claim 9, characterized in that The algorithm model includes an input layer, a network layer and an output layer, wherein the input layer includes one or more intermediate layers or hidden layers, each intermediate layer contains multiple nodes, each node is interconnected with each existing node in the previous layer or the next layer, each node has different calculation weights, and constitutes a complete weighted neural network.

12. The semiconductor processing equipment according to claim 11, characterized in that The weight of the node is obtained by training the data training system and is dynamically updated according to the power signal and the state quantity.

13. The semiconductor processing equipment according to claim 11, characterized in that The input layer is composed of the sampled power signal and state quantity. The output layer is a data set consisting of equipment working status result labels inferred by the neural network.

14. The semiconductor processing equipment according to claim 13, characterized in that The equipment working status result label corresponds to execution measures for eliminating abnormal working status of the semiconductor processing module and / or achieving impedance matching.

15. The semiconductor processing equipment according to claim 1, wherein: The power supply and the current / voltage sensor package form a power module, and the working status sensing and collecting device is arranged on the semiconductor processing module to form a load end.

16. The semiconductor processing equipment according to claim 15, characterized in that The impedance matcher includes an impedance matching network and a communication module. The communication module included in at least one of the impedance matchers is integrated with an industrial Ethernet control master station, which is used to build a self-decision-making communication closed loop based on the impedance matching network, power module, load end and control system.

17. The semiconductor processing equipment according to claim 7, characterized in that The control system / impedance matcher and data training system integrated installation topology is one-to-one or many-to-one.

18. A semiconductor processing equipment control method, characterized in that: include S10: obtaining a power signal of a power source for plasma excitation, and obtaining a state quantity of a working state of a semiconductor processing module, S20: performing impedance matching judgment based on the power signal and the state quantity, and performing root cause analysis when the impedance does not match, wherein the root cause analysis is implemented to judge the cause of the impedance mismatch, wherein the cause includes an abnormal working state of the semiconductor processing module and an error in the matching mode of the impedance matcher. S30 Selectively regulating the working state of the impedance matcher and / or the semiconductor processing module based on the result of the root cause analysis to eliminate the abnormal working state of the semiconductor processing module and / or achieve impedance matching.

19. The semiconductor processing equipment control method according to claim 18, wherein: The power signal includes at least one of amplitude, phase, forward / backward propagation characteristics, and harmonics.

20. The semiconductor processing equipment control method according to claim 18, wherein: The state quantity includes at least one of the electrical, temperature, pressure, and stress states of the semiconductor processing module.

21. The semiconductor processing equipment control method according to claim 18, wherein: The root cause analysis includes: S21 collects the power signal and the state quantity into a data set, and converts the current data set and the historical data set into two dimensions to form a state original signal that conforms to the input layer of the neural network model; S22 extracts the features of the state parameters after the original state signal is processed by the hidden layer convolution of the network layer. S23 When the original state signal passes through the last hidden layer of the network layer, the current working state label is output to the output layer. S24 compares the current working state label with the device working state result label in the output layer to determine the root cause of the device impedance mismatch.

22. The semiconductor processing equipment control method according to claim 21, characterized in that: When the root cause analysis is completed, the historical data set is updated.

23. The semiconductor processing equipment control method according to claim 18, wherein: The abnormal working state of the semiconductor processing module includes plasma leakage, ignition inside the reaction chamber, and plasma extinction.

24. The semiconductor processing equipment control method according to claim 18, wherein: When the working state of the semiconductor processing module is abnormal and sparks are about to occur in the reaction chamber, the power supply is controlled to reduce or shut down the power output, and the concentration of the reaction gas in the reaction chamber is adjusted to avoid the occurrence of sparks.

25. The semiconductor processing equipment control method according to claim 18, wherein: When the semiconductor processing module is in an abnormal working state, such as the plasma in the reaction chamber is extinguished, the control power source increases the output power of the high-frequency radio frequency signal, adjusts the impedance matcher to the impedance matching point, and re-ignites the plasma in the chamber.

26. The semiconductor processing equipment control method according to claim 18, wherein: When the root cause analysis determines that the impedance parameters are incorrect, the impedance matching network is adjusted through the matching mode algorithm, and the impedance adjustment state is monitored in real time. When the impedance is adjusted to the set target, one monitoring control is ended and preparations are made for the next monitoring control.

Images