Method, as well as control device and non-volatile data storage, comprising code segments, for carrying out the same.
The method and control device address the complexity of reactive sputtering by employing a controller to manage multiple variables, achieving precise and stable control of coating processes across multiple chambers with reduced computational demands.
Patent Information
- Application Number
- DE102020124936
- Authority / Receiving Office
- DE · DE
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2020-09-24
- Publication Date
- 2024-11-28
- Estimated Expiration
- 2040-09-24
AI Technical Summary
Controlling reactive sputtering processes is challenging due to complex reaction dynamics, narrow stoichiometry requirements, and unstable reaction modes, which are exacerbated by interactions between multiple coating processes in a highly productive system, leading to difficulties in maintaining precise control and stability.
A method and control device that utilize a controller to manage multiple controlled variables simultaneously, mapping multidimensional vectors to control variables, and incorporate a model-based approach for automatic calibration, allowing for robust and efficient control of reactive coating processes across multiple chambers.
The solution provides precise control with reduced computational power, minimizes over/under-control, and enables scalable and automated calibration, ensuring stable operation across multiple coating processes.
Smart Images

Figure 00000029_0000 
Figure 00000029_0001 
Figure 00000055_0000
Abstract
Description
[0001] Various embodiments relate to a method, as well as a control device and code segments for carrying it out.
[0002] In general, a coating process can be used to deposit a layer onto a substrate. This coating process can be carried out, for example, using a physical vapor deposition process, such as cathode sputtering (also known as sputtering). Modifications of cathode sputtering include reactive sputtering and reactive magnetron sputtering. In reactive sputtering, a working gas (e.g., argon) is used to sputter the cathode (the target material), with the working gas not being incorporated into the layer deposited on the substrate. At least one reactive gas is added, with which the sputtered target material (also called the coating material) reacts chemically, causing the reaction product to be deposited on the substrate.
[0003] In sputtering, a substrate can be coated with a layer by bringing the sputtering setup to and / or maintaining it at an operating point (also called the operating point), at which a layer with the desired properties is obtained. The operating point can define a set of operating parameters of the sputtering setup, the interaction of which influences the properties of the layer.
[0004] In contrast to sputtering without reactive gas, which can generally proceed relatively stably (for example, requiring no further control mechanisms or only minimal intervention), sputtering under reactive gas (also known as reactive magnetron sputtering) may require appropriate control techniques and equipment to manage the complex interplay of operating parameters. This interplay is particularly sensitive with some coating materials, as, for example, the desired layer properties are only achieved within a narrow stoichiometry range, and the reaction dynamics are often unstable (e.g., bistable). Unstable reaction dynamics can, for instance, tend to deviate from the desired operating point on their own and / or oscillate between two self-establishing (stable) reaction modes.
[0005] However, controlling the reaction dynamics becomes more difficult the more complex the coating process is, the more it interacts with other coating processes, and the smaller the acceptable fluctuation. For example, it may be necessary to simultaneously control several controlled variables and manipulated variables for each coating process, while also dealing with multiple disturbances. These variables enclose a multidimensional state space that becomes increasingly difficult to access analytically as the number of independent dimensions grows. Furthermore, there is a high degree of implicit interactions that are difficult to elucidate experimentally, making it nearly impossible to simplify this state space.
[0006] In a high-productivity coating system, for example, several identical coating processes can be carried out sequentially, allowing the substrate to be coated faster and / or with a thicker layer. If space requirements need to be reduced, the gas separation between the coating processes can be decreased. The lower the gas separation, the more the coating processes can influence each other, causing their control loops to act as mutual disturbances, which makes controlling the reaction dynamics more difficult despite the control system.
[0007] In the simplest case, this complexity is addressed by using an individual control loop for each controlled variable. Such individual control loops operate completely independently of one another; that is, their input and output span disjoint hypersurfaces in the state space. This approach intuitively decomposes the state space into individual, more manageable two-dimensional subspaces and is therefore easily recognizable by the fact that the input / output of multiple control loops is completely independent of each other. As a consequence, all process-related interactions are treated as disturbances, so that multiple control loops may, under certain circumstances, work against each other, i.e., attempt to compensate for each other's influence.
[0008] Such an individual control loop can be implemented with a programmable logic controller (PLC), which may even be capable of implementing several such control loops. A so-called PID controller (proportional-integral-derivative controller) used for this purpose feeds back the control error using a P element (proportional control element), an I element (integral control element), and a D element (derivative control element). However, the configuration of the individual control elements is often done by trial and error and frequently requires consolidated empirical data.
[0009] Alternatively, this complexity can be addressed by adding boundary conditions that restrict the state space or simply combine several degrees of freedom. Reference is made to DE 10 2019 123 410 A1 and DE 10 2011 017 583 A1 in this regard.
[0010] According to various embodiments, it has been clearly demonstrated that these conventional concepts are inaccurate and / or prone to over- or under-control. In contrast, various embodiments provide a method, a control device, and code segments that improve the control of one or more coating processes, for example, by being more accurate, less prone to over- / under-control, requiring less computing power, and being easier to calibrate (e.g., automatically). Using the method, the control device, and the code segments, a corresponding controller is implemented and / or calibrated.
[0011] According to various embodiments, several actuators supplying the same coating process are controlled based on the same group of simultaneously acquired control variables (comprising multiple control variables) of the reactive coating process. The provided controller maps a multidimensional vector as an input, whose vector components contain or at least uniquely represent the controlled variables, to a multidimensional vector as an output, whose vector components contain or at least uniquely represent the controlled variables. Each of the control variables thus determined is then fed to a corresponding actuator from a group of actuators supplying the coating process, so that the coating process is influenced by means of the group of actuators.This clearly demonstrates that each controlled variable is taken into account when controlling each actuator, or rather, that the interaction of the multiple controlled variables is taken into account.
[0012] According to various embodiments, a method, a control device and code segments are provided which provide the control of one controlled variable (SISO, single-in singleout) or several controlled variables (MIMO, multiple-in multipleout), e.g. process voltage and / or total pressure, in one or more vacuum-based (e.g. reactive) sputtering processes, e.g. using a method for automatically calibrating the controller, so that automatic adjustment of the controller can take place.
[0013] The controller provided in this way can be used, for example, to control a group of coating processes (also referred to as a process group), taking into account the interaction of these coating processes with one another, e.g., gas flows in neighboring chambers and any potential oscillation effects. As a further aspect, the controller is also integrated to manage the total pressure per vacuum chamber.
[0014] The provided controller is also easier to scale, e.g. by extending it to include additional input variables, e.g. a working gas and two reactive gases, and / or more coating processes.
[0015] The provided controller can, but does not necessarily have to, be implemented for a group of coating processes in multiple processing chambers using only one controller hardware component (e.g., with only one power supply). The coupling of multiple coating processes is taken into account according to various embodiments. Automated calibration without expert knowledge is possible, and the dynamics of the resulting controller can be robust and efficient. The controller is future-proof and expandable, for example, by extending it to include additional reactive gases or by introducing a state monitor.
[0016] A model of the controlled system (also referred to as a control loop model) is provided for one or more vacuum chambers and the coating process performed in each vacuum chamber. Based on this control loop model, controller matrices for a PI output controller are calculated by systematically recording one or more step responses (e.g., a series of step responses) and using numerical methods. These matrices are then implemented using appropriate controller hardware. An autotuning algorithm with preset parameters, using automated experiments and numerical methods while considering the hardware cycle time, can provide a robust and efficient controller.
[0017] They show Fig. 1. A method according to various embodiments in a schematic flowchart; Fig. 2 to 4 each show the procedure according to different embodiments in a schematic control diagram; Fig. 5. The method according to different embodiments in a detailed schematic diagram; Fig. 6. An implementation of the method according to various embodiments in a detailed schematic diagram; Fig. 7. The controlled system according to different embodiments in a detailed schematic diagram; Fig. 8 to 12 each represent a model of different stages according to different embodiments in a detailed schematic assembly diagram; Fig. 13 a method according to different embodiments in a schematic flowchart; Fig. 14 a temporal sequence according to the method according to different embodiments in a schematic diagram; Fig. 15A and Fig. 15B illustrates a process group according to different embodiments in different assembly diagrams; Fig. 16A and Fig. 16B illustrate the control of the process group according to different embodiments in different control diagrams; Fig. 18 a model generator according to various embodiments in a schematic diagram; Fig. 17, Fig. 19 and Fig. 20 each a method according to different embodiments in a schematic flowchart; Fig. 21 Several quantities of a coating process controlled by the method according to different embodiments in different schematic diagrams; and Fig. 22 and Fig. 23 each a model generator according to different embodiments in a schematic network diagram.
[0018] The following detailed description refers to the accompanying drawings, which form part thereof and illustrate specific embodiments in which the invention can be implemented. In this context, directional terminology such as "top," "bottom," "front," "back," "anterior," "rear," etc., is used with reference to the orientation of the described figure(s). Since components of embodiments can be positioned in a number of different orientations, the directional terminology serves only for illustration and is in no way limiting. It is understood that other embodiments may be used and structural or logical modifications may be made without deviating from the scope of protection of the present invention.It is understood that the features of the various exemplary embodiments described herein can be combined with one another, unless specifically stated otherwise. The following detailed description is therefore not to be interpreted in a limiting sense, and the scope of protection of the present invention is defined by the appended claims.
[0019] Within this description, the terms "connected," "connected," and "coupled" are used to describe both direct and indirect connections (e.g., resistive and / or electrically conductive, such as an electrically conductive connection), direct or indirect connections, and direct or indirect couplings. In the figures, identical or similar elements are designated with identical reference numerals where appropriate.
[0020] Depending on the specific embodiment, the term "coupled" or "coupling" can be understood as a connection and / or interaction (e.g., mechanical, hydrostatic, thermal, and / or electrical), whether direct or indirect. For example, several elements can be coupled along an interaction chain, along which the interaction (e.g., a signal) can be transmitted.
[0021] For example, two coupled elements can exchange an interaction, such as a mechanical, hydrostatic, thermal, and / or electrical interaction. Depending on the specific embodiment, "coupled" can be understood as a mechanical (e.g., physical) coupling, for example, by means of direct physical contact. A coupling can be configured to transmit a mechanical interaction (e.g., force, torque, etc.).
[0022] Control can be understood as the intentional influencing of a system. The current state of the system (also called the actual state) can be changed according to a specification (also called the target state). Regulation can be understood as control, but with the additional benefit of counteracting changes in the system's state caused by disturbances. Visually, control can have a forward-directed control path and thus implement a sequential control system that converts an input variable (e.g., the specification) into an output variable. However, the control path can also be part of a control loop, thus implementing a regulation system. In contrast to purely forward-directed sequential control, regulation involves a continuous influence of the output variable on the input variable, which is effected by the control loop (also called feedback).The state of the system (also referred to as the operating point) can be represented by one or more controlled variables, whose actual value represents the current state of the system and whose setpoint (also referred to as the reference value) represents the desired state of the system. In a control system, the actual state of the system (e.g., determined based on a measurement) is compared with the desired state of the system, and the one or more controlled variables are influenced by one or more corresponding manipulated variables or by using one or more actuators in such a way that the deviation of the actual state from the desired state of the system is minimized.
[0023] The system to be controlled can, for example, have one or more coating processes. The state of the system can thus represent the state of each coating process within the system, and vice versa. The current state of each coating process can be detected by means of a measuring element (e.g., having one or more sensors). For each controlled variable, one or more sensors can be used, configured to detect a quantity (also called a measured variable) that represents the controlled variable or that is the controlled variable itself.
[0024] A sensor (also called a detector) can be understood as a transducer designed to detect a property of its environment (e.g., qualitative or quantitative) corresponding to the sensor type, such as a physical property, a chemical property, and / or a material composition. The measured quantity is the physical quantity to which the sensor applies its measurement. Depending on the complexity of the environment being measured, the sensor can be configured to distinguish between only two states of the measured quantity (also called a switching sensor), to distinguish between more than two states, or to detect the measured quantity quantitatively. An example of a quantitatively detected measured quantity is a fluid flow rate (e.g., flow rate), the current state of which can be detected as a value by the sensor.
[0025] A sensor can be part of a measurement chain, which includes the necessary infrastructure (e.g., a processor, storage medium, and / or bus system, etc.). The measurement chain can be configured to control the corresponding sensor (e.g., a gas sensor, pressure sensor, and / or voltage sensor), process its measured value as input, and then provide an electrical signal as output that represents the input value. The measurement chain can be implemented, for example, using a control device.
[0026] According to various embodiments, a multi-variable control system can be provided that jointly and interconnects a coating process or a group of several (e.g., N) coating processes (where, for example, N>1, N>5, or N>10). The multiple coating processes can be identical, for example, in their coating material, reactive gas, and / or the type of coating device (e.g., thermal or atomizing). According to various embodiments, each coating process can be configured to coat at least one substrate (i.e., one or more substrates).
[0027] The following discussion refers, among other things, to reactive coating processes in which a chemical reaction occurs. For example, one or more of the coating processes may involve a reactive sputtering process, which is supplied with one or more reactive gases (e.g., m reactive gases) that chemically bond with the target material, which is then deposited onto a substrate (where, for example, m = 1, m > 1, or m > 2). It should be understood that the coating process does not necessarily have to be reactive. The quantities that represent materials involved in the chemical reaction in a reactive coating process may represent corresponding materials that do not participate in a chemical reaction in a non-reactive coating process. Accordingly, what is described for the reactive coating process can apply analogously to a non-reactive coating process.
[0028] In various embodiments, reference is made to physical vapor deposition (PVD) as an exemplary coating process, e.g., one involving a sputtering process. It can be understood that what is described for PVD can apply analogously to chemical vapor deposition (CVD). In contrast to CVD, in PVD a solid material is first converted into the gas phase (also referred to as gaseous phase or vapor), and a layer is formed using this gas phase. In PVD, the gas phase of the target material can optionally react chemically with a reactive gas to form a chemical compound, which is incorporated into the layer or forms it. Thus, in the chemical reaction of PVD, two or more materials are combined to form the chemical compound.
[0029] In chemical vapor deposition (CVD), a gaseous starting compound (also called precursor or reactant) is split into at least two reaction products, of which at least one is incorporated into the layer and optionally one reaction product is removed from the coating process as an excess (e.g., by means of a pump). Optionally, CVD can be carried out using a plasma in which the splitting of the precursor takes place.
[0030] A plasma can be generated using a working gas. Depending on the specific embodiment, the working gas can be a gaseous material that is unreactive, meaning it participates in few or no chemical reactions. A working gas can be defined by, or adapted to, the target material used. For example, a working gas can be a gas or a gas mixture that does not react with the target material to form a solid. The working gas can be, for example, a noble gas (e.g., helium, neon, argon, krypton, xenon, radon) or several noble gases. The plasma can be generated from the working gas, which essentially causes the target material to be atomized. If a reactive gas is used, it can have a higher chemical reactivity than the working gas, for example, with respect to the target material.In other words, the atomized target material can react faster with the reactive gas (i.e., form more reaction product per unit of time) than with the working gas (e.g., if it reacts chemically with the working gas at all). The reactive gas and the working gas can be supplied together or separately as a process gas (e.g., as a gas mixture), for example, via the gas supply device.
[0031] The reactive gas can be a gaseous material that reacts with the target material (e.g., with the atomized target material) and / or can be incorporated into the deposited layer via a chemical reaction. For example, if a target material is used that can form a nitride (e.g., aluminum nitride), or if a nitride of the target material is to be deposited, the reactive gas can contain nitrogen or be composed of nitrogen. If, for example, a target material is used that can form an oxide (e.g., aluminum oxide), or if an oxide of the target material is to be deposited, the reactive gas can contain oxygen or be composed of oxygen. The reactive gas can, for example, be a gas mixture (reactive gas mixture) of several gases that react with the target material and / or the deposited layer, e.g., oxygen and nitrogen, for example, when an oxynitride (e.g.,The reactive gas mixture can, for example, consist predominantly (i.e., more than 50%) of oxygen, for the deposition of an oxide or oxynitride. Examples of reactive gases include: molecular oxygen, molecular nitrogen, nitric oxide, carbon dioxide, hydrogen sulfide, methane, gaseous hydrocarbons, molecular fluorine, molecular chlorine, ozone, or other gaseous materials.
[0032] The reactive sputtering process can involve sputtering (atomizing) the target material using the working gas and forming a chemical compound (in other words, the reaction product) with the supplied reactive gas, e.g., a metal compound or a metalloid compound, which can be deposited onto a substrate.
[0033] Fundamentally, coating processes can be divided into two types based on their reaction dynamics: stable coating processes and unstable coating processes. In a stable coating process, the controlled variable always assumes the same actual state in response to the same displacement of the coating process from a reference state. In other words, the actual state of the controlled variable (also called the actual controlled variable) is a function of the reference state and the magnitude of the displacement.In an unstable coating process, the actual state of the controlled variable in response to the deviation from the reference state either depends on other influencing factors, such as the history of the coating process or the trajectory along which the deviation occurs, or the controlled variable always assumes an actual state from the same finite set of states (also referred to as stable operating points), regardless of the deviation or the reference state.
[0034] The following refers to a method that involves controlling one or more actuators. In this context, reference is also made to a control device or code segments. The control device can be configured to implement one or more of the methods described herein. For this purpose, the control device can include a processor configured to implement the respective method. For example, the processor can be configured to output corresponding instructions. Alternatively or additionally, the processor can be configured to receive and process corresponding instructions. These instructions can be implemented, for example, by means of code segments stored in non-volatile memory.For example, the code segments may contain instructions which, when executed by the processor, cause the processor to perform one of the procedures.
[0035] The term "control device" can be understood as any type of logic-implementing entity that may, for example, have circuitry and / or a processor capable of executing software stored in a memory medium, firmware, or a combination thereof, and issuing instructions based on that software. The control device may be configured, for example, using code segments (e.g., software). The control device may, for example, be a programmable logic controller (PLC) or be derived from one.
[0036] Depending on the specific embodiment, a data storage device (more generally referred to as a storage medium) can be a non-volatile data storage device. The data storage device can, for example, comprise a hard disk and / or at least one semiconductor memory (such as read-only memory, random access memory, and / or flash memory). The read-only memory can, for example, be a erasable programmable read-only memory (which may also be referred to as an EPROM). The random access memory can be a non-volatile random access memory (which may also be referred to as NVRAM). For example, the data storage device can contain one or more of the following: the code segments representing the process, one or more control elements, one or more tuning parameters, one or more model parameters, or one or more calibration parameters.
[0037] The term "processor" can be understood as any type of entity that allows the processing of data or signals. The data or signals can, for example, be processed according to at least one (i.e., one or more than one) specific function performed by the processor. A processor can be an analog circuit, a digital circuit, a mixed-signal circuit, a logic circuit, a microprocessor, a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), a programmable logic gate array (FPGA), an integrated circuit, or any combination thereof.Any other type of implementation of the respective functions, which are described in more detail below, can also be understood as a processor or logic circuit, for example, virtual processors (or a virtual machine) or a multitude of decentralized processors that are interconnected, for instance, via a network, are arbitrarily geographically distributed, and / or have arbitrary contributions to the implementation of the respective functions (e.g., workload distribution among the processors). The same generally applies to any differently implemented logic for implementing the respective functions. It is understood that one or more of the process steps described in detail herein can be executed (e.g., realized) by a processor through one or more specific functions executed by the processor.
[0038] In this context, a "controller" can be understood as a processor that implements the functions for controlling processes, for example, by converting information about the current state of a coating process into signals for controlling that process. These signals can then be used, for instance, to control one or more actuators that supply the coating process. The processor can be configured as a controller using a control model (e.g., via software) or can provide at least some of the control functions in a hard-coded manner. For example, a control device can incorporate a processor configured as a controller. The controller and the control model refer to the implementation of the same functions for performing the control, with the controller representing the physical implementation and the control model the logical implementation.It can therefore be understood that what has been explained regarding the controller can apply analogously to the control model and vice versa.
[0039] The term "actuator" (e.g., comprising an actuator) can be understood as a transducer designed to influence the state of a process (e.g., a coating process) or a device in response to a control signal. The actuator can convert an input signal (the so-called control signal) into mechanical movements or changes in physical quantities such as pressure or temperature. An electromechanical actuator, for example, can be configured to convert electrical energy into mechanical energy (e.g., through movement) in response to a control signal.
[0040] An actuator can be configured to influence the actual state of one or more controlled variables in the coating process that are supplied by the actuator. This influence can be direct or indirect. For example, the manipulated variable and the controlled variable can differ from each other. Accordingly, the controlled variable can be a function of one or more manipulated variables.
[0041] For example, the actuator can change an electrical voltage as the manipulated variable, which supplies a plasma, so that a atomization rate is consequently changed as the controlled variable. For example, the actuator can change a gas flow rate as the manipulated variable, so that a pressure is consequently changed as the controlled variable. For example, the actuator can change a working gas flow rate as the manipulated variable, so that a atomization rate is consequently changed as the controlled variable. For example, the actuator can change a reactive gas flow rate as the manipulated variable, so that a chemical composition is consequently changed as the controlled variable.
[0042] Examples of actuator components include: a voltage source (if present), a valve (e.g., of a pump assembly and / or gas supply device), a motor (e.g., of a valve or pump), or similar components. The voltage source may, for example, be configured to generate one or more voltages according to the control signal and supply them to the coating process. The pump assembly may, for example, be configured to pump out one or more gases according to the control signal, thereby removing them from the coating process. The gas supply device may, for example, be configured to supply one or more gases according to the control signal to the coating process.
[0043] The following section discusses mathematical relationships and various quantities (e.g., input and / or output quantities), using the following notation: • Simple symbols (for example x, y, ...) represent scalar quantities; • Underlined symbols (e.g. x, y, ...) represent vector quantities; • Bold symbols X, Y, ... represent matrices; • Symbols with tilde x̃, ỹ, ... represent absolute quantities (e.g. a measured quantity such as the process voltage Ũ), • Symbols without a tilde (e.g., U) represent a difference from a constant in the context of absolute quantities; • Symbols with a bar above the symbol (e.g., U) represent a constant; • Symbols or entire expressions in curly braces (e.g. {X}, {X + Y}, ...) are included for completeness only, but are disregarded (i.e., removed from the corresponding relations).
[0044] The mathematical relations and various quantities are listed below in the more easily understood vector notation. Similarly, tensors are written as exemplary mappings in the more easily understood one-dimensional or two-dimensional notation (e.g., as matrices). It can be understood that the relations, mappings, and quantities explained can also be transferred to or written in other mathematical spaces, for example, by means of a corresponding transformation.
[0045] In this context, a model can be understood as a data-based (e.g., digital and / or virtual) representation of an original, such as a process (e.g., a coating process). To create the model (the so-called model building, i.e., mapping the original onto the model), the original can be abstracted, parameterized, and / or simplified. The model can, for example, contain physical information (e.g., length, distance, weight, volume, composition, etc.), motion-related information (e.g., position, orientation, direction of movement, acceleration, speed of movement, etc.), logical information (connections, sequence, couplings, interrelationships, dependencies, etc.), time-related information (e.g., time, total duration, frequency, period, etc.), and / or functional information (e.g., current, effect, characteristic map or curve, operating point space, force, degrees of freedom, etc.) about the original.
[0046] Fig. Figure 1 illustrates a method 100 according to various embodiments in a schematic flowchart. The method 100 can be used to control a (e.g., reactive) coating process, e.g., each coating process of a process group.
[0047] Method 100 comprises, in 101, controlling a first actuator by means of which the coating process is supplied, based on a group of several control variables of the reactive coating process acquired in parallel to each other; and in 103, controlling a second actuator by means of which the coating process is supplied, based on the group of several control variables of the reactive coating process acquired.
[0048] According to various embodiments, the group of several control variables acquired in parallel can include at least one (i.e., one or more than one) control variable per controlled actuator. Alternatively or additionally, the group of several control variables acquired in parallel can include at least two control variables per coating process.
[0049] For example, several controlled variables from a group of multiple controlled variables acquired in parallel (e.g., each controlled variable) can be considered when controlling the first actuator. Similarly, several controlled variables from a group of multiple controlled variables acquired in parallel (e.g., each controlled variable) can be considered when controlling the second actuator.
[0050] The controlled variables can be acquired (in parallel) over time. For example, each controlled variable can be repeatedly acquired using one or more sensors, e.g., according to a clock cycle that defines the acquisition interval. The acquisition of different controlled variables can, but does not necessarily have to, occur at the same clock cycle. For example, they can be acquired simultaneously or at different intervals. The repeated acquisition of the controlled variables can also occur in parallel over time, e.g., over corresponding time periods that overlap.
[0051] The control variables, which are recorded in parallel over time, can be recorded simultaneously, but this is not mandatory. If they are not recorded simultaneously, they can at least be recorded within a period that lies between two immediately successive events. Each of the two immediately successive events could, for example, involve: the activation of the first actuator and / or the activation of the second actuator.
[0052] As a result of acquiring a controlled variable, information about its current state is obtained. This set of acquired current states (e.g., actual pressure and actual voltage) can be taken into account when controlling the first and second actuators. In other words, the same controlled variables (e.g., their current states) that are considered when controlling the first actuator can also be considered when controlling the second actuator.
[0053] The control 101 of the first actuator and the control 103 of the second actuator can be carried out by means of a control model (also called a control algorithm) which implements a mapping of the group of several parallel controlled variables (e.g. their actual states) to a first controlled variable and a second controlled variable.
[0054] For the sake of simplicity, this text refers to a controller that implements the functions of the control model. It can be understood that the controller can be configured using the control model to provide the described functions. Furthermore, it can be understood that the control model can also be stored separately from the execution of procedure 100, for example, on a data storage device, or it can be determined separately, as will be explained in more detail later.
[0055] The first control variable can be supplied to the first actuator for control purposes, and the second control variable can be supplied to the second actuator for control purposes. The control variable is the input variable of the actuator and can be transmitted, for example, by means of a signal (e.g., analog or digital). The controller can be implemented using a control device and / or code segments.
[0056] The controller can, for example, be configured to map a multidimensional control vector, whose vector components represent the relative controlled variables (e.g., their current state), to a multidimensional control vector, whose vector components represent multiple controlled variables. The controller can have at least one (i.e., one or more than one) control element that implements the mapping or at least a part of it.
[0057] For example, a control variable of a coating process can be a parameter (e.g., the position) of an actuator of the coating process, by means of which the supply (e.g., feeding and / or extracting) of the coating process with a medium (electrical energy, material, thermal energy, etc.) is carried out (also referred to as media supply). For example, to control the coating process, one or more of the following parameters of the coating process can be used as a control variable of the coating process: an electrical parameter for supplying the coating process, a temperature of the process gas (containing the working gas and / or the reactive gas), an inflow (e.g., inflow rate and / or inflow distribution) of the process gas (e.g., the spatial distribution of the inflow), an inflow (e.g., inflow rate and / or inflow distribution) of the working gas (e.g., the spatial distribution of the inflow), an inflow (e.g.,The inflow rate and / or inflow distribution of the reactive gas (e.g., the spatial distribution of the inflow), the transport velocity of the substrate, and the emission rate of the coating material are all relevant parameters. The electrical quantity can, for example, represent: electrical power, electrical current (which, for example, flows through the plasma), or electrical voltage (which, for example, is applied across the plasma).
[0058] For example, a parameter of the coating process or its result that changes in response to the media supply can be used as a control variable of a coating process. For example, one or more of the following parameters of the coating process can be used as a control variable to regulate the coating process(es): a pressure of the process gas (and / or its spatial distribution), a partial pressure of the working gas (and / or its spatial distribution); a partial pressure of one or more reactive gases (and / or their spatial distribution), a property (e.g., chemical composition) of the reaction product, an electrical voltage (e.g., that which drops across the plasma), an electrical current (e.g., that is supplied to the plasma).
[0059] Some parameters of the coating process can be used as manipulated variables in some configurations and as controlled variables in others. The actual choice of manipulated and controlled variables therefore depends on the configuration of the coating process and the supply infrastructure, and can be adapted to the specific process. For example, the voltage supplied to a plasma can be used as a manipulated variable and the current drawn by the plasma as a controlled variable, or vice versa. Similarly, the power consumed by the plasma can be used as a controlled variable and the gas flow supplied to the plasma as a manipulated variable, or vice versa.
[0060] The coating process, or processes, controlled by Method 100 can be stable or unstable (e.g., bistable). In other words, the coating process can have at least two stable operating points (e.g., an oxide operating point and a metallic operating point) and a transition region between the two stable operating points. Control can then be achieved, for example, by balancing the process, such as maintaining each coating process in an unstable reaction mode between the stable reaction modes.
[0061] Optionally, a transport device can also be controlled, by means of which a substrate is fed into the coating process, which is then coated by means of the coating process.
[0062] Fig. Figure 2 illustrates the method 100 according to different embodiments in a schematic control diagram 200.
[0063] For the sake of clarity, the following refers to two exemplary elements (numbered 1 and 2) of each instance of the control loop, e.g., two actuators, two controlled variables, two manipulated variables, two control variables, etc. It should be understood that what is described for the two exemplary elements can apply analogously to more than two elements of the instance.
[0064] The coating process 201 can be controlled by several actuators, for example, by the first actuator (actuator 1) and the second actuator (actuator 2). Each of the multiple actuators can be configured to influence a control variable of the coating process 201, for example, based on a control variable supplied to the actuator. The first actuator can influence a first control variable of the coating process 201. The second actuator can influence a second control variable of the coating process 201. The control variable is the output variable of the actuator used, with the help of which a targeted intervention in the controlled system is achieved. The instantaneous value of the control variable (also referred to as the actual state of the control variable) is the so-called manipulated value.
[0065] The supply via the actuator can involve adding and / or removing something (e.g., one or more materials) from the coating process 201, whereby the supply and / or removal can be influenced by the actuator. For example, a material can be supplied to the coating process. In this case, the actuator can, for example, influence a rate (i.e., quantity per unit of time) at which the material is supplied (also referred to as the supply rate). For example, a material can be removed from the coating process. In this case, the actuator can, for example, influence a rate (i.e., quantity per unit of time) at which the material is removed (also referred to as the removal rate).
[0066] If the material is gaseous under standard conditions, for example, the supply rate or the extraction rate can be a gas flow rate (also referred to as flow rate) (the supply rate is then also referred to as inflow rate). The actuator can then, for example, include a valve by which the flow rate is controlled. Alternatively or additionally, the actuator can, for example, include a pump by which the extraction rate is controlled.
[0067] If the material is solid under standard conditions, the feed rate can be either an atomization rate or a vaporization rate. In this case, the actuator can, for example, include a power supply to influence the feed rate. For instance, the electrical power supplied for atomization or vaporization can be varied by the power supply. The atomization rate can also be influenced by the supply rate of the working gas.
[0068] This text refers, among other things, to the inflow rate as an exemplary supply rate, or to the corresponding gas flow (also referred to simply as flow) as an exemplary supply. It can be understood that the explanations given can apply analogously to other supply rates or other forms of supply.
[0069] The coating process 201 can furthermore have at least two (i.e., two or more than two) controlled variables. Of these at least two controlled variables, a group of controlled variables, e.g., the first controlled variable and the second controlled variable, can be acquired in parallel, e.g., clocked and / or repeatingly. In other words, the actual state 211a, 211b of each controlled variable in the group of controlled variables can be acquired in parallel. This can result in an information stream (e.g., clocked) that indicates the actual states 211a, 211b of the group of controlled variables.
[0070] The part of the coating process 201 that converts the manipulated variable(s) into the two controlled variables is also called the controlled system. The controlled system thus refers to that part of a control loop whose output contains one or more controlled physical quantities (controlled variables) that the controller is intended to influence using the manipulated variable. Examples of controlled variables include: temperature, pressure, power, current, voltage, chemical composition, coating rate, spatial position, etc.
[0071] The multiple controlled variables can differ from one another, for example in the material they represent. Each controlled variable can, for instance, represent the presence (e.g., the quantity) of the material in the coating process.
[0072] The first controlled variable can, for example, represent a gas, such as a working gas or a reactive gas, and its partial pressure. The second controlled variable can, for example, represent another gas, such as a working gas or a reactive gas, and its partial pressure.
[0073] Alternatively, the second controlled variable can represent the rate at which the target material is converted from the solid state to the gaseous state (e.g., referred to as the atomization rate or vaporization rate). An example of such a second controlled variable is an electrical quantity used to supply the plasma. Examples of such electrical quantities include: an electrical voltage (e.g., applied to the plasma), an electrical power (e.g., supplied to the plasma), and an electrical current (e.g., supplied to the plasma).
[0074] The controlled system can be understood as an interaction chain of the coating process 201, whose links can often only be partially described analytically or are even experimentally inaccessible. Every output variable of the controlled system, i.e., every controlled variable, can be detected by means of the measuring element (e.g., having one or more sensors). Every input variable of the controlled system, i.e., every manipulated variable, can be adjusted by means of an actuator. The measuring element and the actuator thus form the interfaces between the controlled system and the feedback loop of the control system.
[0075] Mathematically, the controlled system can be viewed as a multidimensional transfer system, as will be described in more detail later. The controlled system can transfer data from a multidimensional space of manipulated variables back into a multidimensional space of controlled variables, which is also referred to as a MIMO controlled system (MIMO = "multiple-input multiple-output system"). The transfer system can exhibit linear and / or nonlinear behavior. The linear transfer system (e.g., a linear time-invariant system) can be described by differential equations (DEs), by transfer functions, by numerical discrete-time methods (difference equations), and / or in state space. For this purpose, a model of the controlled system can be determined.
[0076] The coating process 201 can, for example, be configured such that the first manipulated variable influences the first controlled variable and / or the second controlled variable. The coating process 201 can, for example, be configured such that the second manipulated variable influences the first controlled variable and / or the second controlled variable.
[0077] Alternatively or additionally, the coating process 201 can be set up such that the first controlled variable and / or the second controlled variable influence each other, e.g., are functions of each other. For example, the controlled variables of a group of several controlled variables measured in parallel can interact with each other.
[0078] The actual state 211a, 211b of each controlled variable in the group of parallel controlled variables can thus be converted into several control variables 213, e.g., by means of the controller 203, which has at least one (i.e., one or more than one) control element. The controller 203 can be implemented, for example, by means of the control device. Of the several control variables 213, a first control variable can be supplied to the first actuator to control the first actuator. In other words, the first actuator can be controlled by means of the first control variable. Of the several control variables 213, a second control variable can be supplied to the second actuator. In other words, the second actuator can be controlled by means of the second control variable.
[0079] The controller 203 can be configured such that the actual state 211a, 211b of each controlled variable in the group of parallel controlled variables is converted into the first controlled variable. The controller 203 can be configured such that the actual state 211a, 211b of each controlled variable in the group of parallel controlled variables is converted into the second controlled variable.
[0080] To capture the individual controlled variables, for example, controllability and observability of the coating process can be provided. The controller 203 can, for example, implement a mathematical model of the controlled system (e.g., a linear system of differential equations, possibly provided through linearization).
[0081] According to various embodiments, feedback of all dynamic parameters important for the coating process or the use of a state observer is possible (e.g., using the controller 203). Furthermore, according to various embodiments, feedback of the control deviations is possible (e.g., using the controller 203) for improved dynamic properties, and a compact and uniform mathematical formulation can be implemented.
[0082] In contrast to a simple PID controller, the controller 203 can be configured to control the MIMO control system, taking into account couplings between different input variables of the control system and / or within the control system.
[0083] The following section provides a mathematical explanation of an exemplary implementation of the procedure described above. The mathematical formulation used for the underlying interactions can be understood as an example. For ease of understanding, the quantities representing the elements of each instance of the control loop described above are formulated as vectors. However, it should be understood that any other mathematical space besides a vector space can be suitable for describing the underlying interactions. More generally, tensors and operators can be used, which can also be transformed into systems of linear equations.
[0084] The following example describes the controlled system in state space, specifically using a state-space model in the form of a first-order linear differential equation system. It refers to an internal system state x(t) (also called the state vector), the output vector y(t), and the input vector u(t). The output vector y(t) represents the current state of each controlled variable in the system and can be detected by one or more sensors. The input vector u(t) represents the current state of each manipulated variable in the system and can be influenced by one or more actuators.
[0085] If the controlled system is better described by a nonlinear system of differential equations, it can be linearized around a point of interest (for example, referred to as a representation of the operating point) within the system of differential equations (which, for example, describes a coating process). This linearization can, for instance, represent the actual operating point of the coating process. The linearization then restricts the controller 203 to a subspace of the state space. This may be sufficient, however, since the controller 203 is configured to maintain the system state within this subspace, i.e., to prevent the system state from leaving the subspace. It can be understood that a more complex controlled system can also be linearized and defined piecewise, if necessary. The same principles apply to the linearized controlled system as described for the linear system of differential equations.
[0086] For example, a general linearization of a system of ODEs can be achieved using a Taylor expansion, for example a Taylor expansion up to the first order, which then reads: x_˜˙(t)=f_(x_˜(t),u_˜(t))=f_(x¯_+x_(t),u_¯+u_(t))≈f_(x_¯,u_¯)+Ax_(t)+Bu_(t) y_˜(t)=g_(x_˜(t),u_˜(t))=g_(x¯_+x_(t),u_¯+u_(t))≈g_(x_¯,u_¯)+Cx_(t)+Du_(t) with the model elements (A, B, C) and D, which can be written as: A=(∂ƒ_∂x_˜)x_˜=x_¯,u_˜=u_¯,B=(∂ƒ_∂u˜_)x_˜=x_¯,u_˜=u_¯ ,C=(∂g_∂x_˜)x_˜=x_¯,u_˜=u_¯,D=(∂g_∂u_˜)x_˜=x_¯,u_˜=u_¯ and the deviations x(t) and u(t) from the operating point x¯_, u¯_, which is expressed by: x_(t)=x_˜(t)−x_¯ u_(t)=u_˜(t)−u_¯.
[0087] The model of the controlled system can, for example, have one or more model elements, such as two or more, or three or more model elements (A, B, C). It should be understood that the model described here, which has three model elements, is exemplary and the underlying relationships can also be formulated differently. For example, more or fewer than three model elements can be used. For example, two or more model elements can be combined, for example, by means of a transformation into a higher-dimensional space. For example, a model element can be split, for example, by means of a transformation into two lower-dimensional spaces.
[0088] The implicitly specified operating point y_¯=g_(x_¯,u_¯) with respect to the initial vector (for an asymptotically stable coating process), the difference is defined y_(t)=y_˜(t)−y_¯ and is generally defined by an unknown relation g_(x_¯,u_¯) Given, but can usually be measured directly.
[0089] The choice of an operating point is made in such a way that the controlled system is in equilibrium, which is expressed by: x˙_=f_(x_¯,u_¯)=0 where x_¯ and u_¯ are constant, and causes the deviation x from the operating point to be x_¯ can be dynamically described by the identical system of differential equations, namely according to the relation: x˜˙_(t)=ddtx_˜(t)=ddt(x_¯+x_(t))=ddtx_(t)=x_˙(t).
[0090] The controlled system is approximately in equilibrium when the rate of change of the controlled variables is essentially zero and / or a predefined period has elapsed, e.g., several times the half-life of the response to the deflection. In other words, the step response can involve switching between several setpoint states of the manipulated variable (discrete) as soon as a criterion is met. The criterion might be met, for example, when the rate of change of the actual state of the controlled variable meets a predefined criterion, e.g., when the rate of change falls below a predefined threshold. More generally, the step response implements the fact that the setpoint state of the manipulated variable (also called the target manipulated variable, e.g., a setpoint) is a discontinuous function of the rate of change of the controlled variable.
[0091] Taken together, this results in an approximate model for the deviations x(t), y(t) and u(t) of the controlled system from a correspondingly defined operating point. x_¯,y_¯ and u_¯: y_˜(t)≈g_(x_¯,u_¯)+Cx_(t)+Du_(t)⇒y_˜(t)−g(x_¯,u_¯)=y_˜(t)−y_¯=y_ ≈Cx_(t)+Du_(t) and x˙_(t)=Ax_(t)+Bu_(t){+Ed_(t)}
[0092] The controlled system can be formulated as follows: x˙_(t)=Ax_(t)+Bu_(t){+Ed_(t)} y_(t)=Cx_(t)+Du_(t){+Fd_(t)} with the system matrix A, the input matrix B, the output matrix C, the pass matrix D, the disturbance matrices E and F, and the disturbance variable d(t). As explained above, the matrix formulation is merely an example intended to facilitate understanding due to the more intuitive two-dimensional nature of the matrix. As explained above, expressions in curly brackets ({+E d(t)} and {+F d(t)}) are included only for completeness and will be disregarded in the following discussion.
[0093] The equations of motion for the system state and the output vector can then be expressed as: x_(t)=eAtx_(0)+∫0teA(t−τ)Bu_(τ)dτ y_(t)=CeAtx_(0)+∫0tCeA(t−τ)Bu_(τ)dτ+Du_(t)
[0094] In the following considerations, D = 0 is set, e.g., if the input vector has no passage or instantaneous influence on the output vector in the controlled system under consideration.
[0095] For an asymptotically stable coating process, the final values of the outputs for a constant input u(t) = u can be described using the static matrix, which is: Ks={D}−CA−1B
[0096] Then K applies s u = y(∞).
[0097] The system state x(t) is determined via the drive vector within the framework of state control. u_(t)=−Kx_(t) The data is fed back via a feedback control element K (e.g., at least one controller matrix), which correlates with the term state feedback. The observability and controllability of the coating process facilitate this. The feedback control element K intuitively influences the eigenvalues and thus the dynamics of the overall system (i.e., the control loop).
[0098] The feedback control element K can, for example, have one or more control elements, e.g., two or more control elements, or three or more control elements, as will be described in more detail later. It should be understood that the feedback control element K is exemplary and that the underlying relationships can also be formulated differently. The same applies to other control elements described herein.
[0099] For example, more or fewer than three control elements can be used. For example, two or more control elements can be combined, for example, by means of a transformation into a higher-dimensional space. For example, a control element can be split, for example, by means of a transformation into two lower-dimensional spaces.
[0100] The feedback control element K can be determined using experimental and / or numerical processes, as will be described in more detail later.
[0101] Fig. Figure 3 illustrates the method 100 according to various embodiments in a detailed schematic control diagram 300, which illustrates a state feedback with a pre-filter.
[0102] The pre-filter V can implement the application of setpoint values or setpoint states w(t), which can be written as: u_(t)=−Kx_(t)+Vw_(t)
[0103] This results in static feedforward control and ensures a corresponding setpoint sequence. The homogeneous solution of the differential equation system leads, for example, to x(t → ∞) = 0. The setpoint feedforward, via the particular solution of the differential equation system, provides the transition x(t → ∞) = w without changing the dynamics or the eigenvalues of the control loop.
[0104] The control loop (i.e., comprising the controlled system and the controller) can then be formulated as: x˙_(t)=(A−BK)x_(t)+BVw_(t){+Ed(t)}
[0105] Feedforward control using a pre-filter can result in disturbances or parameter deviations being suppressed only minimally or not at all. Alternatively or additionally to feedforward control, a more flexible PI component can be used in the state feedback. For this purpose, a vector for the control deviations can be used, for example, formulated as... x˙_r(t)=y_(t)−w_(t)=Cx_(t)−w_(t), The equations of motion may be introduced as an additional variable. The resulting system of differential equations can also be formulated as an integral system, which can be written, for example, as: (x˙_(t)x˙_r(t))=(A0C0)(x_(t)x_r(t))+(B0)u_(t)+(0−1)w_(t){+(E0)d_(t)},y_(t)=(C 0)x_(t)⇔x˙I(t)=AIx_l(t)+BIu_(t)+(0−1)w_(t){+(E0)d_(t)},y_(t)=CIx_(t)
[0106] The new variables can be intuitively used as integrators of the setpoint deviation and implemented in the controller using an integral gain K. Ican be traced back to a regular element, which can be written as: u_(t)=−Kx_(t)−Klx_r(t).
[0107] An additional proportional (P) term (which can also be omitted) can improve the dynamic properties of the control loop and introduces design freedom into the controller. The additional P term (if present) optionally provides the effect of static feedforward control, which can be written as: u_(t)=−Kx_(t)−KP(Cx_(t)−w_)−Klx_r(t)
[0108] The interpretation is then carried out using the extended system matrix A. I , the extended input matrix B I and the extended output matrix C I .
[0109] Fig. Figure 4 illustrates the method 100 according to various embodiments in a detailed schematic control diagram 400 (for PI output control), which illustrates a control loop that has an I-extended path and output feedback (and optionally no direct feedback of the output).
[0110] The transition to an output controller (note y(t) = Cx(t)) significantly simplifies the implementation and technical realization of the controller. This transition to an output controller is possible if the controlled system is intuitively benign, i.e., it meets predefined criteria regarding controllability and observability, and C is invertible.
[0111] In this case, the output vector y(t) can be directly fed back (e.g., instead of the system state x(t)), which can be written as: u_(t)=−Kyy_(t)−KP(y_(t)−w_)−Klx_r(t)
[0112] The feedback control element K (e.g., a feedback matrix) can optionally be directly connected to the feedback control element K. y the output vector can be linked, e.g. according to the relation K y = KC -1 .
[0113] A correspondingly extended control element K Integral (e.g., a controller matrix) directly contains the control elements K y , K P and K I , which can be written as: u_(t)=−Kyy_(t)−KP(y_(t)−w_)−KIx_r(t)=−KyCx_(t)−KP(Cx_(t)−w_)−Klx_r(t) =−(K1 K2)︸KIntegral(x_x_r)︸x˙_Integral−(0KP)w_ =−KIntegralx˙_Integral−(0KP)w_ with K1=(Ky+KP)C, and K2=KI.
[0114] The state feedback K Integral The control element of the extended system can be determined using a numerical algorithm (e.g., using pole assignment or a linear-quadratic controller).
[0115] The degrees of freedom of the control elements K y and K P(e.g., controller matrices) can be used to modify the controller. For example, the component K can be y In the case of an asymptotically stable coating process, it should be chosen such that it completely determines the input in the steady state. The accumulated control deviation x r The I-component then only compensates for disturbances and parameter deviations and can, in the best case, even disappear.
[0116] For the calculation of K y For w = constant (e.g., in a steady state), the following relations can apply: y_(∞)=Cx_(∞)=w_⇒x_(∞)=C−1w_ x_˙(∞)=Ax_(∞)+Bu_(∞)=0_ ⇒α) u_(∞)=−B−1Ax_(∞)=−B−1AC−1w_ u_(∞)=−Kyy_(∞)−KP(y_(∞)−w_)−KIx_r(∞)=−Kyw_
[0117] A comparison of b) and c) yields: Ky=B−1AC−1. K P , K I can be determined based on K1, K2, e.g. using KP=K1C−1−Ky,KI=K2.
[0118] It should be noted at this point that this election is equivalent to Ky=Ks−1 is.
[0119] The following section discusses the determination of the control elements (e.g., in the form of controller matrices), whereby what is described can again be understood as an exemplary implementation.
[0120] Fig. Figure 5 illustrates the eigenvalues of the coating process controlled by method 100 according to various embodiments in a schematic diagram 500, where the area for the eigenvalues is illustrated in the complex vector space.
[0121] As a first example of numerically determining a control element, a pole assignment is explained below.
[0122] The pole assignment can be based on a model of the controlled system, e.g. having the model elements (A,B,C).
[0123] The model of the controlled system can, for example, implement linearization around an operating point, as described above. The eigenvalues, and thus the dynamics of the control loop, are predefined. The superposition of all eigenfunctions then intuitively describes the dynamics of the control loop. The imaginary part (Im) of an eigenvalue generates an oscillation, the real part (Re) a decay (for example, the negative real part) or a growth (for example, the positive real part).
[0124] The eigenvalues are given in complex conjugate pairs and can, for example, be chosen to be real and negative (corresponding to stabilizing dynamics without oscillations). This might, for instance, only affect the use of pole assignment. The magnitudes of the eigenvalues depend on the desired dynamics of the control loop (e.g., rapid attainment of setpoints with physically / technically meaningful responses of the manipulated variables). Suitable eigenvalues for input can be more easily determined using simulation. Additionally, one or more numerical algorithms can be used for pole assignment (and thus for determining the feedback matrix of the control loop), e.g., using Python, SciPy, or Matlab extended with the System Control Toolbox.
[0125] If the eigenvalues for pole selection are difficult to predict (e.g., due to lack of experience), optimization for linear-quadratic controllers (also known as LQR optimization) can be performed, as described below.
[0126] As a second example of numerically determining a control element, an LQR optimization (LQR = “Linear-Quadratic Regulator”) is explained below.
[0127] LQR optimization can be based on a model of the controlled system (A,B,C). Intuitively, LQR optimization determines the most optimal control system possible based on solving an optimization problem. This is achieved by minimizing the quadratic performance functional. J=∫0∞[x_T(t)Qx_(t)+u_T(t)Ru_(t)]dt The feedback control element K is determined for the feedback u(t) = -Kx(t) of the control loop. This corresponds to the solution of a Riccati equation. The faster the controlled system is driven to the equilibrium position x = 0, the less contribution the first term x makes. T (t) Q x(t) of the functional. The second summand u T (t) R u(t) intuitively considers the effort of the drive vector u(t) required to drive the system appropriately. The symmetric, positive semidefinite weighting matrix Q and the symmetric, positive definite weighting matrix R determine the performance requirements and how the closed control loop behaves.
[0128] However, choosing the weighting matrices Q and R can be very challenging. An exemplary, yet simple and meaningful, choice of Q can be expressed as Q=CTQyC with Qy=diag(q1,…,qN) This expresses the quality functional with respect to the behavior of the components of the output vector y, which can be written as: J=∫0∞[x_T(t)CTQyCx_(t)+u_T(t)Ru_(t)]dt=∫0∞[(Cx_(t))Tdiag(q1,…,qN)(Cx_(t))+u_T(t)Ru_(t)]dt =∫0∞[∑i=1Nqiyi2(t)+u_T(t)Ru_(t)]dt.
[0129] For example, a boundary condition for optimization may be that there is no slight variation in the manipulated variables while reaching the target values of the output variables as quickly as possible.
[0130] One or more numerical algorithms can be used for LQR optimization (and thus for determining the feedback matrix of the control loop). These algorithms are available in various programming languages, such as Python with the SciPy library or Matlab with the System Control Toolbox. The numerical algorithm for LQR optimization can be called, for example, using an LQR function `LQR("argument")`, which returns the result of the LQR optimization of the argument. This results in K = LQR(Q,R,A,B).
[0131] Optionally, one of the expressions A - α1 (also called shift using the factor α or simply shift) can be used instead of the system matrix A in the context of the continuous-time LQR algorithm, which can then be written as: K=LQR(Q,R,A−α1,B).
[0132] This achieves the requirement of a distance between the eigenvalues and the imaginary axis, which improves stability. This shift approach can, for example, be applied analogously to the sampling system.
[0133] As a first example of an experimental determination of a control element, the determination of a static control element (also called a static element) is explained below (e.g. in the form of a static matrix).
[0134] Experimentally determining the control element does not necessarily require a model of the controlled system (A, B, C). Conversely, the experimental determination of at least one control element is designed to ensure the setpoint sequence and stability. The dynamic behavior can only be influenced imprecisely or not at all.
[0135] Two parameters (also called tuning parameters) are introduced to define at least one control element, and their values are determined experimentally. This can be demonstrated particularly well in the case of an asymptotically stable coating process. If this is not the case, the resulting final values of the output in the steady state can be measured only with difficulty or not at all.
[0136] The structural member K S is obtained through a set of individual experiments. Constant, linearly independent input vectors can be used for this purpose. i for the controlled system, whereby after the coating process has settled, the output vector y i (t = ∞) is measured. For this, a time T is used. meas (also known as settling time) predefined, after which the coating device is considered to be settled, i.e. that y i (t = T meas ) ≈ y i(t = ∞) holds true. The input and output vectors are subsequently determined using the matrices. U=(u_1 u_2…u_n),andY=(y_1(∞)y_2(∞)…y_n(∞)) formally summarized to simplify the notation. In the simplest case, for each change u i Only one component of the input vector needs to be set to a non-zero value, which is also useful for applying automatic calibration (e.g., using a so-called autotuning algorithm), which will be described in more detail later. The static element can then be determined based on the input and output vectors, for example, according to the following relation: Ks=YU−1
[0137] Based on the static element (e.g. its inverse), one or more than one control element (e.g. in the form of a controller matrix) can be determined, for example according to one or more than one of the following relations. Ky=0,KP=bKs−1,K1=aKs−1 with the factors (also called tuning factors) a and b. It is helpful to start with small tuning factors to simplify the process. The tuning factors can be increased to accelerate the settling-in period, as long as the coating process still has a stability reserve and does not oscillate.
[0138] A PI output controller can be formed based on one or more control elements.
[0139] As explained above regarding the PI initial rule, namely regarding the choice of K y , can be used instead of K y = 0 here too Ky=Ks−1 This ensures that when starting procedure 100 (e.g., with the control deviation integrals set to zero), the input is not also set to zero at the start, but is started at a logically meaningful state.
[0140] Fig. Figure 6 illustrates an implementation 600 of the method according to various embodiments in a detailed schematic diagram, which is used to explain, among other things, the discretization (shown here as an example for a discrete PI output controller).
[0141] Discretizing the controller allows one or more sampling effects to be taken into account, e.g., taking into account that the actual states of the controlled variables are determined discretely in time (i.e., determined with a repetition frequency).
[0142] The rectangular (piecewise constant) behavior of the input can be exploited to obtain a discrete mapping for the time evolution of the system state. Discretization can be achieved, for example, by transforming the time-continuous vectors ẋ(t) and y(t) into time-discrete vectors x. k and y k, for example according to the following relation: x˙_(t)=Ax_(t)+Bu_(t)y_(t)=Cx_(t)→sampling time T,t=ktx_k+1=Adx_k+Bdu_ky_k=Cx_k where k is an integer.
[0143] The concrete transformation of the time-continuous vectors ẋ(t) and y(t) into the time-discrete vectors ẋ k and y k This is based on the value of the sampling time T. The sampling time T indicates the time interval between immediately successive points in time at which one or more controlled variables of the coating process are recorded, i.e., at which the coating process is sampled.
[0144] From the analytical solution of the continuous ODE system, using a piecewise constant function of the drive vector u(t), the relations between the model elements (e.g., in the form of model matrices) of the discrete-time dynamic control system can be derived, which can be written as: Ad=eAT,Bd=∫0TeAτdτB
[0145] The matrix exponential function can be calculated, for example, by diagonalizing A = S diag(λ1, ..., λ). n )S -1 This can be done. The discrete-time model elements (e.g., in the form of model matrices) can be expressed as: Ad=eAT=eS diag(λ1,…,λn)S−1T=S ediag(λ1,…,λn)TS−1=S diag(eλ1T,…,eλnT)S−1, Bd=∫0TeAτdτ B=∫0TS diag(eλ1τ,...,eλnτ)S−1dτB =S diag(∫0Teλ1τdτ..,∫0Teλnτdτ)S−1B =S diag(1λ1(eλ1T−1)...,1λn(eλnT−1))S−1B.
[0146] In simpler terms, the numerical implementation of matrix exponentials and integrals is considerably simpler in this form. For example, the expressions in the last few lines can be implemented more easily, or even at all, using software.
[0147] The extension to a discrete-time integral system can be expressed as: (xk+1xr,k+1)=(Ad0C1)(x_kxr,k)+(Bd 0)u_k+(0−1)w_k where the control deviation xr,k+1 = x r,k + y k - w k directly accumulated. The design of a state controller for discrete-time sampling of the coating process (also referred to as a discrete-time control loop) is identical to that of a continuous-time control loop, because: • One or more pole selection algorithms can be used for both continuous-time and discrete-time control systems, since the poles of the sampling system are related to those of the continuous-time control system according to λ. d = e λT ; and • one or more LQR algorithms for sampling systems are known, but do not have to be identical to those of a continuous-time control system.
[0148] Regarding the determination of a discrete-time PI output controller, it should be noted that K yfrom the PI output control described above for the continuous-time control loop, because it is: y_∞=Cy_∞=w_ x_∞=Adx_∞+Bdu_∞=0_ ⇒a) u_∞=−Bd−1(Ad−1)x_∞=−Bd−1(Ad−1)C−1w_ u_∞=Kyy_∞−KP(y_∞−w_)−KIx_r,∞=−Kyw_
[0149] After comparing b) and c) and determining the remaining control elements with respect to K1, K2, the following results: Ky=Bd−1(Ad−1)C−1KP=K1C−1−Ky, and KI=K2.
[0150] Regarding the application of the LQR algorithm, it should be noted that there are differences in solving a continuous-time or a discrete-time Riccati equation, which can be taken into account during numerical implementation or when using appropriate libraries. Furthermore, the application of a predefined degree of stability α differs compared to a continuous-time control system. The generalization to a discrete-time control system is explained below, analogous to the continuous-time control system.
[0151] New sizes are being introduced, namely: x˜_k=eαtx_k,u˜_k=eαtu_k
[0152] In order for the quality functional to assume a finite value, the original quantities x k and u k faster than e -αtThe data falls off. For this, a suitable discrete-time sampling method for the newly introduced quantities is determined from the existing sampling. The multiplication of the discrete-time state vector... x_k+1=Adx_k+Bdu_k with e αkT delivers eαkTx_k+1=e−αTaα(k+1)Tx_k+1︸x˜_k+1=AdeαkTx_k︸x˜_k+BdeαkTu_k︸u˜_k and leads to x˜_k+1=AdeαT︸A˜dx˜_k+BdeαT︸B˜du_˜k.
[0153] Calling the LQR function with LQR(Q, R, A) d e αT , B d e αT ) instead of LQR(Q, R, A d , B d ) then provides the consideration of the required degree of stability analogous to the time-continuous control loop.
[0154] More generally, according to various embodiments, the method provides the modeling of a continuous-time control loop that, for example, takes into account a predefined degree of stability. This results in a significantly more robust controller.
[0155] Fig. Figure 7 illustrates the controlled system 700 according to various embodiments in a detailed schematic diagram, on the basis of which, among other things, an exemplary model of the controlled system (also referred to simply as the process model) is subsequently explained.
[0156] The exemplary process model can facilitate the identification of one or more control elements and / or the execution of numerical simulations. The exemplary process model is explained below using an exemplary coating process that includes a sputtering process.
[0157] The sputtering process involves atomizing a material (also called target material) using a plasma (a process known as sputtering). This atomized target material can then be used to form a layer on a substrate. In other words, the sputtering process involves transforming the material (target material) used for layer formation from the solid phase into the gaseous phase through atomization.
[0158] It can be understood that the sputtering process is only one example of physical vapor deposition (PVD), and that what is described for the sputtering process can apply analogously to other types of PVD where the solid material is converted into a gaseous material in a different way. Examples of this include: electron beam evaporation, thermal evaporation, arc evaporation, molecular beam epitaxy, and ion plating.
[0159] For example, several coating processes can be carried out, each of which deposits at least one layer on the substrate and is to be controlled according to Procedure 100. One or more of the multiple coating processes can, for example, include the sputtering process. Immediately successive coating processes (also referred to as immediately adjacent coating processes) can optionally interact with each other, e.g., by exchanging gases.
[0160] The state vector x, the input vector u, and the output vector y are transformed into one another using the model elements (A, B, C), which represent the controlled system. Analogous to the vectors, the model elements (A, B, C) are denoted as elements of a vector space for the following explanation. The explanation can, of course, be applied analogously in other notations.
[0161] The combination of state vector x, input vector u, and output vector y can take the following form (also called model form): x˙_(t)=Ax_(t)+Bu_(t), undy_(t)=Cx_(t)+{D u_(t)}. A denotes the system matrix, B the input matrix, and C the output matrix. The operator D is omitted in this model but can optionally be included.
[0162] The input vector u represents the current state of the manipulated variable of the coating process, the output vector y represents the current state of the controlled variable of the coating process, and the state vector x represents the current state of the interaction chain of the coating process (i.e. the controlled system), which transforms the input vector u into the output vector y.
[0163] To simplify understanding, the vector components are subsequently labeled R and P. The indices R and P intuitively represent the supply to the coating process. For example, P can represent a first gas, e.g., its flow rate (e.g., standard volumetric flow rate), with which the coating process is supplied. The first gas can, for example, be a working gas or be composed of one. The corresponding first actuator can then, for example, be a first valve by means of which the flow rate of the first gas is controlled. The first target control variable can then be transmitted by means of a first control signal supplied to the valve. The working gas can, for example, be argon or another noble gas or be composed of one.
[0164] For example, R can represent a second gas, such as its inflow rate (e.g., standard volumetric flow rate), which supplies the coating process. This second gas can, for example, be a reactive gas or be composed of one. The corresponding second actuator can then, for example, have a second valve by which the inflow rate of the second gas is controlled. The second target control variable can then be transmitted by means of a second control signal supplied to the second valve. The reactive gas can, for example, be molecular oxygen (O2), molecular nitrogen (N2), or a mixture of O2 and N2, or be composed of these.
[0165] It can be understood that the described process can also apply analogously to more than two gases and / or another gaseous material used to supply the coating process. For example, the gaseous material can also contain the atomized target material (e.g., the target material converted into the gas phase), thus influencing the atomization rate. The corresponding actuator can then, for example, include a voltage supply to provide the plasma with electrical energy.
[0166] In the following, a gas quantity model is used as an approximation for the internal state vector and for decoupled gas reservoirs for each gas type. It should be understood that the corresponding mathematical description is exemplary and was chosen for the sake of simplified understanding. Of course, a different mathematical description of the approximation may be implemented or implemented in the future.
[0167] The state vector x, input vector u, and output vector y can be defined with respect to a specified operating point. The closer the operating point is to the target operating range (e.g., the reference state), the smaller the losses due to the approximation model can be. For simplified illustration, an (N=1) model (also known as a single-chamber model or single-coating-process model) is first explained as an exemplary process model, which is then systematically extended to an N-model, i.e., N chambers or N coating processes, as an exemplary process model.
[0168] Fig. Figure 8 illustrates the (N=1) model 800 as an exemplary process model according to various embodiments in a detailed schematic diagram. The (N=1) model 800 can represent exactly one coating process, which is supplied by the first actuator 801a and the second actuator 801b, is carried out within a vacuum chamber 802 (also referred to as a processing chamber), and coats a substrate 102. The actual state 211a of the first controlled variable of the coating process can be detected by a first sensor 803a. The actual state 211b of the second controlled variable of the coating process can be detected by a second sensor 803b.
[0169] According to various embodiments, a vacuum system by means of which the coating process is carried out can have a vacuum chamber housing in which a vacuum can be generated and / or maintained. The coating of the substrate 102 can, according to various embodiments, take place in the vacuum or the vacuum chamber housing. For this purpose, the vacuum chamber housing can be, for example, airtight, dustproof, and / or vacuum-tight. The vacuum chamber housing can have one or more vacuum chambers, whereby each coating process can be carried out in a separate vacuum chamber, but this is not necessarily required. For example, two or more coating processes can also be carried out in the same vacuum chamber or at least be subject to the same process atmosphere and / or the same pressure.
[0170] Furthermore, the vacuum chamber 802, or each chamber, can be coupled to a pumping system (comprising at least one rough vacuum pump and optionally at least one high vacuum pump) which can provide one or more actuators (for example, the first or second) that supply the coating process. The pumping system can be configured to extract gas from the vacuum chamber, e.g., the coating process carried out therein, so that a vacuum (i.e., a pressure less than 0.3 bar) can be provided or maintained therein, e.g., a pressure in the range of approximately 1 mbar to approximately 10 -3 mbar (in other words, fine vacuum) or less, e.g., a pressure in the range of approximately 10 -3 mbar to approximately 10 -7 mbar (in other words, high vacuum) or less, e.g., a pressure of less than high vacuum, e.g., less than approximately 10 -7 mbar (in other words, ultra-high vacuum).
[0171] Furthermore, each vacuum chamber 802 can be coupled to a gas supply system, which can provide one or more actuators (e.g., the first or second actuator) that supply the coating process. The gas supply system allows one or more gases to be supplied to the vacuum chamber 802, e.g., to the coating process carried out therein, to form a process atmosphere. Depending on the coating process to be performed, the process atmosphere can have a corresponding composition and / or pressure. Examples of the gases to be supplied include a working gas (e.g., an inert gas) and a reactive gas. The pressure to which the coating process is subjected can be determined by an equilibrium of gas supplied by the gas supply system and extracted by the pumping system, i.e., supplied to the coating process.
[0172] Furthermore, a coating device can be arranged in or in each vacuum chamber 802, which can provide one or more (for example, the first or second) actuators that supply the coating process. By means of the coating device, one or more target materials can be supplied to the vacuum chamber 802, e.g., to the coating process carried out therein, for the formation of a coating. The coating device can, for example, include or be formed from a sputtering device.
[0173] In a first exemplary implementation, the (N=1) model 800 considers a process voltage, which supplies the coating process, as an exemplary first control variable and a total pressure to which the coating process is exposed as an exemplary second control variable. The process voltage can, for example, supply the coating device with electrical power. The process voltage can, for example, influence the rate at which the target material is converted into the gaseous state (e.g., vaporized or atomized).
[0174] In a second exemplary implementation, the (N=1) model 800 considers the supply of working gas to the coating process as an exemplary first control variable and the supply of reactive gas to the coating process as an exemplary second control variable.
[0175] In a third exemplary implementation, the (N=1) model 800 takes into account the parallel control of the process voltage U as the first controlled variable and the total pressure p as the second controlled variable in a process chamber 802 and the supply of the coating process with working gas and reactive gas as manipulated variables.
[0176] In general, one or more additional controlled variables can also be considered, for example by extending the state vector x, the input vector u, and the output vector y by one dimension. For each controlled variable, a gas inlet valve (e.g., reactive gas inlet) can be used as an actuator, and one or more sensors configured to determine the ratio between several gases (e.g., reactive gases) can be used (e.g., with respect to their partial pressure and / or flow rate).
[0177] The state vector x represents a difference in the amount of gas of the type used at an unknown operating point n. P , n R , which can be written as: x_=(ΔnPΔnR)
[0178] The output vector y represents a difference between the measured process parameters (also called controlled variables) and a defined operating point, which can be written as: y_=(ΔUΔp) with operating point y¯_=(U¯p¯) and measured variables y˜_=(U˜p˜˜)=y¯_+y_
[0179] The following section describes an exemplary implementation of the initial matrix C in the (N=1) model. This exemplary implementation is derived analytically, although other approaches could also be used, such as employing a different analytical method or fitting model parameters.
[0180] The conversion of gas quantities to partial pressures is done using a factor c. p, which is volume-dependent and temperature-dependent for an ideal gas. Is the factor c p Optionally dependent on the type of gas, individual factors may apply. cPp and cRp The partial pressure differences sum to the total pressure difference, which can be written as: Δp=ΔpA+ΔpR=cPpΔnA+cRpΔnR.
[0181] The approximation that the partial pressure change of a gas causes a proportional, gas-type-dependent stress change to a first order, which can be written as: ΔU=cPUΔpA+cRUΔpR=cPUcPpΔnA+cRUcRpΔnR where CPU and cRU These are proportional factors. This can be summarized as: y_=(ΔUΔp)=(cPUcRU11)(ΔpPΔpR)=(cPUcPpcRUcRpcPpcRp)(ΔnPΔnR)=Cx_
[0182] Based on this, the initial matrix C can be expressed as: C=(cPUcPpcRUcRpcPpcRp).
[0183] The following explains a possible implementation of the input vector u in the (N=1) model. The input vector u represents, for example, the material flows (e.g., gas flows) into vacuum chamber 802, i.e., the material flows supplied to the coating process, as differences from the corresponding initial flows (which could be the material flows in the reference state), which is expressed by: u_=u˜_−u¯_=(ΔvPΔvR)=(working gas MFC−inlet to chamberReactive gas MFC−inlet to chamber)−(working gas MFC−initial flowReactive gas MFC−initial flow),
[0184] The following explains a possible implementation of the input matrix B in the (N=1) model. The input vector u represents a quantity of gas per unit time, which can be written as follows: B=(b00b). a diagonal matrix. The use of standard units (e.g., standard cubic centimeters per minute, i.e., sccm) makes this implementation of B plant-independent and internationally understandable. For a gas type (e.g., the reactive gas or the working gas), one can then write: dxdt=bv=bdndt with the constant gas flow v. The coefficient b only accounts for the conversion of the time unit and can optionally be normalized or omitted. Within the framework of the process model, time can, for example, be specified in seconds (sccm become standard cubic centimeters per second, i.e., sccs), so that b = 1 / 60. Then it can be written that: B=(b00b)
[0185] The following explains a possible implementation of the system matrix A in the (N=1) model. The approximation that the gas reservoirs for working gas and reactive gas do not interact with each other to a first order, or at most only negligibly, can be expressed as a diagonal matrix, for example as: A=(−alocP00−alocR) where the positive parameters a loc represent the decay of the gas quantity through the coating process and the pumping out (i.e., the removal of gas).
[0186] It should be noted here that, alternatively or in addition to the material supplied to the coating process, the material removed from the coating process can also be considered. This is because the amount of material available to the coating process is determined based on the equilibrium between supply and removal. This equilibrium is generally referred to here as "supply." In other words, supplying the coating process with material can involve both adding and / or removing material. The supply and / or removal can be controlled, for example, by means of a valve.
[0187] The diagonal system matrix A leads to decoupling into individual differential equations; the differential equation for each type of gas can then be expressed as: x˙(t)=−alocx_(t)+bu(t), and is analytically solvable for a constant inflow u(t) = u, which can then be written as: x(t)=(ualoc(ealoc t−1)+x(0))e−aloc t.
[0188] From this follows directly the dynamic total pressure change Δp(t) with respect to a gas type, which can be expressed as: Δp(t) = c p x(t).
[0189] To determine the model elements, the structure of this relation can be exploited. A deflection of the input vector results in a deflection of the coating process state, which can be detected. An exemplary jump deflection of the input vector can be expressed as: u(t)={0,t<0Δu,t≥0.
[0190] With a steady-state solution at time t = 0, Δp(0) = c p When x(0) = 0, a pressure difference Δp(∞) arises after the abrupt displacement. The unknown decay coefficient aloc=−ln(1 / 2)τ can be determined experimentally using the half-life τ with respect to Δp(∞), which can be calculated from the pressure signal p(t) = p + Δp(t). Based on the decay coefficient a loc Can a conversion factor be determined with respect to the pressure of the gas under consideration, for example according to the following relationship: cp=Δp(∞)Δualoc
[0191] By analogy, a conversion factor regarding the voltage can be determined based on the resulting difference in the process voltage ΔU(∞), for example according to the relation: cU=ΔU(∞)ΔP(∞)
[0192] The (N=1) model is subsequently extended to an (N=2) model (also known as a two-chamber model or two-coating process model).
[0193] Fig. Figure 9 illustrates the (N=2) model 900 as an exemplary process model according to various embodiments in a detailed schematic diagram. The (N=2) model 900 can represent several coating processes, each of which (e.g., the nth coating process) is supplied by a first actuator 801a, 901a and a second actuator 801b, 901b, is carried out within a vacuum chamber 802 (also referred to as a processing chamber), and coats a substrate 102.
[0194] Adjacent coating processes can interact with each other, e.g. by exchanging gas through an opening 902 (also called substrate transfer opening) through which the substrate is transported between two vacuum chambers 802.
[0195] Subsequently, index 1 refers to a first coating process or a first vacuum chamber in which the first coating process is carried out. Similarly, index 2 refers to a second coating process or a second vacuum chamber in which the second coating process is carried out.
[0196] The generalization of the (N=1) model to the (N=2) model indicates that the interaction between the two coating processes is taken into account by means of the system matrix.
[0197] The state vector x can initially be expressed as: x_=(ΔnP1ΔnP2ΔnR1ΔnR1)
[0198] By analogy, the initial vector y can be expressed as: y_−(ΔU1ΔU2Δp1Δp2) with the working point y_¯=(U¯1U¯2p¯1p¯2) and the measured variables y˜_=(U˜1U˜2p˜1p˜2)=y¯_+y_.
[0199] The initial matrix C can then be expressed as: y_=(ΔU1ΔU2Δp1Δp2)=(cP1UΔpP1+cR1UΔpR1cP2UΔpP2+cR2UΔpR2ΔpP1+ΔpR1ΔpP2+ΔpR2)=(cP 1UcP1p0cR1UcR1p00cP2UcP2p0cR2UcR2pcP1p0cR1p00cP2p0cR2p)(ΔnP1ΔnP2ΔnR1ΔnR2)=Cx_
[0200] The entries in the initial matrix C that take on the value zero, shown in the top two rows, illustrate that partial pressures or gas quantities within a vacuum chamber have no influence on the process stress in the immediately adjacent vacuum chamber(s). Similarly, the entries that take on the value zero in the bottom two rows illustrate that partial pressures or gas quantities within one vacuum chamber have no influence on the total pressures of the immediately adjacent vacuum chamber(s). Written together, this is: C=(cP1UcP1p0cR1UcR1p00cP2UcP2p0cR2UcR2pcP1p0cR1p00cP2p0cR2p).
[0201] The input vector u can then be expressed as: u_=(ΔvP1ΔvP2ΔvR1ΔvR2)
[0202] The input matrix B can then be expressed as: B=(b0000b0000b0000b).
[0203] The following describes a possible implementation of the system matrix A in the (N=2) model. As the first system component, the system matrix A represents the dynamics of each of the two coating processes or within each of the two vacuum chambers 802. Additionally, as the second system component, the system matrix A represents the interaction between the two coating processes or vacuum chambers 802.
[0204] Fundamentally, the two system components are superimposed. To a first approximation, the first and second system components can be superpositioned, which is even exact in the case of linearization. This can be expressed as: A=Alocal+Aexchange. A localIt clearly considers the chain of interactions within the coating process, i.e., clearly and locally (within a single chamber, as in the single-chamber model). The effective parameters αlocPi and αlocRi They take into account the pumping and gas consumption, for example, and also consider the overall impact of adjacent coating processes.
[0205] A local This can be expressed as: Alocal=(−alocP10000−alocP20000−alocR10000−alocR2). A exchange takes into account the interaction between the two coating processes or between the two vacuum chambers. A exchangeThis can be understood as an exemplary implementation of considering the interaction, meaning that this consideration of the interaction can also be done differently. Alternatively or additionally, the interaction between several coating processes can be considered in all control elements.
[0206] The exchange of gas (e.g., continuous gas flow) from the first coating process to the second coating process and vice versa can be expressed as: Aexchange=(−a12Pa21P00a12P−a21P0000−a12Ra21R00a12R−a21R) with the guide values aijP,aijR. The conductance values clearly illustrate how efficiently gas exchange occurs, taking into account both different gases and different flow directions. Entries that are zero illustrate that the flow of a gas of one type (also referred to as gas species) only causes a change in the partial pressure of gas of the same type and has no effect on the partial pressure of a gas of another type.
[0207] In concise terms, this can be expressed as: A=(−alocP1−a12Pa21P00a12P−alocP2−a21P0000−alocR1−a12Ra21R00a12R−alocR2−a21R)
[0208] Fig. Figure 10 illustrates the (N=2) model 900 according to various embodiments in a detailed schematic diagram 1000, in which the components of the system matrix A are shown. The (N=2) model is subsequently extended to an (N=3) model (also known as a three-chamber model or three-coating process model).
[0209] Fig. Figure 11 illustrates an (N=3) model 1100 as an exemplary process model according to different embodiments in two detailed schematic diagrams, in which the components of the system matrix A are again shown.
[0210] Generalizing the (N=2) model to the (N=3) model does not necessarily require consideration of additional influences / interactions. However, the shape of the system matrix changes slightly, since gas exchange can occur between the coating process adjacent to any two other coating processes and each of those two processes.
[0211] The correspondingly extended state vector, extended output vector, extended input vector, extended input matrix, and extended output matrix can then be expressed analogously to the above as: x_=(ΔnP1ΔnP2ΔnP3ΔnR1ΔnR2ΔnR3) C=(cP1UcP1p00cR1UcR1p000cP2UcP2p00cR2UcR2p000cP3UcP3p00cR3UcR3pcP1p00cR1p000cP2p00cR2p000cP3p00cR3p), y_=(ΔU1ΔU2ΔU3Δp1Δp2Δp3) with working point y_¯=(U¯1U¯2U¯3p¯1p¯2p¯3) and measured variables y_˜=(U˜1U˜2U˜3p˜1p˜2p˜3) where y_˜=y_¯+y_, is, u_=(ΔνP1ΔνP2ΔνP3ΔνR1ΔνR2ΔνR3), B=(b000000b000000b000000b000000b000000b).
[0212] The following is an exemplary implementation of the system matrix A in the (N=3) model, which takes into account the dynamics and interaction between the coating processes. This again has the following form: A=Alocal+Aexchange. A local This can also be expanded, which can be expressed as: Alocal=(−alocP1000000−alocP2000000−alocP3000000−alocR1000000−alocR2000000−alocR3)
[0213] The matrix components of A localare illustrated in the diagram above.
[0214] A exchange takes into account gas flows from an i-th vacuum chamber (also referred to as chamber i) into the adjacent vacuum chambers i - 1 and i + 1 and vice versa, Aexchange=(−a12Pa21P0000a12P−a21P−a23Pa32P0000a23P−a32P000000−a12Rα21R0000a12R−a21R−a23Ra32R0000a23P−a32R).
[0215] In this case, A can exchange Optionally, as explained in the example above, it should be taken into account that only an interaction of the coating processes immediately adjacent to each other takes place (also called nearest-neighbor interaction), i.e., intuitively, that the gas flow does not skip a vacuum chamber.
[0216] The matrix components of A exchange are illustrated in the diagram below.
[0217] The system matrix A can then be expressed compactly as: (−alocP1−a12Pa21P0000a12P−alocP2−a21P−a23Pa32P0000a23P−alocP3−a32P000 000−alocR1−a12Ra21R0000a12R−alocP2−a21R−a23Ra32R0000a23R−alocR3−a32R)
[0218] In the same way, the process model can be extended to an N-model, which takes into account more than three coating processes or more than three vacuum chambers, e.g. a number N coating processes or vacuum chambers, as explained below.
[0219] Fig. Figure 12 illustrates the N-model 1200 as an exemplary process model according to various embodiments in a detailed schematic diagram.
[0220] To generalize the (N=3) model, each coating process, or the vacuum chamber in which the coating process is carried out, is numbered below. Visually, the substrate is coated by means of N (e.g., three, four, five, six, or more) coating processes, of which the nth coating process deposits the nth layer on the substrate (over the (n-1)th layer), where 1 ≤ n ≤ N. The number N of coating processes can be four or more, e.g., five or more, e.g., ten or more, e.g., 20 or more, e.g., 30 or more.
[0221] The above interactions are taken into account, so that only the dimension of the matrices / vectors increases according to the number of coating processes.
[0222] The state vector, the output vector, the input vector, the input matrix, and the output matrix can then be expressed as: x_=(ΔnP1…ΔnPNΔnR1…ΔnRN), C=(diag(cP1UcP1p,…,cPNpcPNp)diag(cR1UcR1p,…,cRNpcRNp)diag(cP1p,…,cPNp)diag(cR1p,…,cRNp)), y_=(ΔU1…ΔUNΔp1…ΔpN) with working point y_¯=(U¯1…U¯Np¯1…p¯N) and measured variables y_˜=(U˜1…U˜Np˜1…p˜N), where y_˜=y_¯+y_, is, u_=(ΔvP1⋯ΔvPNΔvR1⋯ΔvRN), B=diag(b,…,b)ist.
[0223] The following describes a possible implementation of the system matrix A in the N-model, which takes into account the dynamics and interaction between the coating processes. In block notation, the system matrix A can be expressed as: A=(AP00AR) where the matrix components of the block matrix A P express as: (Ai,jP)i=1,…,N;j=1,…,N=[−alocPi−(1−δi,1)ai,i−1P−(1−δi,N)ai,i+1P]δi,j+aj,iPδi,j−1+aj,iPδi,j+1 and the matrix components of the block matrix A R express as: (Ai,jR)i=1,…,N;j=1,…,N=[−alocRi−(1−δi,1)ai,i−1R−(1−δi,N)ai,i+1R]δi,j+aj,iRδi,j−1+aj,iRδi,j+1
[0224] The expression δ i,j This refers to the so-called Kronecker Delta, which fulfills the following relation: δi,j={1,i=j0,i≠j.
[0225] Fig. Figure 13 illustrates a method 1300 according to various embodiments in a schematic flowchart. Method 100 can be used to calibrate a (e.g., reactive) coating process. Method 1300 can, for example, be used to calibrate a process group that performs one or more reactive coating processes.
[0226] Method 1300, as described in 1301, involves first determining the response of a first controlled variable and a second controlled variable of the coating process to a deflection of the coating process from a reference state when (e.g., exactly) a first actuator, by which the coating process is supplied, is activated. In other words, the deflection is caused during the first determination by activating the first actuator, by which the coating process is supplied.
[0227] Method 1300, as described in 1303, involves a second determination of the response of the first and second controlled variables of the coating process to a deviation of the coating process from the reference state when (exactly) a second actuator, by which the coating process is supplied, is activated. In other words, the second determination causes the deviation by activating the second actuator, by which the coating process is supplied.
[0228] The reference state can be such that the first controlled variable has a first state and the second controlled variable has a second state. The deflection of the coating process is intuitively determined starting from this same reference state. The deflection of the coating process can be understood as its response to an external stimulus in the form of the activation of the first or second actuator.
[0229] Whether the coating process is in the reference state can be determined, for example, by measuring the first and second controlled variables. Similarly, the response of the first and second controlled variables can be determined by measuring them.
[0230] The deviation of the coating process from the reference state can, for example, occur abruptly (in which case the response is called a step response). This can be achieved by abruptly changing the first and / or the second manipulated variable. It can be understood that what is explained here regarding the exemplary step response can also apply analogously to other (e.g., less abrupt and / or cyclic) deviations of the coating process from the reference state.
[0231] The step response vividly represents two pieces of information about the coating process: the half-life of the response and the convergence value of the response.
[0232] The first and second measurements can be performed sequentially, for example. This makes it easier to capture the corresponding deflection as a response, for instance, when multiple controlled variables interact with each other.
[0233] Procedure 1300 may optionally include the determination of the process model and / or the control model based on the response of the first controlled variable and the response of the second controlled variable. The control model can, for example, be determined based on the process model or directly from the determined response.
[0234] The rule model can then be used to execute procedure 100. For example, procedure 1300 can be used to create and / or update the rule model that is fed into procedure 100.
[0235] Method 1300 makes it easier to determine the control model for one or more than one coating process, e.g., one or more than one element of the control model (also referred to as the control element).
[0236] For example, the model elements (A, B, C), e.g., in the form of model matrices, can be determined wholly or partially experimentally, for instance, if no precise process model of one or more coating processes is available. Alternatively or additionally, the 1300 process can be at least partially automated (e.g., semi-automated), thus eliminating the need for highly qualified and / or experienced personnel. The 1300 process is demonstrably very robust and therefore very user-friendly, even if the operator makes minor errors or inaccuracies, or has little knowledge of the underlying interactions of one or more coating processes.
[0237] This also applies, of course, to highly qualified and / or experienced specialists, for example, when the high complexity of one or more coating processes makes it difficult to guess parameters of the control model, or when an adjustment calculation based on measurement data (e.g. using simplex algorithms) is very unstable and / or unreliable due to the high number of degrees of freedom / parameters.
[0238] In essence, the method 1300 achieves that a process model (also known as an (A,B,C) model) can be approximated with as little measured information as possible, or can be used as an autotuning algorithm.
[0239] For example, choosing the process model described herein allows B to be predefined from the outset, leaving only A and C to be determined. A combination of the experimental determination of K SThe measurement data and the existing model abstraction can be used to determine an approximation for C from the measurement data and for K. S to then immediately determine A, e.g. without requiring user intervention.
[0240] The following is an exemplary implementation of procedure 1300.
[0241] Fig. Figure 14 illustrates a temporal sequence according to the procedure 1300 in a schematic diagram 1400, in which the components of the input vector u(t), e.g. their current state, and the components of the output vector y¯_ e.g. their current state, shown over time t.
[0242] For easier understanding, reference is made again to the relation explained above, which defines the control elements K. y , K P , and K I in relation to each other and to the input vector: u_=−Kyy_(t)−Kp(y_(t)−w_)−Klx_r(t)
[0243] It can be understood that the three control elements K y , K P , and K I These are examples only, and the underlying relationships can also be formulated differently. For instance, more or fewer than three control elements can be used. For example, two or more control elements can be combined, for example, by means of a transformation into a higher-dimensional space. For example, a control element can be split, for example, by means of a transformation into a lower-dimensional space. For example, state feedback K can be used as an alternative or in addition to output feedback K. y be used.
[0244] Method 1300 features bringing the coating process to a reference state in each of phases 1401, 1405, and 1409, for example, by controlling at least one actuator (e.g., the first actuator and / or the second actuator) that supplies the coating process. For example, the control of the at least one actuator can be regulated, e.g., until the actual output vector y¯_ a predefined criterion is met.
[0245] For example, the predefined criterion can be met if the actual output vector y¯_ essentially corresponds to a target output vector (which corresponds to a target operating point as a reference state), i.e., deviates from it as little as possible.
[0246] The target operating point can be determined, for example, by setting the inflow rate. u¯_ as a control variable for each material supplied to the coating process, so that the actual output vector y¯_ It is set implicitly. This can be done, for example, based on hysteresis data near the target operating point.
[0247] The following explains some parameters (also called calibration parameters) according to which the 1300 procedure can be carried out. These parameters clearly define the processing routine, e.g., measurement duration, measurement frequency, magnitude of deflection, etc.
[0248] As an example of a calibration parameter, a settling time T can be used. meas and a value for Δt meas be predefined. T meas is used to reach final values within an experiment / time window or to return to the operating point, with the sensors clocked at intervals of Δt. meas The measurement data will be read out. In other words, the measurement data will be read out with a frequency of 1 / Δt.meas provided. Example values may include: T meas = 90s,Δt meas = 200ms.
[0249] A percentage can be used as an example of a measurement parameter. pTmeasavg the settling time T meas predefined within which the recorded measurement data for the last part of the time window T meas The values are averaged. Averaging is optional and can be used, for example, to compensate for systematically and / or randomly fluctuating measured values (e.g., voltage during target rotation). Example values might include: pTmeasavg=25%,pTmeasavg=55%.
[0250] As an example of a calibration parameter, a step height S can be used for the control values (e.g. for working gas υ). P and reactive gas υ RThe step size can be predefined with respect to the input u. For example, the step size can be as small as possible (so that a good linear model can be determined) but cause a measurable change in the output vector (for example, to suppress measurement uncertainties, target rotation, etc.). Example values might be: υ P = 100 sccm, υ R = 100 sccm.
[0251] As an example of a calibration parameter, a weighting r for the inputs and weightings q for the outputs of the integral system can be predefined. These can serve as the basis for the subsequent determination of Q and R for the discrete-time LQR algorithm. Based on the values obtained after the settling time T has elapsed. meas or within pTmeasavg The recorded measurement data can optionally be normalized for the individual inputs and outputs. In this case, essentially identical weighting factors r and q can be predefined for several coating processes. Example values might include: rP=10, rR=10, qU=1, qUInt=0.001, qp1, qpInt=0.01.
[0252] As an example calibration parameter, a stability margin α can be predefined for the time-discrete LQR algorithm (if used). Example values could be: α = 0.1, α = 0.2.
[0253] Method 1300 includes, in phase 1403, deflecting the coating process from the reference state by controlling the first actuator, by which the coating process is supplied; and in phase 1407, deflecting the coating process from the reference state by controlling the second actuator, by which the coating process is supplied.
[0254] Method 1300 involves capturing the response of the coating process to the deflection in each of phases 1403 and 1407, for example, by recording the change over time of the first and / or second controlled variable. Optionally, the actual state of each coating process in the reference state can be recorded, for example, in each of phases 1401, 1405, and 1409. For instance, one or more sensors can be used to capture the actual state of each of the controlled variables being recorded simultaneously.
[0255] For example, a (for example, automated) measurement can be performed that exhibits measurement data at the defined interval Δt. meas to capture and then save.
[0256] The following describes an example implementation that uses an inflow rate as a control variable. If the coating process is in the reference state, the actual inflow rate is displayed. u_¯ The data is recorded, for example, for one or more materials (e.g., gas) supplied to the coating process. Optionally, the settling time can elapse before the measurement data is recorded. The measurement data can, for example, include a timestamp t. k exhibit and the current state of the input vector u k and / or the current state of the output vector y k This represents the final value (at the end of the respective phase). Capturing the final value (at the end of the respective phase) yields the operating point as the output vector. y_¯ to the set actual inflow rate u_¯.
[0257] In each of phases 1403 and 1407, the input vector can be changed abruptly with respect to an inflow rate, and the settling time can be observed. After the settling time has elapsed, the measurement data can be recorded. Each of phases 1403 and 1407 corresponds intuitively to a single experiment for the corresponding vector component of the input vector u. k .
[0258] In each of the phases 1405, 1409, the inflow rate can u_¯ be brought back to the reference state (i.e., before phase 1405).
[0259] If several adjacent coating processes are to be calibrated, the sequence of phases 1401 to 1409 (also referred to as the calibration sequence) can be performed for each of the coating processes. The calibration sequence can, for example, have at least two phases per control variable or actuator to be calibrated, one phase representing the assumption of the reference state and the other representing the deviation from the reference state.
[0260] Once the calibration sequence 1401 to 1409 for a first coating process is complete, the same sequence can be performed for a second coating process (i.e., consecutively), then for an optional third coating process, and so on. For example, the process gas flow rate is changed for each coating process, followed by the reactive gas flow rate, and then the process continues with the next coating process until the calibration sequence 1401 to 1409 has been performed for each of the coating processes. The measurement data can be saved, for example, in a file (such as a controller calibration file) or in a database. The measurement data (e.g., containing the recorded time series) can, of course, also be saved in other ways, for example, for immediate or later use.
[0261] Based on the input matrix U=(u_1,u_2…u_n) and the initial matrix Y=(y_1(∞)y_2(∞)…y_n(∞)) The response of the coating process can be used to determine the system matrix, for example according to the following relation: KsYU−1
[0262] The following describes an exemplary implementation for determining the model of the controlled system. First, the model element C can be determined (e.g., approximated).
[0263] Determining the model element C may show that an approximation of pressure coefficients cPp,cRp and voltage coefficients CPU, cRU The approximation is determined for each coating process. It can be performed by exploiting the fact that, with respect to the operating point (in equilibrium), only one inflow rate per coating process (e.g., to the respective vacuum chamber) is used. In other words, neglecting self-feedback (gas flowing into neighboring chambers and then back again), the (N=1) model can be used. The signal profile of a single experiment can optionally be smoothed and / or interpolated. Using a numerical zero-finding method, the half-life with respect to the resulting total pressure to which the respective coating process is exposed (e.g., in the vacuum chamber) can be determined. According to the (N=1) model explained above, the coefficients can be determined specifically for the input component.
[0264] Determining the model element C can further involve summarizing the coefficients of the individual components into the matrix C.
[0265] The following is an exemplary implementation for determining the static element K. s explained. Determining the structural element K s This can be done as explained above for the experimental determination of at least one control element. The static element (for example, with respect to the operating point approximation) can be directly determined from the individual experiments. It should be noted that the differences from the target operating point, i.e., u_¯,y_¯, enter.
[0266] The following describes an exemplary implementation for numerically determining the model element A. The model of the controlled system can, by construction, provide the identical static element as explained above, meaning that the static element can be used directly to determine an approximation for the model element A, e.g., according to the relation explained above: Ks=−CA−1B
[0267] If B is already determined independently of the plant (as explained for the (N=1) model), then A = -BK can be directly applied. S -1 C is determined. As a result, the model elements of the controlled system model (A, B, C) are now determined for the operating point. (u_¯,y_¯).
[0268] Optionally, the controlled system can be discretized based on the PLC cycle time. As described above for discretization (see...). Fig. 6) explained, based on the continuous control loop and a cycle time, a time-discrete model (A) can be developed. d ,B d ,C) from cycle to the immediately following cycle. Instructions for determining the matrix exponential function and the integral using diagonalization are also explained in relation to the preceding discretization.
[0269] The following is an exemplary implementation for determining at least one control element.
[0270] Determining at least one control element can involve calculating the weighting matrices Q and R for the discrete-time LQR algorithm. To compensate for differences in the numerical magnitudes of pressure and stress, as well as differences between various coating processes, normalization can be used. Summarizing the initial weighting factors explained above into a vector q yields a weighting matrix from the measurement data, which can be written as: Qy=diag(q_maxk(y_k)−mink(y_k)).
[0271] From the diagonal weighting matrix Q y Can Q = C now be determined? T Q y C is chosen for the LQR algorithm. From the weighting factors for the inputs explained above and the inflow rates (also called flows) set for the individual experiments, a diagonal matrix can be directly derived. R=diag(r_maxk(u_k)−mink(u_k)) be formed.
[0272] Determining at least one control element can show that the control element K can be identified using the LQR algorithm. Integral is determined. For this purpose, for example, a numerical solution of a discrete-time matrix Riccati equation can be used, which is accessible via numerical libraries.
[0273] Determining at least one control element may show that a decomposition of the control element K is required. Integral This is done in several control elements. These multiple control elements can have: Ky=Bd−1(Ad−1)C−1,KP=K1C−1−Ky,KI=K2
[0274] The decomposition of K Integral = (K1 K2) into the multiple control elements can be done as explained above in relation to diagram 400 and discretization.
[0275] The multiple control elements determined in this way can, for example, be fed into a PLC and processed by it. Based on these multiple control elements, the PLC can then execute procedure 100. For example, the group of several control variables acquired in parallel can be mapped to the first and second controlled variables.
[0276] The following describes further aspects that can be used optionally.
[0277] One initial aspect might be that the individual inputs are restricted by means of at least one limit (i.e. having an upper limit and / or lower limit): ui∈[uimin,uimax]
[0278] This ensures that any complications that may arise due to numerical problems, operator errors, or errors in measurement technology are compensated for, or at least their effects are mitigated.
[0279] Should a flow be identified that lies outside the interval [uimin,uimax] is located, the corresponding component can u i The limit should be set to the nearest boundary. This at least one boundary clearly facilitates the process of coating not being operated under highly unfavorable conditions or even failing altogether. Based on this at least one boundary, it can also be determined, for example, whether an integrator windup should be reduced when controlling the coating process, such as when the component u i The limit is set for an extended period of time. The dismantling of the integrator windup is explained below.
[0280] If the controller is given an unreachable target operating point and a significant integral of the control deviation has accumulated (the so-called windup), this can only be reduced once an achievable operating point (e.g., with one or more setpoints) has been specified. The desired controller behavior will then only be restored much later. Two exemplary implementations for suppressing these effects are explained below.
[0281] The first implementation for suppressing these effects can involve pausing the accumulation of the setpoint difference. Should the controller encounter an unattainable operating point, it can be limited to one or more of the calculated input components. iIf an input has a dominant effect on a corresponding output (e.g., working gas on the total pressure or reactive gas on the process voltage of a coating process), the integration / accumulation of the associated setpoint difference can be suspended. As soon as the controller detects another input within the valid interval, the integration can be resumed, and in this case, no windup needs to be cleared.
[0282] The second implementation for suppressing these effects can involve resetting the accumulated setpoint difference. This measure requires no further information about the coating process. The second implementation can also involve determining a suitable criterion that defines the reset time. Since the significant portion of the input to be calculated (see diagram 400) is determined by the output feedback via Ky If an error occurs, the accumulated control deviations can simply be set to zero at a defined time (also called a reset). Since the accumulated control deviations are essentially intended to suppress model deviations and disturbances, the method simply recalibrates after this reset without significantly affecting the behavior of the input u.
[0283] A second aspect may indicate that the coating process is an unstable coating process (e.g., during the deposition of SiO₂). xAn unstable coating process exhibits hysteresis in the response of a controlled variable to a change in the manipulated variable, which makes it difficult to determine the response to a deviation of the coating process from the reference state. For example, the unstable coating process may tend to leave the target operating point on its own and / or oscillate between two self-adjusting (stable) operating points (also referred to as bistable). For instance, the coating process may tip into one of the two bistable operating points and become stuck there.
[0284] Bistable behavior is usually caused by positive feedback within the reaction dynamics, which stabilizes some of the operating points. In reactive sputtering, for example, the sputtering cathode can react with oxygen as a reactive gas, making its surface more resistant to the plasma and thus inhibiting atomization (also known as target poisoning). This reduces oxygen consumption, leading to an excess of oxygen, which in turn oxidizes the surface of the coating material more strongly. Reducing the reactive gas eliminates its excess, thus decreasing the oxidation of the sputtering cathode. This accelerates the atomization of the coating material and therefore also increases the consumption of reactive gas, until its excess is reduced and the reaction dynamics spontaneously shift towards a metallic surface of the sputtering cathode.However, if a reaction product is needed that has reacted incompletely, it is necessary to operate the sputtering process at an unstable operating point between the stable operating points (also known as balance control).
[0285] If the coating process is unstable, the process model can be determined based on two hysteresis curves recorded at different process gas flows. This may involve determining C from several hysteresis values. The main information then lies in the pressure curves on the step response recordings for the individual experiments. The stresses can be generated synthetically by estimating them from the pressures and the coefficients derived from C. The recorded pressures and artificially generated stress measurements can then be processed as described above.
[0286] A second aspect is that the control model can be approximated using machine learning (e.g., multi-layered learning – so-called "deep learning"), as will be described in more detail later. Simulated measurement data from the coating process can be used as input to the machine learning, and the underlying process model can be used as output. Optionally, the simulated measurement data can be made noisy, which makes the result of the machine learning more robust. The noise in the measurement data can be understood as a relatively simple way to simulate one or more real-world disturbances (e.g., a disruptive environmental influence). For example, disturbances such as noise from the measurement equipment or voltage fluctuations due to the rotating target can be simulated.
[0287] More generally, one or more disturbances, each representing (e.g., modeled as) an internal and / or external disturbance of the coating process, can be superimposed on the simulated measurement data. This stimulates the model to suppress and / or ignore such disturbance(s), thus making it more robust.
[0288] The following is an example of a process group to which procedure 100 and / or procedure 1300 can be applied.
[0289] Fig. 15A and Fig. Figure 15B illustrates a process group 1500 according to different embodiments in different assembly diagrams 1500a, 1500b.
[0290] Process group 1500 can perform several (e.g., reactive) coating processes, for example, two or more coating processes (e.g., three, four, five, six, or more). Process group 1500 can also have several coating areas 104a, 104b, ..., 104K, in each of the coating areas 104a, 104b, ..., 104K, where exactly one coating process 304a, 304b, ..., 304K (one controllable coating process 201) can take place.
[0291] Each of the coating areas 104a, 104b, ..., 104K can have a corresponding coating device 114a, 114b, ..., 114K for carrying out the respective coating process therein. Each of the coating areas 104a, 104b, ..., 104K can be provided by means of (e.g., exactly) one vacuum chamber. One or more vacuum chambers can be provided by means of a chamber housing. Several vacuum chambers in a chamber housing can also be referred to as compartments.
[0292] The vacuum chambers can be separated from each other, for example by means of so-called chamber walls, so that they are at least partially (albeit slightly), or even partially or completely, gas-separated from each other in terms of vacuum technology. For example, several coating processes can be controlled that are independent of each other or not coupled to each other, for example to save on controller hardware. However, the gas separation can be so small that the several coating processes exchange a gas with each other (also referred to as gas exchange). For example, a gas flow from a second coating area 104b can pass through a first coating area 104a to a first pumping area 106a.
[0293] Each coating device 114a, 114b, ..., 114K can be configured according to various embodiments for coating at least one substrate 102 (i.e., one or more than one substrate 102), which is transported, for example, through the respective coating area 111. For example, the coating device can be configured to provide a gaseous coating material (material vapor) which can be deposited, for example, on the at least one substrate 102 to form a layer. A coating device can include at least one of the following: a sputtering device or a thermal evaporation device (e.g., a laser beam evaporator, an arc evaporator, an electron beam evaporator, and / or a thermal evaporator). A sputtering device can be configured for atomizing the coating material by means of a plasma (also referred to as sputtering).A thermal evaporation device can be set up to evaporate the coating material using thermal energy.
[0294] A reactive coating process can be supplied with at least one reactive gas with which the emitted (e.g. atomized) coating material reacts chemically, so that the reaction product is deposited on the substrate 102.
[0295] For example, the same working gas and / or reactive gas can be supplied to more than one (e.g., each) coating process in process group 1500. Alternatively or additionally, more than one (e.g., each) coating process can use the same coating material.
[0296] The process group 1500 can further comprise two pumping sections 106a, 106b, between which the several coating sections 104a, 104b, ..., 104K are arranged. Gas can be extracted from each of the two pumping sections 106a, 106b, e.g., more than from each of the coating sections 104a, 104b, ..., 104K. According to various embodiments, the process group 1500 can comprise a pumping arrangement 814 (comprising at least one high-vacuum pump). The pump arrangement 814 can be configured to remove a gas (e.g., the process gas depleted of reactive gas by reaction) from at least the two pumping areas 106a, 106b and optionally from at least one (e.g., each) coating area 104a, 104b, ..., 104K, so that a vacuum (i.e., a pressure less than 0.3 bar) and / or a pressure in a range of approximately 10 -3 millibar (mbar) to approximately 10 -7mbar (in other words, high vacuum) or a pressure smaller than high vacuum, e.g., less than approximately 10 -7 mbar (in other words, ultra-high vacuum) can be provided or made available.
[0297] Using the pump arrangement 814, one or more gases, e.g., a portion of the process atmosphere, can be extracted from each of the coating processes. For this purpose, a corresponding suction capacity L can be applied to each of the coating processes 304a, 304b, ..., 304K. a , L b , ..., L K The reason for the withdrawal.
[0298] Furthermore, the process group 1500 can be set up in such a way that the vacuum conditions (the process conditions) within each of the several coating areas 104a, 104b, ..., 104K (e.g. process pressure, process temperature, chemical composition of the process gas, etc.) can be set or controlled, e.g. during the execution of the coating process (e.g. according to the target operating point) by means of corresponding two or more actuators (first and second actuator) per coating process.
[0299] According to various embodiments, the process group 1500 can have a gas supply 702. The working gas can be supplied to at least one (e.g., each) coating area 104a, 104b, ..., 104K by means of the gas supply 702 to form a process atmosphere in these areas. The process pressure can be established by an equilibrium of process gas, which is supplied by means of the gas supply 702 and extracted by means of the pump arrangement 814.
[0300] Gas supply 702 allows one or more gases to be supplied to each of the coating processes, e.g., to the process atmosphere in which the coating process takes place. For this purpose, a corresponding flow rate (e.g., a standard volume flow rate) F can be set at each of the coating processes 304a, 304b, ..., 304K. a , F b , ..., F K concern with which the supply is made.
[0301] To make process group 1500 as compact as possible, the coating areas 104a, 104b, ..., 104K can optionally be directly adjacent to one another or provided without intervening pumping areas 106a, 106b. This can result in the coating areas 104a, 104b, ..., 104K exchanging more gas with each other.
[0302] For example, there may be less (or no) pumping power between two immediately adjacent coating areas than between either of the two immediately adjacent coating areas.
[0303] The coating of substrate 102 can therefore have several layers, each of which is formed by one of the several coating processes 304a, 304b. For example, each of the layers can contain the coating material and optionally a component of the reactive gas, e.g. a reaction product thereof.
[0304] The control of process group 1500, e.g., every coating process carried out using process group 1500, can be done, for example, using method 100.
[0305] Fig. 16A and Fig. Figure 16B illustrates the control of a process group 1600a, 1600b according to different embodiments in different control diagrams. The variable 306 to be controlled or regulated (also referred to as the controlled variable) of the coating process can be, for example, a pressure and / or a gas composition used by the respective coating process, e.g., within the coating area 104a, 104b in which a coating process takes place.
[0306] According to various interpretations, a quantity can be understood as a physical quantity that represents a quantitatively measurable property of a physical entity (e.g., an object, a process, or a state) (e.g., describes it and / or correlates with it). Its state can optionally depend on time t and / or location, for example, the composition and / or pressure of the process gas.
[0307] The controlled variable 306 can have an actual state, which is detected by a measuring element 304. The measuring element 304 can supply the detected actual state as the first input variable 303 (also referred to as feedback variable 303) to the controller 203 (also referred to as feedback variable 303). More generally, the feedback variable 303 can also be a vector, depending on the number of detected controlled variables. Furthermore, a setpoint state 316 can be supplied to the controller 203 as a second input variable (also referred to as target variable 316), e.g., a target pressure. More generally, the target variable 316 can also be a vector, depending on the number of target values.
[0308] The controller 203 can compare the feedback variable with the setpoint variable 316 and, based on the result of this comparison, control an actuator 302 301. The actuator 801a, 801b can, for example, be the actuator of a valve in the gas supply 702 and / or the pump assembly 814, which carries a standard volume flow containing the reactive gas. The actuator 801a, 801b can be controlled in such a way that a deviation of the setpoint from the actual state of the controlled variable 306 (also referred to as the control deviation) is minimized.
[0309] The manipulated variable is the quantity that is changed to influence the controlled variable.
[0310] The current value of the manipulated variable is the manipulated value (for example, the position of a valve).
[0311] In general, various manipulated variables can be used to influence the same controlled variable. For example, the pressure of a gas can be influenced as a controlled variable by changing the standard volumetric flow rate of gas being supplied and / or extracted, or by changing the temperature of the gas. The standard volumetric flow rate, in turn, can be changed by opening or closing a valve, at least partially (i.e., partially or completely). Similarly, the gas can be compressed and / or expanded to change its pressure. With respect to a valve, the manipulated variable could be, for example, the position of the actuator or a voltage and / or frequency applied to it. Another example of a manipulated variable is the opening state of a valve. In the case of a pneumatic actuator, for example, the pneumatic pressure on the actuator is used as the manipulated variable.
[0312] For example, process group 1600a can perform several coating processes, each of which is controlled by the controller 203, as illustrated in the control diagram. Likewise, process group 1600a can include a control device 228 which implements the controller 203.
[0313] For example, an actual operating point (also called group actual operating point IA) can be defined for each of the several coating processes 304a, 304b. G (designated) are recorded, from which the individual setpoint values to be adjusted are determined.
[0314] Fig. Figure 17 illustrates a method 1700 according to various embodiments in a schematic flowchart.
[0315] Method 1700 may include, in 1701, the simulation of at least one (i.e., one or more than one) coating process. The simulation 1701 of the coating process(s) may, for example, be carried out using a model 1602 of the one or more coating processes (also referred to as a simulation model). The simulation model 1602 may represent the at least one coating process, e.g., the process group and / or the plant (also referred to as the process plant or vacuum plant) by which the at least one coating process is carried out.
[0316] For example, the procedure 1700 may include providing the simulation model 1602, e.g. to determine based on a physical setup of the plant.
[0317] For example, the simulation model 1602 can take into account one or more of the following: - physical information (i.e., the physical architecture, i.e., its structure) of the plant or its components, by means of which at least one coating process is carried out, e.g., vacuum chambers (e.g., their volume), actuators, coating devices, sensors, kinematic chain of the plant, etc.; - Information on the characteristic map (or one or more characteristic curves) of the system or its components, e.g. degrees of freedom, dependencies between current, voltage and / or pressure; - the logical architecture of the system or its components, e.g. the interaction and / or logical coupling of the individual components (e.g. actuators) of the system with each other; - the specific or potentially available configuration of the system (e.g., which components are active or inactive, or which can be switched between); - Parameters of the controlled system, e.g. control variables or regulated variables, input variables of the individual actuators or output variables of the individual sensors; - the coordinate system of the system, e.g., the stacked sub-coordinate systems of the individual actuators of the system, origin of the individual sub-coordinate systems, end positions of the sub-coordinate systems; - permissible operating parameters of the system or their limits, such as operating temperature, flow rate, pressure, suction power, etc.
[0318] The simulation model can be, for example, an object-based model or another explicitly formulated model. The simulation model can, for example, include a formal representation of at least one coating process, or of the components by which the at least one coating process is carried out, e.g., supplied. The simulation model can have a multitude of links between a coating process and the corresponding spatial information of the coating process and can optionally assign corresponding quantities to the coating process that are supplied to and / or withdrawn from the coating process, condition it, inhibit it, promote it, or terminate it. Furthermore, the simulation model can have process logic that logically links several coating processes together and / or that implements a sequence (e.g., a schedule) according to which the execution of each coating process is carried out.The coating process is carried out.
[0319] The simulation model can, for example, be based on explicit information about at least one coating process. This explicit information can specify, for instance, how the coating process is controlled and / or how it interacts. For example, the simulation model can have at least the same input and output variables as the process model, or represent the controlled system in the same way.
[0320] Unlike the process model, the simulation model is not necessarily based on a purely mathematical description but can be based on a hierarchical description, such as a hierarchical database model. The hierarchically ordered objects can then be used as model elements. However, a simulation model can also be used that has one or more of the model elements (A, B, C) described above, such as the system matrix A, the input matrix B, the output matrix C, and / or the pass matrix D. Similarly, a simulation model can be used that has one or more of the control elements described above as a model element, such as K. y , K P , and / or K I .
[0321] As another example, the simulation model or a simulated result trajectory of the simulation model can depict characteristic additional effects from practical problems, e.g., signal noise from the measurement technology or sinusoidal fluctuations in the process voltage due to a rotating target in the coating process. For example, the simulation model can have a mathematically nonlinear description (e.g., a hysteresis model) which is linearized around a point of interest (e.g., corresponding to a suitable operating point) to obtain a controlled system model (e.g., the (A, B, C) model).
[0322] The result of simulation 1701 (also referred to as the simulation result) can accordingly include simulation data 1604 of at least one coating process. The simulation data can, for example, include clearly defined, synthetically generated measurement data that represent the behavior of at least one coating process.
[0323] The simulation data can contain the output variables and / or the input variables of the simulation model. For example, the simulation data can assign values of the input variables and output variables to each other, which are then converted into one another by the simulation model.
[0324] The input variables of the simulation model can, for example, be control variables and / or manipulated variables. The output variables of the simulation model can, for example, be (e.g., measurable) physical quantities of the at least one coating process, such as at least the control variables of the at least one coating process described herein.
[0325] The procedure 1700 can include, in 1703, training a model generator 1606 based on the simulation result, e.g., based on the simulation data 1604. In other words, the training data can include at least the simulation data 1604.
[0326] The model generator can be trained, for example, using an artificial neural network (kNN).
[0327] The artificial neural network (kNN) can be trained using conventional training methods, such as backpropagation. Alternatively or additionally, the training data can be optimized using mathematical optimization and / or machine learning methods. During training, the training vectors are selected according to the desired input parameters, such as the simulation data, the corresponding configuration of the simulation model, etc. It should be noted that both the parameters contained in the kNN's input data and the parameters contained in the kNN's output vector are highly application- or process-dependent and must be selected accordingly.
[0328] This section refers, among other things, to the artificial neural network (kNN). The described principles can be applied analogously if, alternatively or additionally to the kNN, another trainable algorithm is used as a model generator, such as a support vector machine or a long short-term memory (LSTM). The kNN can have a large number of nodes (figuratively, artificial neurons) and a connection network (the mapping of connections to nodes). In a kNN, the processes for information acquisition and processing are modeled analogously to biological neurons. This is achieved through a number of layers of hidden neurons, dependent on the specific circumstances, and the activation functions that transmit the signals.
[0329] The topology of the model generator describes the structure of the interconnection network, i.e., how many nodes are distributed across how many layers, and how these layers are interconnected. For example, multiple nodes can be arranged in successive layers, the first layer of which forms the input of the model generator and the last layer of which forms the output. The last layer whose output is visible outside the interconnection network is called the output layer. The preceding layers are accordingly referred to as hidden layers. Using a graph, the artificial neurons can be represented as nodes and their connections as edges. In this case, the model generator is exemplified as a directed graph with typed nodes.
[0330] After the model generator is constructed, it is first trained (also called the training phase), during which it is adapted to the desired operating mode (figuratively speaking, it "learns"). Data (the training data) is fed into the model generator, which it uses to learn to replicate the desired operating mode. An accurate output from the model generator reinforces what it has learned (i.e., a specific signal path through the network), while an inaccurate output weakens the signal path. In this way, the paths through the network that best describe the desired operating mode gradually emerge.
[0331] During the training phase, one or more of the following operations can occur: modifying the connection network (e.g., creating or deleting connections); changing the weight of nodes; modifying the properties (e.g., thresholds) of the nodes; modifying the number of nodes; modifying the activation, propagation, and / or output function.
[0332] During the training of the model generator (e.g., the kNN), the training data is selected according to the desired input parameters. In one example, the incoming training data (also referred to as input data) consists of simulation data 1604, for which the simulation model, e.g., its architecture, is known. In this case, the model generator 1606 is trained to determine the simulation model that generated a given set of simulation data.
[0333] Optionally, the training data can contain filtered simulation data, either as an alternative or in addition to the simulation data. This can be achieved using a filter, which is fed the simulation data and outputs the filtered simulation data. Examples of filters include a noise filter (configured to add noise to the input data) or a frequency filter (configured to filter the input data according to its frequency).
[0334] Adding noise can be done, for example, by mapping each data point p of the simulation data to a data point p' of the filtered simulation data, e.g., according to the relation p' = f(p, z). R ), where z R denotes a random number, for example from the interval [0,1]. The random number can be determined, for example, using a random number generator, e.g., individually for each data point. For example, f(p, z)R ) are determined according to the relation f(p, zR)=p + 0.1⋅p⋅(zR−0.5).
[0335] This relationship clearly demonstrates a 10% dispersion of the data points.
[0336] More generally, the filter can be configured to increase the variance (e.g., standard deviation) of the simulation data, for example, by adding noise to it. This makes the simulation data more similar to real-world measurements and therefore improves the robustness of the trained model generator.
[0337] Fig. Figure 18 illustrates the model generator 1606 according to various embodiments in a schematic diagram 1800, for example, formed according to method 1700 and / or implemented by means of the control device. The model generator 1606 can be configured using an artificial neural network (kNN). The kNN can, for example, be configured as a convolutional neural network or a feedback neural network. Of course, a differently configured model generator 1606 that provides the corresponding functions can also be used.
[0338] The model generator 1606 to be trained, e.g., its kNN, can have a multitude of layers, e.g., at least one input layer 1801, at least one output layer 1807 (regression layer, also called the final layer), and between these, one or more processing layers 1803, 1805. The input layer 1801 can, for example, contain dimensions of a time series of simulation data or measurement data. A time series can, for example, have one dimension for the time indices and one dimension for the property(ies) of a point (or signal sample). Exemplary implementations show that the property(ies) of a point include a timestamp, flows set at that time, and / or measured values recorded at that time.Exemplary implementations show that the property(ies) of a point include a time / timestamp (or the time difference to the last recorded point), the currently set inputs of the actuators and / or the current measured values.
[0339] When a convolutional neural network is used as a model generator 1606, the one or more processing layers can have at least one convolutional layer 1803 and / or at least one intermediate layer 1805 (fully connected layer). Optionally, a pooling layer (not shown) can be placed between the at least one convolutional layer 1803 and the at least one intermediate layer 1805 (also called a fully connected layer).
[0340] The activity of each node in a convolution layer 1803 can be computed using a discrete convolution. The discrete convolution of each node can be transformed into an output of the node by an activation function. For example, a ReLU function can be used as the activation function. In the context of artificial neural networks or training methods, a ReLU function (also called a rectifier function, rectifier function, or simply ReLU) is an activation function that is essentially defined as the positive part of its argument.
[0341] Several parameters control the size of the output (also called output volume) of each convolution layer 1803, for example the parameters filter count (“depth”), step size (“stride”) and padding (“padding” or “null padding”).
[0342] When a feedback neural network is used as a model generator (1606), the nodes of multiple processing layers can be interconnected, for example, towards the same or a preceding processing layer. In direct feedback, a neuron's own output is used as an additional input. Indirect feedback connects the output of a neuron to a neuron in one or more preceding processing layers. Lateral feedback connects the output of a neuron to another neuron in the same processing layer. Examples of feedback neural networks include: an Elman network, a Jordan network, a Hopfield network, and a fully interconnected neural network.The feedback neural network can be trained, for example, using error feedback (also known as backpropagation) or another supervised learning method. This ensures that the neural network receives feedback based on the target data known from the incoming training data.
[0343] Accordingly, the training data, e.g., the target data of the training data, can contain one or more pieces of information about the simulation model, e.g., about its model elements and / or its configuration. For example, the training data, e.g., the target data of the training data, can contain the model elements of the simulation model, e.g., their configuration.
[0344] The backpropagation process can consist of: supplying the model generator 1606 with the input data; comparing the corresponding output data of the model generator 1606 with the target data, for example, to determine the deviation of the output data from the target data; and supplying the model generator 1606 with information about the result of the comparison, e.g., about the deviation of the output data from the target data. In this process, the weights of the neural connections are changed depending on their influence on the deviation, e.g., such that a subsequent supply of the input data results in the output data approaching the desired target data.
[0345] The simulation data, used as incoming training data, can be multidimensional, for example, represented by a corresponding tensor. In the first dimension, the tensor can reference, for example, a first control variable per coating process, in the second dimension a second control variable per coating process, in the third dimension a first controlled variable per coating process, in the fourth dimension a second controlled variable per coating process, and so on.
[0346] Depending on the dimension of the incoming training data, the nodes of each processing layer 1803, 1805 can also be arranged multidimensionally.
[0347] An exemplary implementation of the kNN network structure involves preprocessing using one or more convolutional layers. The convolutional representations of the input time series are then fed to one or more stacked layers, which are configured to process arbitrary time series, such as LSTM layers (LSTM - "long short-term memory") or TCN layers (TCN - "temporal convolutional network"). Following the time series processing layers, one or more dense layers can be added, culminating in one or more regression layers, which are used to estimate corresponding model parameters. The final dense layers at the end of the kNN can optionally be omitted from the activation function and serve as regression layers.
[0348] The last dense layer can, for example, be or become branched into several dense layers to provide different outputs of the kNN. Exemplary implementations show that the kNN output has a working point. (u_¯,y¯_) and has associated model matrices (A, B, C) or their equivalents for a sampling system. The model generator can optionally have multiple kNNs to determine only a part of the model (e.g., only one operating point, or a subset of the model matrices, or one kNN per model matrix).
[0349] Alternatively or additionally, time series processing can be carried out using a commercially available kNN or an open-source kNN, e.g. using a so-called Transformer Network from Google.
[0350] A TCN layer may already incorporate measures against over-adaptation.
[0351] Exemplary implementations of the model generator according to various embodiments 2200, 2300 using a kNN are each shown in Fig. 22 and Fig. Figure 23 shows the first question mark representing the index of the data set, and the second question mark representing the length of the time series. For example, the kNN is configured to process time series of any length, e.g., 100 points or even 200 points per time series.
[0352] Fig. Figure 19 illustrates a method 1900 according to various embodiments in a schematic flowchart. As described above, simulation and training can be performed, with the difference that the simulation is repeated, e.g., W D -times (W D(is a natural number). Each iteration of the simulation is indexed with w, i.e., the index w=1 indicates a first iteration of the simulation, the index w=2 indicates a second iteration of the simulation, and so on.
[0353] The successive iterations (w=1 to w=W) D The simulation models 1602 differ from each other in their configuration. In other words, after each simulation run 1701, the simulation model 1602 can be reconfigured 1901, for example, by changing one or more model elements. The configuration of the simulation model 1602 in the first run is then a first configuration (A,B,C)1, the configuration of the simulation model 1602 in the second run is then a second configuration (A,B,C)2, and so on.
[0354] The simulation data obtained in this way, 1604 per run, can be stored in a dataset SD. wadded in 1903.
[0355] The result of all (w=1 to w=W) D ) Iterations can create a data set SD w exhibit which has the configuration (A,B,C) w of the simulation model 1602 in the w-th run and the simulation data 1604 of the w-th run. The configuration (A,B,C) w The specification of the simulation model 1602 in the w-th run can, for example, list the model elements of the simulation model 1602 in the w-th run, e.g., their configuration. Of course, other specifications can also be used to indicate the configuration of the simulation model 1602 in the w-th run.
[0356] Each SD data record w The training data from 1902 to 1905 can then be added. The training data from 1902 will then contain a number of W. D Data records on SD w , of which each data record SD w can be used to train the model generator 1606.
[0357] In principle, the model generator 1606 can improve the larger the number of W. D the number of iterations. D for example, it can be 2 or greater, for example, 5 or greater, for example, 10 or greater, for example, 20 or greater, for example, 50 or greater, for example, 100 or greater, for example, 200 or greater.
[0358] Fig. Figure 20 illustrates a procedure 2000 according to various embodiments in a schematic flowchart. Once the training is complete, i.e., the model generator 1606 is fully trained, the controller can be determined based on it.
[0359] The fully trained model generator 1606 is then configured to convert measurement data into the target data used during training, i.e., to determine the target data that was trained on model generator 1602. The following refers to the exemplary case where the model elements (A, B, C) described above were trained on model generator 1606 as target data. In this case, the trained model generator 1606 is configured to determine the process model, or at least components of the process model, namely, in this case, the model elements. It can be understood that what has been explained regarding the model elements (A, B, C) as target data can apply analogously to other target data, for example, if the controller or its components, such as the control elements, are used as target data.
[0360] Method 2000 can, in 2001, determine a response (e.g., a first controlled variable and / or a second controlled variable) per coating process to a deviation of the coating process from a reference state. The deviation can occur by (e.g., precisely) controlling the first actuator that supplies the coating process, and / or by (e.g., precisely) controlling the second actuator that supplies the coating process.
[0361] In principle, the response can be determined as described above, for example, using a step response. The step response can provide a good result for a stable coating process. The step response of the coating process intuitively represents two parameters of the stable coating process: the half-life of the response and the convergence value of the response.
[0362] In the case of an unstable coating process, however, the step response may be less informative. This is because the response of the coating process depends not only, or only minimally, on the reference state or the displacement, or because the instability necessitates the completion of the controller to reliably establish the reference state or to deflect from it. This applies, for example, to a reactive sputtering process with hysteresis, as illustrated in diagram 2011, 306. In this diagram, a controlled variable 306 (e.g., voltage) is plotted against the manipulated variable 2011 (e.g., gas flow). The following discussion refers to this exemplary hysteresis as a representative of unstable behavior, but it should be understood that the explanations can also apply to other types of unstable coating processes.
[0363] If the controller present during commissioning is inadequate (e.g., during the very first operation of the system) or simply not present, it can be replaced with a discontinuous controller (also known as a multi-point controller), for example, a two-point controller. The multi-point controller can switch between several setpoint states of the manipulated variable (discrete) as soon as a criterion is met. The criterion can be met, for example, if the actual state of the controlled variable fulfills a predefined criterion, e.g., if the actual state of the controlled variable leaves a predefined interval [F1, F2], i.e., exceeds the threshold value F1 or falls below the threshold value F2. Instead of the interval, a single threshold value can, of course, also be used, for example, if the rate of change is so large that the amplitude of the actual state of the controlled variable is sufficient within the controller's response time.
[0364] More generally, the multipoint controller implements the fact that the target state of the manipulated variable is a discontinuous function of the position of the actual state of the controlled variable relative to one or more than one threshold value, e.g. taking into account which threshold value the actual state of the controlled variable reaches or passes (e.g. exceeds / falls below).
[0365] This can also be visualized as a step response, although the return to the reference state is not triggered by the rate of change falling below a predefined threshold, but is independent of the rate of change and dependent on the position of the actual state of the controlled variable relative to one or more than one threshold.
[0366] The two-point controller can, for example, be configured such that the setpoint of the manipulated variable is increased, for example to a first setpoint reference state U1, if the actual state of the controlled variable is smaller than the predefined interval [F1, F2], and that the setpoint of the manipulated variable is decreased, for example to a second setpoint reference state U2, if the actual state of the controlled variable is larger than the predefined interval [F1, F2].
[0367] This ensures that the target state of the manipulated variable is repeatedly switched between the first target reference state U1 and the second target reference state U2.
[0368] In response, the actual state of the controlled variable can oscillate between the limits F1 and F2 of the interval [F1, F2]. The respective pairs of actual state of the controlled variable and actual state of the manipulated variable can lie on a trajectory 2005.
[0369] If the target operating point for the subsequent operation of the coating process lies within the hysteresis, the first target reference state U1 and the second target reference state U2 can, for example, be located within the area enclosed by the hysteresis. This ensures that the respective trajectory 2005 lies within the unstable region (also referred to as the transition region) of the coating process, in which the coating process is to be maintained during subsequent operation.
[0370] According to various embodiments, the repeated deflection of the coating process described above, e.g., from the reference state at F1 to the deflected state at F2, can be achieved by controlling only one actuator. If several actuators are to be controlled in parallel, each actuator can be individually controlled by means of a discontinuous controller, which takes as its input the current state of one of the controlled variables. For example, a first actuator can be controlled based on the current state of a first controlled variable, and a second actuator can be controlled based on the current state of a second controlled variable. A chaotic trajectory 2005 can then be followed in a multidimensional subspace of the state space.
[0371] Alternatively, the state space can be traversed section by section (visually “slice by slice”) by performing several passes in succession, whereby in each of the passes only the first actuator is controlled while the second actuator remains unchanged and between the passes the second actuator is changed step by step.
[0372] The result of determining the answer can have one or more than a series of 2005 determined actual states of the coating process (as data points), each of which has, for example, a pair of actual state of the coating process consisting of the manipulated variable and the controlled variable.
[0373] In principle, the state space of each coating process can be traversed sequentially in this way. If several coating processes are run in parallel, this ensures that each series of actual states of the coating process is not only a function of the interaction chain of the coating process itself (see A). local ) but also a function of the interaction between the coating processes (see A exchange ) is.
[0374] The procedure 2000 can include, in 2003, the determination of a control model 2020 using the model generator 1606 based on measurement data (i.e., acquired data) of one or more coating processes. The measurement data, for example, exhibit one or more series 2005 of acquired actual states of the coating process and can be fed into the model generator 1606. Each acquired actual state of the coating process can, for example, each exhibit a pair consisting of an acquired actual state of the manipulated variable (actual manipulated variable) and an acquired actual state of the controlled variable (actual controlled variable).
[0375] For example, the model generator 1606 can be configured to output the process model (e.g., the (N=1) model, (N=2) model, (N=3) model, or more generally the N model) or at least its model elements (A, B, C) based on the measurement data. In that case, the control model 2020 or at least its control elements K can be generated based on the model elements (A, B, C).y , K P , and / or K I can be determined, e.g. as described above, e.g. according to the relations: Ky=B−1A C−1KP=K1C−1−KyKI=K2
[0376] It can be understood that what has been explained regarding the model elements (A, B, C) as target data can apply analogously to other target data, for example, if the control elements K y , K P , and / or K I or other control elements are used as target data. For example, if the model generator 1606 is configured, the control model or at least its control elements K y , K P , and / or K I To output the output, the intermediate step via the process model is not absolutely necessary.
[0377] The following describes an exemplary implementation that incorporates simulation. To train a suitable machine learning algorithm, known pairs of measurement series and control loop models can be used (e.g., supervised learning). The measurement series are generated synthetically using numerical simulations. For this purpose, suitable simulation models (e.g., linear models or known models for sputtering processes with hysteresis) are created using random data, and appropriate measurement series (e.g., based on two-point controllers) are calculated. The synthetically generated measurement series, resulting from the simulation, are then made noisy with further random data or effects known from practice to better imitate real measurement series and to make the machine learning algorithm robust against measurement noise, target rotations, and similar factors.
[0378] In the case of multiple coupled sputtering processes (multi-variable system), a set of two-point controllers per process can be operated simultaneously to obtain suitable measurement series. To better capture crosstalk effects, the setpoints of the two-point controllers could be systematically varied for each process. The simulation models are extended accordingly to train a suitable machine learning algorithm.
[0379] In the case of additional input variables (e.g., process gas for overall pressure control) per coating process, these can be varied during operation of the two-point controller to capture their influence as well. Here, too, the simulation models and the machine learning algorithm would be adjusted accordingly.
[0380] Fig.Figure 21 illustrates several parameters of a coating process controlled by method 100 according to various embodiments 2100 in different schematic diagrams, in which the respective parameter is plotted against time t. The output parameters of the coating process (e.g., measured variables) are represented by y1 and y2. The input parameters of the coating process (e.g., its control variables) are represented by u1 and u2. The output parameters of the coating process are shown once as the measurement series controlled by a two-point controller (also referred to as two-point controlled) (dashed line) and the simulated trajectory of an estimated model (solid line).
[0381] For example, y1 can represent a process voltage and y2 a pressure. The graph shows the deviations of the output variables (i.e., their values) from their respective mean values (here, 0) of the entire recorded measurement series. The first output variable y1 of the coating process is controlled two-point by the first input variable u2 of the coating process, as explained above regarding the two-point controller. The second input variable u1 of the coating process was systematically varied in three stepwise steps.
[0382] Method 100 can be used, for example, to implement automated controller adjustment.
[0383] The method 1300 described herein can, in an exemplary implementation, include: controlling one or more sputtering processes using a two-point controller; time-resolved acquisition of measurement data / series after the initial settling of the two-point controller(s); optional systematic variation of setpoints or additional input variables of one or more sputtering processes; processing of the measurement series using a machine learning algorithm, which outputs a process model (also referred to as a controlled system model) in state-space representation; determining the controller parameters (also referred to as control elements) based on the process model.
[0384] This section explains various implementations of the 2020 control model (e.g., a model representing one or more coating processes). It is important to understand that the 2020 control model does not necessarily have to include the process model. For example, the 2020 control model can include, or be derived from, the specific controller parameters, either as an alternative or in addition to the process model. For instance, the model generator can be trained to determine (e.g., estimate) the controller parameters as a model representing one or more coating processes.
[0385] This section explained various implementations of the process. It can be understood that each process can be configured to control each coating process using only one actuator per process, or to control a group of coating processes using one actuator per coating process.
[0386] The following are various examples that refer to what has been described previously and depicted in the figures.
[0387] Example 1 is a method for controlling a (e.g., reactive) coating process by which a substrate is coated, comprising: controlling a first actuator by which the coating process is supplied, based on a group of several control variables of the (e.g., reactive) coating process acquired in parallel, each of which is taken into account when controlling the first actuator, and optionally based on a control model; controlling a second actuator by which the coating process is supplied, based on the group of several acquired control variables of the (e.g., reactive) coating process, each of which is taken into account when controlling the second actuator, and optionally based on the control model; wherein the controlling of the first actuator and the second actuator is carried out using one or more than one control element (e.g.,(implemented by means of the control model), of which each control element implements a link between the control variable of the first actuator and the control variable of the second actuator.
[0388] Example 2 is a method for controlling a (e.g., reactive) coating process by which a substrate is coated, comprising the following: controlling a first actuator, by which the coating process is supplied, based on a group of several control variables of the (e.g., reactive) coating process acquired in parallel and optionally based on a control model; controlling a second actuator, by which the coating process is supplied, based on the group of several control variables of the (e.g., reactive) coating process and optionally based on the control model; wherein optionally the control variables of the group of several control variables acquired in parallel interact with each other; wherein the controlling of the first actuator and the second actuator is carried out using one or more than one control element (e.g.,(implemented by means of the control model), of which each control element implements a link between the control variable of the first actuator and the control variable of the second actuator.
[0389] Example 3 is the method according to Example 1 or 2, wherein the first actuator is configured to control the supply (e.g., feeding and / or removing) of a first material to the coating process, and wherein the second actuator is configured to control the supply (e.g., feeding and / or removing) of a second material (e.g., gas or atomized material) to the coating process, which differs from the first material.
[0390] Example 4 is the method according to one of Examples 1 to 3, wherein the first actuator and / or the second actuator are set up to supply the coating process with a material (e.g. reactive gas flow or atomization rate) (e.g. adding and / or removing it) which is consumed by the coating process (e.g. incorporated into a layer), e.g. by means of a reaction of the coating process.
[0391] Example 5 is the procedure according to one of Examples 1 to 4, wherein the controlled variables of the group of several parallel controlled variables differ from each other, for example in a material (gas or atomized material) which represents them.
[0392] Example 6 is the method according to one of Examples 1 to 5, wherein of the group of several parallel controlled variables: a first controlled variable represents a first gas pressure (e.g., first partial pressure of the gas) of the coating process; and / or a second controlled variable represents a second gas pressure (e.g., second partial pressure of the gas) of the coating process or an electrical quantity of the coating process.
[0393] Example 7 is the procedure according to one of Examples 1 to 6, wherein the controlled variables of the group of several controlled variables recorded in parallel interact with each other.
[0394] Example 8 is the method according to one of Examples 1 to 7, wherein the coating process takes place in a vacuum.
[0395] Example 9 is the procedure according to one of Examples 1 to 8, wherein the control of a first actuator and the control of the second actuator are linked together (e.g. implemented by means of the control model), e.g. as a function of the group of several acquired control variables of the (e.g. reactive) coating process (i.e. that their linkage together is a function of the group of several acquired control variables of the (e.g. reactive) coating process).
[0396] Example 10 is the procedure according to one of Examples 1 to 9, wherein the control of the first actuator influences each controlled variable of the group of several controlled variables recorded in parallel (e.g., fed back via the control model); and / or wherein the control of the second actuator influences each controlled variable of the group of several controlled variables recorded in parallel (e.g., fed back via the control model).
[0397] Example 11 is the procedure according to one of Examples 1 to 10, wherein the first actuator and the second actuator are controlled using one or more than one control element (e.g., implemented by means of the control model), each of which implements one or more than one of the following: an influence of each controlled variable of the group of several parallel controlled variables on a control variable of the first actuator; and / or (optionally) an influence of each controlled variable of the group of several parallel controlled variables on a control variable of the second actuator (for example, the influence can disappear even if it is taken into account).
[0398] Example 12 is the procedure according to one of Examples 1 to 11, wherein the control of the first actuator and the second actuator is carried out using one or more than one control element (for example, K). I , K P , KY ) occurs, which involves an interaction (for example, A exchange ) of the coating process with an additional coating process by means of which the substrate (e.g. adjacent to the coating process) is coated and / or which is upstream or downstream of the coating process.
[0399] Example 13 is the procedure according to one of Examples 1 to 12, wherein the first actuator and the second actuator are controlled in parallel.
[0400] Example 14 is a method according to one of Examples 1 to 13, further comprising: identifying one or more than one control element (e.g., implemented using the control model), each of which implements one or more than one of the following: an influence of each controlled variable of the group of several parallel controlled variables on a control variable of the first actuator; and / or (optionally) an influence of each controlled variable of the group of several parallel controlled variables on a control variable of the second actuator (for example, the influence can disappear even if it is taken into account); and / or a combination of the control variable of the first actuator and the control variable of the second actuator.
[0401] Example 15 is a method for controlling multiple coating processes, each of which is configured to deposit a layer on a substrate, comprising the method: controlling each of the multiple coating processes by means of the method according to one of Examples 1 to 14, wherein the multiple coating processes optionally comprise three or more coating processes, wherein optionally the controlling of each of the multiple coating processes is linked together (e.g. implemented by means of the control element).
[0402] Example 16 is a method for measuring a (e.g., reactive) coating process by means of which a substrate is coated, comprising the method of first determining a response of a first controlled variable and a second controlled variable of the coating process to a deflection of the coating process from a reference state when (e.g.,exactly) a first actuator, by means of which the coating process is supplied, is controlled (in other words, the displacement is caused by controlling a first actuator, by means of which the coating process is supplied); second, determining a response of the first controlled variable and the second controlled variable of the coating process to a displacement of the coating process from the reference state when (exactly) a second actuator, by means of which the coating process is supplied, is controlled; wherein the first determination and the second determination optionally occur sequentially; wherein optionally the several controlled variables interact with each other; the method optionally further comprising: determining a control model (e.g., the one in the method according to one of Examples 1 to 15) based on the response of the first controlled variable and / or the response of the second controlled variable, wherein, for example, the control model is set up, a specification (e.g.,the control variable) to be determined in the procedure according to one of examples 1 to 15; the procedure optionally further comprising: saving a result of the first determination and / or the second determination.
[0403] Example 17 is the procedure according to Example 16, further comprising: identifying one or more than one control element (e.g., implemented using the control model), each of which implements one or more than one of the following: an influence of each of the multiple controlled variables on a control variable of the first actuator; and / or (optionally) an influence of each of the multiple controlled variables on a control variable of the second actuator (for example, the influence can disappear even if it is taken into account); and / or a combination of the control variable of the first actuator and the control variable of the second actuator.
[0404] Example 18 is a method for providing a model generator, the method set up according to Example 16 and comprising: simulating one or more than one coating process (e.g., based on a model of the one or more than one coating process); training a model generator based on a result of the simulation (e.g., used as input data for the training), wherein, for example, the model generator is trained to map the result of the simulation to specifications of the one or more than one coating process or the model; and / or wherein, for example, the model generator is trained to determine a model (e.g., for control) of the one or more than one coating process.
[0405] Example 19 is the procedure according to Example 18, wherein the training of the model generator is further based on information about the model of one or more than one coating process (which is used, for example, as target data for training).
[0406] Example 20 is the procedure according to Example 18 or 19, wherein the model has one or more than one model member, and wherein the training of the model generator is further based on the one or more than one model member (which is used, for example, as the target data for training).
[0407] Example 21 is the method according to one of Examples 18 to 20, wherein the result of the simulation includes simulated information on controlling one or more than one coating process and / or simulated information on a response of one or more than one coating process to controlling one or more than one coating process.
[0408] Example 22 is the procedure according to one of Examples 18 to 21, wherein the result of the simulation specifies a configuration of the model of one or more than one coating process.
[0409] Example 23 is the procedure according to one of Examples 18 to 22, wherein the simulation includes: reconfiguring the model (e.g., one or more model members thereof) from a first configuration to a second configuration; wherein training the model generator is based on the result of the simulation at the first configuration and at the second configuration.
[0410] Example 24 is the procedure according to one of Examples 18 to 23, wherein the simulation includes: reconfiguring the model once or several times into several different configurations; wherein the training of the model generator is based on the result of the simulation for each of the multiple configurations.
[0411] Example 25 is the method according to any one of Examples 18 to 24, wherein the reconfiguration includes: adding or removing a coating process that represents the model; changing an interaction between at least two coating processes that represents the model; changing a material that supplies the coating process; and / or changing an operating point of one or more coating processes.
[0412] Example 26 is the procedure according to one of Examples 18 to 25, further comprising: filtering (e.g. by means of a noise filter) the result of the simulation; wherein the training of the model generator is based on the result of the filtering; wherein, for example, the filter increases a variance (e.g. standard deviation) of the result of the simulation and / or implements a random number generator.
[0413] Example 27 is the method according to any one of Examples 18 to 26, wherein the simulation comprises simulating a response of one or more than one coating process to a control of one or more than one coating process, for example when one or more than one simulated actuator of one or more than one coating process is controlled.
[0414] Example 28 is the method according to one of Examples 18 to 27, wherein the model generator is set up to determine a process model of one or more than one coating process, or at least components of the process model, based on a response of one or more than one coating process to a control (e.g., taking into account one or more than one manipulated variable and / or one or more than one controlled variable) of one or more than one coating process, for example, when one or more than one simulated actuator of one or more than one coating process is controlled.
[0415] Example 29 is the method according to one of Examples 18 to 28, further comprising: determining a process model and / or a control model (e.g., the one in the method according to one of Examples 1 to 15) of the coating process by means of the model generator based on data (also referred to as measurement data) acquired by one or more coating processes; wherein optionally the measurement data represent a (e.g., repetitive) deviation from a reference state; wherein optionally, for example, the control model is configured to determine control inputs in the method according to one of Examples 1 to 15; wherein optionally the measurement data include information on an actual state of one or more coating processes, e.g., when the coating process is being controlled.
[0416] Example 30 is the method according to Example 29, further comprising: controlling one or more than one coating process by means of the control model, for example in the method according to one of Examples 1 to 15.
[0417] Example 31 is a method for calibrating a coating process by which a substrate is coated, comprising: determining a response (e.g., a first controlled variable and / or a second controlled variable) of the coating process to at least one displacement of the coating process from a reference state when at least one (i.e., one or more than one) actuator by which the coating process is supplied is controlled; determining a control model based on the response and based on a trained model generator (to which the response is, for example, fed or which is configured to process the response); wherein optionally the multiple controlled variables interact with each other; wherein optionally the at least one displacement comprises multiple displacements from the reference state, which are carried out sequentially.
[0418] Example 32 is the procedure according to one of Examples 1 to 31, which is carried out using a computer.
[0419] Example 33 is a control device that is set up (e.g., whose processor is set up) to carry out the procedure according to one of Examples 1 to 32.
[0420] Example 34 is the control device according to Example 33, wherein the control device is configured to control at least one actuator (e.g. the first actuator and the second actuator) by means of which a coating process is supplied, using one or more than one control element (e.g. its actual state).
[0421] Example 35 are code segments that contain instructions which, when performed by a processor, are set up to cause the processor to perform the procedure according to any one of Examples 1 to 32.
[0422] Example 36 is a non-volatile data storage device that includes the code segments according to Example 35 and / or by means of which the control device according to Example 33 is set up.
[0423] Example 37 is the use of simulation data of at least one (e.g., modeled) coating process to train a model generator of Example 18 to determine a model of the at least one coating process.
Claims
[1] Method (100) for controlling a coating process (201) by means of which a substrate (102) is coated, the method (100) comprising: • Controlling (101) a first actuator (801a), by means of which the coating process (201) is supplied, on the basis of a group of several controlled variables of the coating process (201) recorded in parallel, each of which controlled variables is taken into account when controlling the first actuator (801a); and • Controlling (103) a second actuator (801b), by means of which the coating process (201) is supplied, on the basis of the group of several recorded control variables of the coating process (201), each of which control variables is taken into account when controlling the second actuator (801b); • wherein the control of the first actuator (801a) and the second actuator (801b) is carried out using at least one control element which implements a link between a control variable of the first actuator (801a) and a control variable of the second actuator (801b). [2] Method (100) according to claim 1, wherein the first actuator (801a) is configured to influence the supply of the coating process (201) with a first material, wherein the second actuator (801b) is configured to influence the supply of the coating process (201) with a second material which is different from the first material. [3] Method (100) according to claim 1 or 2, wherein the controlled variables of the group of several controlled variables recorded in parallel differ from one another. [4] Method (100) according to one of claims 1 to 3, wherein one or more than one controlled variable of the group of several controlled variables recorded in parallel to one another represents a gas pressure of the coating process (201). [5] Method (100) according to one of claims 1 to 4, wherein at least one of the group of several controlled variables recorded in parallel represents an electrical variable of the coating process (201). [6] Method (100) according to one of claims 1 to 5, wherein the controlled variables of the group of several controlled variables recorded in parallel interact with each other. [7] Method (100) according to one of claims 1 to 6, wherein the coating process (201) is carried out in a vacuum. [8] Method (100) according to one of claims 1 to 7, • wherein the control of the first actuator (801a) influences each of the plurality of controlled variables; and • wherein the control of the second actuator (801b) influences each of the plurality of controlled variables. [9] Method (100) according to one of claims 1 to 8, wherein the control of the first actuator (801a) and the second actuator (801b) is carried out using one or more than one control element, each control element of which implements an influence of each controlled variable of the group of several controlled variables detected in parallel to one another on a control variable of the first actuator (801a). [10] Method (100) according to one of claims 1 to 9, wherein the control of the first actuator (801a) and the second actuator (801b) is carried out using one or more than one control element, each control element of which implements an influence of each controlled variable of the group of several controlled variables detected in parallel to one another on a control variable of the second actuator (801b). [11] Method (100) according to one of claims 1 to 10, wherein the control of the first actuator (801a) and the second actuator (801b) is carried out using one or more than one control element which takes into account an interaction between the coating process (201) and an additional coating process (304a) by means of which the substrate (102) is coated. [12] Method (100) for controlling a coating process (201) by means of which a substrate (102) is coated, the method (100) comprising: • Controlling a first actuator (801a), by means of which the coating process (201) is supplied, on the basis of a group of several control variables of the coating process (201) recorded in parallel; • Controlling a second actuator (801b), by means of which the coating process (201) is supplied, on the basis of the group of several recorded control variables of the coating process (201); • where the controlled variables of the group of several controlled variables recorded in parallel interact with each other; • wherein the control of the first actuator (801a) and the second actuator (801b) is carried out using at least one control element which implements a link between a control variable of the first actuator (801a) and a control variable of the second actuator (801b). [13] Method (1300) for calibrating a coating process (201) by means of which a substrate (102) is coated, the method comprising: • first determining a response of a first controlled variable and a second controlled variable of the coating process (201) to a deflection of the coating process (201) from a reference state when a first actuator (801a), by means of which the coating process (201) is supplied, is controlled; • second determining a response of the first controlled variable and the second controlled variable of the coating process (201) to a deflection of the coating process (201) from the reference state when a second actuator (801b), by means of which the coating process (201) is supplied, is controlled; • where the first determination and the second determination take place one after the other. [14] Method (1700) according to claim 13, • wherein the first determination and / or the second determination are carried out by simulating (1701) the coating process (201); the method further comprising: • Training (1703) a model generator (1606) based on a result of the simulation, wherein the model generator (1606) is trained to determine a model (2020) of the one or more coating processes (201). [15] The method (2000) of claim 13, further comprising: • Determining (2003) a control model (2020) based on the response and based on the trained model generator (1606). [16] Control device (228) configured to carry out the method according to one of claims 1 to 15. [17] A non-volatile data memory comprising code segments comprising instructions arranged, when executed by a processor, to cause the processor to perform the method according to any one of claims 1 to 15.
Citation Information
Patent Citations
Methods for determining process-significant data from a vacuum deposition process and their further processing in measurement or control processes
DE102011017583A1
Method and control device
DE102019123410A1
