Dynamic process control in semiconductor manufacturing

Abstract

To provide methods and systems for advanced monitoring and machine learning in an atomic layer deposition (ALD) process.SOLUTION: A monitoring method 500 for steps or cycles in an ALD process includes: defining a time reference (start) of an ALD cycle (for example, an opening of a certain dose valve); collecting parameter data at regular intervals (for example, at so-called trigger points) based on a collection frequency (for example, at 1 Hz, 10 Hz, 100 Hz, or 1,000 Hz); comparing data for various parameters such as a chamber pressure, a precursor manifold pressure, a purge pressure, RF reflected power, and RF forward power; and generating an output of a curve fitting, or generating a variability report.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] [Priority Claim] This patent application claims the benefit of priority to U.S. Provisional Patent Application No. 62 / 885,667, entitled "Dynamic Process Control In Semiconductor Manufacturing," by Kumar et al., filed August 12, 2019, which is incorporated herein by reference in its entirety.

[0002] The present disclosure relates generally to dynamic process control in substrate processing, and in some examples to systems and methods for advanced monitoring and machine learning in atomic layer deposition (ALD) processes. Some examples also relate to dynamic process control and monitoring for chamber matching and gas line fill times. [Background technology]

[0003] Currently, many, if not most, parameters associated with substrate processing chambers are monitored to operate near the component's setpoint. For example, mass flow controller (MFC) flow and chamber pressure may contain a certain margin of error. To accommodate this error, it is common to allow upper or lower limits for the parameter to be set to a certain value or percentage. For example, in an ALD process, the opening and closing times of valves are monitored and reflected appropriately in the monitored parameters.

[0004] However, current process monitoring methods are generally only suitable for detecting relatively broad or gross malfunctions of a processing chamber or its components. While such gross detection is fine for steady-state or single-step situations, such as chemical vapor deposition (CVD) or plasma-enhanced chemical vapor deposition (PECVD) processes, it has limited use or application in multi-step processes such as ALD, where chamber conditions change on a millisecond basis.

[0005] The background art description provided herein is intended to provide a general context for the present disclosure. Work by the presently named inventors within the scope of what is described in this background art section, as well as aspects of the description that may not otherwise be considered prior art at the time of filing, are not admitted, expressly or impliedly, as prior art against the present disclosure. Summary of the Invention

[0006] The present disclosure generally relates to dynamic process control in semiconductor substrate manufacturing systems. Some examples provide systems and methods for advanced monitoring and machine learning in ALD processes. Some examples also relate to dynamic process control and monitoring for chamber matching and gas line fill times. While some examples are described in the context of semiconductor processing, they are equally applicable to substrate processing outside of such a context, for example, processing of dielectrics such as metals and photomasks.

[0007] In some examples, consecutive ALD cycles are monitored individually, and all steps of the ALD cycles are matched using curve fitting or a defined error margin. All ALD cycles are repeatable in some examples, and time bases can be defined to monitor the repeatability of different variables. In some examples, consecutive ALD cycles may be compared for different measured variables. Examples of variables may include chamber pressure, precursor delivery pressure (or precursor manifold pressure), reflected radio frequency (RF) energy, forward power, and burst purge pressure. Other monitored variables or parameters may also be used.

[0008] Some examples provide tool warning or error messages based on deviations from monitored values. Some examples include algorithms and software to achieve these goals.

[0009] An example embodiment provides a system for monitoring processing cycles in an atomic layer deposition (ALD) semiconductor manufacturing process. The example system includes a processing chamber for the ALD manufacturing process and one or more controllers configured to perform process monitoring operations, including defining a time base for the ALD cycle based on repetitive operations in the manufacturing process, accessing a golden curve including a series of parameter values for a series of data points at periodic time increments based on the time base, accessing a variability or tolerance margin for each data point in the golden curve, collecting parameter data based on the periodic time increments for one cycle of the ALD manufacturing process, dynamically monitoring whether a parameter value of the parameter data at a data point is within the variability or tolerance margin, and, based on determining that a parameter value has fallen outside the variability or tolerance margin, adjusting the manufacturing process so that a parameter value in a subsequent cycle matches the associated parameter value in the golden curve.

[0010] In some examples, the repetitive actions forming the basis of the time reference include opening and closing particular valves supplying the process chamber. In some examples, the parameter data is collected at regular intervals based on a collection frequency, the collection frequency being within a range of 0-1 Hz, 1-10 Hz, 10-100 Hz, or 100-1000 Hz.

[0011] In some examples, the regular intervals are based on trigger points in an ALD manufacturing process, each trigger point defining or based on a point in time in a step within the ALD manufacturing process.

[0012] In some examples, the operations further include comparing the parameter data collected at the trigger point to a corresponding parameter data set in the golden curve, hi some examples, the parameter data includes parameter values related to one or more of a precursor manifold pressure, a purge pressure, a conversion manifold pressure, a chamber pressure, a gas flow, RF reflected power, and RF forward power.

[0013] In a further example, an example system includes a processing chamber for a manufacturing process and one or more controllers configured to perform process monitoring operations, including identifying a parameter of the manufacturing process, generating a first curve of parameter values including the first parameter value based on a first cycle of the manufacturing process, identifying a second value of the parameter for a second cycle of the manufacturing process based on the curve generated for the first cycle, and adjusting the manufacturing process to match the first parameter value with the second parameter value.

[0014] In some examples, the operations further include generating a second curve including a plurality of parameter values derived from a second cycle of the manufacturing process, and curve fitting the second curve of parameter values to the first curve of parameter values.

[0015] In some examples, the curve fitting operation includes fitting a series of parameter value curves based on or derived from third and subsequent cycles of the manufacturing process to the first or second curve to generate a golden curve that defines a set of golden parameter values for the manufacturing process. In some examples, each cycle of the manufacturing process includes multiple steps of an ALD process, and the operation further includes matching the parameter values at each step of the ALD process to the parameter values in the set of golden parameter values.

[0016] In some examples, the golden curve includes golden parameter values for each step of the ALD process. In some examples, the identified parameters are associated with control variables of the manufacturing process, and the operations further include directly or indirectly using the first and second parameter values to identify a matching value between the value of the control variable in the second cycle and the value of the control variable in the first cycle, and adjusting the manufacturing process to match the first parameter value with the second parameter value.

[0017] In a further example, a self-learning system for monitoring processing steps in semiconductor manufacturing cycles is provided. One example system includes a processing chamber housing a semiconductor manufacturing process including a series of iterative semiconductor manufacturing cycles, each cycle including a plurality of processing steps and one or more controllers associated with the processing chamber configured to perform process monitoring operations, the operations including generating a set of reference golden parameter values for each step in the series of iterative cycles based on parameter data collected from the processing chamber, generating a machine learning model based on the set of reference golden values, and using the machine learning model to match parameter values in a second cycle in the series of iterative cycles with corresponding parameter values in a first cycle in the series of iterative cycles.

[0018] In some examples, the semiconductor manufacturing process is an ALD process, and the steps of each cycle in the series of repeating cycles include successive steps including a dose step, a purge step, a conversion step, and a purge step.

[0019] In some examples, the operations further include generating a golden curve including parameter value data at each successive step, and using the golden curve as training data for a machine learning model.

[0020] In some examples, the operations further include repeating and matching each cycle in the series of repeated cycles by matching parameter values in each successive step of the second cycle with corresponding parameter values in each successive step of the first cycle based on the machine learning model.

[0021] In some examples, the parameters include one or more of precursor manifold pressure, purge pressure, chamber pressure, gas flow, chamber temperature, RF reflected power, and RF forward power.

[0022] In some examples, the operations further include collecting performance data from the process chamber, identifying drift in the performance data, generating performance drift data, and incorporating the drift data into training data for a machine learning model.

[0023] In a further example, a system for monitoring process cycles in a semiconductor manufacturing process is provided, where the example system includes a process chamber for the manufacturing process and one or more controllers configured to perform process monitoring operations, the operations including identifying parameters of the manufacturing process and generating a first curve of parameter values based on a first cycle of the manufacturing process, the first curve including a first parameter value, the parameter value including a gas line fill time for a line supplying the processing chamber.

[0024] In some examples, the operations further include calculating a gas line fill time based on a pressure ramp-up time between opening a valve in a line supplying the processing chamber and establishing a constant pressure increase thereafter. In some examples, the operations further include identifying a second value of the parameter for a second cycle of the fabrication process based on the curve generated for the first cycle, generating a second curve including a plurality of parameter values derived from the second cycle of the fabrication process, and curve fitting the second curve of parameter values to the first curve of parameter values.

[0025] In some examples, the curve fitting operation includes fitting a series of parameter value curves based on or derived from third and subsequent cycles of the manufacturing process to the first or second curve to generate a golden curve that defines a set of golden parameter values for the manufacturing process. In some examples, each cycle of the manufacturing process includes multiple steps of an ALD process, and the operation further includes matching the parameter values at each step of the ALD process to the parameter values in the set of golden parameter values.

[0026] In some examples, the golden curve includes golden parameter values for each step of the ALD process.

[0027] In some examples, the identified parameter is associated with a control variable of a manufacturing process, and the operations further include directly or indirectly using the first and second parameter values to identify a matching value between the value of the control variable in the second cycle and the value of the control variable in the first cycle, and adjusting the manufacturing process to match the first parameter value with the second parameter value. [Brief explanation of the drawings]

[0028] Some embodiments will now be described with reference to the accompanying drawings, which are presented by way of example and not limitation.

[0029] [Figure 1] FIG. 1 is a schematic diagram of a reaction chamber to which some examples of the disclosed methods may be applied, according to some examples.

[0030] [Figure 2A] FIG. 2A displays aspects of a conventional method for monitoring chamber parameters, according to an exemplary embodiment. [Figure 2B] FIG. 2B displays aspects of a conventional method for monitoring chamber parameters, according to an exemplary embodiment. [Figure 2C] FIG. 2C displays aspects of a conventional method for monitoring chamber parameters, according to an exemplary embodiment.

[0031] [Figure 3] FIG. 3 is a graph 300 illustrating an exemplary curve fit for 30 ALD cycles, according to an exemplary embodiment.

[0032] [Figure 4] FIG. 4 is an illustration of a graph including tolerance margins, according to an example embodiment.

[0033] [Figure 5] FIG. 5 illustrates a flowchart of exemplary operations in a method, according to an exemplary embodiment.

[0034] [Figure 6] FIG. 6 shows a table of exemplary steps and parameters in an ALD cycle, according to an exemplary embodiment.

[0035] [Figure 7] FIG. 7 shows a table of example golden and corrective values for process control in an ALD cycle, according to an example embodiment.

[0036] [Figure 8] FIG. 8 illustrates an exemplary alignment operation in an alignment method, according to an exemplary embodiment.

[0037] [Figure 9] FIG. 9 illustrates the differences between exemplary chambers, according to an exemplary embodiment.

[0038] [Figure 10] FIG. 10 shows a table including a set of parameters for adjusting precursor manifold pressure in an ALD cycle, according to an example embodiment.

[0039] [Figure 11] FIG. 11 illustrates an exemplary alignment operation in an alignment method, according to an exemplary embodiment.

[0040] [Figure 12] FIG. 12 illustrates an example control operation in a method for intra-chamber dynamic process control, according to an example embodiment.

[0041] [Figure 13] FIG. 13 illustrates a table containing example parameters that can be changed to match different control variables, according to an exemplary embodiment.

[0042] [Figure 14] FIG. 14 illustrates steps in an ALD cycle, according to an example embodiment.

[0043] [Figure 15] FIG. 15 illustrates operations in a self-learning monitoring method, according to an exemplary embodiment.

[0044] [Figure 16] FIG. 16 illustrates operations in a data acquisition method, according to an exemplary embodiment.

[0045] [Figure 17] FIG. 17 illustrates a valve arrangement according to an exemplary embodiment.

[0046] [Figure 18] FIG. 18 is a schematic diagram of a shared split flow configuration of a gas manifold, according to an exemplary embodiment.

[0047] [Figure 19] FIG. 19 is a schematic diagram of an individual station split configuration of a gas manifold, according to an exemplary embodiment.

[0048] [Figure 20] FIG. 20 illustrates aspects of a method for determining gas line fill time, according to an exemplary embodiment.

[0049] [Figure 21] FIG. 21 illustrates aspects of a method for determining gas decay (or residence) time, according to an exemplary embodiment.

[0050] [Figure 22] FIG. 22 illustrates a flowchart of example operations in some methods, according to example embodiments. [Figure 23] FIG. 23 illustrates a flowchart of example operations in some methods, according to example embodiments. [Figure 24] FIG. 24 illustrates a flowchart of example operations in some methods, according to example embodiments. [Figure 25] FIG. 25 illustrates a flowchart of example operations in some methods, according to example embodiments.

[0051] [Figure 26] FIG. 26 is a block diagram illustrating an example of a machine that may implement or control one or more exemplary embodiments. DETAILED DESCRIPTION OF THE INVENTION

[0052] The following description includes systems, methods, techniques, instruction sequences, and computing machine program products that implement exemplary embodiments of the present disclosure. In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the exemplary embodiments. However, it will be apparent to those skilled in the art that the present disclosure may be practiced without these specific details.

[0053] A portion of the disclosure of this patent document may contain material that is subject to copyright protection. The copyright owner has no objection to the patent document or patent disclosure being reproduced by anyone so long as such patent document or patent disclosure remains in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights. The following notice applies to all data described or illustrated below that form part of this document: Copyright Lam Research Corporation, 2019-2020, All Rights Reserved.

[0054] Referring now to FIG. 1 , an example of a plasma-based processing chamber is shown. While the present subject matter can be used for a variety of semiconductor manufacturing and substrate processing operations, the plasma-based processing chamber will be described in the context of plasma-enhanced or radical-enhanced CVD or ALD operations in this example. Those skilled in the art will appreciate that other types of ALD processing techniques (e.g., thermal-assisted ALD operations) are also known, and non-plasma-based processing chambers may also be incorporated. An ALD tool is a specialized CVD processing system in which an ALD reaction occurs between two or more chemical species. The two or more chemical species, called precursor gases, are used to form thin film deposits of materials on substrates, such as silicon substrates used in the semiconductor industry. The precursor gases are sequentially introduced into an ALD processing chamber and react with the substrate surface to form a deposition layer. Typically, the substrate interacts with the precursors repeatedly, gradually depositing one or more films of material on the substrate, with the films gradually becoming thicker. In certain applications, multiple precursor gases can be used to form different types of films during a substrate manufacturing process.

[0055] FIG. 1 illustrates a plasma-based processing chamber 101, which includes a showerhead 103 (which may also be a showerhead electrode) and a substrate support assembly 107. The substrate support assembly 107 may include a pedestal, as described in detail below. Typically, the substrate support assembly 107 is intended to provide a substantially isothermal surface and may function as both a heating element and a heat sink for the substrate 105. The substrate support assembly 107 may also include an electrostatic chuck (ESC), as described above, that includes a heating element to aid in processing the substrate 105. The substrate 105 may include a substrate comprising an elemental semiconductor (e.g., silicon or germanium), a substrate comprising a compound element (e.g., gallium arsenide (GaAs) or gallium nitride (GaN)), or various other types of substrates, including conductive substrates, semiconductor substrates, and non-conductive substrates.

[0056] In operation, a substrate 105 is loaded into the substrate support assembly 107 through the loading port 109. A gas line 113 can supply one or more process gases (e.g., precursor gases) to the showerhead 103, which in turn supplies the one or more process gases to the plasma-based processing chamber 101. A gas source 111 (e.g., one or more precursor gas ampoules) for supplying the one or more process gases is coupled to the gas line 113. In some examples, an RF power source 115 is coupled to the showerhead 103. In other examples, a power source is coupled to the substrate support assembly 107 or the ESC.

[0057] Downstream of the gas line 113 before the inlet of the showerhead 103, a point-of-use (POU) and manifold combination (not shown) controls the entry of one or more process gases into the plasma-based processing chamber 101. For plasma-based processing chamber 101 used to deposit thin films in plasma-enhanced ALD (PEALD) operations, precursor gases may be mixed within the showerhead 103.

[0058] During operation, the plasma-based processing chamber 101 is evacuated by a vacuum pump 117. RF power is capacitively coupled between the showerhead 103 and a lower electrode (not shown) located in or on the substrate support assembly 107. The substrate support assembly 107 is typically supplied with two or more RF frequencies. For example, in various embodiments, the RF frequency may be selected from at least one of approximately 1 MHz, 2 MHz, 13.56 MHz, 27 MHz, 60 MHz, and other desired frequencies. Coils designed to block or partially block particular RF frequencies may be provided as needed. Therefore, the specific frequencies described herein are provided solely for ease of understanding. RF power energizes one or more process gases, forming a plasma in the space between the substrate 105 and the showerhead 103. The generated plasma assists in the deposition of various layers (not shown) on the substrate 105. In other applications, the plasma can be used to etch device features into various layers on the substrate 105. RF power is coupled through at least the substrate support assembly 107. A heater (not shown in FIG. 1) may be incorporated into the substrate support assembly 107. The details of the plasma-based processing chamber 101 may have a variety of designs.

[0059] In some examples, within a given overall cycle, ALD can be considered a multi-step (e.g., primarily four-step) process with dose, purge, conversion, and purge steps occurring within the processing chamber. Other cycles and steps are also applicable. Unlike certain PECVD processes where parameters such as gas flow, chamber pressure, and RF value remain constant throughout the deposition process (cycle), in ALD these process parameters (and others) can vary across the cycle or even within successive cycles at each step.

[0060] Unmonitored variations in parameters at different steps can mask fluctuations in some important control variables. For example, chamber pressure is traditionally required to be maintained at a constant or set pressure. Traditionally, variations in gas flow to the chamber at various steps in a multi-step process have been controlled by throttle valves that move continuously to maintain the set pressure. However, there is an inherent feedback delay problem in varying gas flow and activating throttle valves, which can often lead to control failures during, for example, ALD processes.

[0061] Similarly, during processing, the precursor manifold pressure in the gas line supplying a precursor gas, e.g., argon (Ar), fluctuates, most often due to increased or delayed changes in gas flow as the precursor gas enters or is diverted from the chamber. In other words, the pressure in a given gas line can vary within a certain range depending on whether the gas is flowing into the chamber (e.g., creating a burst purge pressure) or shunting (e.g., creating a flow pressure limit). These fluctuations are evident in many, if not all, ALD cycles and are considered undesirable. Deviations from parameter setpoints can indicate problems with chamber processing and can affect substrate and film properties.

[0062] 2A-2C, conventional methods for monitoring chamber parameters can include (by way of example only) monitoring the chamber pressure ( FIG. 2A ), precursor manifold pressure ( FIG. 2B ), and purge pressure ( FIG. 2C ) around a setpoint 202 and an error band 204 established between an upper limit 206 and a lower limit 208. The error band 204 is often relatively large—indeed, large enough to prevent detection of smaller, but potentially significant, fluctuations that occur during an ALD cycle. As a result, conventional setpoint and error band-based approaches cannot accurately or faithfully monitor more detailed aspects or chamber conditions during an ALD cycle. Recently, substrate manufacturers have been demanding more precise, controllable, and specific chamber control to fabricate high-aspect, nanometer-sized features and semiconductor devices on substrates.

[0063] In this regard, reference is now made to FIG. 3. In some examples, successive dose, purge, conversion, and (again) purge steps within an overall deposition or etch cycle are monitored and aligned using curve fitting. Some examples include defined error margins configured for each specific step, as opposed to a general error band defined for the entire cycle of the type described above. In some examples, each step of an ALD cycle (e.g., each of the four steps described above) is made repeatable based on curve fitting or defined step-specific tolerance margins. Time bases can be defined to monitor the repeatability of different variables.

[0064] In some examples, a reference frame for process parameter monitoring is established for a subsequent step or cycle by a previous or earlier step or cycle, as opposed to, for example, a conventional set point and error band. Indeed, some examples are agnostic (i.e., ignore) with respect to conventional set points or error bands, and operate on the basis of repeating a previous cycle or step in a given substrate manufacturing process. For example, parameters of an ALD cycle may be adjusted by trial and error to form a desired feature on a substrate. In examples of the present disclosure, a processing chamber may be configured to simply repeat successful cycles based on data derived from curve fitting or as defined by step-specific tolerance margins, even if the absolute values of those successful parameters are not necessarily known or can be identified.

[0065] FIG. 3 is a graph 300 depicting an exemplary curve fit for 30 ALD cycles. Chamber pressure (y-axis) was monitored in 100 millisecond (ms) time increments for the duration of each ALD cycle. The elapsed time from the start to the end of each ALD cycle ranged from approximately 1 second to less than 1 second, and chamber pressure data was collected at each time increment during that period, as shown. A set of curves 306 representing chamber pressure fluctuates between data peaks 302 and valleys 304 over the course of successive steps (e.g., dose, purge, conversion, and purge) within each ALD cycle. In the illustrated example, each ALD cycle begins with a dose step.

[0066] A curve fitting operation is performed for each curve 306 in the set of curves, so that successive or subsequent ALD cycles repeat the previous cycle. Each step within an ALD cycle can also repeat the previous cycle. In this regard, step-specific tolerance margins can be established for one or more time increments associated with that step. The tolerance margins can be based on observed or established variability data of the type shown in graph 400 of FIG. 4 . Here, chamber pressure variability (or tolerance margins) can be observed for each time increment. A relatively wide (or loose) tolerance margin 402 can correspond to a relatively wide set of chamber pressures, for example, represented by the set of curves at 0.4 time increments in FIG. 3 . Conversely, a relatively narrow (or strict) tolerance margin 404 can correspond to a relatively narrow (or equal) set of chamber pressures, for example, represented by the set of curves at 0.1 time increments in FIG. 3 . A variability or tolerance margin can be defined for each time increment during a step or ALD cycle. One or more curves 306 within a curve set can also be defined independently or based on tolerance margins. One or more curve sets 306 and / or step-specific tolerance margins can be established and used, for example, for step and cycle monitoring and repeatability during ALD cycles in a substrate processing operation.

[0067] 5 shows a flowchart of example operations of a method 500 for monitoring steps or cycles in an ALD process, for example. In some examples, operation 502 includes defining a time reference (start) of an ALD cycle (e.g., opening of a particular dose valve). In some examples, operation 502 may also include defining a "golden curve" or golden value for a series of data points (e.g., time increments). In some examples, operation 502 further includes defining a variability or tolerance margin for each data point (time increment).

[0068] In some examples, operation 504 includes collecting parameter data. In some examples, the parameter data is collected at regular intervals (e.g., at so-called trigger points) based on a collection frequency (e.g., 1 Hz, 10 Hz, 100 Hz, or 1000 Hz).

[0069] In some examples, operation 506 may include a data comparison, comparing data for various parameters such as chamber pressure, precursor manifold pressure, purge pressure, RF reflected power, and RF forward power. Parameter values for specific time increments (trigger points) are compared to a golden curve or golden value. In some examples, the data comparison can be performed in real time, at the end of a given cycle or step, or at the end of a given substrate manufacturing process. This may be user-defined or based on process needs or optimization.

[0070] In some examples, operation 508 includes generating an algorithm that enables repeatability of the steps and cycles. Exemplary algorithms may include (or be based on) one or more of curve fitting, standard deviation from a golden value, or minimum or maximum ranges for variability or tolerance margins. Other algorithmic factors may also be used.

[0071] In some examples, operation 510 includes reporting (e.g., generating a curve fitting output or generating a variability report). In some examples, operation 510 may include identifying and / or taking corrective action, for example, to adjust a process parameter based on the variability report or to bring the process parameter within a set of fitted curves or tolerance margins. Some examples may include, for example, taking no action if the process parameter falls within a set of fitted curves or tolerance margins. In response, a tool or processing chamber warning may (or may not) be generated.

[0072] Variations in process parameters can also complicate process control and monitoring in other areas. For example, substrate-to-substrate (or lot-to-lot) variations can be caused by chamber heat buildup during substrate processing. Tool-to-tool variations can be caused by differences in pump efficiency. Traditionally, efforts to control variation have focused primarily on monitoring the performance of individual devices. An example of a device and its associated parameters is an MFC flow with a device error limit set at 1% of flow. The MFC flow is monitored to operate within that limit during substrate processing. Devices and parameters can also include valve timing (e.g., ALD valve timing monitored to operate with a 50 ms open time and a 70 ms close time). In another example, a valve may be set to switch between open and closed positions in 25 ms. In another example, pedestal temperature may be controlled using a thermocouple that monitors deviations within a set range. RF power control can include monitoring forward and reflected power. These tools typically have inherent performance or response limitations that can cause poor chamber control and random or variable chamber conditions. Insufficient strictness of monitoring limits, as well as a multitude of process factors and tool limitations, can contribute to this undesirable effect.

[0073] 6, unlike many PECVD processes in which parameters such as gas flow, pressure, and RF power remain generally constant throughout the deposition process, several parameters in each cycle are continuously varied during the ALD process (cycle). Table 600 in FIG. 6 shows exemplary steps and associated parameters in an ALD cycle. These steps may include a dose, a post-dose purge, RF power application, and a purge, as shown.

[0074] Some examples herein address such challenges, including matching measured parameters as opposed to monitoring device parameters. Accordingly, some examples are configured to enable matching of equipment and / or chamber performance, for example, for a given set of process steps or cycles and across a set of process chambers within a processing tool. For example, pressure fluctuations or pressures generated during each ALD step can be matched from cycle to cycle or across tools and chambers by adjusting one or more process parameters occurring in the individual steps of the cycle. For example, a given pressure setpoint for an individual step (or cycle) in a substrate manufacturing process can be matched (or repeated) across steps or cycles by dynamic, real-time adjustment of gas flows.

[0075] Similarly, variations in precursor manifold pressure levels can affect deposition rate (depR) and, in some instances, can be a key variable in chamber matching. Meanwhile, precursor manifold pressure can be affected by one or more exemplary factors, such as precursor gas (e.g., Ar) push flow, precursor flow (or ampoule temperature), pump efficiency due to split flow, and chamber outlet or split flow timing. As discussed above, existing technologies attempt to control chamber parameters to setpoints by controlling and monitoring equipment settings, such as MFC flow, valve open / close times, and power to RF generators. Substrate-to-substrate, lot-to-lot, and tool-to-tool variability is controlled by using and monitoring device parameters. Controlling device parameters is assumed to enable matching. However, the actual conditions within the processing chamber or gas lines are not controlled or matched within the time constraints imposed by ALD steps and cycles.

[0076] Some examples include dynamic process control and monitoring for chamber matching. Examples include achieving matched chamber performance by matching measurement parameters instead of matching device parameters. In some examples, a "golden curve" or golden values may represent a desired set of parameter values for, for example, performing a given or desired substrate formation in a chamber. The chamber may be operated with control variables set to "golden" conditions or "golden" values. In some examples, the "golden curve" may be based on parameter data derived from one step or cycle and then used as a reference or reference point for repeating that step or cycle or for matching parameter values or control variables across, for example, chambers, substrates, and tools.

[0077] Table 700 in FIG. 7 shows an example set of golden values for relevant steps in an ALD cycle. Each pressure and gas flow row contains two values: a “golden” value (denoted simply as pressure or flow) and a “corrected” pressure or flow. In some examples, during operation, the pressure at each step is corrected so that the monitored pressure curve matches the golden value for the golden chamber. Similarly, the gas flow at each step can be corrected to match the golden flow value or golden pressure value. Adjustments in one parameter can cause a direct change in that parameter or can cause an indirect change in another parameter related to the first parameter (e.g., a change in chamber gas flow affects a corresponding change in the pressure of the chamber gas flow).

[0078] In some examples, pressure fluctuations or pressures present in the first step of an ALD cycle are matched in subsequent steps (or cycles) by adjusting pressure setpoints; thus, in some examples, the process is controlled to scale and match previous process parameters rather than by adjusting to given device parameters. In some examples, pressure matching adjustments are made for each individual step or cycle. In some examples, pressure matching is performed directly or indirectly, for example, by matching previous gas flows.

[0079] FIG. 8 illustrates an exemplary matching operation of matching method 800. In graph 802, it can be seen that a tool A pressure curve 804 in a first step or cycle does not match a tool B pressure curve 806 in a second step or cycle, and that the curves for their respective values do not match but are offset relative to one another. In the illustrated example, the tool B pressure value of curve 806 can be said to lag behind the tool A pressure value of curve 804. In some examples, adjustments are made to chamber parameters, such as gas flow or pressure parameters, so that the two curves 804 and 806 substantially match, as shown in graph 808. Accurately matched steps or cycles are useful in ensuring repeatability of chamber conditions and consistency in substrate manufacturing, resulting in improved precision in the fabrication of substrate formations and semiconductor devices.

[0080] In other examples, precursor manifold pressure curves for two different chambers during an ALD cycle may not match even though all relevant equipment on the tool is functioning within specifications. Chamber-to-chamber variability can result from differences in pump efficiency, differences in precursor manifold pressure and chamber valve timing, gas line temperature, etc. FIG. 9 depicts an example of chamber-to-chamber differences. For example, in graph 900, the chamber differences in pressure can be seen during time period 902. In graph 904, the respective (chamber-to-chamber) pressure differences can be seen. The illustrated pressure curves 906 and 908 are offset (mismatched) relative to each other in time and pressure magnitude.

[0081] In some examples, chamber-to-chamber matching issues are addressed. In this regard, Table 1000 in FIG. 10 illustrates a set of parameters for adjusting precursor manifold pressure in an ALD cycle. Exemplary parameters include precursor gas flow (e.g., argon (Ar)), precursor flow, ampoule inlet and outlet valve timing, and chamber inlet and outlet valve timing. Exemplary parameters may each have a corresponding golden value or golden curve. Other parameters may also be used.

[0082] The effect or influence of control on the precursor manifold pressure can be imparted by making specific adjustments to certain parameters, as shown, for example, in the comments section of Table 1000. For example, argon push flow can affect the magnitude of the precursor manifold pressure, shifting the pressure curve vertically. In other words, when matching a precursor manifold pressure curve established in a second or subsequent step (or cycle) with a corresponding curve established (or generated) in a previous step or cycle, if the matching graph (e.g., graph 802) appears "lower" than the previous curve, the pressure curve in subsequent cycles (such as the third and fourth cycles) can be corrected to move upward by adjusting the argon precursor push flow accordingly. Thus, the process chamber conditions can be configured to be essentially constant by monitoring parameter values obtained from previous steps or cycles, rather than based on fixed set points or error bands.

[0083] 11 illustrates an exemplary operation of a matching method 1100, which utilizes variations in parameters such as push flow (per ALD cycle), ampoule temperature, and valve timing to establish matching precursor manifold pressures 1102 and 1104 at each step or cycle of an ALD process. Matched precursor manifold pressures can further match deposition rates between dependent chambers.

[0084] Exemplary operations in a method for dynamic process control may include selecting a controlled variable (e.g., pressure) to match, determining a parameter to change to modify the controlled variable, determining a control increment for the parameter, and calculating the deviation of the controlled variable from the golden curve, etc. 12, examples of control operations in a method 1200 for dynamic process control in a chamber include identifying a controlled variable (e.g., pressure) and establishing a golden curve or value therefor in operation 1202, identifying a first adjustment parameter that can affect the value of the identified controlled variable in operation 1204, determining a magnitude or step increment in the identified parameter for the controlled variable in operation 1206, selecting the parameter increment having the greatest potential impact on the controlled variable in operation 1208, determining whether the controlled variable is within specification in operation 1210, if yes, repeating operations 1206 and 1208 in operation 1212, if no, applying the selected parameter increment in operation 1212, and determining whether the controlled variable is within specification after applying the selected parameter increment in operation 1214. If yes, method 1200 includes repeating operations 1206 and 1208. If no, the example method 1200 includes identifying a second adjustment parameter (e.g., a gas flow setting) that may affect the value of the identified controlled variable in operation 1216, and repeating the control actions summarized above for the second adjustment parameter in operations 1218-1226. Once the selected increments of the first and second adjustment parameters establish a golden value for the identified controlled variable, the chamber is optimized for pressure matching in 1218.

[0085] Some examples include a method for determining deviation from a golden curve in dynamic process control. An example of an operation in this method may include defining a time reference (e.g., a digital output signal for the opening of a chamber dosing valve) at the start of an ALD cycle. Then, all data points collected at regular intervals (e.g., 1 ms, 10 ms, 50 ms, 100 ms) are compared with corresponding data on the golden curve. Deviations or error limits from the golden curve can be predefined based on experimentation or user specifications. Once parameters are optimized for a particular measurement, e.g., chamber pressure, the effect of parameter changes on other control variables, e.g., precursor manifold pressure, is checked and adjusted as necessary. Table 1300 in FIG. 13 shows example parameters for correction to allow for alignment of different control variables.

[0086] Some examples herein achieve substrate-to-substrate, lot-to-lot, and chamber-to-chamber matching by matching dynamic parameter control to actual measurements, such as chamber pressure, gas line pressure, temperature, and delivered RF power. Some examples include parameter optimization to minimize measured deviations from a golden curve or value. Optimized parameters may be different for each chamber. Optimized parameters may be different for each accumulation to account for other factors (e.g., drift). Some examples include software features (described in more detail below) for performing parameter optimization for process control. The process control method can be performed regularly or intermittently, for example, during continuous operation or during tool startup or scheduled maintenance. Parameter optimization can be performed with assigned priorities and, in some examples, based on the control variables that most impact a given process.

[0087] In further aspects, some examples include self-learning techniques for advanced monitoring of processes, particularly ALD and CVD processes. These techniques address monitoring issues that can arise with traditional methods for controlling conventional equipment. For example, MFCs are typically monitored around a setpoint within a set error band. Valve timing control typically monitors the open and close times of ALD valves, which can switch between open and closed states during a given cycle or process in periods as short as 50 ms. RF generation is typically measured and controlled based on forward and reflected power, and then monitored within an error band. Chamber and other temperatures are controlled and monitored around a setpoint within an error band / percentage.

[0088] Conventional approaches to monitoring process parameters (as opposed to device parameters) involve monitoring chamber pressure around a setpoint within a set error band. The error band is often set large enough to ignore the inherent variations that occur during an ALD cycle. As a result, these approaches do not faithfully monitor the finer aspects or chamber conditions during an ALD cycle. Detailed and in-depth chamber control is increasingly being used to enable the fabrication of high-aspect nanosized substrate features and semiconductor devices. Furthermore, precursor manifold and burst purge pressures are also typically monitored around the band. Conventional error bands are typically set too wide to capture the smaller variations during an ALD cycle, resulting in the same problems discussed above.

[0089] Regarding the delivered RF power, a one-time check of the RF power simply checks whether the radio frequency is on or off after the RF strike. A voltage-current (VI) sensor monitors the RF power during the plasma "on" step, simply at a higher frequency (e.g., at 1 kHz) than the RF power. Thus, in a broad sense, existing methods are passive or "silent" monitoring methods based on limit or error band settings and typically rely on limited data. The monitoring bands are generally too wide to address, let alone solve, the ever-increasingly stringent process control challenges demanded in today's semiconductor manufacturing. Yet another drawback is that the same monitoring bands are applied to all tools, with no tool-to-tool modifications or customization. Customization is typically performed manually on an ad hoc basis. Typically, the prior art provides little or no performance comparison between substrates or tools, whether considering accumulation, after preventative maintenance, or after hardware changes.

[0090] As noted above, an ALD process can be viewed as a multi-step process. Referring to FIG. 14 , a typical ALD cycle 1400 includes four major steps: dose 1402, purge 1404, conversion 1406, and purge 1408. In some embodiments, each step of an ALD cycle, and subsequent cycles of a given ALD process, is individually monitored for different variables. The monitored variables are matched for each ALD step and / or cycle using curve fitting or defined error margins in a “smart” self-learning monitoring process. All ALD cycles are made repeatable, and the repeatability of parameters such as chamber pressure, precursor manifold pressure, and temperature can be monitored (and made repeatable) for every cycle or step.

[0091] FIG. 15 illustrates operations 1502-1522 in an example self-learning monitoring method 1500. Method 155 includes, in operation 1502, defining one or more parameters to be monitored, such as valve timing, VI sensor, RF forward direction, RF return direction, pressure (such as chamber pressure, precursor manifold pressure, or burst pressure), and other parameters. Initial data is collected to compare to starting values and define a baseline of quality, and may include a golden value or golden curve. The golden curve may be, for example, as shown in FIG. 3. Operation 1504 includes data analysis and logging of aspects such as mean value, standard deviation, out-of-control rate (OOC), and other statistics. Operation 1504 may further include updating performance data with new data, such as derived statistics. Operation 1506 may include monitoring chamber performance based on deviations from one or more statistics, such as mean performance. Operation 1508 may include determining whether the deviations are within device limits. If yes, previous actions, such as action 1504, may be repeated. Action 1510 may include determining whether the deviation is outside device limits. If no, action 1512 may include issuing a user review alert. If the review results in "no issues," previous actions in the method may be repeated as shown. If "issues exist," action 1522 may include issuing a tool alert. Action 1514 may include saving the tracker data at regular intervals (which may be user-definable). Action 1516 includes comparing the performance data to the tracker data to identify drift in the data. Action 1518 includes identifying whether there is drift in the data based on the comparison. If yes, action 1520 includes issuing a report to obtain user input and corrective response action.

[0092] An example of a self-learning monitoring method may include collecting data from multiple systems to define in-specification chamber performance for chamber-to-chamber and tool-to-tool performance matching. FIG. 16 illustrates exemplary operations 1602-1612 in a data collection method 1600. In operation 1602, multiple systems (e.g., process chambers) are identified. In operation 1604, the monitoring system obtains actual performance or golden curve values of parameters from selected modules within the identified process chambers (or tools). A collection frequency is identified. The collection frequency may be, for example, user-defined, daily, based on tracker data, or weekly.

[0093] In operation 1606, statistical process control (SPC) measurements are obtained from each chamber or tool. Operation 1608 includes performing a chamber-to-chamber (or tool-to-tool) performance comparison and analysis for each relevant parameter and defining mean performance and standard deviation for each chamber (e.g., chambers 1-3 in the figure). Operation 1608 may include generating warnings for outlier performance based on a user-defined criterion (e.g., 3σ) and establishing correlations with the obtained SPC data. Operation 1610 includes performing an overall chamber-to-chamber (or tool-to-tool) comparison and analysis and issuing tool warnings, if appropriate, in operation 1612.

[0094] In some examples, the smart self-learning monitoring system includes a machine learning component that creates a monitoring process based on a machine learning model. The component may also include a component that performs data preprocessing. The preprocessing component receives training data, such as from a processing chamber, or a set of golden values or golden curves (e.g., of the type shown in FIG. 3 ). The preprocessing component preprocesses the training data, including, for example, applying a MapReduce or similar function to the training data. A feature extraction component then extracts multiple features (e.g., process parameters) from the preprocessed training data and provides these features to a machine learning algorithm. The extracted features may be related to one or more of the control variables described above. In some examples, the machine learning algorithm learns weights assigned to each of the features and applies these weights to a function. The function and learned weights may be included in or constitute the machine learning model described above. The one or more machine learning models are stored in a file system and retrieved as needed to analyze chamber performance or perform process monitoring.

[0095] The machine learning algorithm can be selected from a wide variety of supervised or unsupervised machine learning algorithms. Examples of supervised machine learning algorithms include artificial neural networks, Bayesian networks, instance-based learning, support vector machines, random forests, linear classifiers, quadratic classifiers, k-nearest neighbors, decision trees, and hidden Markov models. Examples of unsupervised machine learning algorithms include the expectation-maximization (EM) algorithm, vector quantization, and information bottleneck methods. In one example embodiment, a binary logistic regression model is used. Binary logistic regression deals with situations where the observed outcome of a dependent variable can have only two possible types. Logistic regression is used to predict the odds that one or the other case is true based on the values of the independent variables (predictors). In a further example embodiment, a boosted tree gradient descent process is used for machine learning.

[0096] The functions included in the machine learning model can be evaluated at run time to generate a process match score. The match score is a prediction of the likelihood that an attempt to match conditions in multiple systems will result in a successful match based on evaluating various parameters and applying feature weights learned by the machine learning algorithm to the features. In some examples, the predicted match may include hybrid results or outputs that include parameter adjustments that result in chamber-to-chamber or tool-to-tool matching with improved reliability.

[0097] In a further example, a system and method, including hardware and software, for determining gas line fill time is provided. In some exemplary deposition or etching systems, gas is supplied from a gas box and includes several valves and filters between the gas source and the deposition or etching chamber. In this regard, typically, an MFC, MFC inlet and outlet valves, a filter, and a chamber inlet valve are used. The travel time of gas from a particular valve to the chamber can be defined as the gas line fill time.

[0098] In some instances, gas line fill times can be affected by certain factors. These factors can include MFC response and ramp time, which in some instances takes 1-3 seconds for the corresponding gas flow to reach within + / - 2% of the set point. Fill time factors can also include valve opening time. In some instances, valve opening time is in the range of several milliseconds, including pneumatic and other delays, and some instances include valve opening times in the range of <100 ms for pneumatically actuated valves. Other factors can include gas line conductance, including valve conductance, or pressure loss across filters. Essentially, gas velocity depends on differential pressure, which is affected or determined by gas line conductance. For deposition and etch processes that provide sufficient time for a given gas line fill, any delay in gas line fill will not significantly affect the deposition or etch process, if at all.

[0099] As further described above, an ALD process can include multiple steps that result in film formation on a substrate surface, which can include a dosing step in which precursor gas molecules attach to the substrate surface, a post-dose purge to remove excess precursor gas from the chamber, application of RF power to generate a plasma and convert a monolayer of gas molecules adsorbed on the surface into a film, and an RF purge to remove reaction by-products. The gases used in these steps are typically emitted from a variety of different manifolds.

[0100] Currently, gas line fill times are not measured or monitored. Valve open and close times can be monitored using advanced frequency (kHz) monitoring. ALD valves may include an optical sensor that detects the position of the valve restrictor. The timing of the valve open and close is monitored by the time difference between the command (digital input, DI) to the pneumatic bank that operates the valve and the readback of the optical sensor (digital output, DO) that senses the restrictor movement.

[0101] For example, referring to the valve arrangement 1700 shown in FIG. 17, in a deposition or etching system, gases are typically supplied from a gas box, and several valves and filters can be provided between one or more gas sources and the deposition / etching chamber. Typically, the valves include an MFC, MFC inlet and outlet valves, a filter, and a chamber inlet valve. The time it takes for gas to travel from a particular valve to the chamber can be defined as the gas line fill time. The gas line fill time depends on the MFC response and ramp time. In some instances, a valve typically takes 1 to 3 seconds for the gas flow to reach within + / - 2% of the set point. Valve opening times typically occur in the range of a few milliseconds. Including compressed dry air delays and other delays, the valve opening time can be less than 100 ms for pneumatically actuated valves. Valve conductance can include pressure loss across the filter. Essentially, gas velocity depends on the differential pressure, which in turn is controlled by the conductance of the gas line. While such gas line filling delays may not be a problem in deposition and etch processes where sufficient time is allowed for gas line filling, there are other cases where gas line filling delays are unacceptable. Given the need for reduced ALD cycle times for high throughput (and deep substrate formation capabilities), MFC ramp times or delays in substrate processing operations are unacceptable. Thus, in some instances, gas line fill times are determined to eliminate potential sources of delay or instability. Based on the fill times thus determined, the MFCs supplying the associated gases are configured to operate in a continuous or consistent manner.

[0102] The presence of shunts (which allow gases to bypass the chamber when not needed, for example, to the exhaust line or so-called foreline) can also affect gas line fill times. FIG. 18 is a schematic diagram of a shared shunt configuration 1800 for a gas manifold. Gases are diverted away from the chamber, typically resulting in relatively long gas line fills. FIG. 19 is a schematic diagram of an individual station shunt configuration 1900 for a gas manifold. Gases are diverted closer to the chamber, typically resulting in relatively short gas line fills.

[0103] In an ALD process, an individual step time may include the gas line fill time from the nearest outlet valve to the processing chamber. Thus, some exemplary embodiments address the important, yet complex, need to provide consistent, tool-to-tool, measurable gas line fill times.

[0104] Some exemplary embodiments automatically measure gas line fill times at various times (e.g., at start-up, after preventative maintenance, or at regular intervals). Current measurements are automatically compared to previous measurements, and any deviations in gas line fill times are reported. More broadly, measurements can also be compared to gas line fill times derived from other tools to determine tool-to-tool variations and tool suitability or status.

[0105] In some examples, a chamber vacuum gauge is used to measure or monitor gas line fill times. For example, the time difference between the opening of a manifold outlet valve and the increase in chamber pressure is used to measure or calculate gas line fill times in some examples. Some exemplary methods include establishing a base or constant pressure in the processing chamber supplied by a gas pump. The method includes closing a pump throttle and / or a diverter valve to isolate the chamber from the pump. Gas flow is set to be diverted from the appropriate manifold or gas line for which the line fill time is being measured. The method further includes opening the gas manifold outlet valve, closing the diverter valve, and measuring the increase in chamber pressure. Typically, there is some initial delay before the chamber pressure begins to increase. The pressure ramp curve used in some examples can be used to calculate gas line fill times or delays. In other words, the initial delay of the pressure ramp indicates the time it takes for the gas lines to fill into the chamber. The calculated delay (fill time) can be factored into control and monitoring systems and methods, including gas flow algorithms, to improve steady-state operation and chamber matching of the substrate processing chamber.

[0106] An example method for determining gas line fill time will now be described with reference to graph 2000 in FIG. 20 . Graph 2000 includes line 2002, which represents the movement (closed or open) of a gas valve that supplies gas to or removes gas from a processing chamber. In this case, an example valve is a diversion valve, as shown in the graph key, which controls gas in a line that bypasses the chamber. When the diversion valve is closed, gas is not diverted from the chamber but is instead directed into the chamber. Gas enters the chamber, increasing the pressure within the chamber. Typically, the closure of the diversion valve occurs simultaneously with the opening of a chamber supply valve to allow supply to the chamber. In either case, in this example, a gas line fill time for the diversion valve is established.

[0107] Line 2002 represents the physical closure of the diverter valve from an open (divert) position at position "1" on the y-axis to a closed position at "0" on the same axis. As noted above, when the diverter valve closes, the pressure in the chamber increases, and this increase is represented by pressure line 2004 on graph 2000. After a period of time, the slope of line 2004 becomes constant, i.e., represents the uniform or steady increase in gas pressure that occurs in response to a constant or steady inflow of gas into the chamber. The slope of line 2004 increases at higher gas flows. The gas line fill time is determined by extrapolating the slope of line 2004 until it intersects the x-axis. This intersection is indicated on graph 2000 at position 2006, and the period of time indicated by time line 2008 extending between intersection 2006 and the final closure of the diverter valve at position 2010 is the gas line fill time. In other words, there is a ramp-up or delay time (i.e., gas line fill time) between the closing of the diversion valve (i.e., full entry or maximum fill of gas into the chamber) and the point at which the pressure in the chamber increases at a constant rate.

[0108] In some embodiments, a gas decay time is specified. Closing the gas outlet valve does not mean an immediate cessation of gas flow to the chamber; there is a delay in commanding the valve to close and in the valve physically reaching full closure, as well as delays caused by other components. Even when the valve is closed, gas already in the gas line downstream of the valve still continues to the chamber.

[0109] In some examples, a method for establishing a gas decay (or residence) time may include one or more of the following operations: First, the chamber is loaded at a base or constant pressure. The chamber is supplied by a throttle and slit valve that controls the inflow of gas from a supply pump. The supply throttle and slit valve supplying the chamber are closed to isolate the chamber from the supply pump. Gas flow is set to shunt from the manifold or gas line for which the gas decay time is to be measured. The gas manifold outlet valve is opened and the diversion valve is closed. The rise in chamber pressure is measured. Then, the diversion valve is opened and the gas outlet valve is closed. The method further includes measuring the time for the chamber pressure to maximize or stabilize. This time to maximize or stabilize period is a measure of the time it takes for gas to stop flowing into the chamber (i.e., the gas decay or residence time) after a valve close command is sent to the outlet valve.

[0110] FIG. 21 shows a graph 2100 illustrating an exemplary gas line fill time and an exemplary gas decay time. Line 2102 in the graph represents the opening and closing of a valve supplying a process chamber. The valve is closed for a period indicated by 2104 and opened for a period indicated by 2106, allowing gas to enter the chamber and increasing the pressure within the chamber. The valve is closed again at 2108. The chamber pressure is measured by the chamber gauge and is represented by pressure line 2110. As discussed above, the gas line fill time can be determined from the period or Δ indicated by 2112 (i.e., the period between the intersection of the extrapolated pressure gradient with the x-axis and the moment the valve is commanded to open at 2102). The gas decay time is determined from the time it takes to stabilize after the valve closes at 2108. This gas decay time is more clearly visualized in the zoomed-in view at 2114 (80 ms in this example).

[0111] Thus, some examples provide methods for measuring gas line fill times using chamber pressure and for measuring gas decay (or residence) times using chamber pressure. These values may be integrated into dynamic monitoring processes and software for automated parameter measurement, chamber control, and matching techniques. Fill and decay (residence) values can be important in monitoring tool-to-tool (or chamber-to-chamber) variability and day-to-day drift in gas line fill or decay over time. Such methods are applicable to product lines employing multiple step procedures (or cycles) for deposition or etching, requiring very fast step and cycle times. With faster step and cycle times, gas line fill and decay times can become important. The present embodiments enable their measurement and monitoring.

[0112] 22 , a method 2200 for monitoring processing cycles in an ALD semiconductor manufacturing process includes, in operation 2202, defining a time base for an ALD cycle based on repetitive operations in the manufacturing process, in operation 2204, accessing a golden curve including a series of parameter values for a series of data points at periodic time increments based on the time base, in operation 2206, accessing a variability or tolerance margin for each data point in the golden curve, in operation 2208, collecting parameter data based on one cycle of the periodic time increments in the ALD manufacturing process, in operation 2210, dynamically monitoring whether a parameter value of the parameter data at a data point is within the variability or tolerance margin, and, in operation 2212, adjusting the manufacturing process so that a parameter value in a subsequent cycle matches an associated parameter value in the golden curve based on determining that the parameter value is outside the variability or tolerance margin.

[0113] In some instances, the repetitive actions that form the basis of the time reference include the opening and closing of particular valves that supply a processing chamber.

[0114] In some examples, the parameter data is collected at regular intervals based on a collection frequency, the collection frequency being within the range of 0-1 Hz, 1-10 Hz, 10-100 Hz, or 100-1000 Hz.

[0115] In some examples, the regular intervals are based on trigger points in an ALD manufacturing process, each trigger point defining or based on a point in time in a step within the ALD manufacturing process.

[0116] In some examples, the operations further include comparing the parameter data collected at the trigger point to a corresponding set of parameter data in the golden curve.

[0117] In some examples, the parameter data includes parameter values related to one or more of precursor manifold pressure, purge pressure, conversion manifold pressure, chamber pressure, gas flow, RF reflected power, and RF forward power.

[0118] Referring to FIG. 23 , a method 2300 for monitoring processing cycles in a semiconductor manufacturing process includes, in operation 2302, identifying a parameter of the manufacturing process; in operation 2304, generating a first curve of parameter values including a first parameter value based on a first cycle of the manufacturing process; in operation 2306, identifying a second value of the parameter for a second cycle of the manufacturing process based on the curve generated for the first cycle; and in operation 2308, adjusting the manufacturing process to match the first parameter value with the second parameter value.

[0119] In some examples, the operations further include generating a second curve including a plurality of parameter values derived from a second cycle of the manufacturing process, and curve fitting the second curve of parameter values to the first curve of parameter values.

[0120] In some examples, the curve fitting operation includes fitting a series of parameter value curves based on or derived from the third and subsequent cycles of the manufacturing process to the first or second curve to generate a golden curve that defines a set of golden parameter values for the manufacturing process.

[0121] In some examples, each cycle of the manufacturing process includes multiple steps of an ALD process, and the operations further include matching parameter values in each step of the ALD process to parameter values in a golden parameter value set.

[0122] In some examples, the golden curve includes golden parameter values for each step of the ALD process.

[0123] In some examples, the identified parameter is associated with a control variable of a manufacturing process, and the operations further include directly or indirectly using the first and second parameter values to identify a matching value between the value of the control variable in the second cycle and the value of the control variable in the first cycle, and adjusting the manufacturing process to match the first parameter value with the second parameter value.

[0124] Referring to FIG. 24 , a machine learning method 2400 for monitoring processing steps in a semiconductor manufacturing cycle includes, in operation 2402, generating a set of reference golden parameter values for each step in a series of iterative cycles based on parameter data collected from a processing chamber, in operation 2404, generating a machine learning model based on the set of reference golden values, and in operation 2406, using the machine learning model to match parameter values in a second cycle in the series of iterative cycles with corresponding parameter values in a first cycle in the series of iterative cycles.

[0125] In some examples, the semiconductor manufacturing process is an ALD process, and the steps of each cycle in the series of repeating cycles include successive steps including a dose step, a purge step, a conversion step, and a purge step.

[0126] In some examples, the operations further include generating a golden curve including parameter value data at each successive step, and using the golden curve as training data for a machine learning model.

[0127] In some examples, the operations further include repeating and matching each cycle in the series of repeated cycles by matching parameter values in each successive step of the second cycle with corresponding parameter values in each successive step of the first cycle based on the machine learning model.

[0128] In some examples, the parameters include one or more of precursor manifold pressure, purge pressure, chamber pressure, gas flow, chamber temperature, RF reflected power, and RF forward power.

[0129] In some examples, the operations further include collecting performance data from the process chamber, identifying drift in the performance data, generating performance drift data, and incorporating the drift data into training data for a machine learning model.

[0130] Referring to FIG. 25 , a method 2500 for monitoring processing cycles in a semiconductor manufacturing process includes, in operation 2502, identifying parameters of the manufacturing process, and, in operation 2504, generating a first curve of parameter values based on a first cycle of the manufacturing process, the first curve including a first parameter value, the parameter value including a gas line fill time of a line supplying a processing chamber.

[0131] In some examples, the operations further include calculating a gas line fill time based on a pressure ramp-up time between opening a valve in a line supplying the processing chamber and the subsequent establishment of a constant pressure increase.

[0132] In some examples, the operations further include identifying a second value of the parameter for a second cycle of the manufacturing process based on the curve generated for the first cycle, generating a second curve including a plurality of parameter values derived from the second cycle of the manufacturing process, and curve fitting the second curve of parameter values with the first curve of parameter values.

[0133] In some examples, the curve fitting operation includes fitting a series of parameter value curves based on or derived from the third and subsequent cycles of the manufacturing process to the first or second curve to generate a golden curve that defines a set of golden parameter values for the manufacturing process.

[0134] In some examples, each cycle of the manufacturing process includes multiple steps in an ALD process, and the operations further include matching parameter values in each step of the ALD process to parameter values in a golden parameter value set.

[0135] In some examples, the golden curve includes golden parameter values for each step of the ALD process.

[0136] In some examples, the identified parameter is associated with a control variable of a manufacturing process, and the operations further include directly or indirectly using the first and second parameter values to identify a matching value between the value of the control variable in the second cycle and the value of the control variable in the first cycle, and adjusting the manufacturing process to match the first parameter value with the second parameter value.

[0137] 26 is a block diagram illustrating an example of a machine 2600 that can control one or more example embodiment processes described herein. In alternative embodiments, the machine 2600 may operate as a standalone device or may be connected (e.g., networked) to other machines. In a network deployment, the machine 2600 can operate in the capacity of a server machine, a client machine, or both in a server-client network environment. In one example, the machine 2600 can operate as a peer machine in a peer-to-peer (P2P) (or other distributed) network environment. Furthermore, although only a single machine 2600 is shown, the term “machine” should also be interpreted to include any collection of machines that individually or collectively execute a set (or sets) of instructions to implement any one or more of the methodologies discussed herein, such as via cloud computing, software as a service (SaaS), or other computer cluster configuration. In some examples, as shown in FIG. 26, a non-transitory machine-readable medium includes instructions 2624 that, when read by a machine 2600, cause the machine to control operation in a manner that includes at least the non-limiting example operations summarized above and described herein.

[0138] Examples described herein may include or operate with logic, several components, or mechanisms. A circuit set is a collection of circuits implemented in tangible entities including hardware (e.g., simple circuits, gates, logic, etc.). Membership of a circuit set is flexible over time and to accommodate variability in the underlying hardware. A circuit set includes members that, alone or in combination, can perform a specific operation when in operation. In one example, the hardware of a circuit set may be invariably designed (e.g., hardwired) to perform a specific operation. In one example, the hardware of a circuit set may include variably connected physical components (e.g., execution units, transistors, simple circuits, etc.) that include computer-readable media that have been physically modified (e.g., magnetically, electrically, by a movable arrangement of invariant mass particles, etc.) to encode instructions for a specific operation. When connecting the physical components, the underlying electrical properties of the hardware components are changed (e.g., from an insulator to a conductor or vice versa). The instructions enable embedded hardware (e.g., an execution unit or loading mechanism) to create members of a circuit set within the hardware via variable connections and perform some of the specific operations when in operation. Thus, the computer-readable medium is communicatively coupled to other components of the circuit set when the device is in operation. In one example, any one of the physical components may be used by multiple members of multiple circuit sets. For example, during operation, an execution unit may be used by a first circuit of a first circuit set at one time and reused by a second circuit in the first circuit set or a third circuit in the second circuit set at another time.

[0139] The machine (e.g., computer system) 2600 may include a hardware processor 2602 (e.g., a central processing unit (CPU), a hardware processor core, or any combination thereof), a graphics processing unit (GPU) 2632, a main memory 2604, and a static memory 2606, some or all of which may communicate with each other via an interlink (e.g., a bus) 2608. The machine 2600 may further include a display device 2610, an alphanumeric input device 2612 (e.g., a keyboard), and a user interface (UI) navigation device 2614 (e.g., a mouse). In one example, the display device 2610, the alphanumeric input device 2612, and the UI navigation device 2614 may be touchscreen displays. The machine 2600 may further include a mass storage device (e.g., a drive unit) 2616, a signal generating device 2618 (e.g., a speaker), a network interface device 2620, and one or more sensors 2630 (such as a Global Positioning System (GPS) sensor, a compass, an accelerometer, or another sensor). The machine 2600 may include an output controller 2628, such as a serial (e.g., Universal Serial Bus (USB)), parallel, or other wired or wireless (e.g., infrared (IR), near field communication (NFC), etc.) connection, for communicating with or controlling one or more peripheral devices (e.g., a printer, a card reader, etc.).

[0140] Mass storage device 2616 may include machine-readable medium 2622. This machine-readable medium 2622 stores one or more sets of data structures or instructions 2624 (e.g., software) that embody or are utilized by any one or more of the techniques or functions described herein. Also, as shown, instructions 2624 may reside, completely or at least partially, within main memory 2604, static memory 2606, hardware processor 2602, or GPU 2632 during execution by machine 2600. In one example, machine-readable medium 2622 may be constituted by any one of hardware processor 2602, GPU 2632, main memory 2604, static memory 2606, or mass storage device 2616, or any combination thereof.

[0141] Although machine-readable medium 2622 is shown as a single medium, the term "machine-readable medium" may include a single medium or multiple media (e.g., a centralized or distributed database and / or associated caches and servers) configured to store one or more instructions 2624.

[0142] The term "machine-readable medium" may include any medium capable of storing, encoding, or carrying instructions 2624 for execution by machine 2600 and causing machine 2600 to perform any one or more of the techniques of this disclosure, or any medium capable of storing, encoding, or carrying data structures used by or related to such instructions 2624. Non-limiting examples of machine-readable media may include solid-state memory, optical media, and magnetic media. In one example, a high-capacity machine-readable medium includes a machine-readable medium 2622 having a plurality of particles with an unchanging (e.g., stationary) mass. Thus, a high-capacity machine-readable medium is not a transitory, propagating signal. Specific examples of high-capacity machine-readable media may include non-volatile memory such as semiconductor memory devices (e.g., electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM)) and flash memory devices, magnetic disks such as internal hard disks and removable disks, magneto-optical disks, and CD-ROM and DVD-ROM disks. Additionally, the instructions 2624 may be transmitted or received over a communications network 2626 via the network interface device 2620 using a transmission medium.

[0143] While examples have been described with reference to certain exemplary embodiments or methods, it will be apparent that various modifications and changes can be made to these embodiments without departing from the broader scope of the embodiments. Accordingly, the specification and drawings are to be considered in an illustrative and not a restrictive sense. The accompanying drawings, which form a part of this specification, show, by way of example, and not by way of limitation, specific embodiments in which the subject matter may be practiced. The illustrated embodiments are described in sufficient detail to enable those skilled in the art to practice the teachings disclosed herein. Other embodiments may be utilized and derived from the teachings disclosed herein, such that structural and logical substitutions and changes can be made without departing from the scope of the present disclosure. Therefore, this detailed description is not to be construed in a limiting sense, and the scope of various embodiments is defined only by the appended claims, along with the full range of equivalents to which such claims are entitled.

[0144] Although such embodiments of the inventive subject matter may be individually and / or collectively referred to herein by the term "invention," this is merely a matter of convenience and is not intended to voluntarily limit the scope of this application to any single invention or inventive concept (if in fact multiple are disclosed). Accordingly, although specific embodiments have been illustrated and described herein, it should be recognized that any configurations calculated to achieve the same purpose may be substituted for the specific embodiment shown. The present disclosure is intended to cover any and all adaptations or variations of various embodiments. Combinations of the above embodiments with other embodiments not specifically described herein will be apparent to those of skill in the art upon reviewing the above description.

Claims

1. 1. A system for monitoring processing cycles in a manufacturing process, comprising: a processing chamber for the manufacturing process; and one or more controllers configured to perform process monitoring operations; The operation is identifying parameters of the manufacturing process; generating a first curve of parameter values based on a first cycle of the manufacturing process, the first curve including a first parameter value, the parameter value including a gas line fill time of a line supplying the processing chamber; system.

2. 2. The system of claim 1, wherein the operation comprises: The system further includes calculating the gas line fill time based on a pressure ramp-up time between opening a valve in the line supplying the processing chamber and establishing a constant pressure increase thereafter.

3. 3. The system of claim 2, wherein the operation comprises: Based on the curve generated for the first cycle, identifying a second value of the parameter for a second cycle of the manufacturing process; generating a second curve comprising a plurality of parameter values derived from the second cycle of the manufacturing process; and curve fitting the second curve of parameter values to the first curve of parameter values; The system further includes:

4. 4. The system of claim 3, wherein the curve fitting operation comprises: fitting a series of parameter value curves based on or derived from third and subsequent cycles of the manufacturing process to the first or second curve to generate a golden curve defining a set of golden parameter values for the manufacturing process. Including, the system.

5. 5. The system of claim 4, wherein each cycle of the manufacturing process includes multiple steps in an ALD process; The operation is Matching the parameter values at each step of the ALD process with the parameter values in the golden parameter value set. The system further includes:

6. 5. The system of claim 4, wherein the golden curve comprises golden parameter values for each step of the ALD process.

7. 6. The system of claim 5, wherein the identified parameters are associated with control variables of the manufacturing process; The operation is directly or indirectly using the first and second parameter values to identify a match between the value of the control variable in the second cycle and the value of the control variable in the first cycle; and adjusting the manufacturing process to match the first parameter value with the second parameter value; The system further includes:

8. 1. A method for monitoring a processing cycle in a manufacturing process, comprising: identifying parameters of the manufacturing process; generating a first curve of parameter values based on a first cycle of the manufacturing process, the first curve including a first parameter value, the parameter value including a gas line fill time of a line supplying the processing chamber; method.

9. 9. The method of claim 8, wherein the action comprises: calculating the gas line fill time based on a pressure ramp-up time between opening a valve in the line supplying the processing chamber and establishing a constant pressure increase thereafter; The method further comprises:

10. 10. The method of claim 9, wherein the action comprises: Based on the curve generated for the first cycle, identifying a second value of the parameter for a second cycle of the manufacturing process; generating a second curve comprising a plurality of parameter values derived from the second cycle of the manufacturing process; and curve fitting the second curve of parameter values to the first curve of parameter values; The method further comprises:

11. 11. The method of claim 10, wherein the curve fitting operation comprises: fitting a series of parameter value curves based on or derived from third and subsequent cycles of the manufacturing process to the first or second curve to generate a golden curve defining a set of golden parameter values for the manufacturing process.

12. 12. The method of claim 11, wherein each cycle of the manufacturing process comprises multiple steps in an ALD process; The operation is Matching the parameter values at each step of the ALD process with the parameter values in the golden parameter value set. The method further comprises:

13. 12. The method of claim 11, wherein the golden curve comprises golden parameter values for each step of the ALD process.

14. 13. The method of claim 12, wherein the identified parameters are associated with control variables of the manufacturing process; The operation is directly or indirectly using the first and second parameter values to identify a match between the value of the control variable in the second cycle and the value of the control variable in the first cycle; and adjusting the manufacturing process to match the first parameter value with the second parameter value; The method further comprises:

15. A machine-readable medium that, when read by a machine, Identifying manufacturing process parameters; instructions to execute an operation of generating a first curve of parameter values based on a first cycle of the manufacturing process, the first curve including a first parameter value, the parameter value including a gas line fill time of a line supplying the processing chamber; Machine-readable medium.

16. 16. The medium of claim 15, wherein the action comprises: calculating the gas line fill time based on a pressure ramp-up time between opening a valve in the line supplying the processing chamber and establishing a constant pressure increase thereafter; The medium further includes:

17. 17. The medium of claim 16, wherein the action comprises: identifying a second value of the parameter for a second cycle of the manufacturing process based on the curve generated for the first cycle; generating a second curve comprising a plurality of parameter values derived from the second cycle of the manufacturing process; and curve fitting the second curve of parameter values to the first curve of parameter values; The medium further includes:

18. 18. The medium of claim 17, wherein the curve fitting operation comprises: fitting a series of parameter value curves based on or derived from third and subsequent cycles of the manufacturing process to the first or second curve to generate a golden curve defining a set of golden parameter values for the manufacturing process.

19. 20. The medium of claim 18, wherein each cycle of the manufacturing process includes multiple steps in an ALD process; The operation is Matching the parameter values at each step of the ALD process with the parameter values in the golden parameter value set. The medium further includes:

20. 20. The medium of claim 18, wherein the golden curve comprises golden parameter values for each step of the ALD process.

21. 16. The medium of claim 15, wherein the identified parameters are associated with control variables of the manufacturing process; The operation is directly or indirectly using the first and second parameter values to identify a match between the value of the control variable in the second cycle and the value of the control variable in the first cycle; and adjusting the manufacturing process to match the first parameter value with the second parameter value; The medium further includes:

Images