DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claims 1-27 have been presented for examination based on the amendment filed on 06/29/2023.
Claims 2-3 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite.
Claim(s) 1-6, 21-27 is/are rejected under 35 U.S.C. 102(a)(2) as being anticipated by US Patent No. 8036869B2 by Strang & Mitrovic.
Claim(s) 7-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over US Patent No. 8036869B2 by Strang & Mitrovic in further view of US Patent No. 20200192325 by (Sadeghi et al.)
This action is made Non-Final.
---- This page is left blank after this line ----
Claim Objections
Claim 1 objected to because of the following informalities: Non-obvious acronym "HFS" needs to be shown in expanded form.. Appropriate correction is required.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 2-3 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Regarding claim 2, the term "different model type" is a relative term that renders the claim indefinite. This term relies on the list of model types provided in claim 1. However, claim 1 introduces these types with the phrase "that is one of," which does not clearly indicate if the list is exhaustive or merely exemplary. (2173.05(d)) If the list is not exhaustive, the scope of what constitutes a "different" model type is unclear, and a person of ordinary skill in the art would not be reasonably apprised of the boundaries of the claim. (See MPEP § 2173.05(h), subsection I)
Regarding claim 3, the term "different class of physical phenomena" is similarly indefinite. It relies on a list in claim 1 that is likewise non-exhaustive ("that is one of"). (2173.05(d)) Without a clearly circumscribed set of physical phenomena classes, the metes and bounds of the requirement for the models to represent "different" classes cannot be determined with reasonable certainty. (See MPEP § 2173.05(h), subsection I)
Claim Rejections - 35 USC § 102
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Claim(s) 1-6, 21-27 is/are rejected under 35 U.S.C. 102(a)(2) as being anticipated by US Patent No. by Strang & Mitrovic.
Regarding Claim 1
Strang & Mitrovic teaches A digital twin of a process chamber ([Col 17 Line 27-52]: “As seen in FIG. 8, the system includes a model analysis processor 840, which is coupled to the simulation module 806 and configured to receive a simulation result from the module 806. In the embodiment of FIG. 8, model analysis includes the construction of an empirical model from non-dimensionalization of the simulation results. As simulation results are received on a run-to-run or batch-to-batch basis, an empirical model is constructed and stored in the empirical model library 842. For example, the process tool 802 undergoes a history of process cycles ranging from process development through yield ramp to Volume production. During these process cycles, a process chamber of the tool evolves from a “clean' chamber through chamber qualification and seasoning, to an “aged chamber preceding chamber cleaning and maintenance. After several maintenance cycles, an empirical model can evolve to include a statistically sufficient sample of the parameter space corresponding to the specific process tool and process associated therewith. In other words, through cleaning cycles, process cycles, and maintenance cycles, the tool 802 (with the aid of the simulation module) inherently determines the bounds of the parameter space. Ultimately, the evolved empirical model stored in library 842 can supersede the generally more intensive process model based on first principles simulation and can provide input to the APC con troller for process adjustment/correction.” [Col 5 Line 39-64]: “First principles physical model 106 is a model of the physical attributes of the tool and tool environment as well as the fundamental equations necessary to perform first principles simulation and provide a simulation result for facilitating a process performed by the semiconductor processing tool. Thus, the first principles physical model 106 depends to some extent on the type of semiconductor processing tool 102 analyzed as well as the process performed in the tool. For example, the physical model 106 may include a spatially resolved model of the physical geometry of the tool 102. which is different, for example, for a chemical vapor deposition (CVD) chamber and a diffusion furnace. Similarly, the first principles equations necessary to compute flow fields are quite different than those necessary to compute temperature fields. The physical model 106 may be a model as implemented in commercially available software, such as ANSYS, of ANSYS Inc., Southpointe, 275 Technology Drive Canons burg, Pa. 15317, FLUENT, of Fluent Inc., 10 Cavendish Ct. Centerra Park, Lebanon, N.H. 03766, or CFD-ACE--, of CFD Research Corp., 215 Wynn Dr., Huntsville, Ala. 35805, to compute flow fields, electro-magnetic fields, temperature fields, chemistry, Surface chemistry (i.e. etch Surface chemistry or deposition Surface chemistry). However, special purpose or custom models developed from first principles to resolve these and other details within the processing system may also be used.”
[FIG. 8]:
PNG
media_image1.png
808
945
media_image1.png
Greyscale
[FIG. 13]:
PNG
media_image2.png
998
943
media_image2.png
Greyscale
of semiconductor manufacturing equipment, ([Abstract]: “A method, system and computer readable medium for controlling a process performed by a semiconductor processing tool. The method includes inputting data relating to a process performed by the semiconductor processing tool, inputting a first principles physical model relating to the semiconductor processing tool, performing first principles simulation using the input data and the physical model to provide a first principles simulation result. The first principles simulation result is used to build an empirical model, and at least one of the first principles simulation result and the empirical model is selected to control the process performed by the semiconductor processing tool.” The examiner interprets where A digital twin of a process chamber of semiconductor manufacturing equipment is shown in a system for using first-principles simulation to control a semiconductor manufacturing process. The "digital twin1" under BRI refers to a model of a process chamber comprising multiple models. Strang & Mitrovic’s "first principles physical model" represents the tool and environment.) comprising one or more non-transitory machine-readable media comprising logic configured to implement: ([Col 3 Line 9-21]: “In still another aspect of the invention, a computer readable medium contains program instructions for execution on a processor, which when executed by the computer system, cause the processor to perform the steps of inputting data relating to a process performed by the semiconductor processing tool, inputting a first principles physical model relating to the semiconductor processing tool, and performing first principles simulation using the input data and the physical model to provide a first principles simulation result. The first principles simulation result is used to build an empirical model, and at least one of the first principles simulation result and the empirical model is selected to control the process performed by the semiconductor processing tool.” The examiner interprets where comprising one or more non-transitory machine-readable media comprising logic configured to implement is shown in a computer-readable medium containing program instructions (logic) for execution on a processor.) a first model of a first location of the process chamber; ([Col 5 Line 39-64]: “ First principles physical model 106 is a model of the physical attributes of the tool and tool environment as well as the fundamental equations necessary to perform first principles simulation and provide a simulation result for facilitating a process performed by the semiconductor processing tool. Thus, the first principles physical model 106 depends to some extent on the type of semiconductor processing tool 102 analyzed as well as the process performed in the tool. For example, the physical model 106 may include a spatially resolved model of the physical geometry of the tool 102, which is different, for example, for a chemical vapor deposition (CVD) chamber and a diffusion furnace. Similarly, the first principles equations necessary to compute flow fields are quite different than those necessary to compute temperature fields. The physical model 106 may be a model as implemented in commercially available software, such as ANSYS, of ANSYS Inc., Southpointe, 275 Technology Drive Canonsburg, Pa. 15317, FLUENT, of Fluent Inc., 10 Cavendish Ct. Centerra Park, Lebanon, N.H. 03766, or CFD-ACE+, of CFD Research Corp., 215 Wynn Dr., Huntsville, Ala. 35805, to compute flow fields, electro-magnetic fields, temperature fields, chemistry, surface chemistry (i.e. etch surface chemistry or deposition surface chemistry). However, special purpose or custom models developed from first principles to resolve these and other details within the processing system may also be used.” [Col 4 Line 45- Col 5 Line 3]: “Data input device 104 is a device for collecting data relating to a process performed by the semiconductor processing tool 102 and inputting the collected data to the first principles simulation processor 106. The process performed by the semiconductor process tool 102 may be a characterization process (i.e. process design or development), a cleaning process, a production process, or any other process performed by the semiconductor processing tool. In one embodiment, the data input device 104 may be implemented as a physical sensor for collecting data about the semiconductor processing tool 102 itself, and/or the environment contained within a chamber of the tool. Such data may include fluid mechanic data such as gas velocities and pressures at various locations within the process chamber, electrical data such as voltage, current, and impedance at various locations within the electrical system of the process chamber, chemical data such as specie concentrations and reaction chemistries at various locations within the process chamber, thermal data such as gas temperature, surface temperature, and surface heat flux at various locations within the process chamber, plasma processing data (when plasma is utilized) such as a plasma density (obtained, for example, from a Langmuir probe), an ion energy (obtained, for example, from an ion energy spectrum analyzer), and mechanical data such as pressure, deflection, stress, and strain at various locations within the process chamber.” The examiner interprets where a first model of a first location of the process chamber is shown in a "spatially resolved model of a physical geometry" of the tool & collecting data at "various locations within the process chamber".) and a second model of a second location of the process chamber, ([Col.10 Lines 30-51]: “In one example of using metrology data as input data for obtaining a virtual sensor reading, metrology data pertaining to an etch mask pattern and underlying film thickness can serve as input to a first principles etch process model, and subsequently performed etch process. Prior to performing the etch process, measurements of the mask pattern including pattern critical dimension(s) and mask film thickness at one or more locations (e.g., center and edge) on a given substrate for a given substrate lot can be provided as input to the etch process model. Moreover, measurements of the underlying film thickness (i.e., film thickness of the film to be etched) can also serve as input to the etch process model. Following execution of the first principles etch process model for a specified process recipe, and the above identified metrology input data, the time for completing the etch process at, for example, the center and edge can be calculated as output, and this output can be utilized to determine an over-etch period and any process adjustment necessary to preserve, for example, feature critical dimensions center-to-edge. Thereafter, these results can be utilized to adjust the process recipe for the current or upcoming substrate lot.” [Col 25 Line 33- Col 26 Line 13]: “ A method of controlling a process performed by a semiconductor processing tool, comprising: inputting a first principles physical model including a set of computer-encoded differential equations, the first principles physical model describing at least one of a basic physical or chemical attribute of the semiconductor processing tool and including 1) a spatially resolved model of a physical geometry of the semiconductor processing tool and 2) a grid set addressing the semiconductor processing tool or a geometry of the semiconductor processing tool; inputting process data related to an actual process being performed by the semiconductor processing tool; setting boundary conditions for the spatially resolved model of a physical geometry of the semiconductor processing tool based on said process data related to the actual process being performed by the semiconductor processing tool; storing in a fab-level library known simulation results obtained from simulation modules in a device manufacturing fab and distributing the known simulation results to other semiconductor processing tools in the device manufacturing fab; solving the computer-encoded differential equations of the first principles physical model for the spatially resolved model concurrently with the actual process being performed and in a time frame shorter in time than the actual process being performed to produce a first principles simulation by: using code parallelization techniques on multiple simulation modules in the device manufacturing fab, and re-using known simulation solutions as initial conditions for the first principles simulation, wherein re-using known simulation solutions comprises searching in the fab-level library for a closest fitting solution which if used for the initial condition would reduce the number of iterations required by the simulation module; providing from the solution of the computer-encoded differential equations solved concurrently with the actual process being performed a first principles simulation result; and using the first principles simulation result obtained during performance of the actual process to build an empirical model; and selecting at least one of the first principles simulation result and the empirical model to control the actual process being performed by the semiconductor processing tool.”
The examiner interprets where a second model of a second location of the process chamber is shown in the first locations (center) simulated by process model as first model and second location (edge) simulated by process model as second model.) wherein the first model of the first location of the process chamber is coupled to the second model of the second location of the process chamber, and (([Col.10 Lines 30-51]: “In one example of using metrology data as input data for obtaining a virtual sensor reading, metrology data pertaining to an etch mask pattern and underlying film thickness can serve as input to a first principles etch process model, and subsequently performed etch process. Prior to performing the etch process, measurements of the mask pattern including pattern critical dimension(s) and mask film thickness at one or more locations (e.g., center and edge) on a given substrate for a given substrate lot can be provided as input to the etch process model. Moreover, measurements of the underlying film thickness (i.e., film thickness of the film to be etched) can also serve as input to the etch process model. Following execution of the first principles etch process model for a specified process recipe, and the above identified metrology input data, the time for completing the etch process at, for example, the center and edge can be calculated as output, and this output can be utilized to determine an over-etch period and any process adjustment necessary to preserve, for example, feature critical dimensions center-to-edge. Thereafter, these results can be utilized to adjust the process recipe for the current or upcoming substrate lot.” The examiner interprets where the first model of the first location of the process chamber is coupled to the second model of the second location of the process chamber is shown in center to edge critical dimensions.) wherein the first model of the first location of the process chamber and the second model of the second location of the process chamber are each of a model type that is one of: 1) an AI/ML model; 2) an HFS model; or 3) a closed-form solution, (See [Abstract], [Col 14 Line 6-30]: “However, for these statistical methods to be able to reliably sense and control the tool under widely varying operating conditions, the database must be broad-enough to cover all operating conditions, which makes the database a burden to produce. The on-tool first principles simulation capability of the present invention does not require the creation of any such database because tool response to process conditions is predicted from physical first principles directly and accurately, given accurate working models and accurate input data. However, statistical methods can still be used to refine working models and input data as more run-time information under different operating conditions becomes available, but having such information is not required by the present invention for process sensing and control capability. Indeed, the process model can provide a basis upon which the process can be empirically controlled by using the process model to extend those known empirical solutions to "solutions" where empirical results have not been physically made. Hence, the present invention in one embodiment empirically characterizes the process tool by supplementing the known (i.e. physically observed) solutions with first principle simulation module solutions, the simulation module solutions being consistent with the known solutions. Eventually, as better statistics develop, the simulation module solutions can be superseded by the database of empirical solutions.” The examiner interprets where wherein the first model and second model are each of a model type that is an HFS model as shown in "first principles physical models" (corresponds to HFS)) and wherein the first model of the first location of the process chamber and the second model of the second location of the process chamber each represent a class of physical phenomena that is one of: 1) thermal characteristics; ([Col 5 Line 53-64]: “..The physical model 106 may be a model as implemented in commercially available software, such as ANSYS, of ANSYS Inc., Southpointe, 275 Technology Drive Canonsburg, Pa. 15317, FLUENT, of Fluent Inc., 10 Cavendish Ct. Centerra Park, Lebanon, N.H. 03766, or CFD-ACE+, of CFD Research Corp., 215 Wynn Dr., Huntsville, Ala. 35805, to compute flow fields, electro-magnetic fields, temperature fields, chemistry, surface chemistry (i.e. etch surface chemistry or deposition surface chemistry). However, special purpose or custom models developed from first principles to resolve these and other details within the processing system may also be used.”) 2) plasma characteristics; ([Col 5 Line 53-64]: “..The physical model 106 may be a model as implemented in commercially available software, such as ANSYS, of ANSYS Inc., Southpointe, 275 Technology Drive Canonsburg, Pa. 15317, FLUENT, of Fluent Inc., 10 Cavendish Ct. Centerra Park, Lebanon, N.H. 03766, or CFD-ACE+, of CFD Research Corp., 215 Wynn Dr., Huntsville, Ala. 35805, to compute flow fields, electro-magnetic fields, temperature fields, chemistry, surface chemistry (i.e. etch surface chemistry or deposition surface chemistry). However, special purpose or custom models developed from first principles to resolve these and other details within the processing system may also be used.”) 3) fluid dynamics; ([Col 4 Line 56- Col 5 Line 3]: “Such data may include fluid mechanic data such as gas velocities and pressures at various locations within the process chamber, electrical data such as voltage, current, and impedance at various locations within the electrical system of the process chamber, chemical data such as specie concentrations and reaction chemistries at various locations within the process chamber, thermal data such as gas temperature, surface temperature, and surface heat flux at various locations within the process chamber, plasma processing data (when plasma is utilized) such as a plasma density (obtained, for example, from a Langmuir probe), an ion energy (obtained, for example, from an ion energy spectrum analyzer), and mechanical data such as pressure, deflection, stress, and strain at various locations within the process chamber.”) 4) structural characteristics; ([Col 4 Line 56- Col 5 Line 3]: “Such data may include fluid mechanic data such as gas velocities and pressures at various locations within the process chamber, electrical data such as voltage, current, and impedance at various locations within the electrical system of the process chamber, chemical data such as specie concentrations and reaction chemistries at various locations within the process chamber, thermal data such as gas temperature, surface temperature, and surface heat flux at various locations within the process chamber, plasma processing data (when plasma is utilized) such as a plasma density (obtained, for example, from a Langmuir probe), an ion energy (obtained, for example, from an ion energy spectrum analyzer), and mechanical data such as pressure, deflection, stress, and strain at various locations within the process chamber.”) or 5) chemical reactions. ; ([Col 4 Line 56- Col 5 Line 3]: “Such data may include fluid mechanic data such as gas velocities and pressures at various locations within the process chamber, electrical data such as voltage, current, and impedance at various locations within the electrical system of the process chamber, chemical data such as specie concentrations and reaction chemistries at various locations within the process chamber, thermal data such as gas temperature, surface temperature, and surface heat flux at various locations within the process chamber, plasma processing data (when plasma is utilized) such as a plasma density (obtained, for example, from a Langmuir probe), an ion energy (obtained, for example, from an ion energy spectrum analyzer), and mechanical data such as pressure, deflection, stress, and strain at various locations within the process chamber.”)
Regarding Claim 2
Strang & Mitrovic teaches The digital twin of claim 1, (See claim 1) wherein the first model of the first location of the process chamber is of a different model type (See [Col 5 Line 39-64]: The examiner interprets where the first model of the first location of the process chamber is of a different model type is shown in a "first principles physical model" (which maps to HFS under BRI).) than the second model of the second location of the process chamber. (See [Col 25 Line 33- Col 26 Line 13], [Col 20 Line 22-Col 21 Line 21]: “FIG. 12 is a flow chart showing a process for using first principles simulation techniques to detect a fault and control a process performed by a semiconductor processing tool in accordance with an embodiment of the present invention. The flow chart is presented beginning with step 1202 for processing a substrate or batch of substrates within a process tool, such as the process tool 1002. At step 1204, tool data is measured and provided as input to a simulation module such as simulation module 1006. Boundary conditions and initial conditions are then imposed on the physical model of the simulation module to set up the model as shown in step 1206. At step 1208, the first principles physical model is executed to perform first principles simulation results that are output to a controller such as the APC controller 1008 of FIG. 10. At any time, for example, from run-to-run or batch-to-batch, the operator has the opportunity to select the control model to be employed within the APC controller. For example, the APC controller can utilize either the process model perturbation results, or the PCA model results. In either run-to-run or batch-to-batch, the process can be adjusted/corrected by the controller using model output. At step 1010, the process model output serves as input to the PLS model in the fault detector 1040, permitting a fault to be detected and classified at step 1012. For example, as described above, a difference between the real process performance Yreal and the simulated (or predicted) process performance for the given process condition (i.e. set of input control variables) Ysim can be utilized to determine the existence of a process fault, wherein Yreal is measured using either a physical sensor, or a metrology tool, and Ysim is determined by executing a simulation provided the input for the current process condition. If the difference (or variance, root mean square, or other statistic) between the real and simulated results exceeds a predetermined threshold, then a fault can be predicted to have occurred. The predetermined threshold can, for example, comprise a fraction of the mean value for the specific data, i.e. 5%, 10%, 15%, or it can be a multiple of a root mean square of the data, i.e. 1δ, 2 δ, 3. δ. Once a fault is detected, it can be classified using PLS analysis. For example, a sensitivity matrix X has been determined (and, possibly, stored in library 1010) for a given input condition (i.e. set of input control variables). Either the tool perturbation data (sensitivity matrix) is determined in-situ, centered on the current model solution, or determined a priori within the n-dimensional solution space using the process model. Using the sensitivity matrix and the difference between the real and simulated results, equation (1) can be solved using PLS analysis to identify those control variables (input parameters) that exhibit the greatest correlation with the observed difference between the real and simulated results. Using the example provided above, the process performance may be summarized by a profile of static pressure across the space overlying the substrate. The real result Yreal represents the measured profile of pressure, and Ysim represents the simulated profile of pressure. Let's assume a gas flow rate is set, however, the mass flow controller doubles the flow rate (yet reports the set value). One would expect to see a difference between the simulated and measured (real) profiles of pressure; i.e. the flow rate is off by a factor of two between the real and simulated cases. The difference between the real and simulated results would be sufficiently large to exceed a predetermined threshold. Using the PLS analysis, those parameters which tend to affect the profile in pressure the greatest would be identified, such as a gas flow rate. The presence of a fault and its characterization can be reported to an operator as process tool fault status or can cause the APC controller to perform control of the process tool (such as shut down) in response to the fault detection.”
[FIG. 12]:
PNG
media_image3.png
1081
487
media_image3.png
Greyscale
[FIG. 10]:
PNG
media_image4.png
941
936
media_image4.png
Greyscale
The examiner interprets where than the second model of the second location of the process chamber is shown in a second, "empirical model" built from results of the first model. These are different types: physics-based (HFS) vs. data-driven (Empirical/ML).)
Regarding Claim 3
Strang & Mitrovic teaches The digital twin of claim 1, (See Claim 3) wherein the first model of the first location of the process chamber represents a different class of physical phenomena ([Col 21 Line 58- Col 22 Line 6]: “A diagnostic controller 1324 can be coupled to each of the sensors described above and can be configured to provide measurements from these sensors to the simulation module described above. For the exemplary system of FIG. 13, the model executed on the simulation module can, for example, include three components, namely, a thermal component, a gas dynamic component, and a chemistry component. In the first component, the gas-gap pressure field can be determined, followed by a calculation of the gas-gap thermal conductance. Thereafter, the spatially resolved temperature field for the substrate (and substrate holder) can be determined by properly setting boundary conditions (and internal conditions) such as boundary temperature, or boundary heat flux, power deposited in resistance heating elements, power removed in cooling elements, heat flux at substrate surface due to the presence of plasma, etc.”
[FIG. 13]:
PNG
media_image5.png
991
937
media_image5.png
Greyscale
The examiner interprets where the first model of the first location of the process chamber represents a different class of physical phenomena is shown in a multi-component model including a "thermal component" (thermal characteristics) for a specific location such as a "substrate holder".) than the second model of the second location of the process chamber. ([See FIG.13], [Col 21 Line 58- Col 22 Line 6], [Col 21 Line 22-57]: “FIG. 13 is a block diagram of a vacuum processing system, to which a process control embodiment of the present invention may be applied. The vacuum processing system depicted in FIG. 13 is provided for illustrative purposes and does not limit the scope of the present invention in any way. The vacuum processing system includes a process tool 1302 having a substrate holder 1304 for supporting a substrate 1305, a gas injection system 1306, and a vacuum pumping system. The gas injection system 1306 can include a gas inject plate, a gas injection plenum, and one or more gas injection baffle plates within the gas injection plenum. The gas injection plenum can be coupled to one or more gas supplies such as gas A and gas B, wherein the mass flow rate of gas A and gas B into the processing system is affected by two mass flow controllers MFCA 1308 and MFCB 1310. Furthermore, a pressure sensor 1312 for measuring a pressure P1 can be coupled to the gas injection plenum. The substrate holder can, for example, include a plurality of components including but not limited to a helium gas supply for improving the gas-gap thermal conductance between the substrate and the substrate holder, an electrostatic clamping system, temperature control elements including cooling elements and heating elements, and lift pins for lifting the substrate to and from the surface of the substrate holder. Additionally, the substrate holder can include a temperature sensor 1314 for measuring the substrate holder temperature (T1) or substrate temperature, and a temperature sensor 1316 for measuring the coolant temperature (T3). As described above, helium gas is supplied to the backside of the substrate, wherein the gas-gap pressure (P(He)) can be varied at one or more locations. Furthermore, another pressure sensor 1318 can be coupled to the process tool to measure chamber pressure (P2), another temperature sensor 1320 can be coupled to the process tool to measure a surface temperature (T2), and another pressure sensor 1322 can be coupled to the inlet of the vacuum pumping system to measure an inlet pressure (P3).” The examiner interprets where the second model of the second location of the process chamber is shown in a "gas dynamic component" (fluid dynamics) for a second location like a "gas injection plenum" and a "chemistry component" for the "substrate surface".
Regarding Claim 4
Strang & Mitrovic teaches The digital twin of claim 1, (See Claim 1) wherein the first location is one of: (See [Col 5 Line 39-64], [Col 21 Line 58- Col 22 Line 6], [FIG. 13]: The examiner interprets where the first location is one of is shown as modeling "various locations within the process chamber" and provides "spatially resolved" models. 1) a pedestal of an ESC; 2) a showerhead; 3) a gap between the pedestal and the showerhead; 4) a chamber wall; (See [Col 4 Line 45- Col 5 Line 3]: The examiner interprets where a chamber wall is shown in collecting data about "the tool itself, and/or the environment contained within a chamber of the tool”) or 5) a surface of a wafer fabricated by the process chamber.
Regarding Claim 5
Strang & Mitrovic teaches The digital twin of claim 1, (See claim 1) wherein the first model of the first location of the process chamber being coupled to the second model of the second location of the process chamber comprises (See [Col 21 Line 58- Col 22 Line 6]: The examiner interprets where wherein the first model of the first location... being coupled to the second model of the second location... comprises is shown in a simulation model that can "include three components, namely, a thermal component, a gas dynamic component, and a chemistry component".) the first model of the first location of the process chamber providing outputs to the second model of the second location of the process chamber ([Col 22 Line 7-26]: “In one example of the present invention, ANSYS is utilized to compute the temperature field. Utilizing the second component of the process model, (i.e. the gas dynamic component), the gas pressure field and velocity field can be determined using the surface temperatures computed in the thermal component, and several of the aforementioned measurements. For example, the mass flow rate and pressure (P1) can be utilized to determine an inlet condition, and the pressure (P3) can be utilized to determine an outlet condition, and CFD-ACE+ can be utilized to compute the gas pressure and velocity fields. Utilizing the chemistry model (i.e., the third component), the previously computed velocity, pressure, and temperature fields can be utilized as inputs to a chemistry model to compute, for example, an etch rate. Depending on the complexity of the process tool geometry, each of these model components can be executed on a time scale within a batch-to-batch process cycle. Any one of these components can, for example, be utilized to provide spatial uniformity data as input to the process control, methodology, process characterization, and/or fault detection/classification.” The examiner interprets where the first model of the first location of the process chamber providing outputs to the second model of the second location of the process chamber is shown in the gas dynamic component (second model) determining fields "using the surface temperatures computed in the thermal component" (first model/first location).) for use by the second model of the second location of the process chamber. (See [Col 22 Line 7-26]: The examiner interprets where for use by the second model of the second location of the process chamber is shown in utilizing previously computed velocity, pressure, and temperature fields (outputs from first/second models) as inputs to a chemistry model (third model).)
Regarding Claim 6
Strang & Mitrovic teaches The digital twin of claim 5, (See claim 5) wherein the first model of the first location of the process chamber being coupled to the second model of the second location of the process chamber comprises (See [Col 21 Line 58- Col 22 Line 6]: The examiner interprets where wherein the first model of the first location of the process chamber being coupled to the second model of the second location of the process chamber comprises is shown in a simulation model that can "include three components, namely, a thermal component, a gas dynamic component, and a chemistry component".) the first model of the first location of the process chamber receiving outputs from the second model of the second location of the process chamber (See [Col 21 Line 58- Col 22 Line 6] & [Col 21 Line 58- Col 22 Line 6]: The examiner interprets where the first model of the first location of the process chamber receiving outputs from the second model of the second location of the process chamber is shown in a "first component" determining a gas-gap pressure field (second model/location), which is used to calculate gas-gap thermal conductance. Thereafter, the temperature field (first model/location) is determined.) for use by the first model of the first location of the process chamber. (See [Col 21 Line 58- Col 22 Line 6]: The examiner interprets where for use by the first model of the first location of the process chamber is shown in the spatially resolved temperature field for the substrate holder (first model) is determined using the calculated gas-gap thermal conductance (output from the second model).)
Regarding Claim 21
Strang & Mitrovic teaches A computer program product for using digital twins of process chambers, the computer program product comprising a non-transitory computer readable medium on which is provided computer-executable instructions for: ([Col 30 Line 6-55: “At least one of non-volatile media and volatile media containing program instructions for execution on a processor, which when executed by the computer system, cause the processor to perform the steps of: inputting a first principles physical model including a set of computer-encoded differential equations, the first principles physical model describing at least one o f a basic physical or chemical attribute of the semiconductor processing tool and including 1) a spatially resolved model of a physical geometry of the semiconductor processing tool and 2) a grid set addressing the semiconductor processing tool or a geometry of the semiconductor processing tool; inputting process data related to an actual process being performed by the semiconductor processing tool; setting boundary conditions for the spatially resolved model of a physical geometry of the semiconductor processing tool based on said process data related to the actual process being performed by the semiconductor processing tool; storing in a fab-level library known simulation results obtained from simulation modules in a device manufacturing fab and distributing the known simulation results to other semiconductor processing tools in the device manufacturing fab; solving the computer-encoded differential equations of the first principles physical model for the spatially resolved model concurrently with the actual process being performed and in a time frame shorter in time than the actual process being performed to produce a first principles simulation by: using code parallelization techniques on multiple simulation modules in the device manufacturing fab, and re-using known simulation solutions as initial conditions for the first principles simulation, wherein re-using known simulation solutions comprises searching in the fab-level library for a closest fitting solution which if used for the initial condition would reduce the number of iterations required by the simulation module; providing from the solution of the computer-encoded differential equations solved concurrently with the actual process being performed a first principles simulation result; and using the first principles simulation result obtained during performance of the actual process to build an empirical model; and selecting at least one of the first principles simulation result and the empirical model to control the actual process being performed by the semiconductor processing tool.” The examiner interprets where A computer program product for using digital twins... non-transitory computer readable medium... computer-executable instructions for is shown in a non-volatile media containing program instructions for execution on a processor to perform process control steps.) identifying a plurality of inputs to a digital twin of a process chamber, (See [Col 3 Line 9-21] & [Col 2 Line 63- Col 3 Line 8]: “In yet another aspect of the invention, a system for facilitating a process performed by a semiconductor processing tool includes means for inputting data relating to a process performed by the semiconductor processing tool, means for inputting a first principles physical model relating to the semiconductor processing tool, and means for performing first principles simulation using the input data and the physical model to provide a first principles simulation result. Also included is means for using the first principles simulation result to build an empirical model and means for selecting at least one of the first principles simulation result and the empirical model to control the process performed by the semiconductor processing tool.” The examiner interprets where identifying a plurality of inputs to a digital twin of a process chamber is shown in means for inputting data relating to a process performed by a semiconductor processing tool.) wherein the digital twin comprises a first model of a first location of the process chamber and a second model of a second location of the process chamber, (See [FIG. 12], [Col 20 Line 22-Col 21 Line 21], [Col 21 Line 58- Col 22 Line 6]: The examiner interprets where the digital twin comprises a first model of a first location... and a second model of a second location... is shown in a simulation model including a thermal component for a substrate holder (location 1) and a gas dynamic component for the tool environment (location 2).) and wherein the first model of the first location of the process chamber and the second model of the second location of the process chamber are coupled, (See [Col 21 Line 58- Col 22 Line 6]: The examiner interprets where the first model... and the second model... are coupled is shown in a second component (gas dynamic) determining fields using the results (outputs) computed in the first (thermal) component.) and wherein the plurality of inputs represent operating conditions of the process chamber; (See [Col 4 Line 45- Col 5 Line 3], [Col 4 Line 45- Col 5 Line 3], [Col 4 Line 45- Col 5 Line 3], [Col 4 Line 45- Col 5 Line 3], [Col 4 Line 22-32], [FIG. 1], [FIG. 13], [FIG. 1]: The examiner interprets where the plurality of inputs represent operating conditions of the process chamber is shown in inputs such as "gas velocities and pressures," "voltage, current," "temperature," and "plasma density".) providing the plurality of inputs to the digital twin; ([Col 5 Line 65- Col 6 Line 25]: “First principles simulation processor 108 is a processing device that applies data input from the data input device 104 to the first principles physical model 108 to execute a first principles simulation. Specifically, the first principles simulation processor 108 may use the data provided by the data input device 104 to set initial conditions and/or boundary conditions for the first principles physical model 106, which is then executed by the simulation module. First principles simulations in the present invention include, but are not limited to, simulations of electro-magnetic fields derived from Maxwell's equations, continuum simulations, for example, for mass, momentum, and energy transport derived from continuity, the Navier-Stokes equation and the First Law of Thermodynamics, as well as atomistic simulations derived from the Boltzmann equation, such as for example Monte Carlo simulations of rarefied gases (see Bird, G. A. 1994. Molecular gas dynamics and the direct simulation of gas flows, Clarendon Press). First principles simulation processor 108 may be implemented as a processor or workstation physically integrated with the semiconductor processing tool 102, or as a general purpose computer system such as the computer system 1401 of FIG. 14. The output of the first principles simulation processor 108 is a simulation result that is used to facilitate a process performed by the semiconductor processing tool 102. For example, the simulation result may be used to facilitate process development, process control and fault detection as well as to provide virtual sensor outputs that facilitate tool processes, as will be further described below.” The examiner interprets where providing the plurality of inputs to the digital twin is shown in applying data input from the data input device to the physical model to execute a simulation.) and generating predicted wafer characteristics of a simulated wafer using the digital twin. (See [Col 22 Line 7-26]: The examiner interprets where generating predicted wafer characteristics of a simulated wafer using the digital twin is shown in using the coupled models to compute an "etch rate" at the substrate surface and provide "spatial uniformity data".)
Regarding Claim 22
Strang & Mitrovic teaches The computer program product of claim 21, (See claim 21) wherein the first model of the first location of the process chamber includes specifications of a component of the process chamber, (See [Col 25 Line 33- Col 26 Line 13] & [Col 27 Line 50- Col 28 Line 35]: The examiner interprets where the first model of the first location... includes specifications of a component of the process chamber is shown in a "first principles physical model" including a "spatially resolved model of a physical geometry" of the tool.) and further comprising computer-executable instructions for validating the specifications of the component based on the predicted wafer characteristics. (See [Col 5 Line 65- Col 6 Line 25] & [Col 11 Line 57-Col 12 Line 12]: The examiner interprets where further comprising... instructions for validating the specifications of the component based on the predicted wafer characteristics is shown in using simulation to facilitate "process development" and "what-if analysis on the tool itself" to alter the "physical model itself".)
Regarding Claim 23
Strang & Mitrovic teaches The computer program product of claim 21, (See claim 21) wherein the plurality of inputs include parameters of a recipe implemented by the process chamber, (See [Col 15 Line 25-62] & [Col 25 Line 33- Col 26 Line 13]: The examiner interprets where the plurality of inputs include parameters of a recipe implemented by the process chamber is shown in inputting "process data related to an actual process" such as "pressure, a power... a gas flow rate". Under BRI, these are recipe parameters.) and further comprising computer-executable instructions for validating at least one parameter of the recipe based on the predicted wafer characteristics. (See [Col 10 Line 61-Col 11 Line 9], [Col 11 Line 57-Col 12 Line 12], [Col 17 Line 28-52]: Th examiner interprets where further comprising... instructions for validating at least one parameter of the recipe based on the predicted wafer characteristics is shown in using simulation results (predicted data) to "facilitate process development" and performing "what-if analysis" to "alter the input data".)
Regarding Claim 24
Strang & Mitrovic teaches The computer program product of claim 21, (See claim 21) wherein the predicted wafer characteristics (See [FIG.12] & [Col 20 Line 22-Col 21 Line 21]: The examiner interprets where the predicted wafer characteristics is shown in a simulation module executing coupled model components to provide simulation results.) comprise an indication of a defect (See [Col 15 Line 25-62], [FIG. 10], [Col 18 Line 43-61]: “In yet another embodiment of the present invention, a fault detector/classifier may be used in conjunction with the first principles simulation to provide control of a process performed by the process tool. FIG. 10 is a block diagram of a system for using first principles simulation techniques and a fault detector to control a process performed by a semiconductor processing tool in accordance with an embodiment of the present invention. As seen in this figure, the system includes a process tool 1002 coupled to an advanced process control (APC) infrastructure 1004, which includes a simulation module 1006 and an APC controller 1008 and library 1010. While not shown in FIG. 10, the library 1010 includes a solutions database and a grid database. Also coupled to the APC infrastructure 1004 is a metrology tool 1012 and remote controller 1014. These items are similar to those corresponding items discussed with respect to FIG. 6, except the items of FIG. 10 are further configured to function in consideration of fault detection. Thus, these similar items are not described with respect to FIG. 10.” The examiner interprets where an indication of a defect is shown in a "fault detector" that predicts the occurrence of a "process fault" when simulated results deviate from a threshold. Mentions detecting "process non-uniformity".) of the simulated wafer. (See [Col 20 Line 22-Col 21 Line 21] & [Col 18 Line 62- Col 19 Line 5]: The examiner interprets where of the simulated wafer is shown in determining faults by comparing simulated performance (Ysim) to real performance (Yreal) for specific substrates.)
Regarding Claim 25
Strang & Mitrovic teaches The computer program product of claim 21, (See claim 21) further comprising computer-executable instructions for identifying a recommendation to modify (See [Col 14 Line 65- Col 15 Line 24], [Col 27 Line 37-49], [Col 27 Line 37-49]: The examiner interprets where further comprising computer-executable instructions for identifying a recommendation to modify is shown in an APC controller 608 that utilizes simulation results to "implement a control methodology for process.) at least one operating condition of the operating conditions (See [Col 15 Line 25-62]: Th examiner interprets where at least one operating condition of the operating conditions is shown in adjusting process parameters such as "pressure, a power... a gas flow rate". These constitute operating conditions.) based on the predicted wafer characteristics. (See [Col 27 Line 18-36]: The examiner interprets where based on the predicted wafer characteristics is shown in process adjustment/correction performed based on "simulation result[s]", such as calculated "etch rate[s]" or "spatial uniformity data".)
Regarding Claim 26
Strang & Mitrovic teaches The computer program product of claim 25, (See Claim 25) wherein the recommendation is identified in response to determining (See [Col 20 Line 22-Col 21 Line 21] & [Col 15 Line 63- Col 16 Line 17]: The examiner interprets where the recommendation is identified in response to determining is shown in "the process can be adjusted/corrected by the controller using model output"; which determines a "correction" to minimize non-uniformity.) that the predicted wafer characteristics indicate a defect (See [Col 18 Line 62- Col 19 Line 5]: The examiner interprets where the predicted wafer characteristics indicate a defect is shown in a "fault detector" that predicts the existence of a "process fault" if simulated results deviate from a threshold. Mentions "non-uniformity".) of the simulated wafer. (See [Col 20 Line 22-Col 21 Line 21]: The examiner interprets where the simulated wafer is show in determining faults/uniformity for specific substrates using simulated process performance (Ysim).)
Regarding Claim 27
Strang & Mitrovic teaches The computer program product of claim 25, (See claim 25) wherein the recommendation is identified in response to determining (See [Col 15 Line 63- Col 16 Line 17]: The examiner interprets where the recommendation is identified in response to determining is shown the detection of a fault "can cause the APC controller to perform control of the process tool" by determining a correction.) that at least one of the first model and the second model has generated values (Col 29 Line 39-48]: “42. The system of claim 41, wherein said processor is configured to use the first principles simulation result to control by controlling at least one of a chemical vapor deposition system and a physical vapor deposition system.
43. The system of claim 24, wherein said processor is configured to input at least one of etch rate, deposition rate, etch selectivity, an etch critical dimension, an etch feature anisotropy, a film property, a plasma density, an ion energy, a concentration of a chemical specie, a photoresist mask film thickness, a photoresist pattern dimension.” The examiner interprets where at least one of the first model and the second model has generated values is shown in that a simulation module executing coupled model components (e.g., thermal and gas dynamic) to provide simulation results (Ysim).) that indicate anomalous operating conditions of the process chamber. (See [Col 18 Line 43-61]: The examiner interprets where that indicate anomalous operating conditions of the process chamber is shown in a "fault detector" that determines the existence of a "process fault" when the difference between real and simulated performance exceeds a threshold. Under BRI, a process fault is an anomalous operating condition.)
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claim(s) 7-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over US Patent No. 8036869B2 by Strang & Mitrovic in further view of US Patent No. 20200192325 by (Sadeghi et al.)
Regarding Claim 7
Strang & Mitrovic teaches A computer program product for generating digital twins of process chambers, the computer program product comprising a non-transitory computer readable medium on which is provided computer-executable instructions for: (See [Col 3 Line 9-21] & [Col 25 Line 33- Col 26 Line 13]: The examiner interprets where A computer program product for generating digital twins... comprising a non-transitory computer readable medium... computer-executable instructions for is shown in a computer-readable medium containing program instructions for execution on a processor to perform process control steps.)
generating a digital twin by: ([Col 4 Line 22-32]: “Referring now to the drawings, wherein like reference numerals designate identical or corresponding parts throughout the several views, FIG. 1 is a block diagram of a system for using first principles simulation techniques to facilitate a process performed by a semiconductor processing tool in accordance with an embodiment of the present invention. As seen in FIG. 1, the system includes a semiconductor processing tool 102, a data input device 104, a first principles physical model 106, and a first principles simulation processor 108. The system of FIG. 1 may also include a tool level library 110 as shown in phantom.”
[FIG. 1]:
PNG
media_image6.png
730
938
media_image6.png
Greyscale
The examiner interprets where generating a digital twin is shown in a simulation system that facilitates processes via spatially resolved models of tool geometry and environments.) generating, for a first location of a process chamber, a plurality of High Fidelity Simulation (HFS) values using an HFS model of the first location of the process chamber; (See [Col 5 Line 39-64] & [Col 25 Line 33- Col 26 Line 13]: The examiner interprets where generating, for a first location... a plurality of High Fidelity Simulation (HFS) values using an HFS model... is shown in executing a "first principles physical model" (HFS) to provide simulation results for specific "spatially resolved" locations.) receiving a plurality of sensor measurements corresponding to the first location of the process chamber; (See [Col 4 Line 45- Col 5 Line 3], [Col 25 Line 33- Col 26 Line 13], [Col 28 Line 36-39]: “25. The system of claim 24, wherein said input device comprises at least one of a physical sensor and a metrology tool physically mounted on the semiconductor processing tool.” The examiner interprets where receiving a plurality of sensor measurements corresponding to the first location... is shown in receiving tool data from physical sensors at various locations within the process chamber.) training an Artificial Intelligence/Machine Learning (AI/ML) model of the first location of the process chamber using at least one of the plurality of HFS values and the plurality of sensor measurements; (See [Abstract], [Col 2 Lin 32-48], [Col 25 Line 33- Col 26 Line 13] : The examiner interprets where training an AI/ML model of the first location... using at least one of the plurality of HFS values and the plurality of sensor measurements is shown in using first principles simulation results (HFS values) and tool data (sensor measurements) to "build an empirical model".) and coupling the trained AI/ML model of the first location of the process chamber to a model of a second location of the process chamber, (See [Col 21 Line 58- Col 22 Line 6] & [Col 21 Line 58- Col 22 Line 6] [Col.10 Lines 30-51]: “In one example of using metrology data as input data for obtaining a virtual sensor reading, metrology data pertaining to an etch mask pattern and underlying film thickness can serve as input to a first principles etch process model, and subsequently performed etch process. Prior to performing the etch process, measurements of the mask pattern including pattern critical dimension(s) and mask film thickness at one or more locations (e.g., center and edge) on a given substrate for a given substrate lot can be provided as input to the etch process model. Moreover, measurements of the underlying film thickness (i.e., film thickness of the film to be etched) can also serve as input to the etch process model. Following execution of the first principles etch process model for a specified process recipe, and the above identified metrology input data, the time for completing the etch process at, for example, the center and edge can be calculated as output, and this output can be utilized to determine an over-etch period and any process adjustment necessary to preserve, for example, feature critical dimensions center-to-edge. Thereafter, these results can be utilized to adjust the process recipe for the current or upcoming substrate lot.” The examiner interprets where coupling the trained AI/ML model of the first location to a model of a second location... is shown in a model with multiple components (e.g., thermal, gas dynamic) where outputs from one (e.g., thermal at location 1) are inputs for another (e.g., gas dynamic at location 2) & in center to edge critical dimensions.) wherein the digital twin of the process chamber is comprised of the trained AI/ML model of the first location of the process chamber and the model of the second location of the process chamber. ([Col.10 Lines 30-51]: “In one example of using metrology data as input data for obtaining a virtual sensor reading, metrology data pertaining to an etch mask pattern and underlying film thickness can serve as input to a first principles etch process model, and subsequently performed etch process. Prior to performing the etch process, measurements of the mask pattern including pattern critical dimension(s) and mask film thickness at one or more locations (e.g., center and edge) on a given substrate for a given substrate lot can be provided as input to the etch process model. Moreover, measurements of the underlying film thickness (i.e., film thickness of the film to be etched) can also serve as input to the etch process model. Following execution of the first principles etch process model for a specified process recipe, and the above identified metrology input data, the time for completing the etch process at, for example, the center and edge can be calculated as output, and this output can be utilized to determine an over-etch period and any process adjustment necessary to preserve, for example, feature critical dimensions center-to-edge. Thereafter, these results can be utilized to adjust the process recipe for the current or upcoming substrate lot.” [Col 15 Line 25-62]: “The APC controller 608 is coupled to the simulation module 606 in order to receive a simulation result from the simulation module 606 and to utilize the simulation result to implement a control methodology for process adjustment/correction of a process performed on the tool 602. For example, an adjustment can be made to correct process non-uniformities. In one embodiment of the present invention, one or more perturbation solutions are executed on the simulation module 606, centered on a process solution for a process currently run on the process tool 602. The perturbation solutions can then be utilized with, for instance, a nonlinear optimization scheme such as the method of steepest descent (Numerical Methods, Dahlquist & Bjorck, Prentice-Hall, Inc., Englewood Cliffs, N.J., 1974, p. 441; Numerical Recipes, Press et al., Cambridge University Press, Cambridge, 1989, pp. 289-306) to determine a direction within an n-dimensional space for applying the correction. The correction can then be implemented on the process tool 602 by the APC controller 608. For example, at least one of tool data (i.e. physical sensor data), or results from a current execution of the simulation can indicate that the processing system exhibits a non-uniform static pressure field overlying the substrate given the current initial/boundary conditions. The non-uniformity can, in turn, contribute to an observed non-uniformity of a metric used to quantify the performance of the substrate process, measured by the metrology tool, on the substrate, i.e. a critical dimension, feature depth, film thickness, etc. By perturbing the input parameters to the current execution of the simulation, a set of perturbation solutions can be obtained in order to determine the best "route" to take in order to remove, or reduce, the static pressure non-uniformity. For example, the input parameters for the process can include a pressure, a power (delivered to an electrode for generating plasma), a gas flow rate, etc. While perturbing one input parameter at a time and holding all other input parameters constant, a sensitivity matrix can be formed that may be employed with the above identified optimization scheme to derive a correction suitable for correcting the process non-uniformity.” The examiner interprets where the digital twin... is comprised of the trained AI/ML model... and the model of the second location is shown in the simulation module executing the coupled model components to provide spatial uniformity and facilitate the tool process & measurements of the mask pattern including pattern critical dimension(s) and mask film thickness at one or more locations (e.g., center and edge))
Strang does not explicitly use the term "training an AI/ML model" . However, Sadeghi teaches training an Artificial Intelligence/Machine Learning (AI/ML) model of the first location of the process chamber using at least one of the plurality of HFS values and the plurality of sensor measurements; ([0047]: “In conjunction with the machine detection module 500, a machine-learning module 504 is also contemplated to be used to predict with greater accuracy and confidence the component needing maintenance and that nature and extent thereof. The machine-learning module 504 may be implemented by computer 103 or it may be implemented at a remote server that communicates with computer 103. In any case, the machine-learning module 504 may use a supervised learning algorithm that detects features associated with the various door state data 503 and the sensor data 505 for classification by a maintenance detection model. The machine-learning module 504 may be provided with training data that provides ground truth relative to what features a set of sensor data and door state data will have when various components are in need of replacement, for example, due to wear and tear, degradation, contamination, deterioration, etc.”
[FIG. 5]:
PNG
media_image7.png
1528
1154
media_image7.png
Greyscale
The examiner interprets where Training AI/ML model is shown in a "machine-learning module" (AI/ML) provided with "training data" (training) including "sensor data".)
It would have been obvious to a POSITA before the effective filing date of the invention to implement Strang's self-correcting empirical model using the specific AI/ML training techniques of Sadeghi. Furthermore, because Strang teaches a "spatially resolved model" partitioned into a grid addressing "various locations", it would have been an obvious matter of routine engineering design to train these models locally for a first location and then couple them to models of second locations as taught in Strang's distributed simulation framework. Such a combination would predictably result in a more efficient and accurate digital twin that balances physical theory with empirical ground truth.
Regarding Claim 8
Strang & Mitrovic in combination with Sadeghi teaches The computer program product of claim 7, (See claim 7) wherein the second model of the second location of the process chamber is one of: (See [Col 21 Line 58- Col 22 Line 6]: The examiner interprets where the second model of the second location of the process chamber is one of is shown in a simulation model comprising multiple "components" (models) for different locations.) 1) an AI/ML model; 2) an HFS model; (See [Col 5 Line 39-64] & [Col 21 Line 58- Col 22 Line 6]: The examiner interprets where an HFS model is shown in "first principles physical models" using fundamental equations (HFS) for tool components.) or 3) a closed-form solution.
Strang does not explicitly use the term an AI/ML model" . However, Sadeghi teaches an AI/ML model (See [0047], [0005]: “Methods and systems for real-time tool health monitoring are provided. Embodiments described include methods and system that use in-situ sensors to monitor airborne particles, the measurement data of which is used to monitor the maintenance needs of a semiconductor processing system. Further, door state data relating to timing various doors open/close operation within the semiconductor processing system is used to identify the source of the airborne particles to provide a recommendation as to a maintenance procedure.” The examiner interprets where Second model is AI/ML is shown in a "machine learning module" (AI/ML) utilized to monitor tool health and predict states.)
It would have been obvious to a POSITA before the effective filing date of the invention to represent a second location of the chamber using either an AI/ML model (as taught by Sadeghi) or an HFS model (as taught by Strang) based on the specific physical stability or computational requirements of that second region. The choice between these known model types for different segments of a "spatially resolved" system is a predictable application of known modeling techniques to achieve the optimal balance between speed and fidelity as recognized by Strang. Therefore, the selection of the second model type from the claimed group is an obvious matter of design choice.
Regarding Claim 9
Strang & Mitrovic in combination with Sadeghi teaches The computer program product of claim 7. (See Claim 7) Strang & Mitrovic teaches wherein the HFS model of the first location of the process chamber (See [Col 25 Line 33- Col 26 Line 13] & [Col 25 Line 33- Col 26 Line 13]: The examiner interprets where the HFS model of the first location of the process chamber is shown in a "first principles physical model" (HFS) comprising a "spatially resolved model of a physical geometry" of the tool.) and the AI/ML model of the first location of the process chamber (See [FIG. 10] & [Col 18 Line 62- Col 19 Line 5]: “As seen in FIG. 10, the system includes a fault detector 1040 coupled to the simulation module 1006 and configured to receive a simulation result from the module 1006. For example, the output of the simulation module 1006 can include a profile of data. The profile of data can then serve as input to multivariate analysis such as partial least squares (PLS) performed in the fault detection device 1040. In the PLS analysis, a set of loading (or correlation) coefficients can be defined which relate tool perturbation data ( X) to process performance data ( Y) describing a difference between simulated results Ysim and actual results Yreal.” The examiner interprets where the AI/ML model of the first location of the process chamber is shown in building an "empirical model" from first principles results and tool data using statistical multivariate analysis. Under BRI, this is an AI/ML model.) both model a same class of physical phenomena. (See [Col 14 Line 6-30], [Col 21 Line 58- Col 22 Line 6], [Col 21 Line 58- Col 22 Line 6], [Col 17 Line 28-52]: “As seen in FIG. 8, the system includes a model analysis processor 840, which is coupled to the simulation module 806 and configured to receive a simulation result from the module 806. In the embodiment of FIG. 8, model analysis includes the construction of an empirical model from non-dimensionalization of the simulation results. As simulation results are received on a run-to-run or batch-to-batch basis, an empirical model is constructed and stored in the empirical model library 842. For example, the process tool 802 undergoes a history of process cycles ranging from process development through yield ramp to volume production. During these process cycles, a process chamber of the tool evolves from a "clean" chamber through chamber qualification and seasoning, to an "aged" chamber preceding chamber cleaning and maintenance. After several maintenance cycles, an empirical model can evolve to include a statistically sufficient sample of the parameter space corresponding to the specific process tool and process associated therewith. In other words, through cleaning cycles, process cycles, and maintenance cycles, the tool 802 (with the aid of the simulation module) inherently determines the bounds of the parameter space. Ultimately, the evolved empirical model stored in library 842 can supersede the generally more intensive process model based on first principles simulation and can provide input to the APC controller for process adjustment/correction.”
[FIG. 8]:
PNG
media_image8.png
800
938
media_image8.png
Greyscale
The examiner interprets where both model of the same class of physical phenomena is shown in the empirical model constructed to characterize the tool/process behavior simulated by the HFS model. For example, thermal results from the HFS simulation inform the empirical models that control the process.)
It would have been obvious to a POSITA before the effective filing date of the invention to implement the "empirical model" of Strang using the trained "machine-learning module" of Sadeghi. Because Strang teaches building the empirical model directly from the simulation results of a specific physical component, it is a predictable and necessary result of this modeling process that both the source HFS model and the resulting trained AI/ML model would represent the same class of physical phenomena (e.g., both modeling thermal behavior for a pedestal). Aligning the surrogate model with the physics of the HFS model is a routine design choice to ensure that the empirical model can predictably facilitate or control the process as intended by Strang.
Regarding Claim 10
Strang & Mitrovic in combination with Sadeghi teaches The computer program product of claim 7. (See Claim 7) Strang & Mitrovic teaches wherein the trained AI/ML model of the first location of the process chamber (See [Col 18 Line 62- Col 19 Line 5] & [Col 22 Line 27-38]: “From the derived models and analysis of the process in response to changes in processing conditions and/or effects such as reactor aging, an empirical model can be assimilated over time. As such, when the number of repetitions on the reactor becomes statistically significant as determined by standard statistical analysis programs, the process control evolves to a control which is empirically based for those processes which are essentially "repeats" of previously run operations. Yet, according to the present invention, the process control returns the capability to perform first-principles simulation if necessary to accommodate new processes or alterations in the process geometry.” The examiner interprets where the trained AI/ML model of the first location of the process chamber building an "empirical model" (AI/ML) for a specific tool environment using simulation results and tool data.) and the model of the second location of the process chamber (See [Col 21 Line 58- Col 22 Line 6]:The examiner interprets where the model of the second location of the process chamber is shown in a simulation model comprising multiple "components" (models) for different locations, such as the chamber environment.) each model a class of physical phenomena. (See [Col 5 Line 39-64]: The examiner interprets where each model a class of physical phenomena is shown in models/components representing and computing "thermal fields" (thermal), "flow fields" (fluid dynamics), and "chemistry" (chemical reactions).)
Strang does not explicitly use the term “trained AI/ML model " . However, Sadeghi teaches an AI/ML model (See [0047] & [0005]: The examiner interprets where 1st location AI/ML models a physical class is shown in using a trained ML module to represent tool behavior. Implementing Strang's phenomenon-specific surrogate via Sadeghi's ML results in an AI/ML model modeling a class of physical phenomena.
It would have been obvious to a POSITA before the effective filing date of the invention to implement Strang's empirical model using the specific trained "machine-learning module" taught by Sadeghi. Because Strang teaches building the empirical surrogate directly from the results of a specific phenomenon-based component (e.g., a thermal solver), it is a predictable and routine result that the resulting trained AI/ML model would be specialized to model that same class of physical phenomena as taught by Strang & Mitrovic. Furthermore, since Strang teaches coupling these phenomenon-specific components together (e.g., temperature outputs from a thermal model serving as inputs to a gas dynamic model), the resulting digital twin would predictably be comprised of coupled models that each represent a distinct class of physical phenomena. This modular, domain-specific approach is a known technique for managing the computational complexity of modeling semiconductor process chambers.
Regarding Claim 11
Strang & Mitrovic in combination with Sadeghi teaches The computer program product of claim 10. (See claim 10) Strang & Mitrovic teaches wherein the class of physical phenomena is one of: (See [Col 4 Line 45- Col 5 Line 3] & [Col 5 Line 39-64]: The examiner interprets wherein the class of physical phenomena is one of is shown in models/components representing and computing various physical attributes of the tool and environment.) thermal characteristics, plasma characteristics, fluid dynamics, (See [Col 4 Line 45- Col 5 Line 3]: The examiner interprets fluid dynamics shown in measuring and modeling "fluid mechanic data such as gas velocities and pressures".) structural characteristics, or chemical reactions.
It would have been obvious to a POSITA before the effective filing date of the invention to implement Strang's phenomenon-specific model components using the trained "machine-learning module" taught by Sadeghi. Because Strang already identifies these specific classes of physical phenomena as the essential functional partitions for tool modeling and control, selecting one of these known classes (e.g., thermal characteristics) as the domain for a trained AI/ML surrogate is a simple substitution of a modern modeling technique for a traditional one to obtain the predictable result of increased computational efficiency as sought by both prior art. Therefore, specifying the subject matter of the models from this known list of tool phenomena is an obvious matter of routine engineering choice.
Regarding Claim 12
Strang & Mitrovic in combination with Sadeghi teaches The computer program product of claim 10. (See claim 10) Strang & Mitrovic teaches wherein the trained AI/ML model of the first location of the process chamber (See [Col 21 Line 58- Col 22 Line 6]: The examiner interprets where the trained AI/ML model of the first location of the process chamber is shown in building/training an "empirical model" for a location (e.g., substrate holder) using simulation results and tool data. Under BRI, this is a trained AI/ML model.) and the model of the second location of the process chamber (See [Col 21 Line 58- Col 22 Line 6]:The examiner interprets where the model of the second location of the process chamber is shown in a simulation model comprising multiple "components" (models) for different locations, such as the tool environment or plenum.) model different classes of physical phenomena. (See [Col 21 Line 58- Col 22 Line 6]: The examiner interprets where model different classes of physical phenomena is shown in coupled models/components representing distinct phenomena: a thermal component (thermal characteristics) and a gas dynamic component (fluid dynamics).)
Strang does not explicitly use the term “trained AI/ML model " . However, Sadeghi teaches wherein the trained AI/ML model of the first location of the process chamber (See [0047]: The examiner interprets where AI/ML model of 1st location is shown in a "machine-learning module" (AI/ML) trained with sensor data for specific tool parts.)
It would have been obvious to a POSITA before the effective filing date of the invention to implement the empirical surrogate models suggested by Strang using the trained machine-learning module of Sadeghi. Because Strang already recognizes that these models must be coupled across different physical domains (e.g., thermal coupled to chemical) to accurately facilitate process control, the resulting digital twin would predictably comprise a trained AI/ML model of one phenomena coupled to a model of a different class of phenomena. This configuration is a predictable application of known hybrid and modular modeling techniques used to manage the multi-physics complexity of semiconductor processing tool.
Regarding Claim 13
Strang & Mitrovic in combination with Sadeghi teaches The computer program product of claim 7. (See claim 7). Strang & Mitrovic teaches wherein the HFS model of the first location of the process chamber generates simulation values with a timestep that is shorter than a timestep of the AI/ML model of the first location of the process chamber (See [Col 5 Line 39-64], [Col 25 Line 33- Col 26 Line 13], [Col 11 Line 28-56]: “Where the first principles simulation is run concurrently with the process performed by the semiconductor tool, the data input in step 401 may be data from physical sensors mounted on the semiconductor processing tool to sense a predetermined parameter during the process run by the tool. In this embodiment, steady-state simulations are repeatedly run concurrently with the process by using the physical sensor measurements to repeatedly update boundary conditions of the first principles simulation model. The virtual measurement data generated is useful for monitoring by tool operators, and in no way differs from measurements made by physical sensors. However, the simulation is preferably capable of running fast so virtual measurements can be updated at a reasonable rate (e.g. "sampling rate"). The first principles simulation may also be run concurrently without the use of physical sensor input data. In this embodiment, initial and boundary conditions for the simulation are set based on the initial setting of the tool prior to a tool process and the readings of physical sensors prior to the run; a full time-dependent simulation is then run during, but independent of, the tool process. The obtained virtual measurements can be displayed to and analyzed by the operator like any other actually measured tool parameter. If the simulation runs faster than the wafer process, then simulation results are known ahead of the corresponding actual measurements made during the wafer process. Knowing the measurements ahead of time allows the implementation of various feed-forward control functions based on these measurements as will be further described below.” [Col 7 Line 25-38]: “In step 205, the first principles simulation processor 108 uses the input data of step 201 and the first principles physical model of step 203 to execute a first principles simulation and provide a simulation result. Step 205 may be performed either concurrently with or not concurrently with the process performed by the semiconductor processing tool. For example, simulations that can be performed at short solution times may be run concurrently with a tool process, and results used to control the process. More computationally intensive simulations may be performed not concurrently with the tool process and the simulation result may be stored in a library for later retrieval. In one embodiment, step 205 includes using the input data of step 201 to set initial and/or boundary conditions for the physical model provided in step 205.” [Col 18 Line 9-43]: “At any time, for example from run-to-run or from batch-to-batch, the operator has the opportunity to select process control based on the first principles simulation or the empirical model. At some point in the building of the empirical model, the operator may select to override the first principles simulation altogether in favor of the empirical model which at that point can use a library of data and interpolation/extrapolation schemes to rapidly extract controller input for a given set of tool data. Thus, decision block 912 determines whether the first principles simulation or the empirical model is used to control the process. Where no override is determined in step 912, the process continues at step 914 with the APC controller determining a control signal from the simulation result. Where model override is selected, the APC controller determines a control signal from the empirical model as shown in step 916. In another embodiment, a combination of first principles simulation results and empirical modeling can be used by the APC controller to control the process. As shown by step 918, the process can be adjusted/corrected by the controller using either the model output shown in step 914 or the empirical model output shown in step 916. Thus, the process of FIG. 9 shows a method of in-situ construction of an empirical model, and, once statistically significant, the empirical model can override the computationally intensive simulation process model. During process control, a filter, such as an exponentially weighted moving average (EWMA) filter, can be employed in order to impart only a fraction of the requested correction. For example, the application of the filter can take the form Xnew=(1- λ)Xold+ λ(Xpredicted-Xold), wherein Xnew is the new value for the given input parameter (control variable), Xold is the old (or previously used) value for the given input parameter, Xpredicted is the predicted value for the input parameter based upon one of the above described techniques, and λ is the filter coefficient ranging from 0 to 1.”) The examiner interprets where the HFS model of the first location... generates simulation values with a timestep is shown in HFS models (first principles) for spatially resolved locations that produce simulation results & mentions "time-dependent simulation" & where shorter than a timestep of the AI/ML model of the first location of the process chamber is shown in "computationally intensive" HFS models vs. empirical models that "rapidly extract controller input".
It would have been obvious to a POSITA before the effective filing date of the invention to implement the training process of Sadeghi using an HFS model with a shorter timestep than the resulting AI/ML model. It is well-known in the art of computational physics that high-fidelity solvers require fine temporal discretization (short timesteps) for accuracy and numerical stability, whereas empirical or machine-learning surrogates are intentionally designed for efficiency, operating at macro-level process time scales (longer timesteps). Therefore, using models with different temporal resolutions to balance accuracy and speed is a predictable application of known multi-scale modeling techniques and a matter of routine engineering design choice.
Regarding Claim 14
Strang & Mitrovic in combination with Sadeghi teaches The computer program product of claim 7. (See claim 7) Strang & Mitrovic teaches wherein the first location of the process chamber is one of: (See [Col 4 Line 45- Col 5 Line 3]: The examiner interprets where the first location of the process chamber is one of is shown in modeling "various locations within the process chamber".) 1) a pedestal of an electrostatic chuck (ESC); 2) a showerhead; 3) a gap between the showerhead and the pedestal; 4) a chamber wall; or 5) a surface of a wafer fabricated by the process chamber. (See [Col 21 Line 58- Col 22 Line 6]: The examiner interprets where the surface of a wafer fabricated by the process chamber is shown in a "substrate 1305" and a "substrate surface". Teaches computing an "etch rate" at that surface.)
It would have been obvious to a POSITA before the effective filing date of the invention to select one of the locations taught by Strang(such as a showerhead or ESC pedestal) as the "first location" for the digital twin generation process of claim 7. These components are well-known in the art as the critical hardware interfaces influencing process physics (thermal, gas dynamic, and chemical) and are identified by the prior art as the primary targets for modeling and health monitoring. The selection of a specific location from this known group of tool parts to implement the hybrid modeling technique is an obvious matter of routine engineering design and yields predictable results in tool control.
Regarding Claim 15
Strang & Mitrovic in combination with Sadeghi teaches The computer program product of claim 7. (See claim 7) Strang & Mitrovic teaches wherein coupling the trained AI/ML model of the first location of the process chamber to the model of the second location of the process chamber comprises ([Col 14 Line 65- Col 15 Line 24]: “The simulation module 606 is a computer, workstation, or other processing device capable of executing first principles simulation techniques to control a process performed by the tool 602, and therefore may be implemented as the simulation module 302 described with respect to FIG. 3. Thus, the simulation module 602 includes the first principles physical model 106 and the first principles simulation processor 108 described with respect to FIG. 1, as well as any other hardware and/or software that may be helpful for executing first principles simulations to control a process. In the embodiment of FIG. 6, the simulation module 606 is configured to receive tool data from one or more diagnostics on the tool 602 for processing and subsequent use during simulation model execution. The tool data may include the aforementioned fluid mechanic data, electrical data, chemical data, thermal, and mechanical data, or any of input data described with respect to FIGS. 1 and 2 above. In the embodiment of FIG. 6, the tool data can be utilized to determine boundary conditions and initial conditions for a model to be executed on the simulation module 606. The model can, for example, include the aforementioned ANSYS, FLUENT, or CFD-ACE+codes, to compute flow fields, electro-magnetic fields, temperature fields, chemistry, surface chemistry (i.e. etch surface chemistry or deposition surface chemistry), etc. The models developed from first principles can resolve details within the processing system in order to provide an input for process control of the tool.”
[FIG. 2]:
PNG
media_image9.png
1074
706
media_image9.png
Greyscale
[FIG. 3]:
PNG
media_image10.png
591
788
media_image10.png
Greyscale
[FIG. 6]:
PNG
media_image11.png
772
901
media_image11.png
Greyscale
[Col 15 Line 63- Col 16 Line 17]: “In another embodiment of the present invention, the simulation results are utilized in conjunction with a principal components analysis (PCA) model formulated as described in pending U.S. patent Application Ser. No. 60/343,174, entitled "Method of detecting, identifying, and correcting process performance," the contents of which are incorporated herein by reference. Therein, a relationship can be determined between a simulated signature (i.e. spatial components of the simulation model results) and a set of at least one controllable process parameter using multivariate analysis (i.e. PCA). This relationship can be utilized to improve the data profile corresponding to a process performance parameter (i.e., a model result). The principle components analysis determines a relationship between spatial components of a result (or predicted output) of a simulation of the semiconductor processing tool and a set of at least one control variable (or input parameter). The determined relationship is utilized to determine a correction to the at least one control variable (or input parameter) in order to cause a minimization of the magnitude of the spatial components in order to improve (or decrease) the non-uniformity of the simulated result (or measured result if available).” [Col 26 Line 61- Col 27 Line 7]: “14. The method of claim 13, wherein said sharing simulation information comprises distributing simulation results among the interconnected resources to reduce redundant execution of substantially similar first principles simulations by different resources.
15. The method of claim 13, wherein said sharing simulation information comprises distributing model changes among the interconnected resources to reduce redundant refinements of first principles simulations by different resources.
16. The method of claim 11, further comprising using remote resources via a wide area network to control the semiconductor process performed by the semiconductor processing tool.” [Col 28 Line 65- Col 29 Line 11]: “33. The system of claim 24, further comprising a network of interconnected resources connected to said processor and configured to assist said processor in performing at least one of the inputting a first principles simulation model and performing a first principles simulation.
34. The system of claim 33, wherein said network of interconnected resources is configured to use code parallelization with said processor to share the computational load of the first principles simulation.
35. The system of claim 33, wherein said network of interconnected resources is configured to share simulation information with said processor to facilitate said process performed by the semiconductor processing tool.” [Col 16 Line 18-40]: “As noted above, the library 610 coupled to the simulation module 606 in FIG. 6 is configured to include a solution database 616 and a grid database 618. The solution database 616 can include a coarse n-dimensional database of solutions, whereby the order n of the n-dimensional space is governed by the number of independent parameters for the given solution algorithm. When the simulation module 606 retrieves the tool data for a given process run, the library 610 can be searched based upon model input to determine the closest fitting solution. This solution can be used according to the present invention as an initial condition for subsequent first principles simulation, thereby reducing the number of iterations required to be performed by the simulation module to provide a simulation result. With each model execution, the new solution can be added to the solution database 616. Additionally, the grid database 618 can include one or more grid sets, whereby each grid set addresses a given process tool or process tool geometry. Each grid set can include one or more grids with different grid resolutions, ranging from coarse to fine. The selection of grids can be utilized to reduce solution time by performing multi-grid solution techniques (i.e. solve for a simulation result on coarse grid, followed by solution on finer grid, finest grid, etc.)” The examiner interprets where coupling the trained AI/ML model of the first location... to the model of the second location... comprises is shown in a framework where "simulation results are utilized in conjunction with" multivariate analysis models and where results are used as "initial conditions for subsequent... simulation".) providing a plurality of outputs of the trained AI/ML model of the first location of the process chamber ([Col 4 Line 5-8]: “FIG. 11 is a schematic representation of the data inputs, X and Y, to a PLS analysis and the corresponding outputs T, P, Ū, C, W, Ē, F, H and variable importance in the projection (VIP)”
[FIG. 11]:
PNG
media_image12.png
795
925
media_image12.png
Greyscale
The examiner interprets where providing a plurality of outputs of the trained AI/ML model of the first location... is shown in a Partial Least Squares (PLS) analysis model that generates a plurality of outputs (e.g., T, P, U, C, W, E, F, H).) to the model of the second location of the process chamber. ([Col 27 Line 37-49]: “22. The method of claim 1, wherein said inputting data comprises: inputting physical geometric parameters of at least one of a material processing system, an etch system, a photoresist spin coating system, a lithography system, a dielectric coating system, a deposition system, a rapid thermal processing system for thermal annealing, and a batch diffusion furnace.
23. The method of claim 1, wherein said using the first principles simulation result to control comprises: controlling the semiconductor processing tool by using empirical model output to adjust said process performed by the semiconductor processing tool.” [Col 29 Line 49-Col 30 Line 5]: “44. The system of claim 24, wherein said processor is configured to input physical geometric parameters of at least one of a material processing system, an etch system, a photoresist spin coating system, a lithography system, a dielectric coating system, a deposition system, a rapid thermal processing system for thermal annealing, and a batch diffusion furnace.
45. The system of claim 24, wherein said processor is configured to use the first principles simulation result to control by controlling the semiconductor processing tool by using empirical model output to adjust said process performed by the semiconductor processing tool.” The examiner interprets where to the model of the second location of the process chamber is shown in Discloses using known simulation solutions as "initial conditions" for subsequent simulations of tool geometry and using empirical model outputs to "adjust said process".)
Strang does not explicitly use the term “trained AI/ML model " . However, Sadeghi teaches providing a plurality of outputs of the trained AI/ML model of the first location of the process chamber to the model of the second location of the process chamber ([0047]: The examiner interprets where Trained AI/ML model to second model a "trained" machine-learning module.)
It would have been obvious to a POSITA before the effective filing date of the invention to implement the empirical model of Strang using the trained machine-learning module of Sadeghi. Because Strang already recognizes that "coupling" in a spatially resolved tool model requires the transfer of multiple physical parameters (outputs) between locations, it is a predictable and routine application of known modeling principles to provide the outputs of the trained surrogate to the model of the second location. This configuration yields a functional, modular digital twin where the data dependencies between different parts of the chamber are satisfied via standard data hand-offs as taught by Strang & Mitrovic.
Regarding Claim 16
Strang & Mitrovic in combination with Sadeghi teaches The computer program product of claim 15 (See claim 15). Strang & Mitrovic teaches wherein providing the plurality of outputs of the trained AI/ML model of the first location of the process chamber to the model of the second location of the process chamber comprises: (See [FIG. 3 & 11], [Col 4 Line 5-8], [Col 9 Line 4-26]: “The present inventors have also discovered that the network architecture of FIG. 3 provides the ability to distribute model results done at one processing tool 102 for one condition set, to other similar or identical tools operating later under the same or similar conditions, so redundant simulations are eliminated. Running simulations only for unique processing conditions at on-tool and standalone modules and re-using results from similar tools that have already known simulated solutions allows for rapid development of lookup libraries containing results that can be used for diagnostics and control over a large range of processing conditions. Further, the reuse of the known solutions as initial conditions for first principles simulation reduces the computational requirements and facilitates the production of simulated solutions in a time frame consistent with on-line control. Similarly, the network architecture of FIG. 3 also provides the ability to propagate changes and refinements made to physical models and model input parameters from one simulation module to others in the network. For example, if during process runs and parallel executions of a model it is determined that some input parameters need to be changed, then these changes can be propagated to all other simulation modules and tools via the network.” The examiner interprets where providing the plurality of outputs of the trained AI/ML model... to the model of the second location... comprises is shown in generating a plurality of results (e.g., PLS model outputs T, P, U, C, W, E, F, H) and distributing them via a network for use as initial conditions for other tools.) waiting until the plurality of outputs of the trained AI/ML model of the first location of the process chamber have been received; (Col 22 Line 7- 26]: “In one example of the present invention, ANSYS is utilized to compute the temperature field. Utilizing the second component of the process model, (i.e. the gas dynamic component), the gas pressure field and velocity field can be determined using the surface temperatures computed in the thermal component, and several of the aforementioned measurements. For example, the mass flow rate and pressure (P1) can be utilized to determine an inlet condition, and the pressure (P3) can be utilized to determine an outlet condition, and CFD-ACE+ can be utilized to compute the gas pressure and velocity fields. Utilizing the chemistry model (i.e., the third component), the previously computed velocity, pressure, and temperature fields can be utilized as inputs to a chemistry model to compute, for example, an etch rate. Depending on the complexity of the process tool geometry, each of these model components can be executed on a time scale within a batch-to-batch process cycle. Any one of these components can, for example, be utilized to provide spatial uniformity data as input to the process control, methodology, process characterization, and/or fault detection/classification.” The examiner interprets where Waiting until outputs received is shown where in computer programming of modular simulations, the use of "previously computed" data implicitly requires a "wait" or synchronization state to ensure the first module has finished and the data is available before the second module starts.) and transmitting the plurality of outputs to the model of the second location of the process chamber. (See [Col 28 Line 65- Col 29 Line 11] & [Col 22 Line 7- 26]: The examiner interprets where transmitting the plurality of outputs to the model of the second location of the process chamber is shown in distributing simulation results via a network to other simulation modules/tools to be used as initial conditions & explicitly teaches that computed fields are "utilized as inputs" to a subsequent model, which requires the transmission of those data values between software modules.)
Strang does not explicitly use the term “trained AI/ML model ". However, Sadeghi teaches wherein providing the plurality of outputs of the trained AI/ML model of the first location of the process chamber to the model of the second location of the process chamber comprises ([0047]: The examiner interprets where Trained AI/ML model to second model a "trained" machine-learning module.)
It would have been obvious to a POSITA before the effective filing date of the invention to implement Strang's data utilization process by "waiting" until the plurality of outputs from the first module (which could be the trained AI/ML model of Sadeghi) have been received and then "transmitting" them to the model of the second location. The requirement for a "previously computed" result (Strang) inherently necessitates a synchronization point where the second model waits for the completion of the first. Managing such data dependencies via waiting and transmission is a well-known, routine technique in modular software design. Combining this standard programming practice with the modeling architecture of Strang and the AI/ML implementation of Sadeghi yields the predictable result of a functional, synchronized digital twin. Therefore, the additional steps of waiting and transmitting are an obvious matter of routine engineering design choice.
Regarding Claim 17
Strang & Mitrovic in combination with Sadeghi teaches The computer program product of claim 7. (See claim 7) Strang & Mitrovic teaches wherein coupling the trained AI/ML model of the first location of the process chamber to the model of the second location of the process chamber comprises (See [Col 14 Line 65- Col 15 Line 24]: The examiner interprets where coupling the trained AI/ML model of the first location... to the model of the second location... comprises is shown in a modular simulation environment where different components (thermal, gas, chemical) are executed for specific tool locations and environments.) providing a plurality of outputs of the model of the second location of the process chamber to the trained AI/ML model of the first location of the process chamber. (See [Col 21 Line 58- Col 22 Line 6] & [Col 21 Line 58- Col 22 Line 6]: The examiner interprets where providing a plurality of outputs of the model of the second location... to the trained AI/ML model of the first location is shown in a model component determining a gas-gap pressure field (Location 2), which is used to calculate thermal conductance provided to and used by the thermal component (Location 1) to determine a temperature field.)
It would have been obvious to a POSITA before the effective filing date of the invention to implement Strang's coupling process by providing the plurality of outputs from the model of the second location back to the trained AI/ML model of the first location. In the art of computational physics, physical domains like heat transfer and fluid dynamics are frequently interdependent, requiring "two-way coupling" where data is iteratively passed between modules to achieve a converged, accurate solution. Because Strang directs the POSITA to use solvers designed for such complex interactions, implementing a bidirectional data hand-off is a predictable application of known modeling techniques. This configuration yields a synchronized digital twin where the trained surrogate for the first location (per Sadeghi) is properly constrained by the physical results of the second location as taught Strang & Mitrovic. Therefore, providing outputs back to the first model is an obvious matter of routine engineering design choice.
Regarding Claim 18
Strang & Mitrovic in combination with Sadeghi teaches The computer program product of claim 7. (See claim 7) Strang & Mitrovic teaches further comprising computer-executable instructions for validating a performance of the trained AI/ML model of the first location of the process chamber (See [FIG. 3], [Col 10 Line 61-Col 11 Line 9]: “In addition to inputting the input data, the first principles simulation processor 108 also inputs the first principles physical model for emulating a physical sensor as shown by step 403. Step 403 includes inputting the physical attributes of the tool modeled by the model, as well as the first principles fundamental equations necessary to perform a first principles simulation to obtain a virtual sensor reading that can substitute for a physical sensor reading relating to the process performed by the semiconductor processing tool 102. The first principles physical model of step 403 may be input to the processor from an external memory or an internal memory device integral to the processor. Moreover, while step 403 is shown in FIG. 4 as following step 401, it is to be understood that the first principles simulation processor 104 may perform these steps simultaneously or in reverse of the order shown in FIG. 4.”
[FIG. 4]:
PNG
media_image13.png
1068
710
media_image13.png
Greyscale
[Col 11 Line 57-Col 12 Line 12]: “In yet another embodiment of the process of FIG. 4, the first principles simulation may be performed in a self-correcting mode by comparing virtual sensor measurements to corresponding physical sensor measurements. For example, during the first run with a certain process recipe/tool condition, the tool operator would use the "then best-known input parameters" for the model. During and after each simulation run, the simulation module(s) can compare the predicted "measurements" to the actual measurements, at locations where actual measurements from physical sensors, are made. If a significant difference is detected, optimization and statistical methods may be used to alter the input data and/or the first principles physical model itself, until better agreement of predicted and actual measured data is achieved. Depending on the situation, these additional refinement simulation runs may be made concurrent with the next wafer/wafer cassettes, or when the tool is off-line. Once refined input parameters are known, they can be stored in a library for later use, eliminating the need for subsequent input parameter and model refinements for the same process condition. Furthermore, refinements of the model and input data can be distributed via the network setup of FIG. 3 to other tools, eliminating the need for self-correcting runs in those other tools.” The examiner interprets where further comprising computer-executable instructions for validating a performance of the trained AI/ML model of the first location... is shown in "the simulation module(s) comparing the predicted 'measurements' to the actual measurements... If a significant difference is detected, optimization and statistical methods may be used to alter... the physical model itself".) after inclusion of the trained AI/ML model of the first location of the process chamber in the digital twin. (See [Col 11 Line 57-Col 12 Line 12]: The examiner interprets where after inclusion of the trained AI/ML model... in the digital twin is shown in validation (comparison) occurring "during and after each simulation run" once the model is implemented in the simulation module.)
Strang does not explicitly use the term “trained AI/ML model ". However, Sadeghi teaches further comprising computer-executable instructions for validating a performance of the trained AI/ML model of the first location of the process chamber ([0048]: “Any suitable machine-learning algorithm may be implemented to meet the needs of the operator and the machine-learning module 504. Some may include, but are not limited to a Bayesian network, linear regression, decision trees, neural networks, k-means clustering, and the like. The machine-learning is contemplated to be “supervised” because when the operator is prompted for some maintenance, the operator may then evaluate the accuracy of the machine-learned prompt while performing the maintenance. If the prompt is accurate, the operator may label the prompt “true,” and if the prompt is not accurate, the operator may label the prompt as “false,” along with any notations indicating his or her findings as to the actual need for maintenance. In this fashion, the machine-learning algorithm may learn from the operator of a particular machine, and if the machine-learning algorithm is distributed across many computers 103, the machine-learning algorithm may learn from thousands of semiconductor processing machines and operators. In certain embodiments, the maintenance detection module 500 and or the machine-learning module 504 may be provided as a service to operators of semiconductor processing system.” The examiner interprets where Validating performance of trained model is shown in an operator “evaluate[s] the accuracy” of model prompts.) The examiner interprets where Validating performance of trained model is shown in an operator "evaluate[s] the accuracy" of model prompts.) after inclusion of the trained AI/ML model of the first location of the process chamber in the digital twin. (See [0048]: The examiner interprets where After inclusion in digital twin is shown in validation by the operator while "performing the maintenance" prompted by the deployed model.)
It would have been obvious to a POSITA before the effective filing date of the invention to implement Strang's self-correcting validation instructions using the trained AI/ML performance evaluation techniques of Sadeghi. Including instructions to validate model performance after inclusion in the operational digital twin is a predictable application of known control-theory fundamentals used to ensure that a data-driven surrogate continues to accurately match physical reality. Such a combination represents a routine engineering choice to achieve the goal of accurate on-tool simulation and control recognized by both references.
Regarding Claim 19
Strang & Mitrovic in combination of Sadeghi teaches The computer program product of claim 18. (See claim 18) Strang & Mitrovic teaches wherein validating the performance of the trained AI/ML model comprises: (See [Col 11 Line 57-Col 12 Line 12]: The examiner interprets where validating the performance of the trained AI/ML model comprises is shown in comparing predicted "measurements" to actual measurements from physical sensors to "alter the input data and/or the first principles physical model itself".) generating simulated data using the digital twin that includes the trained AI/ML model of the first location of the process chamber and the model of the second location of the process chamber; (See [Col 21 Line 58- Col 22 Line 6], [Col 25 Line 33- Col 26 Line 13], [Col 27 Line 50- Col 28 Line 35]: “24. A system comprising: a semiconductor processing tool configured to perform a process; a fab-level library storing known simulation results obtained from simulation modules in a device manufacturing fab; a fab-level process controller distributing the known simulation results to other semiconductor processing tools in the device manufacturing fab; a first principles simulation processor configured to input a first principles physical model including a set of computer-encoded differential equations describing at least one of a basic physical or chemical attribute of the semiconductor processing tool and including 1) a spatially resolved model of a physical geometry of the semiconductor processing tool and 2) a grid set addressing the semiconductor processing tool or a geometry of the semiconductor processing tool; an input device configured to input process data related to an actual process being performed by the semiconductor processing tool; and said first principles simulation processor further configured to: set boundary conditions for the spatially resolved model of a physical geometry of the semiconductor processing tool based on said process data related to the actual process being performed by the semiconductor processing tool, solve the computer-encoded differential equations of the first principles physical model for the spatially resolved model concurrently with the actual process being performed and in a time frame shorter in time than the actual process being performed to produce a first principles simulation by: using code parallelization techniques on multiple simulation modules in the device manufacturing fab, and re-using known simulation solutions as initial conditions for the first principles simulation, wherein re-using known simulation solutions comprises searching in the fab-level library for a closest fitting solution which if used for the initial condition would reduce the number of iterations required by the simulation module, and provide from the solution of the computer-encoded differential equations solved concurrently with the actual process being performed a first principles simulation result, and use the first principles simulation result obtained during performance of the actual process to build an empirical model, wherein at least one of said first principles simulation result and said empirical model is selected to control the actual process being performed by the semiconductor processing tool.” The examiner interprets where generating simulated data using the digital twin that includes the trained AI/ML model of the first location... and the model of the second location... is shown in a simulation module executing coupled model components (e.g., thermal and gas dynamic) to provide simulation results used for tool control.) and comparing the simulated data to experimental data collected using a plurality of sensors associated with a physical process chamber. (See [Col 20 Line 22-Col 21 Line 21]: The examiner interprets where comparing the simulated data to experimental data collected using a plurality of sensors associated with a physical process chamber is shown in a difference between the real performance (Yreal) and simulated performance (Ysim) is used for fault detection, where Yreal is measured by physical sensors.)
It would have been obvious to a POSITA before the effective filing date of the invention to implement Strang's self-correcting comparison process using the trained machine-learning module of Sadeghi. Comparing the integrated outputs of a modular twin against real-world hardware sensors is the standard, predictable method for ensuring that a data-driven model (AI/ML) remains accurate and representative of the physical process. Such a combination represents a routine engineering choice to achieve the goal of high-fidelity, on-tool simulation and control as sought by both references.
Regarding Claim 20
Strang & Mitrovic in combination with Sadeghi teaches The computer program product of claim 7. (See Claim 7) Strang & Mitrovic teaches wherein the model of the second location of the process chamber is an HFS model, (See [Col 14 Line 65- Col 15 Line 24] & [Col 27 Line 50- Col 28 Line 35]: The examiner interprets where the model of the second location of the process chamber is an HFS model is shown in a simulation module executing a "first principles physical model" (HFS) for a tool environment (second location).) and further comprising computer-executable instructions for replacing the HFS model of the second location of the process chamber (See [Col 17 Line 28-52]: The examiner interprets where further comprising computer-executable instructions for replacing the HFS model of the second location... is shown in an empirical model assimilated over time; which explicitly teaches that the empirical model "can supersede the generally more intensive process model based on first principles simulation".) with a trained AI/ML model of the second location in the digital twin. (See [Col 25 Line 33- Col 26 Line 13]: The examiner interprets where with a trained AI/ML model of the second location in the digital twin is shown in constructing an empirical model (AI/ML) from simulation results to provide input for process adjustment,. Under BRI, "superseding" the simulation constitutes "replacement".)
Strang does not explicitly use the term “trained AI/ML model ". However, Sadeghi teaches with a trained AI/ML model of the second location in the digital twin (See [0047]: The examiner interprets where Trained AI/ML surrogate is shown in the specific implementation of the surrogate as a "machine-learning module" (AI/ML) that is specifically "trained" using sensor data.)
It would have been obvious to a POSITA before the effective filing date of the invention to implement the "empirical model" of Strang using the trained "machine-learning module" taught by Sadeghi. Because Strang already teaches the logical step of substituting a slow physics simulation for a fast empirical surrogate to facilitate on-line control, providing specific instructions to perform this replacement with a trained AI/ML model is nothing more than the application of a modern modeling technique to the architecture and goals of the primary reference. Such a combination yields the predictable result of an optimized digital twin that balances initial physical accuracy with long-term computational speed. Therefore, the transition between model types is a routine matter of engineering design.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to AARIC RAYJEE MARKS whose telephone number is (571)467-6372. The examiner can normally be reached Monday-Friday 8am-5pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Ryan Pitaro can be reached at (571) 272-4071. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/AARIC R MARKS/
/RYAN F PITARO/ Supervisory Patent Examiner, Art Unit 2188
1 See Spec [0051]: “A “digital twin” of a process chamber or other type of digital equipment as used herein refers to a model of an entire process chamber. In some embodiments, a digital twin can be made up of multiple models of different types, where each model represents a different class of physical phenomenon and/or a different location of the process chamber. For example, a digital twin can include a structural model of a showerhead, a thermal model of the showerhead, a chemistry model of a gap between the showerhead and a pedestal, a Computational Fluid Dynamics (CFD) model of the gap between the showerhead and the pedestal, etc. In some embodiments, each model that makes up the digital twin can be one of: 1) a closed-form solution; 2) an AI/ML model; and 3) an HFS model. In other words, a digital twin may be comprised of any of any combinations of a closed-form solution, an AI/ML model, and/or an HFS model”
Prosecution Insights