System and method for accelerating physical simulation models during microelectronic device fabrication
A method for accelerating physical simulation models during microelectronic device fabrication may include, but is not limited to, running a physical simulation model at a first grid size having a first resolution, generating outputs at the first grid size having the first resolution from the physical simulation model, inputting the outputs at the first grid size having the first resolution from the physical simulation model into a resolution enhancement model (REM), and generating outputs at a second grid size having a second resolution via the REM to reduce at least one of a computational time cost or a computational resource cost of a controller running the physical simulation model. The physical simulation model may simulate an on-wafer performance of at least one microelectronics fabrication process. The second grid size may be smaller than the first grid size. The second resolution may be higher than the first resolution.
1 . A system, comprising:
a controller, wherein the controller includes one or more processors and memory, wherein the memory is configured to store program instructions, wherein the one or more processors are configured to execute the program instructions causing the one or more processors to:
run a physical simulation model at a first grid size having a first resolution, wherein the physical simulation model simulates an on-wafer performance of at least one microelectronics fabrication process;
generate outputs at the first grid size having the first resolution from the physical simulation model;
input the outputs at the first grid size having the first resolution from the physical simulation model into a resolution enhancement model (REM); and
generate outputs at a second grid size having a second resolution via the REM, wherein the second grid size is smaller than the first grid size, wherein the second resolution is higher than the first resolution; and
adjusting the REM to achieve a desired degree of similarity of the outputs of the REM model,
wherein the generation of the outputs at the second grid size having the second resolution via the REM based on the outputs at the first grid size having the first resolution from the physical simulation model reduces at least one of a computational time cost or a computational resource cost of the controller to run the physical simulation model.
2 . The system of claim 1 , wherein the outputs at the first grid size having the first resolution from the physical simulation model include one or more images having the first resolution, wherein the outputs at the second grid size having the second resolution generated via the REM include one or more images having the second resolution.
3 . The system of claim 1 , wherein the first grid size is Δ, wherein the second grid size is fΔ, wherein f is less than 1.0.
4 . The system of claim 1 , wherein the controller is further configured to:
run the physical simulation model at the first grid size having the first resolution;
run the physical simulation model at the second grid size having the second resolution; and
develop the REM based on outputs from the physical simulation model run at the first grid size having the first resolution and the physical simulation model run at the second grid size having the second resolution.
5 . The system of claim 4 , wherein the controller is further configured to:
develop the REM through a training process of at least one of a deep learning model, a machine learning model, or a heuristics-based model.
6 . The system of claim 5 , wherein the controller is further configured to:
run the physical simulation model at the first grid size having the first resolution for an n number of trials; and
generate an m number of outputs at the second grid size having the second resolution from the REM, wherein m is less than n.
7 . The system of claim 5 , wherein a subset of an n number of trials equates to a single output through the REM.
8 . The system of claim 1 , wherein the controller is further configured to:
compare the outputs at the second grid size having the second resolution generated by the REM to the set of reference data to determine whether the outputs at the second grid size having the second resolution are within a select mismatch tolerance; and
iteratively adjust one or more parameters of the physical simulation model when the outputs at the second grid size having the second resolution are outside of the select mismatch tolerance.
9 . The system of claim 8 , wherein the controller is further configured to:
obtain the set of reference data from a metrology subsystem.
10 . The system of claim 1 , wherein the controller is further configured to:
run a plurality of trials of the physical simulation model at the first grid size having the first resolution to produce a probabilistic distribution of the plurality of trials;
generate a plurality of outputs at the first grid size having the first resolution from the plurality of trials of the physical simulation model;
run the plurality of outputs at the first grid size having the first resolution through a post-processing model;
generate a plurality of post-processing model outputs at the first grid size having the first resolution via the post-processing model;
run the plurality of post-processing model outputs at the first grid size having the first resolution through the REM; and
generate a plurality of outputs at the second grid size having the second resolution from the REM.
11 . The system of claim 10 , wherein the post-processing model includes a Monte Carlo simulation.
12 . A method comprising:
running a physical simulation model at a first grid size having a first resolution, wherein the physical simulation model simulates an on-wafer performance of at least one microelectronics fabrication process;
generating outputs at the first grid size having the first resolution from the physical simulation model;
inputting the outputs at the first grid size having the first resolution from the physical simulation model into a resolution enhancement model (REM); and
generating outputs at a second grid size having a second resolution via the REM, wherein the second grid size is smaller than the first grid size, wherein the second resolution is higher than the first resolution; and
adjusting the REM to achieve a desired degree of similarity of the outputs of the REM model,
wherein the generation of the outputs at the second grid size having the second resolution via the REM based on the outputs at the first grid size having the first resolution from the physical simulation model reduces at least one of a computational time cost or a computational resource cost of a controller running the physical simulation model.
13 . The method of claim 12 , wherein the outputs at the first grid size having the first resolution from the physical simulation model include one or more images having the first resolution, wherein the outputs at the second grid size having the second resolution generated via the REM include one or more images having the second resolution.
14 . The method of claim 12 , wherein the first grid size is Δ, wherein the second grid size is fΔ, wherein f is less than 1.0.
15 . The method of claim 12 , further comprising:
running a physical simulation model at the first grid size having the first resolution;
running the physical simulation model at the second grid size having the second resolution; and
developing the REM based on outputs from the physical simulation model run at the first grid size having the first resolution and the physical simulation model run at the second grid size having the second resolution.
16 . The method of claim 15 , further comprising:
developing the REM through a training process of at least one of a deep learning model, a machine learning model, or a heuristics-based model.
17 . The method of claim 16 , further comprising:
running a physical simulation model at the first grid size having the first resolution for an n number of trials; and
generating an m number of outputs at the second grid size having the second resolution from the REM, wherein m is less than n.
18 . The method of claim 16 , wherein a subset of an n number of trials equates to a single output through the REM.
19 . The method of claim 12 , further comprising:
comparing the outputs at the second grid size having the second resolution generated by the REM to the set of reference data to determine whether the outputs at the second grid size having the second resolution are within a select mismatch tolerance; and
iteratively adjusting one or more parameters of the physical simulation model when the outputs at the second grid size having the second resolution are outside of the select mismatch tolerance.
20 . The method of claim 19 , further comprising:
obtaining the set of reference data from a metrology subsystem.
21 . The method of claim 12 , further comprising:
running a plurality of trials of the physical simulation model at the first grid size having the first resolution to produce a probabilistic distribution of the plurality of trials;
generating a plurality of outputs at the first grid size having the first resolution from the plurality of trials of the physical simulation model;
running the plurality of outputs at the first grid size having the first resolution through a post-processing model;
generating a plurality of post-processing model outputs at the first grid size having the first resolution via the post-processing model;
running the plurality of post-processing model outputs at the first grid size having the first resolution through the REM; and
generating a plurality of outputs at the second grid size having the second resolution from the REM.
22 . The method of claim 21 , wherein the post-processing model includes a Monte Carlo simulation.
23 . A system, comprising:
a metrology subsystem configured to acquire one or more measurements of a portion of a sample; and
a controller communicatively coupled to, wherein the controller includes one or more processors and memory, wherein the memory is configured to store program instructions, wherein the one or more processors are configured to execute the program instructions causing the one or more processors to:
obtain one or more inputs from the metrology subsystem, wherein the one or more inputs include images having a first resolution;
run the physical simulation model at a first grid size having the first resolution, wherein the physical simulation model simulates an on-wafer performance of at least one microelectronics fabrication process;
generate outputs at the first grid size having the first resolution from the physical simulation model;
input the outputs at the first grid size having the first resolution from the physical simulation model into a resolution enhancement model (REM); and
generate outputs at a second grid size having a second resolution via the REM, wherein the second grid size is smaller than the first grid size, wherein the second resolution is higher than the first resolution; and
adjusting the REM to achieve a desired degree of similarity of the outputs of the REM model,
wherein the generation of the outputs at the second grid size having the second resolution via the REM based on the outputs at the first grid size having the first resolution from the physical simulation model reduces at least one of a computational time cost or a computational resource cost of the controller to run the physical simulation model.
24 . The system of claim 23 , wherein the outputs at the first grid size having the first resolution from the physical simulation model include one or more images having the first resolution, wherein the outputs at the second grid size having the second resolution generated via the REM include one or more images having the second resolution.
25 . The system of claim 23 , wherein the first grid size is Δ, wherein the second grid size is fΔ, wherein f is less than 1.0.
26 . The system of claim 23 , wherein the controller is further configured to:
run the physical simulation model at the first grid size having the first resolution;
run the physical simulation model at the second grid size having the second resolution; and
develop the REM based on outputs from the physical simulation model run at the first grid size having the first resolution and the physical simulation model run at the second grid size having the second resolution.
27 . The system of claim 26 , wherein the controller is further configured to:
develop the REM through a training process of at least one of a deep learning model, a machine learning model, or a heuristics-based model.
28 . The system of claim 27 , wherein the controller is further configured to:
run the physical simulation model at the first grid size having the first resolution for an n number of trials; and
generate an m number of outputs at the second grid size having the second resolution from the REM, wherein m is less than n.
29 . The system of claim 27 , wherein a subset of an n number of trials equates to a single output through the REM.
30 . The system of claim 23 , wherein the controller is further configured to:
compare the outputs at the second grid size having the second resolution generated by the REM to the set of reference data to determine whether the outputs at the second grid size having the second resolution are within a select mismatch tolerance; and
iteratively adjust one or more parameters of the physical simulation model when the outputs at the second grid size having the second resolution are outside of the select mismatch tolerance.
31 . The system of claim 30 , wherein the controller is further configured to:
obtain the set of reference data from the metrology subsystem.
32 . The system of claim 23 , wherein the controller is further configured to:
run a plurality of trials of the physical simulation model at the first grid size having the first resolution to produce a probabilistic distribution of the plurality of trials;
generate a plurality of outputs at the first grid size having the first resolution from the plurality of trials of the physical simulation model;
run the plurality of outputs at the first grid size having the first resolution through a post-processing model;
generate a plurality of post-processing model outputs at the first grid size having the first resolution via the post-processing model;
run the plurality of post-processing model outputs at the first grid size having the first resolution through the REM; and
generate a plurality of outputs at the second grid size having the second resolution from the REM.
33 . The system of claim 32 , wherein the post-processing model includes a Monte Carlo simulation.