Tool control using multistage LSTM for predicting on-wafer measurements
A method for process control using predictive long short term memory includes obtaining historical post-process measurements taken on prior products of the manufacturing process; obtaining historical in-process measurements taken on prior workpieces during the manufacturing process; training a neural network to predict each of the historical post-process measurements, in response to the corresponding historical in-process measurements and preceding historical post-process measurements; obtaining present in-process measurements on a present workpiece during the manufacturing process; predicting a future post-process measurement for the present workpiece, by providing the present in-process measurements and the historical post-process measurements as inputs to the neural network; and adjusting at least one controllable variable of the manufacturing process in response to the prediction of the future post-process measurement.
1. A method for controlling a manufacturing process in real-time during a current process step of the manufacturing process, the method comprising:
obtaining historical post-process measurements taken on prior products of the manufacturing process;
obtaining historical in-process measurements taken on prior workpieces during prior iterations of the current process step;
training a neural network to predict at least one of the historical post-process measurements in response to the corresponding historical in-process measurements and in response to a subset of the historical post-process measurements that preceded the at least one of the historical post-process measurements to be predicted;
obtaining present in-process measurements on a present workpiece during the current process step;
predicting a future post-process measurement for the present workpiece, by providing the present in-process measurements and the historical post-process measurements as inputs to the neural network, wherein predicting the future post-process measurement comprises a first long short term memory layer taking as inputs the present in-process measurements, and producing as an output a naïve prediction of the future post-process measurement, wherein predicting the future post-process measurement also comprises a second long short term memory layer taking as inputs historical post-process measurements and the naïve prediction of the post-process measurement, and passing an output to a fully connected layer which produces an informed prediction of the future post-process measurement; and
adjusting at least one controllable variable of the manufacturing process in response to the prediction of the future post-process measurement.
2. The method of claim 1 wherein adjusting the at least one controllable variable includes repeatedly obtaining at least one in-process measurement and predicting the future post-process measurement.
3. The method of claim 1 wherein adjusting the at least one controllable variable includes changing the at least one controllable variable consistent with an empirical error function for the future post-process measurement relative to the at least one controllable variable.
4. A method for controlling a manufacturing process in real-time during a current process step of the manufacturing process, the method comprising:
obtaining historical post-process measurements taken on prior products of the manufacturing process;
obtaining historical in-process measurements taken on prior workpieces during prior iterations of the current process step;
training a neural network to predict at least one of the historical post-process measurements in response to the corresponding historical in-process measurements and in response to a subset of the historical post-process measurements that preceded the at least one of the historical post-process measurements to be predicted;
obtaining present in-process measurements on a present workpiece during the current process step;
predicting a future post-process measurement for the present workpiece, by providing the present in-process measurements and the historical post-process measurements as inputs to the neural network; and
adjusting at least one controllable variable of the manufacturing process in response to the prediction of the future post-process measurement, wherein predicting the future post-process measurement comprises a first long short term memory layer taking as inputs the present in-process measurements, and producing as an output a first naïve prediction of the future post-process measurement, and wherein the neural network incorporates a second long short term memory layer that takes as inputs in-process measurements from a previous process step that preceded the current process step on the present workpiece, and that produces as an output a second naïve prediction of the future post-process measurement, and wherein the neural network further incorporates a third long short term memory layer that takes as inputs the first naïve prediction, the second naïve prediction, and historical post-process measurements and that passes an output to a fully connected layer which produces as an output an informed prediction of the future post-process measurement.
5. A non-transitory computer readable medium embodying computer executable instructions which when executed by a computer cause the computer to facilitate a method for controlling a manufacturing process in real-time during a current process step of the manufacturing process, the method comprising:
obtaining historical post-process measurements taken on prior products of the manufacturing process;
obtaining historical in-process measurements taken on prior workpieces during prior iterations of the current process step;
training a neural network to predict at least one of the historical post-process measurements in response to the corresponding historical in-process measurements and in response to a subset of the historical post-process measurements that preceded the at least one of the historical post-process measurements to be predicted;
obtaining present in-process measurements on a present workpiece during the current process step;
predicting a future post-process measurement for the present workpiece, by providing the present in-process measurements and the historical post-process measurements as inputs to the neural network, wherein predicting the future post-process measurement comprises a first long short term memory layer taking as inputs the present in-process measurements, and producing as an output a naïve prediction of the future post-process measurement, wherein predicting the future post-process measurement also comprises a second long short term memory layer taking as inputs historical post-process measurements and the naïve prediction of the post-process measurement, and passing an output to a fully connected layer which produces an informed prediction of the future post-process measurement; and
adjusting at least one controllable variable of the manufacturing process in response to the prediction of the future post-process measurement.
6. The method of claim 5 wherein adjusting the at least one controllable variable includes repeatedly obtaining at least one in-process measurement and predicting the future post-process measurement.
7. A non-transitory computer readable medium embodying computer executable instructions which when executed by a computer cause the computer to facilitate a method for controlling a manufacturing process in real-time during a current process step of the manufacturing process, the method comprising:
obtaining historical post-process measurements taken on prior products of the manufacturing process;
obtaining historical in-process measurements taken on prior workpieces during prior iterations of the current process step;
training a neural network to predict at least one of the historical post-process measurements in response to the corresponding historical in-process measurements and in response to a subset of the historical post-process measurements that preceded the at least one of the historical post-process measurements to be predicted;
obtaining present in-process measurements on a present workpiece during the current process step;
predicting a future post-process measurement for the present workpiece, by providing the present in-process measurements and the historical post-process measurements as inputs to the neural network; and
adjusting at least one controllable variable of the manufacturing process in response to the prediction of the future post-process measurement, wherein predicting the future post-process measurement comprises a first long short term memory layer taking as inputs the present in-process measurements, and producing as an output a first naïve prediction of the future post-process measurement, and wherein the neural network incorporates a second long short term memory layer that takes as inputs in-process measurements from a previous process step that preceded the current process step on the present workpiece, and that produces as an output a second naïve prediction of the future post-process measurement, and wherein the neural network further incorporates a third long short term memory layer that takes as inputs the first naïve prediction, the second naïve prediction, and historical post-process measurements and that passes an output to a fully connected layer which produces as an output an informed prediction of the future post-process measurement.
8. A process control system comprising:
a memory embodying computer executable instructions; and
at least one processor, coupled to the memory, and operative by the computer executable instructions to facilitate a method for controlling a manufacturing process in real-time during a current process step of the manufacturing process, the method comprising:
obtaining historical post-process measurements taken on prior products of the manufacturing process;
obtaining historical in-process measurements taken on prior workpieces during prior iterations of the current process step;
training a neural network to predict each of the historical post-process measurements in response to the corresponding historical in-process measurements and in response to a subset of the historical post-process measurements that preceded the at least one of the historical post-process measurements to be predicted;
obtaining present in-process measurements on a present workpiece during the current process step;
predicting a future post-process measurement for the present workpiece, by providing the present in-process measurements and the historical post-process measurements as inputs to the neural network, wherein predicting the future post-process measurement comprises a first long short term memory layer taking as inputs the present in-process measurements, and producing as an output a naïve prediction of the future post-process measurement, wherein predicting the future post-process measurement also comprises a second long short term memory layer taking as inputs historical post-process measurements and the naïve prediction of the post-process measurement, and passing an output to a fully connected layer which produces an informed prediction of the future post-process measurement; and
adjusting at least one controllable variable of the manufacturing process in response to the prediction of the future post-process measurement.
9. The system of claim 8 wherein adjusting the at least one controllable variable includes repeatedly obtaining at least one in-process measurement and predicting the future post-process measurement.
10. The system of claim 9 wherein adjusting the at least one controllable variable includes changing the at least one controllable variable consistent with an empirical error function for the post-process measurement relative to the at least one controllable variable.
11. A process control system comprising:
a memory embodying computer executable instructions; and
at least one processor, coupled to the memory, and operative by the computer executable instructions to facilitate a method for controlling a manufacturing process in real-time during a current process step of the manufacturing process, the method comprising:
obtaining historical post-process measurements taken on prior products of the manufacturing process;
obtaining historical in-process measurements taken on prior workpieces during prior iterations of the current process step;
training a neural network to predict each of the historical post-process measurements in response to the corresponding historical in-process measurements and in response to a subset of the historical post-process measurements that preceded the at least one of the historical post-process measurements to be predicted;
obtaining present in-process measurements on a present workpiece during the current process step;
predicting a future post-process measurement for the present workpiece, by providing the present in-process measurements and the historical post-process measurements as inputs to the neural network; and
adjusting at least one controllable variable of the manufacturing process in response to the prediction of the future post-process measurement, wherein predicting the future post-process measurement comprises a first long short term memory layer taking as inputs the present in-process measurements, and producing as an output a first naïve prediction of the future post-process measurement, and wherein the neural network incorporates a second long short term memory layer that takes as inputs in-process measurements from a previous process step that preceded the current process step on the present workpiece, and that produces as an output a second naïve prediction of the future post-process measurement, and wherein the neural network further incorporates a third long short term memory layer that takes as inputs the first naïve prediction, the second naïve prediction, and historical post-process measurements and that passes an output to a fully connected layer which produces as an output an informed prediction of the future post-process measurement.