Semiconductor manufacturing using artificial intelligence
In one embodiment, a method, a method of manufacturing a semiconductor is disclosed. A monitored semiconductor manufacturing system (monitored system) is operated over a period of time, the monitored system comprising an impedance matching network coupled between a radio frequency (RF) source and a plasma chamber. First values for a parameter of the monitored system are received, the first values comprising different values for the parameter over the time period of operation of the monitored system, and a learning model is trained using the first values for the parameter. A substrate is then placed in a plasma chamber of a controlled semiconductor manufacturing system (controlled system). A characteristic of the controlled system is determined using a current value of the parameter and the trained learning model. An action is then taken upon the controlled system to address the determined characteristic.
1. An impedance matching network comprising:
a radio frequency (RF) input configured to operably couple to an RF source of a semiconductor manufacturing system, the system configured to operate over a period of time;
an RF output configured to operably couple to a plasma chamber of the system;
a variable reactance element (VRE), the VRE having different positions for providing different reactances; and
a control circuit configured to carry out the operations of:
receiving first values for a parameter of the system, the first values comprising different values for the parameter over the time period of operation of the system;
training a learning model using the first values for the parameter;
determining, based on a received signal, a current value of the parameter for the system;
determining a characteristic of the system using the current value and the trained learning model; and
causing an action to be performed upon the system to address the determined characteristic.
2. A non-transitory computer-readable storage medium encoded with instructions which, when executed on a processor, perform a method of:
a) operating a monitored semiconductor manufacturing system (monitored system) over a period of time, the monitored system comprising an impedance matching network coupled between a radio frequency (RF) source and a plasma chamber, the matching network comprising a variable reactance element (VRE), the VRE having different positions for providing different reactances; and
b) receiving first values for a parameter of the monitored system, the first values comprising different values for the parameter over the time period of operation of the monitored system;
c) training a learning model using the first values for the parameter;
d) placing a substrate in a plasma chamber of a controlled semiconductor manufacturing system (controlled system), the controlled system comprising an impedance matching network coupled between an RF source and the plasma chamber, the matching network of the controlled system comprising a VRE having different positions for providing different reactances, and the plasma chamber depositing a material layer onto the substrate or etching a material layer from the substrate;
e) determining, based on a received signal, a current value of the parameter for the controlled system; and
f) determining a characteristic of the controlled system using the current value and the trained learning model; and
g) taking an action upon the controlled system to address the determined characteristic.
3. The medium of claim 2 wherein the monitored system and the controlled system are the same system.
4. The medium of claim 2 wherein step b) is performed by the matching network of the monitored system, and steps e) and f) are performed by the matching network of the controlled system.
5. The medium of claim 2 wherein the controlled system forms part of a plurality of controlled systems each controlled using the trained learning model.
6. A method of manufacturing a semiconductor, the method comprising:
a) operating a monitored semiconductor manufacturing system (monitored system) over a period of time, the monitored system comprising an impedance matching network coupled between a radio frequency (RF) source and a plasma chamber, the matching network comprising a variable reactance element (VRE), the VRE having different positions for providing different reactances; and
b) receiving first values for a parameter of the monitored system, the first values comprising different values for the parameter over the time period of operation of the monitored system;
c) training a learning model using the first values for the parameter;
d) placing a substrate in a plasma chamber of a controlled semiconductor manufacturing system (controlled system), the controlled system comprising an impedance matching network coupled between an RF source and the plasma chamber, the matching network of the controlled system comprising a VRE having different positions for providing different reactances, and the plasma chamber depositing a material layer onto the substrate or etching a material layer from the substrate;
e) determining, based on a received signal, a current value of the parameter for the controlled system; and
f) determining a characteristic of the controlled system using the current value and the trained learning model; and
g) taking an action upon the controlled system to address the determined characteristic;
wherein steps b) to g) are performed by one or more processors.
7. The method of claim 6 wherein the monitored system and the controlled system are the same system.
8. The method of claim 6 wherein step b) is performed by the matching network of the monitored system, and steps e) and f) are performed by the matching network of the controlled system.
9. The method of claim 6 wherein the controlled system forms part of a plurality of controlled systems each controlled using the trained learning model.
10. The method of claim 6 wherein the current value of the parameter is provided to the learning model to continue the training of the learning model.
11. The method of claim 6 wherein the learning model receives the current value of the parameter and determines the characteristic of the system.
12. The method of claim 6 wherein the characteristic is determined by comparing the current value to an output of the learning model.
13. The method of claim 6 further comprising:
for each of the first values of the parameter, further receiving a corresponding second value for a second parameter; and
further training the learning model using the second values and their correspondence with the first values;
wherein the action taken is, using the learning model and the current value of the parameter, altering the second parameter of the controlled system to a new value determined by the learning model to return the parameter of the monitored system to a predetermined valid range of values.
14. The method of claim 6 further comprising:
for each of the first values of the parameter, further receiving a corresponding VRE position for the monitored system; and
further training the learning model using the VRE positions corresponding with the first values;
wherein the training of the learning model comprises the learning model determining a valid range of values for the parameter based on the received first values, and learning which of the VRE positions for the monitored system will cause the parameter to return to the valid range of values; and
wherein the action taken is, using the learning model and the current value of the parameter, altering the VRE of the controlled system to a new position determined by the learning model to return the parameter of the monitored system to the valid range of values.
15. The method of claim 6 wherein the determined characteristic of the controlled system is that there is a problem with the plasma chamber, that the plasma chamber needs to be cleaned, or that the plasma chamber has reached a particular stage in processing a silicon wafer.
16. The method of claim 6 wherein the parameter for each of the monitored system and the controlled system is at least one of:
a current position of the VRE from among the possible positions of the VRE;
a load impedance at an RF output of the matching network or at a node internal to the matching network;
a matching tune time;
a reflected RF power or a reflection coefficient at an RF input of the matching network;
a peak-to-peak voltage at an RF output of the matching network;
a DC voltage at an RF output of the matching network;
a DC current at an RF output of the matching network; or
a phase angle of the load impedance at an RF output of the matching network.
17. The method of claim 6 :
wherein the parameter is a power setpoint of the RF source;
wherein the determined characteristic is a change in the power setpoint; and
wherein the action taken upon the controlled system is starting a switching process for the VRE of the controlled system, wherein the VRE is electronically variable such that its reactance is varied by switches carrying out the switching process.
18. The method of claim 6 :
wherein the determined characteristic of the controlled system comprises a determination that etching in the plasma chamber has reached a certain layer of a silicon wafer, and the action taken comprises stopping etching; and
wherein the determined characteristic of the controlled system further comprises a determination that depositing of a layer on a silicon wafer in the plasma chamber has reached a predetermined thickness, and the action taken further comprises stopping deposition of the layer.
19. The method of claim 6 :
wherein the training of the learning model comprises the learning model determining a valid range of values for the parameter based on the received first values for the monitored system; and
wherein determination of the characteristic of the controlled system is based on a determination that the current value is outside the valid range of values.
20. The method of claim 19 wherein the valid range of values is based on values for the monitored system determined over the time period of operation for:
an amount of current drawn by each of a plurality of switches of the VRE of the monitored system, the VRE being electronically variable;
an amount of current drawn by at least one fan of the matching network of the monitored system;
a number of switching operations by each of a plurality of switches of the VRE of the monitored system, the VRE being electronically variable;
a number of motor reversals and a duration of each motor operation for a motor of the matching network of the monitored system, the VRE being mechanically variable using the motor;
an amount of DC or AC current drawn from an input power supply of the monitored system; or
an AC or DC voltage at a point inside the matching network of the monitored system.