Stabilizing GAN Training by Regulating Learning Rates in Semiconductor Workpiece Inspection Models
Systems and methods for inspecting semiconductor workpieces are provided. In one example, a method includes obtaining workpiece data for a semiconductor workpiece. The method includes providing the workpiece data as input to an inspection model, the inspection model being a stabilized learning generative adversarial network (SLGAN) trained model, wherein the SLGAN trained model is associated with a regulated learning rate for one or more of a discriminator network or a generator network. The method also includes obtaining an output from the inspection model, the output associated with one or more characteristics of the semiconductor workpiece.
1 . A method for inspecting semiconductor workpieces, the method comprising:
obtaining workpiece data for a semiconductor workpiece;
providing the workpiece data as input to an inspection model, the inspection model being a stabilized learning generative adversarial network (SLGAN) trained model, wherein the SLGAN trained model is associated with a regulated learning rate for one or more of a discriminator network or a generator network; and
obtaining an output from the inspection model, the output associated with one or more characteristics of the semiconductor workpiece.
2 . The method of claim 1 , wherein the inspection model comprises a machine-learned autoencoder model, the machine-learned autoencoder model comprising an encoding portion and a decoding portion, wherein the decoding portion of the autoencoder model generates a target image, wherein the decoding portion of the autoencoder is trained at least in part using the discriminator network.
3 . The method of claim 2 , wherein the output comprises an encoding from the encoding portion of the machine-learned autoencoder model.
4 . The method of claim 3 , wherein the encoding is indicative of a similarity of the semiconductor workpiece or an anomaly of the semiconductor workpiece.
5 . The method of claim 3 , wherein the encoding is indicative of a feature or a feature distribution of the semiconductor workpiece.
6 . The method of claim 5 , wherein the feature is one or more of a threading edge dislocation, basal plan dislocation, super screw dislocation, micropipe, mixed dislocation, hexagonal void, stacking fault, or scratch.
7 . The method of claim 1 , wherein the workpiece data comprises image data of at least a portion of the semiconductor workpiece.
8 . The method of claim 7 , wherein the image data comprises one or more of an optical surface microscopy image, photoluminescence (PL) microscopy image, cross-polarized light imaging image, x-ray topography image, or a scanning electron microscopy image.
9 . The method of claim 1 , wherein the output is a feature detection output from the inspection model, wherein the feature detection output comprises a target image, the target image comprising one or more pixels associated with a feature or feature distribution.
10 . The method of claim 9 , wherein the feature detection output comprises data indicative of one or more locations of the feature or feature distribution, classification of the feature or feature distribution, size of the feature or feature distribution, or shape of the feature or feature distribution.
11 . The method of claim 1 , wherein the output is an image translation output providing second image data that is different from the image data of at least a portion of the workpiece.
12 . The method of claim 1 , wherein the discriminator network has a first learning rate that is different than a second learning rate of the generator network.
13 . The method of claim 12 , wherein the first learning rate of the discriminator network is a regulated learning rate based at least in part on an adversarial ratio, the adversarial ratio determined based on a ratio of a first loss of the generator network to a second loss of the discriminator network.
14 . The method of claim 13 , wherein when the adversarial ratio is greater than a threshold for a training epoch, one or more gradients for a next training period for the discriminator network are frozen relative to one or more gradients for the next training period for the generator network.
15 . The method of claim 14 , wherein the one or more gradients for the discriminator network remain frozen until the adversarial ratio is less than or equal to the threshold.
16 . The method of claim 1 , wherein the semiconductor workpiece comprises a silicon carbide semiconductor wafer.
17 . The method of claim 1 , wherein the method comprises determining one or more characteristics of the semiconductor workpiece based at least in part on the output.
18 . The method of claim 1 , wherein the method comprises modifying a semiconductor manufacturing process based at least in part on the output.
19 . A method for training a machine-learned model comprising a generative adversarial network, the method comprising:
conducting a first training epoch for a generative network;
determining a first loss for the generative network;
conducting a second training epoch for a discriminator network;
determining a second loss for the discriminator network; and
regulating a learning rate for one or more of the generative network or the discriminator network based at least in part on the first loss for the generative network and the second loss for the discriminator network.
20 . A system for inspection of a semiconductor workpiece, the system comprising:
one or more imaging devices configured to capture image data of at least a portion of the semiconductor workpiece;
processing circuitry configured to perform operations, the operations comprising:
providing workpiece data as input to an inspection model, the inspection model being a generative adversarial network (GAN) trained model, wherein the GAN trained model is associated with a regulated learning rate for one or more of a discriminator network or a generator network; and
obtaining an output from the inspection model, the output associated with one or more characteristics of the semiconductor workpiece.