Methods and systems for registering images for electronic designs
View Patent ↗Systems for training a convolutional neural network to register images for masks or wafers in semiconductor manufacturing include a computer processor configured to receive a first pair of images aligned in a first modality and a second pair of images aligned in a second modality. Images in the first pair of images and the second pair of images are a computer aided design (CAD) image pre-aligned with a scanning electron microscope (SEM) image. An affine transformation is generated with a convolutional neural network, using one image from the first pair of images and one image from the second pair of images. The one image from the first pair of images is in the first modality and the one image from the second pair of images is in the second modality. Systems for registering images for masks or wafers in semiconductor manufacturing use the trained convolutional neural network.
1 . A system for training a convolutional neural network to register images for masks or wafers in semiconductor manufacturing, the system comprising:
a computer processor configured to receive a first pair of images aligned in a first modality and a second pair of images aligned in a second modality, wherein images in the first pair of images and the second pair of images are a computer aided design (CAD) image pre-aligned with a scanning electron microscope (SEM) image; and
a computer processor configured to generate an affine transformation with a convolutional neural network, using one image from the first pair of images and one image from the second pair of images, wherein the one image from the first pair of images is in the first modality and the one image from the second pair of images is in the second modality.
2 . The system of claim 1 , further comprising:
a computer processor configured to register the second pair of images with the affine transformation; and
a computer processor configured to verify the registering of the second pair of images against the first pair of images using normalized cross correlation (NCC) loss.
3 . The system of claim 2 , wherein the convolutional neural network comprises an encoder with a kernel comprising weights, and wherein the weights are adjusted based on the NCC loss.
4 . The system of claim 2 , wherein the first pair of images and the second pair of images each comprise a CAD image and a SEM image.
5 . The system of claim 4 , wherein:
the one image from the second pair of images is a CAD image;
the affine transformation is applied only on the CAD image of the second pair of images; and
the NCC loss is calculated with only the CAD image of the first pair of images and the CAD image of the second pair of images after affine transformation.
6 . The system of claim 4 , wherein the CAD image of the first pair of images and the second pair of images has been created using lithography simulation.
7 . The system of claim 6 , wherein the first pair of images and the second pair of images have been synthesized using a digital twin using affine transformation.
8 . The system of claim 1 , wherein the convolutional neural network further comprises a fully connected layer and an initialization of the affine transformation.
9 . The system of claim 1 , wherein the affine transformation consists of only scaling, translation and rotation.
10 . The system of claim 1 , wherein a pair of patches is taken from the first pair of images and the second pair of images; wherein the affine transformation is applied to the second pair of images.
11 . The system of claim 10 , wherein the pair of patches has a size that is smaller than the first pair of images and the second pair of images.