Implementation of deep neural networks for testing and quality control in the production of memory devices
Techniques are presented for the application of neural networks to the fabrication of integrated circuits and electronic devices, where example are given for the fabrication of non-volatile memory circuits and the mounting of circuit components on the printed circuit board of a solid state drive (SSD). The techniques include the generation of high precision masks suitable for analyzing electron microscope images of feature of integrated circuits and of handling the training of the neural network when the available training data set is sparse through use of a generative adversary network (GAN).
1. A method, comprising:
preparing a first plurality of samples of an integrated circuit for imaging;
generating electron microscope images of the first plurality of prepared samples of the integrated circuit;
training a neural network to generate pixel level masks of features of the first plurality of the samples of the integrated circuit from the electron microscope images, comprising:
generating pixel level feature maps of the first plurality of the samples of the integrated circuit from the electron microscope images;
generating bounding box predictions from the pixel level feature maps;
introducing added error into the bounding box predictions;
locating regions of interest within the pixel level feature maps; and
generating the pixel level masks of features of the first plurality of the samples of the integrated circuits for the located regions of interest from the bounding box predictions with the introduced added error;
applying the trained neural network to one or more additional samples of the integrated circuit to generate pixel level masks of features of the one or more additional samples of the integrated circuit from the bounding box predictions with the introduced added error;
analyzing results of the application of the trained neural network to the one or more additional samples of the integrated circuit;
adjusting processing for fabrication of the integrated circuit based upon the analyzing of the results of the application of the trained neural network to the one or more additional samples of the integrated circuit; and
fabricating the integrated circuit by the adjusted processing.
2. The method of claim 1 , wherein preparing the first plurality of samples of an integrated circuit for imaging includes:
fabricating a plurality of integrated circuits each having a plurality of layers formed upon a substrate and a plurality of holes extending through one or more of the layers, wherein the features include uniformity of the holes.
3. The method of claim 1 , wherein preparing a first plurality of samples of an integrated circuit for imaging includes:
paring down the fabricated plurality of the integrated circuits to a desired level.
4. The method of claim 1 , wherein the integrated circuit is formed to be one of a bonded die pair and the features include contacts for exchanging electrical signals between dies of the bonded die pair.
5. The method of claim 1 , wherein preparing a first plurality of samples of an integrated circuit for imaging includes:
forming a photoresist on a plurality of integrated circuits, wherein the features include a dimension of the photoresist.
6. The method of claim 1 , wherein preparing a first plurality of samples of an integrated circuit for imaging includes:
fabricating a plurality of integrated circuits each having a plurality of layers formed upon a substrate and a plurality of holes extending through one or more of the layers, wherein the features include the shape of the holes.
7. The method of claim 1 , wherein preparing a first plurality of samples of an integrated circuit for imaging includes:
fabricating a plurality of integrated circuits each having a plurality transistors formed thereon, wherein the features include uniformity of a gate of the transistors.
8. The method of claim 1 , wherein applying the trained neural network to one or more additional samples of the integrated circuit is performed as part of a failure analysis process.
9. The method of claim 1 , further comprising:
subsequent to adjusting processing for fabrication of the integrated circuit, applying the trained neural network to one or more samples of the integrated circuit fabricated according to the adjusted processing to generate pixel level masks of features of the one or more samples of the integrated circuit fabricated according to the adjusted processing; and
analyzing results of the application of the trained neural network to the one or more additional samples of the integrated circuit fabricated according to the adjusted processing.
10. The method of claim 9 , further comprising:
further adjusting processing for fabrication of the integrated circuit based upon the analyzing of the results of the application of the trained neural network to the one or more additional samples of the integrated circuit fabricated according to the adjusted processing.
11. The method of claim 1 , wherein preparing a first plurality of samples of an integrated circuit for imaging includes:
augmenting the samples of the integrated circuits with images generated by a generative adversary network.
12. The method of claim 11 , wherein augmenting the samples of the integrated circuits is performed in response to determining that the first plurality of samples of the integrated circuit are a sparse set of sample.
13. A system, comprising:
one or more processors, the one or more processors configured to:
receive electron microscope images of a first plurality of prepared samples of an integrated circuit manufactured in a fabrication facility according to a first set of processing parameters;
train a neural network to generate pixel level masks of features of the first plurality of the samples of the integrated circuit from the electron microscope images, where, to train the neural network, the one or more processors are configured to:
generate pixel level feature maps of the first plurality of the samples of the integrated circuit from the electron microscope images;
generate bounding box predictions from the pixel level feature maps;
introduce added error into the bounding box predictions;
locate regions of interest within the pixel level feature maps; and
generate the pixel level masks of features of the first plurality of the samples of the integrated circuits for the located regions of interest from the bounding box predictions with the introduced added error;
apply the trained neural network to one or more additional samples of the integrated circuit to generate pixel level masks of features of the one or more additional samples of the integrated circuit from the bounding box predictions with the introduced added error;
analyze results of the application of the trained neural network to the one or more additional samples of the integrated circuit;
adjust values for the first set of processing parameters for fabrication of the integrated circuit based upon the analyzing of the results of the application of the trained neural network to the one or more additional samples of the integrated circuit; and
provide the adjusted values for the first set of process parameters to the fabrication facility for fabricating the integrated circuit according to the adjusted values for the first set of process parameters.
14. The system of claim 13 , wherein the first plurality of prepared samples of the integrated circuit includes a plurality of integrated circuits each having a plurality of layers formed upon a substrate and a plurality of holes extending through one or more of the layers, wherein the features include uniformity of the holes.
15. The system of claim 13 , wherein the integrated circuit is formed to be one of a bonded die pair and the features include contacts for exchanging electrical signals between dies of the bonded die pair.
16. The system of claim 13 , wherein the prepared samples of an integrated circuit are prepared by paring down the fabricated plurality of the integrated circuits to a desired level.
17. The system of claim 13 , wherein the first plurality of prepared samples of the integrated circuit includes a plurality of integrated circuits include a photoresist formed on a plurality of integrated circuits, wherein the features include a dimension of the photoresist.
18. The system of claim 13 , wherein the first plurality of prepared samples of the integrated circuit includes a plurality of integrated circuits prepared by fabricating the plurality of integrated circuits such that each has a plurality of layers formed upon a substrate and a plurality of holes extending through one or more of the layers, wherein the features include the shape of the holes.
19. The system of claim 13 , wherein the prepared samples of an integrated circuit are prepared by fabricating a plurality of integrated circuits each having a plurality transistors formed thereon, and wherein the features include uniformity of a gate of the transistors.
20. The system of claim 13 , wherein applying the trained neural network to one or more additional samples of the integrated circuit is performed as part of a failure analysis process.