Inspection recipe optimization for semiconductor specimens

Patent metadata

Field	Value
Publication number	US-12480889-B2
Application number	US-202217845953-A
Country	US
Kind code	B2
Filing date	Jun 21, 2022
Priority date	Jun 21, 2022
Publication date	Nov 25, 2025
Grant date	Nov 25, 2025

Abstract

There is provided a system and method of optimizing an inspection recipe for inspecting a semiconductor specimen. The method includes obtaining test data from a test performed after inspection, the test data indicative of functional defectivity of the specimen with respect to at least one structural feature at a suspected layer; retrieving inspection data of the suspected layer including a set of inspection images and a set of defect maps of the plurality of processing steps of the suspected layer; correlating the test data and the set of defect maps of the suspected layer to identify one or more structural features of the suspected layer with unmatched defectivity; for each of the identified structural features, including at least part of the inspection images corresponding to the structural feature in a training set; and using the training set to train a machine learning (ML) model in the inspection recipe.

First claim

What is claimed is:

1 . A computerized system of optimizing an inspection recipe for inspecting a semiconductor specimen, the system comprising a processing and memory circuitry (PMC) configured to:
obtain test data from a test performed on the semiconductor specimen after inspection thereof, the semiconductor specimen comprising one or more layers, each layer comprising structural features manufactured by a plurality of processing steps, the test data indicative of functional defectivity of the semiconductor specimen associated with at least one structural feature located on a suspected layer of the one or more layers; retrieve inspection data of the suspected layer acquired during the inspection of the semiconductor specimen, the inspection data including a set of inspection images corresponding to at least a sampled set of the plurality of processing steps of the structural features of the suspected layer, and a set of defect maps corresponding to the set of inspection images and indicative of inspection defectivity with respect to the sampled set of the plurality of processing steps; correlate the test data and the set of defect maps of the suspected layer to identify one or more structural features of the suspected layer with unmatched defectivity between the functional defectivity indicated by the test data and the inspection defectivity indicated by the set of defect maps; prepare a training set, wherein, for each structural feature of the identified one or more of the structural features, at least part of the set of inspection images corresponding to the structural feature and ground truth data thereof, as indicated by the test data, are included in the training set; and use the training set to train a machine learning (ML) model in the inspection recipe usable for inspection of a subsequent semiconductor specimen.

2 . The computerized system according to claim 1 , wherein the test is one of: an electrical test, a destructive test, or a reliability test.

3 . The computerized system according to claim 1 , wherein the inspection is in line inspection performed during a fabrication process of the semiconductor specimen, and the test is an end-of-line (EOL) test performed upon completion of the fabrication process or part thereof.

4 . The computerized system according to claim 1 , wherein structural features in a layer of the one or more layers are of a type of: a contact, a via, a gate, a shallow trench isolation (STI), or a metal wire.

5 . The computerized system according to claim 1 , wherein the plurality of processing steps includes one or more of the following: lithography, etching, filling, depositing, polishing, recessing, planarization, growth, or implantation, and the sampled set includes a subset of the plurality of processing steps that are selected to be inspected during the inspection.

6 . The computerized system according to claim 1 , wherein the correlation is performed by aligning the test data with the set of defect maps, and comparing the functional defectivity and the inspection defectivity for corresponding locations.

7 . The computerized system according to claim 1 , wherein the one or more structural features with unmatched defectivity include a first structural feature indicated as defective by the test data, but not defective by the set of defect maps, wherein the PMC is further configured to analyze the set of inspection images of the suspected layer at a location of the first structural feature to identify one or more processing steps in the at least sampled set with high probability of causing defectivity of the first structural feature, and wherein the including comprises including, in the training set, at least part of one or more inspection images of the one or more processing steps corresponding to the first structural feature with an associated label of defect of interest (DOI).

8 . The computerized system according to claim 7 , wherein the including further comprises including design data of the one or more processing steps corresponding to the first structural feature, in addition to the at least part of the one or more inspection images, in the training set.

9 . The computerized system according to claim 1 , wherein the one or more structural features with unmatched defectivity include a second structural feature indicated as defective by at least one defect map in the set of defect maps, but not defective by the test data, wherein the including comprises including, in the training set, at least part of at least one inspection image corresponding to the at least one defect map with an associated label of nuisance, the at least part corresponding to the second structure feature.

10 . The computerized system according to claim 7 , wherein the one or more structural features with unmatched defectivity further include a second structural feature indicated as defective by at least one defect map in the set of defect maps, but not defective by the test data, wherein the including comprises including, in the training set, at least part of at least one inspection image corresponding to the at least one defect map with an associated label of nuisance, the at least part corresponding to the second structure feature.

11 . The computerized system according to claim 1 , wherein the ML model, upon being trained, is capable of improving capture rate of defects of interest (DOIs) which were previously missed by the inspection recipe, and reducing false alarm rate of nuisances which were previously incorrectly detected by the inspection recipe as DOIs.

12 . A computerized method of optimizing an inspection recipe for inspecting a semiconductor specimen, the method comprising:
obtaining test data from a test performed on the semiconductor specimen after inspection thereof, the semiconductor specimen comprising one or more layers, each layer comprising structural features manufactured by a plurality of processing steps, the test data indicative of functional defectivity of the semiconductor specimen associated with at least one structural feature located on a suspected layer of the one or more layers; retrieving inspection data of the suspected layer acquired during the inspection of the semiconductor specimen, the inspection data including a set of inspection images corresponding to at least a sampled set of the plurality of processing steps of the structural features of the suspected layer, and a set of defect maps corresponding to the set of inspection images and indicative of inspection defectivity with respect to the sampled set of the plurality of processing steps; correlating the test data and the set of defect maps of the suspected layer to identify one or more structural features of the suspected layer with unmatched defectivity between the functional defectivity indicated by the test data and the inspection defectivity indicated by the set of defect maps; preparing a training set, wherein, for each structural feature of the identified one or more of the structural features, at least part of the set of inspection images corresponding to the structural feature, and ground truth data thereof, as indicated by the test data, are included in the training set; and using the training set to train a machine learning (ML) model in the inspection recipe usable for inspection of a subsequent semiconductor specimen.

13 . The computerized method according to claim 12 , wherein the inspection is in-line inspection performed during a fabrication process of the semiconductor specimen, and the test is an end-of-line (EOL) test performed upon completion of the fabrication process or part thereof.

14 . The co

Assignees

• Applied Materials Israel Ltd

Inventors

• Yabbo Paz
• Dudovich Boaz
• Ghai Bhavna
• Bar Amir

Classifications

• H10P74/207

  Electrical properties, e.g. testing or measuring of resistance, deep levels or capacitance-voltage characteristics · CPC title

• H10P74/23

  characterised by multiple measurements, corrections, marking or sorting processes · CPC title

• H10P74/203

  Structural properties, e.g. testing or measuring thicknesses, line widths, warpage, bond strengths or physical defects · CPC title

• G06T7/0004 Primary

  Industrial image inspection · CPC title

• G06T2207/30148

  Semiconductor; IC; Wafer · CPC title