Building analytical platform to enable device fabrication

Patent metadata

Field	Value
Publication number	US-11734580-B2
Application number	US-202117323760-A
Country	US
Kind code	B2
Filing date	May 18, 2021
Priority date	May 19, 2020
Publication date	Aug 22, 2023
Grant date	Aug 22, 2023

Abstract

This disclosure relates generally to methods and systems for building an intelligent analytical platform to enable a device fabrication in material science. Material engineers and design engineers may face various challenges with existing knowledge, as more time and efforts are required in finding a relevant knowledge from the existing knowledge, mainly due to the unstructured form, for fabricating new devices. The present disclosure solves the technical problem of finding the relevant knowledge out of the existing knowledge, in a structured form by building an analytical platform. The unstructured format of the existing knowledge of the fabrication process is transformed into a structured format in terms of operation sequence knowledge graphs, using a set of artificial intelligence (AI) and machine learning models, and a knowledge representation model of the fabrication process. The structured format of the existing knowledge is hierarchically arranged to build the analytical platform.

First claim

What is claimed is:

1. A processor-implemented method comprising the steps of:
receiving, via one or more hardware processors, (i) a device category associated with a device fabrication, (ii) one or more device fabrication knowledge documents associated with the device category from a device fabrication knowledge repository, wherein each device fabrication knowledge document comprises a plurality of document paragraphs, wherein each document paragraph comprises one or more paragraph sentences, and each paragraph sentence comprises a plurality of paragraph sentence words;
pre-processing, via the one or more hardware processors, each device fabrication knowledge document of the one or more device fabrication knowledge documents, to obtain: (i) a plurality of pre-processed document paragraphs, in a plain text format, (ii) a section header for each pre-processed document paragraph of the plurality of pre-processed document paragraphs, wherein each pre-processed document paragraph comprises a plurality of pre-processed paragraph sentences, and wherein each pre-processed paragraph sentence comprises a plurality of pre-processed paragraph sentence words;
identifying, via the one or more hardware processors, one or more fabrication procedure paragraphs out of the plurality of pre-processed document paragraphs, by a trained fabrication procedure paragraph classification model, using the section header for each pre-processed document paragraph of the plurality of pre-processed document paragraphs, wherein each pre-processed document paragraph of the plurality of pre-processed document paragraphs is a fabrication procedure paragraph, if the pre-processed document paragraph comprises information related to a fabrication procedure;
identifying, via the one or more hardware processors, one or more entities, for each pre-processed paragraph sentence of the plurality of pre-processed paragraph sentences associated with each fabrication procedure paragraph of the one or more fabrication procedure paragraphs, by a trained named entity identification model, wherein the plurality of entities are associated with a plurality of predefined concepts related to the fabrication procedure;
identifying, via the one or more hardware processors, (i) one or more first predefined relations out of a first set of predefined relations, and (ii) one or more second predefined relations out of a second set of predefined relations, for each pre-processed paragraph sentence of the plurality of pre-processed paragraph sentences associated with each fabrication procedure paragraph of the one or more fabrication procedure paragraphs, by (i) a trained relation identification model, and (ii) a set of predefined pattern expressions, respectively, using the one or more entities identified for each pre-processed paragraph sentence;
identifying, via the one or more hardware processors, a device fabrication procedure for each fabrication procedure paragraph of the one or more fabrication procedure paragraphs, by a rule based unsupervised algorithm, using the one or more entities identified for each pre-processed paragraph sentence of the plurality of pre-processed paragraph sentences associated with each fabrication procedure paragraph, wherein the device fabrication procedure for each fabrication procedure paragraph, comprises a sequence of operations; and
generating, via the one or more hardware processors, an operation sequence knowledge graph for each fabrication procedure paragraph of the one or more fabrication procedure paragraphs, using (i) the one or more entities identified for each pre-processed paragraph sentence of the plurality of pre-processed paragraph sentences associated with each fabrication procedure paragraph of the one or more fabrication procedure paragraphs, (ii) (a) the one or more first predefined relations out of the first set of predefined relations, and (b) the one or more second predefined relations out of the second set of predefined relations, identified for each pre-processed paragraph sentence of the plurality of pre-processed paragraph sentences associated with each fabrication procedure paragraph of the one or more fabrication procedure paragraphs, and (iii) the device fabrication procedure for each fabrication procedure paragraph of the one or more fabrication procedure paragraphs, using a graph database tool.

2. The method of claim 1 , further comprising the step of building, via the one or more hardware processors, an analytical platform to enable the device fabrication, using the operation sequence knowledge graph for each fabrication procedure paragraph of the one or more fabrication procedure paragraphs, in a graph search engine.

3. The method of claim 1 , wherein the trained fabrication procedure paragraph classification model is obtained by:
receiving (i) a plurality of training document paragraphs associated with the device category, in the plain text format, (ii) the section header for each training document paragraph of the plurality of training document paragraphs, and (iii) an annotation class for each training document paragraph of the plurality of training document paragraphs, wherein each training document paragraph comprises a plurality of training document paragraph sentences, and each training document paragraph sentence comprises a plurality of training document paragraph sentence words, and wherein the annotation class for each training document paragraph is one of: (a) the fabrication procedure paragraph, and (b) a non-fabrication procedure paragraph;
obtaining: (a) a paragraph text vector, (b) a dictionary feature vector, and (c) a section header feature vector, for each training document paragraph of the plurality of training document paragraphs, wherein:
(a) the paragraph text vector for each training document paragraph comprises an embedding for each training document paragraph sentence word of the plurality of training document paragraph sentence words corresponding to each training document paragraph sentence of the plurality of training document paragraph sentences corresponding to the training document paragraph;
(b) the dictionary feature vector for each training document paragraph is obtained by: (i) defining an initial dictionary feature vector with a plurality of predefined keywords, and (ii) assigning a Boolean value for each predefined keyword of the plurality of predefined keywords defined in the initial dictionary feature vector, based on presence of the predefined keyword in the training document paragraph; and
(c) the section header feature vector for each training document paragraph of the plurality of training document paragraphs, is obtained by: (i) defining an initial section header feature vector with a plurality of predefined section headers, and (ii) assigning the Boolean value for each predefined section header of the plurality of predefined section headers defined in the initial section header feature vector, based on matching of the predefined section header with the section header of the training document paragraph; and
training a first bi-directional long short term memory (BiLSTM) network with: (i) (a) the paragraph text vector, (b) the dictionary feature vector, and (c) the section header feature vector, for each training document paragraph, at a time, of the plurality of training document paragraphs, and (ii) the annotation class for each training document paragraph of the plurality of training document paragraphs, to obtain the trained fabrication procedure paragraph classification model, wherein training the first BiLSTM network with each training document paragraph comprises:
passing the paragraph text vector corresponding to the training document paragraph, as an input to a bi-directional long short term memory (BiLSTM) layer of the first BiLSTM network, to learn a hidden state of a first training document paragraph senten

Assignees

• Tata Consultancy Services Ltd

Inventors

• Jain Deepak
• Shah Sapankumar Hiteshchandra
• Rai Beena
• Mitra Pritwish
• Reddy Sreedhar
• Wadhwa Neelanshi
• Sasidharan Sarath

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N5/022 Primary

  Knowledge engineering; Knowledge acquisition · CPC title

• G06F16/9024

  Graphs; Linked lists (G06F16/9027 takes precedence) · CPC title

• G06F16/93

  Document management systems · CPC title