# US 12,111,355 B2 · Semiconductor substrate yield prediction based on spectra data from multiple substrate dies > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,111,355 B2 - Document ID: US12111355B2 - Status: Granted - Kind: B2 - Application number: 17/532,700 - Filed: 2021-11-22 - Published: 2023-05-25 - Granted: 2024-10-08 - Earliest claimed: 2021-11-22 - Assignees: Onto Innovation Inc. - Applicants: Onto Innovation Inc. - Inventors: Xin Song; Jian Lu - Web page: https://patentgrep.com/patent/US12111355B2 ## CPC - G01R 31/318511: {Wafer Test} - H10P 74/23: {characterised by multiple measurements, corrections, marking or sorting processes} - G01R 31/2642: {Testing semiconductor operation lifetime or reliability, e.g. by accelerated life tests} - G01R 31/2831: {Testing of materials or semi-finished products, e.g. semiconductor wafers or substrates (G01R31/318511 takes precedence; testing during manufacture H10P74/00)} - G01R 31/2886: {Features relating to contacting the IC under test, e.g. probe heads; chucks (G01R31/2865 takes precedence, test connections, e.g. test sockets, or probes per se, G01R1/04 or G01R1/06)} - H10P 74/207: {Electrical properties, e.g. testing or measuring of resistance, deep levels or capacitance-voltage characteristics} - G01R 1/07342: {the body of the probe being at an angle other than perpendicular to test object, e.g. probe card} - G01R 31/2894: {Aspects of quality control [QC] (G01R31/31718 takes precedence; program control for QC G05B19/41875)} - G01R 31/308: using non-ionising electromagnetic radiation, e.g. optical radiation {(investigating or analysing materials by the use of optical means G01N21/00; image analysis G06T7/00)} - G06F 30/00: Computer-aided design [CAD] - H10P 74/277: {Circuits for electrically characterising or monitoring manufacturing processes, e.g. circuits in tested chips or circuits in testing wafers} - G01R: MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES (indicating correct tuning of resonant circuits H03J3/12) - H10P: GENERIC PROCESSES OR APPARATUS FOR THE MANUFACTURE OR TREATMENT OF DEVICES COVERED BY CLASS H10 - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract Systems and methods for improving substrate fabrication are provided. Subsets of dies of substrates may be inspected at various points in the fabrication process to generate spectra data. The spectra data can be used to generate data that are input to a machine learning model to predict yields for the substrates. ## Claims ### Claim 1 A method for predicting yield of a substrate, the method comprising: combining first spectra data for a first die of a plurality of dies on the substrate and second spectra data for the first die of the plurality of dies on the substrate into combined data, the first spectra data for the first die being generated from a first inspection of only the first die of the plurality of dies at a first time, the second spectra data for the first die being generated from a second inspection of only the first die of the plurality of dies at a second time, the second time being different from the first time; and generating, based on the combined data, a predicted yield for the substrate, including providing the combined data as input to a trained machine learning model. ### Claim 2 (depends on 1) The method of claim 1, wherein the machine learning model is a convolutional neural network. ### Claim 3 (depends on 1) The method of claim 1, wherein the combined data includes a first subset of data that represents the first spectra data for the first die of the plurality of dies. ### Claim 4 (depends on 3) The method of claim 3, wherein the combined data is organized so that the first subset of data is differentiated from a second subset of data that represents second spectra data for a second die of the plurality of dies. ### Claim 5 (depends on 4) The method of claim 4, wherein each data value of the first subset of data is associated with a wavelength of the first spectra data. ### Claim 6 (depends on 5) The method of claim 5, wherein a data value in the first subset of data represents a signal strength of the first spectra data, at the wavelength corresponding to the data value, for the first die of the plurality of dies. ### Claim 7 (depends on 1) The method of claim 1, wherein the first time occurs after a first manufacturing step is performed on the substrate and before a second manufacturing step is performed on the substrate; and wherein the second time occurs after the second manufacturing step is performed on the substrate. ### Claim 8 (depends on 1) The method of claim 1, further comprising: augmenting the combined data by adding third spectra data for the first die of the plurality of dies into the combined data to form augmented combined data, the third spectra data being generated from a third inspection of the first die of the plurality of dies; and generating another predicted yield for the substrate, including providing the augmented combined data as input to the trained machine learning model, wherein the first time occurs after a first manufacturing step is performed on the substrate and before a second manufacturing step is performed on the substrate; wherein the second time occurs after the second manufacturing step is performed on the substrate and before a third manufacturing step is performed on the substrate; and wherein the third time occurs after the third manufacturing step is performed on the substrate. ### Claim 9 (depends on 1) The method of claim 1, wherein each of the first spectra data and the second spectra data includes additional spectra data for each of the plurality of dies, and wherein the plurality of dies are less than all dies on the substrate. ### Claim 10 (depends on 1) The method of claim 1, wherein the combined data is augmented combined data; wherein the predicted yield is a modified predicted yield, the method further comprising, before the second inspection: generating non-augmented combined data including the first spectra data; and providing the non-augmented combined data as input to the trained machine learning model to generate a first predicted yield for the substrate, the method further comprising, after the second inspection: modifying the first predicted yield to generate the modified predicted yield. ### Claim 11 (depends on 1) The method of claim 1, further comprising: generating a remediation recommendation based on the predicted yield. ### Claim 12 (depends on 11) The method of claim 11, wherein the remediation recommendation includes a recommendation to adjust a substrate manufacturing tool. ### Claim 13 (depends on 1) The method of claim 1, further comprising, prior to the combining, generating the first spectra data and the second spectra data. ### Claim 14 (depends on 9) The method of claim 9, wherein the plurality of dies is in a range from 1/1000to 1/50,000of a total number of dies of the substrate. ### Claim 15 (depends on 6) The method of claim 6, further comprising training a machine learning model to provide the trained machine learning model, including inputting training matrices to the machine learning model, each of the training matrices including a first dimension corresponding to different training dies, a second dimension corresponding to different spectra wavelengths, and a third dimension corresponding to different inspection channels, the training matrices including, for each of the different channels, values of reflected signal strengths of the different training dies at the different spectra wavelengths. ### Claim 16 (depends on 8) The method of claim 8, wherein the augmenting includes adding one or more rows of a data corresponding to the third spectra data to a matrix. ### Claim 17 A non-transient computer-readable medium containing instructions for causing a computing device to: combine first spectra data for a first die of a plurality of dies of a substrate and second spectra data for the first die of the plurality of dies on the substrate into combined data, the first spectra data for the first die being generated from a first inspection of only the first die of the plurality of dies at a first time, the second spectra data being generated from a second inspection of only the first die of the plurality of dies at a second time, the second time being different from the first time; and generate, based on the combined data, a predicted yield for the substrate, including to provide the combined data as input to a trained machine learning model. ### Claim 18 (depends on 17) The non-transient computer-readable medium of claim 17, wherein the machine learning model is a convolutional neural network. ### Claim 19 (depends on 17) The non-transient computer-readable medium of claim 17, wherein the combined data includes a first subset of data that represents the first spectra data for the first die of the plurality of dies; wherein the combined data is organized so that the first subset of data is differentiated from a second subset of data that represents second spectra data for a second die of the plurality of dies; wherein each data value of the first subset of data is associated with a wavelength of the first spectra data; and wherein a data value in the first subset of data represents a signal strength of the first spectra data, at the wavelength corresponding to the data value, for the first die of the plurality of dies. ### Claim 20 (depends on 17) The non-transient computer-readable medium of claim 17, wherein the first time occurs after a first manufacturing step is performed on the substrate and before a second manufacturing step is performed on the substrate; and wherein the second time occurs after the second manufacturing step is performed on the substrate. ### Claim 21 (depends on 17) The non-transient computer-readable medium of claim 17, containing further instructions configured for causing the computing device to: augment the combined data by adding third spectra data for the first die of the plurality of dies into the combined data to form augmented combined data, the third spectra data being generated from a third inspection of the first die of the plurality of the dies; and generate another predicted yield for the substrate, including to provide the augmented combined data as input to the trained machine learning model, wherein the first time occurs after a first manufacturing step is performed on the substrate and before a second manufacturing step is performed on the substrate; wherein the second time occurs after the second manufacturing step is performed on the substrate and before a third manufacturing step is performed on the substrate; and wherein the third time occurs after the third manufacturing step is performed on the substrate. ### Claim 22 (depends on 17) The non-transient computer-readable medium of claim 17, wherein each of the first spectra data and the second spectra data includes additional spectra data for each of the plurality of dies, and wherein the plurality of dies are less than all dies on the substrate. ### Claim 23 (depends on 17) The non-transient computer-readable medium of claim 17, wherein the combined data is augmented combined data; wherein the predicted yield is a modified predicted yield, the non-transient computer-readable medium containing further instructions configured for causing the computing device to, before the second inspection: generate non-augmented combined data including the first spectra data; and provide the non-augmented combined data as input to the trained machine learning model to generate a first predicted yield for the substrate, the non-transient computer-readable medium containing further instructions configured for causing the computing device to, after the second inspection; modify the first predicted yield to generate the modified predicted yield. ### Claim 24 (depends on 17) The non-transient computer-readable medium of claim 17, containing further instructions configured for causing the computing device to: generate a remediation recommendation based on the predicted yield. ### Claim 25 (depends on 24) The non-transient computer-readable medium of claim 24, wherein the remediation recommendation includes a recommendation to adjust a substrate manufacturing tool. ### Claim 26 (depends on 22) The non-transient computer-readable medium of claim 22, wherein the plurality of dies is in a range from 1/1000to 1/50,000of a total number of dies of the substrate. ### Claim 27 (depends on 19) The non-transient computer-readable medium of claim 19, containing further instructions configured for causing the computing device to: train a convolutional neural network machine learning model to provide the trained machine learning model, including to input training matrices to the machine learning model, each of the training matrices including a first dimension corresponding to different training dies, a second dimension corresponding to different spectra wavelengths, and a third dimension corresponding to different inspection channels, the training matrices including, for each of the different channels, values of reflected signal strengths of the different training dies at the different spectra wavelengths. ### Claim 28 (depends on 21) The non-transient computer-readable medium of claim 21, containing further instructions configured for causing the computing device to add one or more rows of data corresponding to the third spectra data to a matrix.