公开(公告)号：US20190121317A1

公开(公告)日：2019-04-25

申请号：US15792436

申请日：2017-10-24

Applicant: Applied Materials, Inc.

Inventor： Heng HAO ,  James Tom PYE ,  Sreekar BHAVIRIPUDI

IPC: G05B19/4063 ,  G05B19/18

Abstract: A method for detecting an anomaly in sensor data generated in a substrate processing apparatus is disclosed herein. A plurality of data sets is received. A first data set from a first sensor and second data set from a second sensor are selected. The first second sensors are defined as a sensor pair. A reference correlation is generated by selecting a subset of values in each data set for each of the first and second data sets. A difference of remaining data correlation outside the subset of values in each data set to the reference correlation is normalized. The normalized data set is filtered to smooth the normalized difference to avoid isolated outliers with high chance of false positive candidates. One or more anomalies are identified. Process parameters of the substrate processing apparatus are adjusted, based on the one or more identified anomalies from the filtered data set.

公开(公告)号：US20200082245A1

公开(公告)日：2020-03-12

申请号：US16545908

申请日：2019-08-20

Applicant: Applied Materials, Inc.

Inventor： Heng HAO ,  Sreekar BHAVIRIPUDI ,  Shreekant GAYAKA

IPC: G06N3/04 ,  G06N3/08 ,  G05B19/418

Abstract: Implementations described herein generally relate to a method for detecting anomalies in time-series traces received from sensors of manufacturing tools. A server feeds a set of training time-series traces to a neural network configured to derive a model of the training time-series traces that minimizes reconstruction error of the training time-series traces. The server extracts a set of input time-series traces from one or more sensors associated with one or more manufacturing tools configured to produce a silicon substrate. The server feeds the set of input time-series traces to the trained neural network to produce a set of output time series traces reconstructed based on the model. The server calculates a mean square error between a first input time series trace of the set of input time series traces and a corresponding first output time series trace of the set of output time-series traces. The server declares the sensor corresponding to the first input time-series trace as having an anomaly when the mean square error exceeds a pre-determined value.

公开(公告)号：US20200243359A1

公开(公告)日：2020-07-30

申请号：US16261041

申请日：2019-01-29

Applicant: Applied Materials, Inc.

Inventor： Heng HAO ,  Tianqing LIAO ,  Sima DIDARI ,  Harikrishnan RAJAGOPAL

IPC: H01L21/67 ,  H01J37/32

Abstract: A server trains a neural network by feeding a first set of input time-series data of one or more sensors of a first processing chamber that is within specification to the neural network to produce a corresponding first set of output time-series data. The server calculates a first error. The server feeds a second set of input time-series data from corresponding one or more sensors associated with a second processing chamber under test to the trained neural network to produce a corresponding second set of output time-series data. The server calculates a second error. Responsive to the difference between a second error between the second set of input time-series data and the corresponding second set of output time-series data and a first error between the first set of input time-series data and the corresponding first set of output time-series data being equal to or exceeding a threshold amount, the server declares that the second processing chamber under test mismatches the first processing chamber that is within specifications.

公开(公告)号：US20230153574A1

公开(公告)日：2023-05-18

申请号：US18151156

申请日：2023-01-06

Applicant: Applied Materials, Inc.

Inventor： Heng HAO ,  Sreekar BHAVIRIPUDI ,  Shreekant GAYAKA

IPC: G06N3/04 ,  G06N3/08 ,  G05B19/418 ,  G06V10/82

CPC classification number: G06N3/04 ,  G06N3/08 ,  G05B19/4189 ,  G06V10/82 ,  G05B2219/32335 ,  G05B2219/45031

Abstract: Implementations described herein generally relate to a method for detecting anomalies in time-series traces received from sensors of manufacturing tools. A server feeds a set of training time-series traces to a neural network configured to derive a model of the training time-series traces that minimizes reconstruction error of the training time-series traces. The server extracts a set of input time-series traces from one or more sensors associated with one or more manufacturing tools configured to produce a silicon substrate. The server feeds the set of input time-series traces to the trained neural network to produce a set of output time series traces reconstructed based on the model. The server calculates a mean square error between a first input time series trace of the set of input time series traces and a corresponding first output time series trace of the set of output time-series traces. The server declares the sensor corresponding to the first input time-series trace as having an anomaly when the mean square error exceeds a pre-determined value.

公开(公告)号：US20170343999A1

公开(公告)日：2017-11-30

申请号：US15610280

申请日：2017-05-31

Applicant: Applied Materials, Inc.

Inventor： Jimmy ISKANDAR ,  Michael D. ARMACOST ,  Heng HAO

IPC: G05B23/02 ,  H01L21/67

CPC classification number: G05B23/0291 ,  G05B23/0243

Abstract: Embodiments provide techniques for compressing sensor data collected within a manufacturing environment. One embodiment monitors a plurality of runs of a recipe for fabricating one or more semiconductor devices within a manufacturing environment to collect runtime data from a plurality of sensors within the manufacturing environment. The collected runtime data is compressed by generating, for each of the plurality of sensors and for each of the plurality of runs, a respective representation of the corresponding runtime data that describes a shape of the corresponding runtime data and a magnitude of the corresponding runtime data. A query specifying one or more runtime data attributes is received and executed against the compressed runtime data to generate query results, by comparing the one or more runtime data attributes to at least one of the generated representations of runtime data.