# US 12,001,973 B2 · Machine learning-based adjustments in volume diagnosis procedures for determination of root cause distributions

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,001,973 B2
- Document ID: US12001973B2
- Status: Granted
- Kind: B2
- Application number: 16/361,915
- Filed: 2019-03-22
- Published: 2020-09-24
- Granted: 2024-06-04
- Earliest claimed: 2019-03-22
- Assignees: Siemens Industry Software Inc.
- Applicants: Siemens Industry Software Inc.
- Inventors: Gaurav Veda; Wu-Tung Cheng; Manish Sharma; Huaxing Tang; Yue Tian
- Web page: https://patentgrep.com/patent/US12001973B2

## CPC

- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G01R 31/2894: {Aspects of quality control [QC] (G01R31/31718 takes precedence; program control for QC G05B19/41875)}
- G01R 31/318566: {Comparators; Diagnosing the device under test}
- G06F 30/367: Design verification, e.g. using simulation, simulation program with integrated circuit emphasis [SPICE], direct methods or relaxation methods
- G06N 20/00: Machine learning
- G06F 30/27: using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
- G06F 30/333: Design for testability [DFT], e.g. scan chain or built-in self-test [BIST]
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G01R: MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES (indicating correct tuning of resonant circuits H03J3/12)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A computing system may include a model training engine configured to train a supervised learning model with a training set comprising training probability distributions computed for training dies through a local phase of a volume diagnosis procedure. The computing system may also include a volume diagnosis adjustment engine configured to access a diagnosis report for a given circuit die that has failed scan testing and compute, through the local phase of the volume diagnosis procedure, a probability distribution for the given circuit die from the diagnosis report. The volume diagnosis adjustment engine may also adjust the probability distribution into an adjusted probability distribution using the supervised learning model and provide the adjusted probability distribution for the given circuit die as an input to a global phase of the volume diagnosis procedure to determine a global root cause distribution for multiple circuit dies that have failed the scan testing.

## Claims

### Claim 1

A method comprising:
by a computing system:
accessing a diagnosis report for a given circuit die that has failed scan testing;
computing, through a local phase of a volume diagnosis procedure, a probability distribution for the given circuit die from the diagnosis report, wherein the probability distribution specifies probabilities for different suspect root causes as having caused the given circuit die to fail;
adjusting the probability distribution into an adjusted probability distribution using a supervised learning model, including by, through the supervised learning model, changing at least one of the probabilities for the different suspect root causes specified in the probability distribution as having caused the given circuit die to fail through the supervised learning model, the supervised learning model trained with a training set comprising training probability distributions of suspect root causes computed from training dies through the local phase of the volume diagnosis procedure, the training dies injected with particular root causes that cause the training dies to fail, and each training probability distribution of suspect root causes of the training set labeled with an actual root cause, wherein the actual root cause specifies the particular root cause injected into a given training die that caused the given training die to fail;
providing the adjusted probability distribution for the given circuit die as an input to a global phase of the volume diagnosis procedure to determine a global root cause distribution for multiple circuit dies that have failed the scan testing;
performing the global phase of the volume diagnosis procedure using, as inputs, a plurality of adjusted probability distributions from the multiple circuit dies that have failed the scan testing, including the adjusted probability distribution for the given circuit die adjusted through the supervised learning model; and:
generating the supervised learning model, including by
accessing the training dies, wherein each training die has been injected with a respective specific root cause to actually cause a scan test failure;
generating diagnosis reports for each of the training dies;
computing, through the local phase of a volume diagnosis procedure, the training probability distributions of suspect root causes from the diagnosis reports generated for the training dies, each of the training probability distributions of suspect root causes respectively corresponding to one of the training dies;
labeling each of the training probability distributions of suspect root causes with the respective specific root cause for the training die corresponding to the training probability distribution of suspect root causes, wherein the respective specific root cause specifies the actual root cause injected into the training die to cause the training die to fail; and
providing, as the training set, the labeled training probability distributions to train the supervised learning model.

### Claim 2 (depends on 1)

The method of claim 1, wherein the volume diagnosis procedure utilizes an unsupervised learning model to compute the probability distribution for the given circuit die, determine the global root cause distribution for multiple circuit dies, or both.

### Claim 3 (depends on 1)

The method of claim 1, wherein the global phase of the volume diagnosis procedure is performed using a root cause deconvolution (RCD) model.

### Claim 4 (depends on 1)

The method of claim 1, wherein the training dies are generated via simulation, emulation, or a combination of both.

### Claim 5 (depends on 1)

The method of claim 1, wherein the supervised learning model comprises a linear function that linearly adjusts the probability distribution computed for the given circuit die.

### Claim 6 (depends on 5)

The method of claim 5, wherein the linear function comprises an adjustment matrix that linearly adjusts an input probability distribution; and
wherein the adjustment matrix has dimensions of ‘N’בN’, wherein ‘N’ is a number of different root causes in probability distributions computed by the local phase of a volume diagnosis procedure.

### Claim 7

A system comprising:
a model training engine, comprising hardware and programming, configured to train a supervised learning model with a training set comprising training probability distributions of suspect root causes computed for training dies through a local phase of a volume diagnosis procedure, the training dies intentionally injected with root causes that cause the training dies to fail, wherein:
each given training probability distribution of suspect root causes specifies probabilities for different suspect root causes as having caused a given training die to fail as computed by the volume diagnosis procedure; and
each given training probability distribution of suspect root causes is labeled with an actual root cause wherein the actual root cause specifies a particular root cause injected into the given training die that caused the given training die to fail, and
wherein the model training engine is configured to train the supervised learning model by:
accessing the training dies, wherein each training die has been injected with a respective specific root cause to actually cause a scan test failure;
generating diagnosis reports for each of the training dies;
computing, through the local phase of a volume diagnosis procedure, the training probability distributions of suspect root causes from the diagnosis reports generated for the training dies, each of the training probability distributions of suspect root causes respectively corresponding to one of the training dies;
labeling each of the training probability distributions of suspect root causes with the respective specific root cause for the training die corresponding to the training probability distribution of suspect root causes, wherein the respective specific root cause specifies the actual root cause injected into the training die to cause the training die to fail; and
providing, as the training set, the labeled training probability distributions to train the supervised learning model; and
a volume diagnosis adjustment engine, comprising hardware and programming, configured to:
access a diagnosis report for a given circuit die that has failed scan testing;
compute, through the local phase of the volume diagnosis procedure, a probability distribution for the given circuit die from the diagnosis report, wherein the probability distribution specifies probabilities for different suspect root causes as having caused the given circuit die to fail;
adjust the probability distribution into an adjusted probability distribution using the supervised learning model including by, through the supervised learning model, changing at least one of the probabilities for the different suspect root causes specified in the probability distribution as having caused the given circuit die;
provide the adjusted probability distribution for the given circuit die as an input to a global phase of the volume diagnosis procedure to determine a global root cause distribution for multiple circuit dies that have failed the scan testing; and
perform the global phase of the volume diagnosis procedure using, as inputs, a plurality of adjusted probability distributions from the multiple circuit dies that have failed the scan testing, including the adjusted probability distribution for the given circuit die adjusted through the supervised learning model.

### Claim 8 (depends on 7)

The system of claim 7, wherein the volume diagnosis procedure utilizes an unsupervised learning model to compute the probability distribution for the given circuit die, determine the global root cause distribution for multiple circuit dies, or both.

### Claim 9 (depends on 7)

The system of claim 7, wherein the global phase of the volume diagnosis procedure is performed using a root cause deconvolution (RCD) model.

### Claim 10 (depends on 7)

The system of claim 7, wherein the training dies are generated via simulation, emulation, or a combination of both.

### Claim 11 (depends on 7)

The system of claim 7, wherein the supervised learning model comprises a linear function that linearly adjusts the probability distribution computed for the given circuit die.

### Claim 12 (depends on 11)

The system of claim 11, wherein the linear function comprises an adjustment matrix that linearly adjusts an input probability distribution; and
wherein the adjustment matrix has dimensions of ‘N’בN’, wherein ‘N’ is a number of different root causes in probability distributions computed by the local phase of a volume diagnosis procedure.

### Claim 13

A non-transitory machine-readable medium comprising instructions that, when executed by a processor, cause a computing system to:
access a diagnosis report for a given circuit die that has failed scan testing;
compute, through a local phase of a volume diagnosis procedure, a probability distribution for the given circuit die from the diagnosis report, wherein the probability distribution specifies probabilities for different suspect root causes as having caused the given circuit die to fail;
adjust the probability distribution into an adjusted probability distribution using a supervised learning model, including by, through the supervised learning model, changing at least one of the probabilities for the different suspect root causes specified in the probability distribution as having caused the given circuit die to fail through the supervised learning model, the supervised learning model trained with a training set comprising training probability distributions of suspect root causes computed from training dies through the local phase of the volume diagnosis procedure, the training dies injected with particular root causes that cause the training dies to fail, and each training probability distribution of suspect root causes of the training set labeled with an actual root cause, wherein the actual root cause specifies the particular root cause injected into a given training die that caused the given training die to fail;
provide the adjusted probability distribution for the given circuit die as an input to a global phase of the volume diagnosis procedure to determine a global root cause distribution for multiple circuit dies that have failed the scan testing;
perform the global phase of the volume diagnosis procedure using, as inputs, a plurality of adjusted probability distributions from the multiple circuit dies that have failed the scan testing, including the adjusted probability distribution for the given circuit die adjusted through the supervised learning model; and
generate the supervised learning model, including by:
accessing the training dies, wherein each training die has been injected with a respective specific root cause to actually cause a scan test failure;
generating diagnosis reports for each of the training dies;
computing, through the local phase of a volume diagnosis procedure, the training probability distributions of suspect root causes from the diagnosis reports generated for the training dies, each of the training probability distributions of suspect root causes respectively corresponding to one of the training dies;
labeling each of the training probability distributions of suspect root causes with the respective specific root cause for the training die corresponding to the training probability distribution of suspect root causes, wherein the respective specific root cause specifies the actual root cause intentionally injected into the training die to cause the training die to fail; and
providing, as the training set, the labeled training probability distributions to train the supervised learning model.

### Claim 14 (depends on 13)

The non-transitory machine-readable medium of claim 13, wherein the volume diagnosis procedure utilizes an unsupervised learning model to compute the probability distribution for the given circuit die, determine the global root cause distribution for multiple circuit dies, or both.

### Claim 15 (depends on 13)

The non-transitory machine-readable medium of claim 13, wherein the training dies are generated via simulation, emulation, or a combination of both.

### Claim 16 (depends on 13)

The non-transitory machine-readable medium of claim 13, wherein the supervised learning model comprises a linear function that linearly adjusts the probability distribution computed for the given circuit die.

### Claim 17 (depends on 16)

The non-transitory machine-readable medium of claim 16, wherein the linear function comprises an adjustment matrix that linearly adjusts an input probability distribution; and
wherein the adjustment matrix has dimensions of ‘N’בN’, wherein ‘N’ is a number of different root causes in probability distributions computed by the local phase of a volume diagnosis procedure.