# US 2025/0148273 A1 · System and method for utilizing transformer deep learning based outlier ic detection

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

- Number: US 2025/0148273 A1
- Document ID: US20250148273A1
- Status: Published application
- Kind: A1
- Application number: 18/926,397
- Filed: 2024-10-25
- Published: 2025-05-08
- Earliest claimed: 2023-11-03
- Assignees: Mediatek Inc.
- Applicants: Mediatek Inc.
- Inventors: Khim Jun Koh; Chi-Ming Lee; Yi-Ju Ting; Chung-Kai Chang; Po-Chao Tsao; Chin-Wei Lin; Yu-Lin Yang; Tung-Hsing Lee; Chin-Tang Lai
- Web page: https://patentgrep.com/patent/US20250148273A1

## CPC

- G06N 3/0499: Feedforward networks
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

In an aspect of the disclosure, a method for detecting outlier integrated circuits on a wafer is provided. The method comprises: operating multiple test items for each IC on the wafer to generate measured values of the multiple test items for each IC; selecting a target IC and neighboring ICs on the wafer repeatedly. each time after selecting the target IC executes the following steps: selecting a measured value of the target IC as a target measured value and selecting measured values of the target IC and the neighboring ICs as feature values of the target IC and the neighboring ICs; executing a transformer deep learning model to generate a predicted value of the target measured value; and identifying outlier ICs according to the predicted values of all the target ICs and the corresponding target measured values of all the target ICs.

## Claims

### Claim 1

A method for detecting outlier integrated circuits (ICs) on a wafer, comprising:
operating a plurality of test items for each of a plurality of ICs on the wafer to generate measured values of the plurality of test items for each of the plurality of ICs;
selecting a target IC and neighboring ICs adjacent to the target IC from the plurality of ICs on the wafer, repeatedly, wherein each time the target IC is selected, the following steps are executed:
selecting a measured value from the measure values of the target IC as a target measured value and selecting measured values of the target IC and the neighboring ICs which are related to the target measured value as feature values of the target IC and the neighboring ICs; and
executing a transformer deep learning model to generate a predicted value of the target measured value according to the feature values of the target IC and the neighboring ICs; and
identifying outlier ICs according to the predicted values of all the target ICs and the corresponding target measured values of all the target ICs after generating predicted values for all the target ICs.

### Claim 2 (depends on 1)

The method according to claim 1, wherein the step of executing a transformer deep learning model comprising:
obtaining by a feed-forward network of the transformer deep learning model, the non-linear relationships between the feature values of the target IC and corresponding feature values of each of the neighboring ICs; and
generating by a feed-forward network of the transformer deep learning model, a predicted value corresponding to the target measured value of the target IC according to the non-linear relationships.

### Claim 3 (depends on 1)

The method according to claim 1, wherein the step of executing a transformer deep learning model comprising:
obtaining by a self-attention mechanism of the transformer deep learning model, dependency information between the target IC and the neighboring ICs according to the feature values of the target IC and the corresponding feature values of the neighboring ICs;
obtaining by a feed-forward network of the transformer deep learning model, the non-linear relationships between the feature values of the target IC and corresponding feature values of each of the neighboring ICs according to the dependency information; and
generating by the feed-forward network of the transformer deep learning model, a predicted value corresponding to the target measured value of the target IC according to the non-linear relationships.

### Claim 4 (depends on 3)

The method according to claim 3, wherein for each target IC, the self-attention mechanism and the feed-forward network repeat N iterations to keep updating the dependency information and the non-linear relationships,
wherein N is a positive integer, and each attention loop of the N iterations focuses different aspects of the feature values or different group of neighboring ICs.

### Claim 5 (depends on 4)

The method according to claim 4, wherein the predicted value is the final output of the last one of the N iterations from the feed-forward network.

### Claim 6 (depends on 3)

The method according to claim 3, wherein before executing the transformer deep learning model, the transformer deep learning model is trained by employing a standard Mean Squared Error (MSE) loss function to optimize weights of the self-attention mechanism and the feed-forward network by backpropagation.

### Claim 7 (depends on 1)

The method according to claim 1, wherein the plurality of test items for each of the plurality of ICs on the wafer are related to current leakage, minimum operating voltage, or on-chip sensors.

### Claim 8 (depends on 1)

The method according to claim 1, wherein before executing a transformer deep learning model, the feature values of the target IC and the neighboring ICs are organized into a 2-dimensional vector, wherein the feature values of the target IC form a first dimension of the 2-dimensional vector and the corresponding feature values of the neighboring ICs form a second dimension of the 2-dimensional vector.

### Claim 9 (depends on 1)

The method according to claim 1, wherein the neighboring ICs of the target IC are all or a subset of the plurality of ICs excluding the target IC, on the wafer.

### Claim 10 (depends on 1)

The method according to claim 1, wherein the step of identifying outlier ICs comprising:
obtaining a Mahalanobis distance for each target IC according to the predicted value of the target IC and the target measured value of the target IC, and identifying outlier IC(s) according to the Mahalanobis distances of all the target ICs.

### Claim 11 (depends on 10)

The method according to claim 10, wherein the step of identifying outlier IC(s) according to the Mahalanobis distances of all the target ICs, further comprising:
Identifying the target IC as an outlier IC when the Mahalanobis distance of the target IC occurs at an outlier point outside a predetermined specified range; and
Identifying the target IC as a normal IC when the Mahalanobis distance of the target IC occurs inside the predetermined specified range.

### Claim 12

A system for detecting outlier ICs on a wafer, comprising:
an IC test module, configured to operate a plurality of test items for each of a plurality of ICs on the wafer to generate measured values of the plurality of test items for each of the plurality of ICs;
a target designating module, configured to select a target IC and neighboring ICs adjacent to the target IC, from a plurality of ICs on the wafer repeatedly, and select a measured value from the measure values of the target IC as a target measured value and select measured values of the target IC and the neighboring ICs which are related to the target measured value as feature values of the target IC and the neighboring IC, wherein a different target IC is selected at each time;
a model execute module, configured to execute a transformer deep learning model to generate a predicted value of the target measured value according to the feature values of the target IC and the neighboring ICs; and
a detecting module, configured to identify outlier ICs according to the predicted values of all the target ICs and the corresponding target measured values of all the target ICs after generating predicted values for all the target ICs.

### Claim 13 (depends on 12)

The system according to claim 12, wherein when executing the transformer deep learning model, the model execute module is further configured to obtain by a feed-forward network of the transformer deep learning model, the non-linear relationships between the feature values of the target IC and corresponding feature values of each of the neighboring ICs, and generate by the feed-forward network of the transformer deep learning model, a predicted value corresponding to the target measured value of the target IC according to the non-linear relationships.

### Claim 14 (depends on 12)

The system according to claim 12, wherein when executing the transformer deep learning model, the model execute module is further configured to:
obtain by a self-attention mechanism of the transformer deep learning model, dependency information between the target IC and the neighboring ICs according to the feature values of the target IC and the corresponding feature values of the neighboring ICs;
obtain by a feed-forward network of the transformer deep learning model, the non-linear relationships between the feature values of the target IC and corresponding feature values of each of the neighboring ICs according to the dependency information; and
generate by the feed-forward network of the transformer deep learning model, a predicted value corresponding to the target measured value of the target IC according to the non-linear relationships.

### Claim 15 (depends on 14)

The system according to claim 14, wherein for each target IC, the self-attention mechanism and the feed-forward network repeat N iterations to keep updating the dependency information and the non-linear relationships,
wherein N is a positive integer, and each attention loop of the N iterations focuses different aspects of the feature values or different group of neighboring ICs;
wherein the predicted value is the final output of last one of the N iterations from the feed-forward network.

### Claim 16 (depends on 14)

The system according to claim 14, further comprising:
a model training module, configured to train the transformer deep learning model by employing a standard Mean Squared Error (MSE) loss function to optimize weights of the self-attention mechanism and the feed-forward network by backpropagation.

### Claim 17 (depends on 12)

The system according to claim 12, wherein the target designating module is further configured to organize the feature values of the target IC and the neighboring ICs into a 2-dimensional vector before the model execute module execute the transformer deep learning model, wherein the feature values of the target IC form a first dimension of the 2-dimensional vector and the corresponding feature values of the neighboring ICs form a second dimension of the 2-dimensional vector.

### Claim 18 (depends on 12)

The system according to claim 12, wherein the neighboring ICs of the target IC are all or a subset of the plurality of ICs excluding the target IC, on the wafer.

### Claim 19 (depends on 12)

The system according to claim 12, wherein when identifying outlier ICs according to the predicted values of all the target ICs and the corresponding target measured values of all the target ICs, the detecting module is further configured to:
obtain a Mahalanobis distance for each target IC according to the predicted value of the target IC and the target measured value of the target IC, and identify outlier IC(s) according to the Mahalanobis distances of all the target ICs.

### Claim 20 (depends on 19)

The system according to claim 19, wherein when identifying outlier IC(s) according to the Mahalanobis distances of all the target ICs, the detecting module is further configured to:
identify the target IC as an outlier IC when the Mahalanobis distance of the target IC occurs at an outlier point outside a predetermined specified range; and
identify the target IC as a normal IC when the Mahalanobis distance of the target IC occurs inside the predetermined specified range.
