Memory Testing Using Neural Network Yield Estimation

Solution Overview

Problem

Current memory testing methods are time-consuming, requiring extensive re-execution of test programs to adjust sequences or modify conditions, leading to significant expenditure of testing time.

Innovation Solution

A memory testing method and apparatus that uses a test program group to generate independent test data for each die, followed by a neural network operation to estimate die yields, thereby reducing testing time and improving data effectiveness.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If all test programs are executed to ensure comprehensive testing, then testing completeness is improved, but testing time increases significantly

Engineering Contradiction:

Improvetesting completenessVSAvoidtesting time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary testing with a first test program to obtain initial test data before executing the complete test program group. This preliminary action allows the system to identify and exclude test programs that will not affect the final yield result, thereby reducing the overall testing time while maintaining testing completeness for critical parameters.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent divides the complete test program group into multiple segments: a first test program executed preliminarily, and a remaining set of test programs executed subsequently. By segmenting the testing process and selectively executing only necessary programs based on preliminary results, the system reduces redundant testing while ensuring all critical yield-affecting parameters are tested.

Inventive Principle:

Principle #1Segmentation

2Measurement precision

If test program sequences are adjusted or conditions modified to address yield loss, then testing effectiveness is improved, but extensive re-execution of all test programs is required

Engineering Contradiction:

Improvetesting effectivenessVSAvoidre-execution time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary testing with a first test program to obtain initial test data before executing the complete test program group. This preliminary action allows the system to identify and exclude test programs that will not affect the final yield result, thereby reducing the overall testing time while maintaining testing completeness for critical parameters.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent divides the complete test program group into multiple segments: a first test program executed preliminarily, and a remaining set of test programs executed subsequently. By segmenting the testing process and selectively executing only necessary programs based on preliminary results, the system reduces redundant testing while ensuring all critical yield-affecting parameters are tested.

Inventive Principle:

Principle #1Segmentation

3Measurement precision

If neural network operation is performed on independent test data, then yield estimation accuracy is improved, but data processing complexity increases

Engineering Contradiction:

Improveyield estimation accuracyVSAvoiddata processing complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts only the necessary independent test data from the test program group that actually affects yield results, excluding redundant data from test programs that are excluded based on preliminary testing. This extraction reduces the volume of data requiring neural network processing while maintaining the accuracy needed for reliable yield estimation.

Inventive Principle:

Principle #2Taking out (Extraction)

Data Source

PatentUS12283334B2Method and apparatus for memory testing

Publication Date: 2025.04.22 WINBOND ELECTRONICS CORP

AI summary

Disclosed are a method and an apparatus for memory testing. The method comprises following steps: using a test program group including N test programs to test M dies respectively to generate independent N test data, wherein N and M are positive integers greater than 1; and executing a neural network operation on the N test data to estimate a yield of M dies passing the test program group.