Power Spectrum Analysis for IC Defect Screening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current techniques for defect screening in integrated circuit devices are complex, time-consuming, and costly, and fail to detect latent defects until they develop into observable failures, posing a reliability risk.
Innovation Solution
A method involving power spectrum analysis (PSA) where a time-varying signal is applied to an integrated circuit device, and the power spectrum is measured and compared to known defect status spectra using Principal Component Analysis (PCA) to identify defects non-destructively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional electrical, optical, and thermal techniques are used for defect identification, then defect detection capability is improved, but device complexity and testing cost increase
Solution Approach 1:
The patent extracts and measures only the power spectrum characteristics from the device response, separating the essential defect information from complex time-domain signals. By transforming the full signal analysis into targeted power spectrum measurement at specific frequencies, the method simplifies the testing process while maintaining defect detection capability.
Solution Approach 2:
The patent performs power spectrum analysis on test devices before final product deployment to identify latent defects early. By conducting preliminary power spectrum measurements and comparing them against reference spectra from known good devices, the method detects potential failures before they manifest as actual defects, reducing downstream testing complexity.
2Measurement precision
If reliability testing with high voltage and temperature is applied, then latent defect detection is improved, but production time and cost increase
Solution Approach 1:
The patent applies power spectrum analysis as a preliminary screening method before reliability testing. By measuring power spectra under normal operating conditions and comparing against reference spectra, latent defects are identified early in the production process, reducing or eliminating the need for time-consuming high-voltage/high-temperature reliability testing for all devices.
Solution Approach 2:
The patent changes the measurement parameters from extreme stress conditions (high voltage, high temperature) to normal operating conditions with power spectrum analysis. This parameter change allows latent defect detection without subjecting devices to accelerated stress testing, significantly reducing production time while maintaining detection effectiveness.
3Measurement precision
If extensive failure analysis is performed on ICs with millions of transistors, then defect identification accuracy is improved, but testing duration increases
Solution Approach 1:
The patent extracts characteristic power spectrum features that specifically indicate defect presence, separating relevant defect information from the complex behavior of millions of transistors. By focusing measurement on power spectrum characteristics at specific frequencies rather than analyzing entire device behavior, the method achieves high defect identification accuracy rapidly.
Solution Approach 2:
The patent uses power spectrum analysis to detect changes in the electrical 'color' or frequency characteristics of the device. Defects manifest as specific spectral signatures or frequency components in the power spectrum, allowing rapid identification without extensive analysis of individual transistor behavior, thus reducing testing duration while maintaining accuracy.
Data Source
AI summary
A method involving the non-destructive testing of a sample electrical or electronic device is provided. The method includes measuring a power spectrum of the device and performing a Principal Component Analysis on the power spectrum, thereby to obtain a set of principal components of the power spectrum. The method further includes selecting a subset consisting of some of the principal components, and comparing the subset to stored reference data that include representations in terms of principal components of one or more reference populations of devices. Based at least partly on the comparison, the sample device is classified relative to the reference populations.