| Abstract: |
To guarantee the reliability and functionality of semiconductor devices, risk chips, so-called Mavericks, have to be identified and eliminated. In this context, the most sensitive period is the early life stage, where failures are likely to occur and can lead to fatal consequences. Consequently, to identify potential early life failures, a pre-defined stress e.g. operation at elevated temperature, is applied to sort out weak devices during their initial lifetime. This method is commonly known as Burn-In process. Nevertheless, this method is expensive and therefore a reduction of the number of devices to be Burn-In tested is desirable. More cost- and also time-efficient are methods based on statistical analysis, e.g. Screening methods, which are applied to distinguish between good chips and Mavericks.
Unfortunately, due to miniaturization of devices, the efficiency of currently applied Screening methods, like Part Average Testing (PAT) and Good Die in Bad Neighborhood (GDBN) techniques decreases. To compensate for this, advanced Screening methods are required. One promising strategy is an algorithm which combines the method of Independent Component Analysis (ICA) and Nearest Neighbor Residuals (NNR) and is therefore investigated. First evaluations confirm that this approach is highly promising to provide a clearer distinction between good chips and Mavericks, implying reduced Burn-In effort and therewith a relevant reduction of time and money.
|