The project has been successfully completed by 30 June 2016 with the Doctoral Thesis "Device Level Maverick Screening", AAU Klagenfurt.
In semiconductor industry the reliability demands on chips grow as they are used in safety relevant applications. To guarantee their functionality, verification measurements are done. Based on this data, devices failing the pre-defined specifications are scrapped. Nevertheless, among the remaining devices are still devices at risk, so-called Mavericks, which are responsible for early life failures. To increase the reliability of devices, risk devices need to be identified and excluded before delivered to the customers.
Proposed solution: apply a high-order data transformation to the measurements to find irregularities and reveal potential masked information indicating Mavericks.
A common method to detect Mavericks is the Burn-In. This procedure is an endurance test under increased stress conditions to detect risk chips. Due to unwanted side effects of Burn-In, e.g. possible pre-damage and high costs, less expensive screening methods capable to reduce the need for Burn-In are desired. With decreasing chip size, the accuracy of the state-of-the-art Device Screening methods will decrease as well; therefore, new methods to identify Mavericks are needed.
Using the already performed measurements to find irregularities in the data is convenient, but these data are often inconspicuous (see Figure 2, left side), implying either that no information is given (all devices are at low risk to fail early) or that the meaningful information is masked. Therefore a reliable data transformation to reveal potential masked information in these measurements is needed, e.g. Independent Component Analysis (ICA).
Our investigations on advanced Device Screening methods started 2013 in cooperation with the Alpen-Adria-Universität Klagenfurt. Within this time KAI developed competences in these areas:
- Independent Component Analysis (ICA)
- Principal Component Analysis (PCA)
- State-of-the-art Device Screening methods (PAT, D-PAT, NNR)
- Measures for non-Gaussianity (Negentropy, statistical tests)