| Summary: | 博士 === 國立成功大學 === 電機工程學系 === 106 === With the shrinking manufacturing process and increasing design complexity, the defect behaviors in contemporary integrated circuits have become much more complex than ever. It is generally realized that the test set for a simple, single fault model such as the stuck-at fault model or transition delay fault model is insufficient to detect all defects. Hence, it is necessary to use more accurate, and often more than one fault model in order to avoid test escapes and increase defect coverage, resulting in long test time for whole test and diagnosis flow. The low defect coverage also causes that the diagnosis analysis tool cannot identify the fault site and location of real defect efficiently.
In this dissertation, we propose the test and diagnosis methodology to deal with various fault models to reduce the cost problems due to large amount of test and diagnosis data volume. We apply our methodology to various fault models, including time-dependent fault models, time-independent fault models and undefined defects. By using a commercial ATPG tool, a compact set of test and diagnosis pattern can be generated to detect various fault models and distinguish the fault pair that consists of two faults. Not only speeding up the whole test and diagnosis flow but also increasing the efficiency of diagnosis analysis (identify the defect location and defect type), resulting in much lower test and diagnosis cost.
The generated diagnosis patterns can help us identify the real defect efficiently. However, due to the limitation of logic circuit, there are still many functionally-equivalent faults that cannot be distinguished by any possible test. In our dissertation, we develop a repair-for-diagnosis architecture to help us distinguish those functionally-equivalent faults. Bu adding the redundant transistors or redundant gates into the fault site of those functionally-equivalent faults, we can identify the location and type of real defect if repairing is successful. We can apply this repair-for-diagnosis architecture to new advance manufacturing process to identify the real defect quickly and improve chip yields efficiently.
|
|---|
