Résumé
Diagnosis is becoming a major concern with the rapid development of semiconductor memories. It provides information about the location of manufacturing defects in the memory, and its effectiveness allows a fast yield ramp up. Most of existing diagnosis methods uses a fault dictionary to provide detailed information of fault localization. However, these solutions are most of the time unable to distinguish between all faults, and more importantly often fail to identify the actual faulty block of the memory. Identifying which block of a memory (core-cell array, write drivers, address decoders, pre-charge circuits, etc ...) is defective allows saving considerable amount of time during the ramp up phase. In this paper, we propose a very low cost Design-for-Diagnosis (DfD) solution for identifying faulty write drivers. It consists in verifying logic and analog conditions that guarantee the fault-free behavior of the write driver. The proposed solution allows a fast diagnosis (only three consecutive write operations are needed to fully diagnose the write driver) and induces a low area overhead (about 0.5% for a 512x512 SRAM). Beside diagnosis, an additional interest of such a solution is its usefulness during a post-silicon characterization process, where it can be used to extract the main features of write drivers (logic and analog levels on bit lines).