On intrinsic failure rate of products with error correction

Abstract

The bathtub curve is widely used in the reliability world to illustrate three characteristics of the system (or an IC) failure rate curve: the early failure rate during initial phase of life, the intrinsic constant failure rate in the use life, and the wear-out phase at the end-of-life (Reddy, 2003). The bottom part is usually characterised by a constant failure rate expressed in FIT (failure in time, which is 1 failure in 109 device hours). As systems become more and more complex, more and more redundancy and EDAC (error detection and correction, also called error correction code ECC) are implemented in order to improve the system robustness and reliability (Slayman, 2003 and Ziegler and Puchner, 2004). By studying the data retention failure rates of Flash products with ECC, we found that the bottom part of the bathtub (used to be characterized by random constant failure rate) is not flat. We suggest replace the FIT number by a cumulative failure fraction with a certain time stamp.