Technology scaling reduces the size of memory cells, continuing to deliver dramatic improvements in memory density. Increasing memory density, however, also increases susceptibility to known failure types. This paper highlights some of the common failure modes in today’s memory technologies as well as uncommon failure modes that may grow in significance on future technology nodes. A number of approaches are described to efficiently compensate for failure-prone memory, and argue that a key ingredient in resilient systems is the ability to compensate for unanticipated memory failures.