Reliable design of high-speed cache and control store memories

Abstract

The design of the cache and control store memories of the Tandem NonStop VLX processor is discussed. Service costs are reduced by using hot-standby sparing to improve the reliability of the large static RAM arrays. Detection, isolation, and spare substitution of failed RAMs are performed automatically without the disruption of normal processing. A control store design with sparing is described. A mathematical model is used to predict reliability improvements for the multiple arrays for each processor board. The model takes into account the selected repair policy which calls for replacing a board only on spare exhaustion or on the failure of nonspared logic. The success of the chosen approach is illustrated through model predictions as well as through field failure data.<>