Holistic energy efficient crosstalk mitigation in DRAM

Abstract

The scaling of DRAM to increasingly small geometries has resulted in considerable challenges to both reliability and energy consumption of memory systems. For example, this aggressive scaling has resulted in increased vulnerability to both bitline and wordline crosstalk. Moreover, deep scaling has also introduced an understudied implication of dramatically increased embodied energy, or energy due to manufacturing memory integrated circuits. While many correction schemes have been proposed targeting, often independently, metrics of reliability and operational energy, recent studies have demonstrated that the impacts of manufacturing on reliability and holistic energy consumption must also be considered. In this work, we propose a technique to evaluate memory systems and their tradeoffs for reliability, embodied energy, and operational energy. We use this technique to examine several proposed correction schemes for DRAM faults. Further, we study a novel bitline crosstalk error correction scheme, Periodic Flip Encoding, which has considerable advantages in sustainability and reliability at high error rates.