Sudden Power-Outage Resilient In-Processor Checkpointing for Energy-Harvesting Nonvolatile Processors

This paper introduces a sudden power-outage resilient in-processor checkpointing for energy-harvesting nonvolatile processors. In energy harvesting applications, a power supply generated from a renewable power source is unstable that may induce frequent sudden power outages, causing the inconsistency among distributed nonvolatile flip-flops (NVFFs) and hence failure rollbacks in conventional nonvolatile processors. To realize continuous operations upon the frequent sudden power outages, the proposed in-processor checkpointing technique fixes the inconsistency using time-reminding redundant NVFFs (TM-RNVFFs). The TM-RNVFFs store the current and the past few data with the timing information of storing. If several NVFFs fail to store the current data due to the sudden power outages, the proposed in-processor checkpointing technique exploits the timing information to find the common newest state among distributed NVFFs, leading to correct rollbacks to the state with consistency. The sudden power-outage effect is modeled to perform design space explorations at different configurations, such as redundancy and checkpointing period. Nonvolatile ARM Cortex-M0 processors are designed using hybrid 90 nm CMOS and 70 nm magnetic tunnel junction (MTJ) technologies. Based on the design space explorations, the proposed nonvolatile processor achieves a several order-of magnitude reduction in rollback error probability with a power dissipation overhead of 11.6 percent and an area overhead of 52.1 percent in comparison with the conventional nonvolatile processor.