A Probability-Based Data Allocation Strategy for Hybrid DRAM/NVM Memory in Real-Time Embedded Systems

Abstract

Non-volatile memory (NVM)has emerged as a promising DRAM alternative due to its high density, and zero leakage power. Nevertheless, it suffers from higher write energy. According to a given data-access frequencies, the previous studies focus on data allocation technique for utilizing the benefits of both NVM and DRAM. However, data-access frequencies is often obtained with probability, which could not be effectively applied on the previous techniques. To address this issue, this paper proposes a probability-based data allocation strategy for hybrid DRAM/NVM memory in real-time embedded systems. The basic idea is first to obtain the probability-based data-access frequencies of a given embedded program by exploiting its application-specific feature. Combining with the maximum data-access frequencies and the data-access frequency expectations of each data in a given program, this paper proposes a novel and simple data allocation algorithm, named PBDA, to minimize the energy consumption of real-time embedded system. Finally, compared to the Greedy algorithm and an existing optimal data allocation algorithm, the experiments show that our technology can reduce energy consumption by 45.01% and 10.49% on average.