Abstract:

As DRAM memory scales down to smaller technology nodes, row hammering becomes a critical problem that not only degrades system reliability but also opens up security vulnerabilities. This paper proposes a new row-refresh hammering prevention solution, named Recent Activation Counter based Probabilistic Refresh（RACPR）, which exploits the recent activation information of victim rows. This paper first makes a key observation that a passive activation to victim rows has the same effect of preventing row hammering compared to a dedicated proactive refresh. This paper then proposes Recent Activation Counter（RAC） to track the timing information of activation per row. This paper optimizes the structure and operation of RAC so as to reduce its complexity and overheads. RACPR strategy provides robust row hammering protection at negligible performance overhead and 0.006% area cost. Experimental results show that the approach further reduces the number of proactive refresh by 75% on average, and provides better average performance compared to previous probabilistic approaches.

Key words: DRAM, row hammering, probabilistic refresh, recent activation counter

摘要：

随着DRAM存储器制程的发展，存储单元尺寸更小，行扰动成为一个降低DRAM可靠性和暴露安全漏洞的关键性问题。提出一种基于内存行最近激活信息（RACPR）的解决DRAM行扰动问题的方法，主要思想是基于受害行的最近激活信息，利用被动激活操作与专用的主动刷新操作具有相同的防止行扰动的效果，减少不必要的主动刷新操作。提出最近激活计数器（RAC）来跟踪内存行的被动激活信息，并优化RAC的结构和操作，以降低其复杂性和开销。提出的RACPR方法性能开销可忽略，面积成本为0.006%。实验结果表明：与以往的概率方法相比，RACPR方法平均减少75%的主动刷新次数，并提供更好的平均内存访问性能。

关键词: DRAM存储器, 行扰动, 概率刷新, 最近激活计数器