Cellular automata based multi-bit stuck-at fault diagnosis for resistive memory

This paper presents a group-based dynamic stuck-at fault diagnosis scheme intended for resistive random-access memory (ReRAM). Traditional static random-access memory, dynamic random-access memory, NAND, and NOR flash memory are limited by their scalability, power, package density, and so forth. Next-generation memory types like ReRAMs are considered to have various advantages such as high package density, non-volatility, scalability, and low power consumption, but cell reliability has been a problem. Unreliable memory operation is caused by permanent stuck-at faults due to extensive use of write- or memory-intensive workloads. An increased number of stuck-at faults also prematurely limit chip lifetime. Therefore, a cellular automaton (CA) based dynamic stuck-at fault-tolerant design is proposed here to combat unreliable cell functioning and variable cell lifetime issues. A scalable, block-level fault diagnosis and recovery scheme is introduced to ensure readable data despite multi-bit stuck-at faults. The scheme is a novel approach because its goal is to remove all the restrictions on the number and nature of stuck-at faults in general fault conditions. The proposed scheme is based on Wolfram’s null boundary and periodic boundary CA theory. Various special classes of CAs are introduced for 100% fault tolerance: single-length-cycle single-attractor cellular automata (SACAs), single-length-cycle two-attractor cellular automata (TACAs), and single-length-cycle multiple-attractor cellular automata (MACAs). The target micro-architectural unit is designed with optimal space overhead.