PETRA: Powerful Early Termination-Based Redundancy Analysis

In dynamic random-access memory (DRAM), memory redundancy analysis (RA) is a crucial process for enhancing memory yield and reducing production costs. It finds a memory repair solution by efficiently allocating the limited number of spare cell lines that replace a faulty cell. However, it is challenging to quickly find a memory repair solution because RA is an NP-complete problem. To address this issue more effectively, we present a powerful early termination-based high-speed RA method. This method rapidly assesses memory repairability, terminating the RA process early in cases where repair is impossible, or a solution can be easily found. Additionally, by dividing faulty cells into several groups, the proposed RA method finds fast and approximate albeit nonoptimal solution sets for each group. This facilitates the rapid acquisition of a memory repair solution without the need to search for all the optimal solution sets. These features enable RA to be promptly executed while ensuring the repair solution for any repairable memory. Experimental results demonstrate that the proposed RA method can find a repair solution faster than the existing RA methods.