隨著奈米製程與系統晶片的發展,高密度記憶體的良率往往不到一半,因此修復成為必需。本團隊提出之超方體式修復法具有近100%的修復率,但存取時間受限。本文提出整合三值內容可定址記憶體及位址位元集中器的重新映射結構,得以使其存取時間大幅下降;結果使得數個備份列便能近100%修復記憶體,並在合理存取時間內操作,對智慧電子與物聯網來臨的時代,有莫大的貢獻。 因此本篇論文針對這些記憶體修復的文獻加以探討並提出了改良記憶體修復技術,以降低修復時的存取時間。在我們提出的超方體式記憶體修復技術中之快速重新映射架構中以使用預先計算的方式,降低位址滑動所需的時間,並以本團隊快速且低面積之內建自我繞線器作為位址集中器來代替先前所使用的平行排序器為例,來證明提出的記憶體修復時是可以有效降低存取時間。
Hypercube-based architectures with clustered faults can be most efficiently repaired by a few spare subcubes including topological networks and multidimensional memory. Channel concentrators are can be efficiently employed as a hypercube remapping architecture. However the architecture in prior work suffers extra remapping time. In this paper we proposed a precomputation-based remapping architecture for reducing the access time in normal operations. As a key component a precomputation-based address concentrator is presented. The address concentrators are also precomputed in the remapping architecture. The two-level precomputation makes the time penalty reduced to only a searching and gating time. Proposed architecture and presented concentrator are compared with previous work in our experiments.