CharTM: The dynamic stability characterization for memory based on tail distribution modeling. (March 2023)

• Record Type: Journal Article
• Authors: Pang, Liang; Yao, Mengyun; Shi, Xiao; Yan, Hao; Shi, Longxin
• Is Part Of: Microelectronics journal. Volume 133(2023)
• Publication Date: 2023-03
• Subjects: Memory -- Yield analysis -- Dynamic stability -- High-dimensional meta-model -- Tail distribution modeling; Microelectronics -- Periodicals; Microélectronique -- Périodiques; Microelectronics; Electronic journals; Journals - contents and abstracts; Periodicals; 621.3805
• DOI: 10.1016/j.mejo.2023.105697
• Languages: English
• ISSNs: 0959-8324

Abstract

Abstract: Dynamic stability has become a major constraint to static random access memory (SRAM) circuits design. In this paper, the relationship between the read/write delay and corresponding yield will be given by our CharTM, an unified memory yield analysis method by modeling the tail of performance distribution. In charTM, we design an efficient weighted regression-based high-dimensional meta-model to collect the tail samples without running simulations and construct the tail distribution by generalized Pareto distribution (GPD) with an adaptive tail fraction selection strategy to improve the accuracy in the high-sigma region. The effectiveness of CharTM is verified on bit cell, read, and write critical paths of an SRAM. The results show that the efficiency of CharTM exceeds that of a commercial yield analysis tool by 5.3–16.5 times on our validation cases with the same level of accuracy, and exceeds that of state-of-the-art methods by 13–82 times with much higher accuracy, especially on read and write paths.