Variation Aware Evaluation Approach and Design Methodology for SOT-MRAM

• Published in: IEEE transactions on circuits and systems. I, Regular papers Vol. 71; no. 4; pp. 1651 - 1664
• Main Authors: Wang, Chao, Wang, Zhaohao, Li, Shixing, Zhang, Zhongkui, Zhang, Youguang
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.04.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: bit-cell; Error correction codes; high-density; Layout; magnetic tunnel junction (MTJ); Magnetic tunneling; Optimization; Random access memory; Semiconductor device modeling; SOT-MRAM; Spintronics; Transistors; variation
• ISSN: 1549-8328; 1558-0806
• DOI: 10.1109/TCSI.2024.3361470

Spin-orbit torque magnetic random access memory (SOT-MRAM), which exhibits sub-nanosecond write speed and high reliability, is a promising candidate for the future high-level cache. However, SOT-MRAM faces the problem of large bit-cell layout area due to its structural characteristics and write performance requirements, therefore it is necessary to explore the bit-cell design with optimal overall performance under the unified bit-cell area. In this paper, we propose a comprehensive variation aware evaluation approach for the area, latency, and energy of SOT-MRAM under the uniform yield standard. Based on this, the mainstream SOT-MRAM bit-cell designs with high-density method and multi-finger configuration are evaluated, meanwhile bit-cell designs with excellent write performance and their optimum area ranges are identified. Moreover, the source line read (SLR) mode with higher robustness against transistor variation is proposed to improve the read performance, and the dual SL (DSL) method is proposed to further reduce the read latency and write energy. With the DSL method, the read latency and write energy of 2-word-line (WL)-type bit-cells can be reduced by up to 36.5% and 12.6%, respectively. In addition, the DSL method can solve the shunt current issue of 1WL-type bit-cells and reduce the read latency and write energy by up to 43.6% and 17.4%, respectively.