Single-Reference-Based Multi-Level Cell Sensing Scheme for Spin-Orbit Torque MRAM

Spin-orbit torque magnetic random access memory (SOT-MRAM) is a strong candidate to replace static RAM and dynamic RAM due to its high speed, near-zero standby power, and high density. Moreover, employing multi-level cell (MLC) techniques can further increase storage density. However, there are several challenges when sensing MLC: 1) The sensing margin is too small. 2) Three references and several different sensing are required to distinguish four states. As a result, significant read yield degradation, area overhead, and performance penalty are unavoidable. There are two conceptual methods for sensing MLC. The one-step parallel sensing scheme (SS) allows for data reading at once, but suffers from low read yield due to the offset of sensing circuit. On the other hand, the two-step SS using offset-cancellation technique can achieve a higher read yield but requires twice the time. To address these challenges, the single-reference-based MLC-SS using offset-cancellation technique is proposed. HSPICE simulation results, based on industry-compatible 28 nm model parameters demonstrate that the proposed MLC-SS can achieve a comparable read yield of 94.19% compared to 86.17% for the one-step parallel SS and 96.77% for the two-step SS, while requiring only one reference and incurring just a 20% increase in sensing time.