Thermal stability analysis and modelling of advanced perpendicular magnetic tunnel junctions

Simon Van Beek,Yueh Chang Wu,Woojin Kim,Guido Groeseneken,Koen Martens,Philippe Roussel,Davide Crotti,Johan Swerts,Dimitri Linten,Gouri Sankar Kar,Siddharth Rao

doi:10.1063/1.5007690

Abstract

STT-MRAM is a promising non-volatile memory for high speed applications. The thermal stability factor (Δ = Eb/kT) is a measure for the information retention time, and an accurate determination of the thermal stability is crucial. Recent studies show that a significant error is made using the conventional methods for Δ extraction. We investigate the origin of the low accuracy. To reduce the error down to 5%, 1000 cycles or multiple ramp rates are necessary. Furthermore, the thermal stabilities extracted from current switching and magnetic field switching appear to be uncorrelated and this cannot be explained by a macrospin model. Measurements at different temperatures show that self-heating together with a domain wall model can explain these uncorrelated Δ. Characterizing self-heating properties is therefore crucial to correctly determine the thermal stability.

Highlights

Spin-transfer torque magnetic random access memory (STT-MRAM) has high endurance, high speed and is compatible with CMOS.[1]
One CoFeB layer is pinned to a high anisotropy Co/Pt synthetic antiferromagnet to create a fixed reference layer (RL) [see Fig.1(a)]
Methods have been developed to accelerate switching and extrapolate ∆: based on current pulses,[4] magnetic field sweeps[5] or temperature.[6]. In these methods Eb is expressed by a simplified generalization of the Stoner-Wohlfarth model, macrospin model (MS), even though switching by the formation of a domain wall (DW) is thought to occur.[7]

Summary

INTRODUCTION

Spin-transfer torque magnetic random access memory (STT-MRAM) has high endurance, high speed and is compatible with CMOS.[1]. Methods have been developed to accelerate switching and extrapolate ∆: based on current pulses,[4] magnetic field sweeps[5] or temperature.[6] In these methods Eb is expressed by a simplified generalization of the Stoner-Wohlfarth model, macrospin model (MS), even though switching by the formation of a domain wall (DW) is thought to occur.[7] the extracted parameters from the domain wall (DW) and macrospin (MS) model are not consistent.[8] the conventional magnetic field sweep method leads to a significant error.[9] different methods result in different ∆ values. We find that for current switching self-heating will play a significant role

EXPERIMENTAL

Magnetic field acceleration

Current acceleration and influence of self-heating

CONCLUSIONS