D I S S E R T A T I O N

Computation of Torques
in Magnetic Tunnel Junctions


ausgeführt zum Zwecke der Erlangung des akademischen Grades
eines Doktors der technischen Wissenschaften

unter der Betreuung von

Privatdoz. MSc PhD Viktor Sverdlov
O.Univ.Prof. Dipl.-Ing. Dr.techn. Dr.h.c. Siegfried Selberherr

eingereicht an der Technischen Universität Wien
Fakultät für Elektrotechnik und Informationstechnik
von

Simone Fiorentini, MSc.
Matrikelnummer: 11839446

Wien, im Juli 2023  

Abstract

Non-volatility is an emerging solution to stand-by power leakages caused by the down-scaling of traditional semiconductor components. Spin-transfer torque magnetoresistive random access memory (STT-MRAM) is a viable nonvolatile candidate, thanks to its simple structure and compatibility with CMOS technology. It possesses high speed and excellent endurance, being thus promising for applications ranging from IoT and automotive uses to embedded DRAM and last level caches. Accurate simulation tools offer valuable insight in the design of STT-MRAM devices. The description of the temporal evolution of the magnetization is given by solving the Landau-Lifshitz-Gilbert (LLG) equation. The LLG equation can be enriched with terms describing the torque generated by an electric current flowing through the device, responsible for STT-MRAM operation. The present work focuses on the development and calibration of simulation approaches for dealing with the sources of torque acting in STT-MRAM devices.

First, an in-house solver based on the finite difference method, with the STT included using the Slonczewski expression, is generalized for performing switching simulations of an MRAM cell. The solver is updated to allow running switching simulations with three different approaches: with uniform and constant current density, with more realistic non-uniform current density and constant total current, and with constant voltage. The validity of the description with the fixed current density for predicting the switching time is tested by comparing its results with the ones of the other two approaches, and the switching time dependence on several system parameters is evaluated. The obtained results show that a correction on the value of the applied current allows for all the approaches to deliver compatible results.

To accurately evaluate torques beyond the interface Slonczewski approximation in composite layered structures, consisting of several ferromagnetic parts separated by nonmagnetic metal spacers and tunnel barriers, an approach based on the solution of coupled spin and charge transport equations to determine the non-equilibrium spin accumulation is adopted and implemented in a finite element-based framework. The formalism is extended to describe transport through magnetic tunnel junctions. The tunneling magnetoresistance (TMR) effect is included by modeling the tunnel barrier as a poor conductor, whose conductivity is locally dependent on the relative magnetization orientation of the ferromagnetic layers. The tunneling spin current polarization is included via appropriate boundary conditions at the interface between oxide and ferromagnetic layers. The proposed approach successfully reproduces angular and voltage dependencies of the torque exerted in MTJs, and allows to evaluate the interplay between the interface and bulk sources of torque. The approach is successfully applied to switching simulations and design of ultra-scaled STT-MRAM cells.

Kurzfassung

Nichtflüchtigkeit ist eine aufkommende Lösung für Standby-Leistungsverluste, die durch das Herunterskalieren herkömmlicher Halbleiterkomponenten verursacht werden. Magnetoresistiver Spin-Transfer-Torque-Direktzugriffsspeicher (STT-MRAM) ist dank seiner einfachen Struktur und Kompatibilität mit der CMOS-Technologie ein geeigneter nichtflüchtiger Kandidat. Er verfügt über eine hohe Geschwindigkeit und eine hervorragende Lebensdauer und ist daher vielversprechend für die Verwendung in IoT- und Automobilanwendungen bis hin zu eingebettetem DRAM und Last-Level-Caches. Präzise Simulationswerkzeuge bieten wertvolle Einblicke in das Design von STT-MRAM-Bauelementen. Die Beschreibung der zeitlichen Entwicklung der Magnetisierung erfolgt durch Lösung der Landau-Lifshitz-Gilbert (LLG)-Gleichung. Die LLG-Gleichung kann mit Termen erweitert werden, die das Drehmoment beschreiben, das durch einen fließenden elektrischen Strom erzeugt wird, und für das Funktionieren des STT-MRAM verantwortlich ist. Die vorliegende Arbeit konzentriert sich auf die Entwicklung und Kalibrierung von Simulationsansätzen zum Umgang mit den Drehmomentquellen, die in STT-MRAM-Bauelementen wirken.

Im ersten Teil der Arbeit werden die Ergebnisse von Schaltsimulationen von MRAM-Zellen beschrieben, welche mit einem hauseigenen Solver durchgeführt wurden, der auf der Finite-Differenzen Methode basiert und der das STT unter Verwendung des Slonczewski-Ausdrucks berücksichtigt. Der Solver wird erweitert, um das Ausführen von Schaltsimulationen mit drei verschiedenen Ansätzen zu ermöglichen: mit gleichmäßig verteilter und konstanter Stromdichte, mit realistischer ungleichmäßig verteilter Stromdichte und konstantem Gesamtstrom und mit konstanter Spannung. Die Gültigkeit der Beschreibung mit der festen Stromdichte zur Abschätzung der Schaltzeit wird durch Vergleich der Ergebnisse mit denen der anderen zwei Ansätze getestet. Die Abhängigkeit der Schaltzeit von mehreren Systemparametern wird analysiert, was zeigte, dass eine Korrektur des Werts des angelegten Stroms es ermöglicht, dass alle Ansätze vergleichbare Ergebnisse liefern.

Um Drehmomente jenseits der Slonczewski-Näherung der Grenzfläche in zusammengesetzten Schichtstrukturen genau zu bewerten, die aus mehreren ferromagnetischen Teilen bestehen und die durch nichtmagnetische Metallabstandshalter und Tunnelbarrieren getrennt sind, wird ein Ansatz in einem Finite-Elemente-basierten Rahmenwerk übernommen und implementiert, der auf der Lösung gekoppelter Spin- und Ladungstransportgleichungen zur Bestimmung der Nichtgleichgewichtsspinakkumulation basiert ist. Der Formalismus wird erweitert, um den Transport durch magnetische Tunnelkontakte zu beschreiben. Der magnetoresistive Tunneleffekt (TMR) wird miteinbezogen, indem die Tunnelbarriere als schlechter Leiter modelliert wird, deren Leitfähigkeit lokal abhängig von der relativen Magnetisierungsorientierung der ferromagnetischen Schichten ist. Die Polarisation des Tunnelspinstroms wird über geeignete Randbedingungen an der Grenzfläche zwischen Oxid und ferromagnetischer Schicht berücksichtigt. Der vorgeschlagene Ansatz reproduziert Winkel- und Spannungsabhängigkeiten des in MTJs ausgeübten Drehmoments erfolgreich und ermöglicht die Bewertung des Zusammenspiels zwischen die Drehmomentquellen der Grenzfläche und des Bulks. Der Ansatz wird auf Schaltsimulationen und das Design ultraskalierter STT-MRAM-Zellen erfolgreich angewendet.

Sommario

La non volatilità è una soluzione emergente per le perdite di potenza in stand-by causate dal ridimensionamento dei tradizionali componenti a semiconduttore. La memoria ad accesso casuale magnetoresistiva a torsione da trasferimento di spin (STT-MRAM) è un valido candidato non volatile, grazie alla sua struttura semplice e alla compatibilità con la tecnologia CMOS. Possiede alta velocità ed eccellente durabilità, risultando quindi promettente per applicazioni che vanno dall’IoT e dagli usi automobilistici alla DRAM incorporata e alle cache di ultimo livello. Strumenti di simulazione accurati offrono informazioni preziose nella progettazione di dispositivi STT-MRAM. La descrizione dell’evoluzione temporale della magnetizzazione è fornita dall’equazione di Landau-Lifshitz-Gilbert (LLG). L’equazione LLG può essere arricchita con termini che descrivono la torsione generata da una corrente elettrica che scorre attraverso il dispositivo, responsabile delle operazioni di STT-MRAM. Questa tesi si concentra sullo sviluppo e la calibrazione di approcci di simulazione per descrivere le fonti di torsione che agiscono nei dispositivi STT-MRAM.

In primo luogo, un risolutore ad uso interno basato sul metodo delle differenze finite, con il contributo di STT incluso utilizzando l’espressione di Slonczewski, è generalizzato per eseguire simulazioni di switching di una cella MRAM. Il risolutore è aggiornato per consentire l’esecuzione di simulazioni di switching con tre diversi approcci: con densità di corrente uniforme e costante, con una più realistica densità di corrente non uniforme e corrente totale costante, e con tensione costante. La validità della descrizione con la densità di corrente fissa per la stima del tempo di switching è verificata confrontando i suoi risultati con gli altri due approcci. È inoltre valutata la dipendenza del tempo di switching da diversi parametri del sistema, dimostrando che una correzione sul valore della corrente applicata permette di ottenere risultati compatibili con tutti e tre gli approcci.

Per valutare con precisione la torsione, andando oltre l’approssimazione di Slonczewski all’interfaccia, in strutture stratificate composite costituite da diverse sezioni ferromagnetiche separate da distanziatori metallici non magnetici e barriere a effeto tunnel, è utilizzato un approccio basato sulla soluzione delle equazioni di trasporto di spin e carica per determinare l’accumulo di spin, implementato in un framework basato sul metodo agli elementi finiti. Il formalismo è esteso per descrivere il trasporto attraverso giunzioni a effetto tunnel magnetiche. L’effetto della tunnel magnetoresistance (TMR) è incluso modellando la barriera ad effetto tunnel come un cattivo conduttore, la cui conduttività dipende localmente dall’orientazione relativa della magnetizzazione negli strati ferromagnetici. La polarizzazione della corrente di spin trasmessa per effetto tunnel è inclusa tramite opportune condizioni al contorno all’interfaccia tra ossido e strati ferromagnetici. L’approccio proposto riproduce con successo la dipendenza, sia angolare che dalla tensione applicata, della torsione presente nelle MTJ, e consente di valutare l’interazione tra le sorgenti di torsione all’interfaccia e nel bulk. L’approccio è applicatio con successo a simulazioni di switching e design di celle STT-MRAM ultra ridimensionate.

Acknowledgement

I would like to thank Dr. Viktor Sverdlov for the great opportunity to pursue a doctorate under his supervision, for his patience, guidance and support throughout my stay at the Institute of Microelectronics, and for his very successful approach to unite the demands of industry and research in the Christian Doppler Laboratory for Nonvolatile Magnetoresistive Memory and Logic.

I would like to express my gratitude to Prof. Siegfried Selberherr, who founded the Institute more than 30 years ago, for providing his full support and guidance on all my efforts and for always maintaining an excellent working environment.

I thank my colleagues of the Christian Doppler Laboratory for Nonvolatile Magnetoresistive Memory and Logic Roberto, Johannes, Tomáš, Nils, Mario and Wilton for their friendship, help, and for the always interesting discussions. Furthermore, I would like to also thank Wolfgang Goes for the successful collaboration with our Laboratory.

I thank all the staff of the Institute for Microelectronics for making the years together a very enjoyable experience. In particular, I would like to thank Diana for the constant support on all bureaucratic matters, and Cerv for the help on fixing issues with my computer.

I would like to thank my family, for their unfaltering support during all my studies which allowed me to be where I am now, and all my friends, for always being there for me.

Finally, I would like to give a special thanks to Valeria, my beloved wife, for her constant support and encouragement, especially during the long pandemic months. Without her, everything would have been much harder.

Contents

List of Abbreviations

.
RAM

Random-access memory

SRAM

Static random-access memory

DRAM

Dynamic random-access memory

MRAM

Magnetoresistive random-access memory

STT

Spin-transfer torque

MTJ

Magnetic tunnel junction

FM

Ferromagnetic

NM

Nonmagnetic

AFM

Antiferromagnetic

TB

Tunnel barrier

RL

Reference layer

FL

Free layer

P

Parallel

AP

Anti-parallel

LLG

Landau-Lifshitz-Gilbert

GMR

Giant magnetoresistance

TMR

Tunnel magnetoresistance

FE

Finite element

FD

Finite difference

PL

Pinning layer

NMS

Nonmagnetic and conductive spacer

AMR

Anisotropic magnetoresistance

DOS

Density of states

pMTJ

Perpendicular magnetic tunnel junction

DD

Drift-Diffusion

PDE

Partial differential equation

DL

Damping-like

FL

Field-like

ZL

Zhang-Li


List of Symbols

.

\(M_\text {S}\) (A/m)

Saturation magnetization

\(\mu _0\) (N/A\(^2\))

Vacuum permeability

\(\gamma \) (rad/(T s))

Gyromagnetic ratio

\(e\) (C)

Elementary charge

\(h\) (J s)

Planck constant

\(\hbar \) (J s)

Reduced Planck constant

\(\mu _B\) (J/T)

Bohr magneton

\(g\)

Electron g-factor

\(\alpha \)

Gilbert damping constant

\(\eta \)

Spin-transfer torque efficiency

\(\vb {m}\)

Unit magnetization vector

\(\gamma _0 = -\gamma \,\mu _0\) (m/(A s))

Rescaled gyromagnetic ratio

\(A\) (J/m)

Exchange coefficient

\(K\) (J/m\(^3\))

Anisotropy coefficient

\(K_\text {int}\) (J/m\(^2\))

Interface anisotropy coefficient

\(\vb {H}_\text {eff}\) (A/m)

Effective magnetic field vector

\(\vb {J}_\text {C}\) (A/m\(^2\))

Charge current vector

\(I_\text {C}\) (A)

Total charge current

\(\vb {T}_\text {S}\) (A/(m s))

Spin torque vector

\(P\)

Slonczewski polarization factor

\(\vb {E}\) (V/m)

Electric field vector

\(\sigma \) (S/m)

Electrical conductivity

\(G\) (S)

Elctrical conductance

\(R\) (\(\Omega \))

Electrical resistance

\(V\) (V)

Electrical potential

\(\vb {S}\) (A/m)

Spin accumulation vector

\(D_\text {e}\) (m\(^2\)/s)

Electron diffusion coefficient

\(\lambda _\text {J}\) (m)

Exchange length

\(\lambda _{\varphi }\) (m)

Dephasing length

\(\beta _\sigma \)

Conductivity polarization

\(\beta _\text {D}\)

Diffusion polarization


.

\(\tilde {\vb {J}}_\text {S}\) (A/s)

Spin current polarization density tensor

\(\lambda _\text {sf}\) (m)

Spin-flip length

\(\lambda \) (m)

Momentum relaxation length

\(D_\text {S}\) (m\(^2\)/s)

Diffusion coefficient in the TB

\(a\)

In-plane torque reduction

\(P^\eta \)

Out-of-plane polarization

List of Figures

List of Tables