Magnetoresistance and Spin-Transfer Torque in Magnetic Tunnel Junctions

Magnetoresistance and spin-transfer torque in magnetic tunnel junctions

S. Yuasa

IEEE Distinguished Lecturer 2012

E-mail: [email protected]

Collaborations

Toshiba Corp. (H. Yoda) (STT-MRAM)

Canon Anelva Corp. (sputtering deposition process)

Osaka University (Y. Suzuki) (rf & high-speed experiments)

CNRS/Thales(A.Fert, V.Cros, J.Grollier) (spin-torque oscillator) Outline

(1) Spintronics

(2) Tunnel magnetoresistance (TMR) ・Magnetoresistance ・Tunnel magnetoresistance in magnetic tunnel junction (MTJ) ・Giant TMR in MgO-based MTJ ・CoFeB/ MgO/ CoFeB structure for device applications

(3) Spin-transfer torque (STT) ・Physics of spin-transfer torque ・Spin-transfer torque MRAM (STT-RAM or Spin-RAM) ・Microwave applications non-volatile Hard Disk Drive (HDD) (HDD) Drive Disk Hard magnetic recording recording magnetic

Magnetics ・ (amagnet) small ? of of Spin N spin and e - spintronics STT TMR Electron & S charge new functions.

Spintronics electrons are utilized for utilized are electrons Both Both What is What is Electric Charge power amplification operation logic basically volatile basically

Electronics ・・ Transistor, LSI LSI Transistor, Difference between conventional magnetics and spintronics ?

Magnetics Coupling between charge and spin by induction coil Most of the energy is wasted.

N Spintronics Coupling between charge and spin -e by quantum mechanical effects (e.g. tunnel magnetoresistance, spin-transfer torque) S Highly efficient ! Outline

(1) Spintronics

(3) Spin-transfer torque (STT) ・Physics of spin-transfer torque ・Spin-transfer torque MRAM (STT-RAM or Spin-RAM) ・Microwave applications Magneto-Resistance (MR)

Change in electric resistance induced by magnetic field. R Magneto-resistance ratio (MR ratio)

Magnetic field H Resistance, Resistance, required to induce MR

0 Magnetic field, H

MR converts magnetic signals into electric signals. (cf. STT converts electric signals into magnetic signals.)

MR ratio at RT & a low H (~1 mT) is important for device applications. Magnetoresistance MR ratio @RT&low H Year AMR effect 1857 Lord Kelvin MR = 1 - 2%

1985 GMR effect A. Fert, P. Grünberg MR = 5 - 15% (Nobel Prize 2007) 1990

TMR effect 1995 (Al-O barrier) T.Miyazaki, J. Moodera MR = 20 - 70%

2000

2005

2010 Outline

(1) Spintronics

(3) Spin-transfer torque (STT) ・Physics of spin-transfer torque ・Spin-transfer torque MRAM (STT-RAM or Spin-RAM) ・Microwave applications Tunnel MagnetoResistance (TMR) effect in magnetic tunnel junction (MT J)

Ferromagnetic electrode e e e e e e Insulator (nm –thick) (tunnel barrier) Ferromagnetic electrode Parallel (P) state Antiparallel (AP) state

Resistance RP : low Resistance RAP : high

MR ratio  (RAP – RP)/RP×100% (performance index)

3d ferromagnet Amorphous Al-O MR ratios of 20 – 70% at RT 3d ferromagnet Read head of hard disk drive (HDD) Sense current

Magnetic field signal Recording media Read head Head Medium （magnetic sensor）） 2 AMR GMR TMR 1000 Higher MR ratio GMRヘッドの出現 ■ ■ is required for 100 ■ ■■ Gbit/in ■ （ ■■ > 200 Gbit/inch2. ■■■ 10 ■ ■ ■ ■ TMR head 1 ■ (amorphous ■ ■ tunnel barrier) GMR head 0.1 ■

0.01

Recording density 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 Year Non-volatile Magnetoresistive Random Access Memory (MRAM)

Digit line Bit line Bit line MTJ Parallel ：“0” or Antiparallel：“1” MTJ Write line Non-volatile Digit line magnetic memory CMOS + n+ p n CMOS

Read out Writing

Review: Science 294, 1488 (2001). Non-volatile Magnetoresistive Random Access Memory (MRAM)

Freescale (US)’s 4 Mbit – MRAM based on Al-O MTJs (volume production since 2006)

Non-volatile, high speed, write endurance > 1016

High-density MRAM is difficult to develop. Three important properties for memory device: speed, density, and write endurance

1 Write endurance > 1016 (virtually unlimited) Work Volatile SRAM memory 10 MRAM DRAM Non-volatile FeRAM Write endurance <1012 NOR 100 PRAM Storage NAND Access speed (nsec)

HDD 1000 1 Mbit 1 Gbit 1 Tbit Density (bit/chip) Three important properties for memory device: speed, density, and write endurance

1 Write endurance > 1016 High-speed Work (virtually unlimited) MRAM Volatile SRAM memory 10 MRAM DRAMGbit-scale MRAM Non-volatile FeRAM Write endurance <1012 NOR 100 PRAM Storage NAND Access speed (nsec)

HDD 1000 1 Mbit 1 Gbit 1 Tbit Density (bit/chip) How can we develop Gbit-scale MRAM ?

Major requirement for Gbit-MRAM ・Large output voltage MR ratio >> 100%  Breakthrough ・Reduction of writing power  spin-torque switching

1G DRAM

1M MRAM Capacity of MRAM cannot

Capacity (Capacity bit ) go beyond 64 Mbit.

2000 2002 2004 2006 2008 Year MR effects Device applications MR ratio (RT & low H) Year AMR effect HDD head 1857 MR = 1 ~ 2 % Inductive head 1985 GMR effect MR = 5 ~ 15 % 1990

MR head TMR effect 1995 (Al-O barrier) MR = 20 ~ 70 % GMR head 2000 Memory

TMR head 2005 MRAM

Much higher MR ratio is required 2010 for next-generation devices. Simple model for TMR effect : Julliere’s model

Tunnel Tunnel FM 1 Barrier FM 2 FM 1Barrier FM 2

e e

DOS DOS DOS DOS

E F EF EF EF

D D1 D2 2 D D1 D D D1  1 2 2

P alignment AP alignment

2 ↑ ↓ ≡ , ,1,2 1 ↑ ↓ Spin Polarization, P How can we attain giant MR ratio ? (1) Electrode material with full spin-polarization

Energy |P| = 1

EF “Half Metal”

(E.g.) some Heusler alloys, Fe3O4, CrO2, LaSrMnO3 perovskite

Room-temperature MR ratios for half-metal electrodes have never exceeded those for simple 3d alloys such as Co-Fe.

(2) Crystalline tunnel barrier such as MgO(001) Outline

(1) Spintronics

(3) Spin-transfer torque (STT) ・Physics of spin-transfer torque ・Spin-transfer torque MRAM (STT-RAM or Spin-RAM) ・Microwave applications Amorphous Al-O barrier vs. crystal MgO barrier (theory)

Amorphous Al-O barrier Crystal MgO(001) barrier (conventional MTJ)  2  2 5 5 Fe(001) 1 Fe(001) 1

Amorphous MgO(001) 1 Al-O

Fe(001) 1

Incoherent tunneling of Dominant tunneling of fully various Bloch states. spin-polarized 1 Bloch states

MR < 100% at RT MR >> 1000% (theory) Butler et al. PRB 2001; Mathon ibid 2001. What we learn about “tunneling effect” at an undergraduate course

Barrier Potential barrier is width, t assumed to be vacuum. Free-electron wave exp(ik·z ) Barrier height 

Electrode Tunnel Electrode barrier

Exponential decay of DOS

Tunneling transmittance (T) decays exponentially as a function of t. (WKB approx.)  2× T exp( 8m/  t) m: effective mass Realistic tunneling effect

Conduction Bloch states band  Band EF gap

Valence band Electrode Barrier Electrode

Evanescent states in the band gap

Both Bloch states and evanescent states have (i) specific orbital symmetry & (ii) specific band dispersion. (complex wave vector k = kr + i )

Bloch states and evanescent states couple at interface. Decay of evanescent state largely depends on orbital symmetry.

1 (spd)

5 (pd) Fe MgO Fe

2’ (d)

Butler (2001). 1 : s + pz + d 2 z 2 – x 2 –y 2 Coupling between Bloch states and evanescent states

Ideal coherent tunneling for k// = 0 direction

Fe(001) MgO(001) Fe(001)

1 1  1 k

k// 5 5 5

2 2’ 2 Fully spin-polarized 1 band in bcc Fe(001) 1.5

1 majority spin 1.0 minority spin  ( eV ( eV ) 0.5 1 F E -

E 0.0 EF

-0.5  (001) direction H

Fully spin-polarized 1 band Giant MR ratio is theoretically expected. Not only bcc Fe but also many other bcc alloys based

on Fe or Co have fully spin-polarized 1 band. (e.g. bcc Fe1-xCox , some Heusler alloys) Importance of interface (theory)

X.-G. Zhang, et al., PRB 68, 092402 (2003).

Excess O atom k

LDOS of  states k// 1

Ideal interface Oxidized interface

Fe- 1 states couple with Fe- 1 states do not couple MgO-1 states at k// . with MgO-1 states at k// = 0. MR ratio > 1000% MR ratio < 100% Fully epitaxial Fe/MgO/Fe MTJ grown by MBE

Yuasa et al., Nature Mater. 3, 868 (2004).

Fe(001) (Pinned layer)

MgO(001)

Fe(001) (Free layer)

2 nm TEM image Experimental demonstrations of giant TMR

Single-crystal MgO(001) barrier Textured MgO(001) barrier

300 180 T = 20 K 160 MR = 247% 140 200 120 T = 293 K 100 MR = 180% 80

100 MR ratio(%) 60 MR ratio(%) 40 20 0 0 -200 -100 0 100 200 -1000 -500 0 H (Oe) H (Oe)

Yuasa et al., Nature Mater. 3, 868 (2004). Parkin (IBM), Nature Mater. 3, 862 (2004). 600

500 “Giant TMR effect”

400 Crystal MgO(001) tunnel barrier

300 MgO(001) FeCo(001)

MR ratioat RT (%) IBM [3] 200 Textured MTJ AIST [2] MgO(001) Amorphous Al-O AIST [1] 100 tunnel barrier Fe(001) Fully epitaxial MTJ 0 1995 2000 2005 Year

[1] Yuasa, Jpn.J.Appl.Phys.43,L558 (2004). [2] Yuasa, Nature Mater.3,868 (2004). [3] Parkin, Nature Mater.3, 862 (2004). Fundamental problem on thin film growth

MTJ structure for practical applications

Free layer or MTJ Tunnel barrier

Pinned layer Ru This structure is FM (Co-Fe) based on fcc (111). AF layer (Pt-Mn or Ir-Mn) for exchange biasing

MgO(001) cannot be grown on fcc (111). 4-fold symmetry 3-fold symmetry 600 Crystal MgO(001) barrier 500 : Single crystal MgO(001) : Poly-crystal MgO(001) 400 : CoFeB/MgO/CoFeB

300 Canon-Anelva & AIST [4] MgO(001) FeCo(001)

MR ratioat RT (%) IBM [3] 200 Textured MTJ AIST [2] MgO(001) AIST [1] 100 AmorphousAl-O Fe(001) tunnel barrier Fully epitaxial MTJ 0 1995 2000 2005 Year

[1] Yuasa, Jpn.J.Appl.Phys.43,L558 (2004). [2] Yuasa, Nature Mater.3,868 (2004). [3] Parkin, Nature Mater.3, 862 (2004). [4] Djayaprawira, SY, APL 86,092502 (2005). Outline

(1) Spintronics

(3) Spin-transfer torque (STT) ・Physics of spin-transfer torque ・Spin-transfer torque MRAM (STT-RAM or Spin-RAM) ・Microwave applications CoFeB/MgO/CoFeB structure for device applications Canon-Anelva, AIST Djayaprawira, SY, Appl. Phys. Lett. 86, 092502 (2005). Yuasa & Djayaprawira, J. Phys. D: Appl. Phys. 40, R337 (2007).

Annealing Amorphous Poly-crystal above 250ºC CoFeB bcc CoFeB(001)

Poly-crystal Crystal- Poly-crystal MgO(001) lization MgO(001)

Amorphous Poly-crystal 1 nm 1 nm CoFeB bcc CoFeB(001) As-grown state After annealing

MgO-MTJ (4-foldsymmetry) Core technology for device applications Practical bottom structure:fcc(111) (3-fold symmetry) Mass-manufacturing technology for HDD industry

Sputtering Sputtering

Etching Oxidation

RF DC

RF DC Sputtering Sputtering

Load lock Load lock Canon-Anelva C-7100 sputtering system  200 – 300 mm wafer All the HDD manufacturers use this type of sputtering machine for the production of HDD magnetic heads. MR effects Industrial applications MR ratio (RT & low H) Year HDD head AMR effect 1857 MR = 1 - 2 % Inductive head 1985 GMR effect MR = 5 - 15 % : commercialized 1990 MR head : R & D level TMR effect 1995 Al-O barrier MR = 20 - 70 % GMR head 2000 Memory

Giant TMR effect TMR head 2005 MgO(001) barrier MRAM Novel MR = 150 - 600 % MgO-TMR head devices 2010 STT-RAM (Spin-RAM) Microwave MgO-TMR head for ultrahigh-density HDD

MgO-TMR head

Top lead

◆Volume production since 2007. ◆700 Gbit/ inch2 achieved (5 increase). MgO MTJ ◆Applicable up to 1 – 2 Tbit / inch2. 20 nm Bottom lead ◆World market of HDD: 25 billion USD head: 5 billion USD TEM image Ultrathin CoFeB electrode can have perpendicular magnetization.

In-plane magnetization Perpendicularmagnetization

 1.5 nm or CoFeB or < 1.5 nm MgO barrier

< 1.5 nm  1.5 nm CoFeB

S.Ikeda, H. Ohno et al., Since Djayaprawira, SY, et al., Nature Mater. 9, 721 (2010). APL 86, 092502 (2005).

H  plane

) H // plane T ( M

tCoFeB = 1.3 nm

0H (T) Outline

(1) Spintronics

(3) Spin-transfer torque (STT) ・Physics of spin-transfer torque ・Spin-transfer torque MRAM (STT-RAM or Spin-RAM) ・Microwave applications Spin-transfer torque in magnetic nano-pillar

Spin-transfer torque

Spin momentumNM transfer

FM2 Free layer  S2

NM spacer layer

Local spin FM1 (pinned layer) Conduction S electron 1 I 

Theory of spin-transfer torque J. Slonczewski, J. Magn. Magn. Mater. 159, L1 (1996). L. Berger, Phys. Rev. B 54, 9353 (1996).

S 2  S2 Heff  sˆ 2 S 2  ST Iesˆ2 (sˆ2 sˆ1) FT Iesˆ2 Precession Damping Spin-transfer Pinned Free When spin torque>damping torque, Heff Damping the precession angle is amplified, and S1 S2 the free-layer moment (S2) switches. Current I sˆ 2 Precession Spin-transfer S torque S1 2

S2 Outline

(1) Spintronics

(3) Spin-transfer torque (STT) ・Physics of spin-transfer torque ・Spin-transfer torque MRAM (STT-RAM or Spin-RAM) ・Microwave applications Writing (magnetization switching) by spin-transfer torque

Ipulse Ipulse

Switching by field0.3 Switching by Spin torque 0.2 Not scalable ! Ideal for high-density MRAM cells 0.1 Switching current density

Writing power Writing 5 2 0 JC0 = 510 A/cm 10 100 1000 is required for application. Cell size （nm） Gbit-MRAM In-plane magnetization vs. Perpendicular magnetization In-plane Perpendicular magnetization magnetization

・CoFeB ・L10-ordered alloy (e.g. FePt) ・Co-Fe ・Multilayer, superlattice ・Ni-Fe ・RE-TM alloy (e.g. Tb-Co) ・HCP alloy (e.g. Co-Cr) ・ultrathin CoFeB Development of Spin-RAM

2005 Sony (IEDM 2005) - in-plane MTJ cells -4 kb

2007 Hitachi/Tohoku U. (ISSCC 2007) - in-plane MTJ cells -2 Mb

2008 Toshiba, AIST etc. (IEDM 2008) - p-MTJ cells -1 kb

2010 Toshiba (ISSCC 2010) - p-MTJ cells - 64 Mb, 65 nm Canon-Anelva C-7100 sputtering system installed at AIST

4PVD

8PVD ・RT – 350ºC ・Co-sputtering from 3 targets.

Plasma LL LL oxidation / 8PVD Etching ・RT – 250ºC ・Co-sputtering from 3 targets. DC RF New perpendicular magnetic material : superlattice film

Yakushiji, SY et al., Appl. Phys. Express3, 053003 (2010). Yakushiji, SY et al., Appl. Phys. Lett., 97, 232508 (2010).

fcc(111)-based cf. conventional magnetic multilayer superlattice film with thick Pt(Pd) layers

Co (~1 ML) Co (2-3 ML) Pt(Pd) (~1 ML) Pt(Pd) (6-8 ML) fcc(111) or hcp-c buffer

Ru cap

Pt [Co/Pt]6 Co

Ru buffer 3 nm

TEM HAADF-STEM Magnetic superlattice vs. conventional multilayer

Superlattice Conventional film multilayer film Thickness of Ultra-thin Relatively thick total stack ( < 1.2 nm possible) (> 3 nm)

Structure Artifical alloy Mutilayer

Up to 12 Merg/cc K ~ 5 Merg/cc u (tunable)

Annealing stability Very good Poor > 370ºC ~ 200ºC

Origin of perp. Magneto-crystalline Interfacial anisotropy magnetic anisotroy anisotropy Basic requirements for Gbit-scale Spin-RAM

(1)    /kBT > 60–80 for cell size < 50 nm



(2) MR ratio > 100 – 150% and low RA product

5 2 (3) Switching current density, JC0 = 510 A/cm

(4) Switching speed < 20 ns to replace DRAM < 1 – 3 ns to replace SRAM Thermal stability of MTJ,  = KuV/kBT

When cell size is smaller than 50 nm, the uniaxial shape anisotropy cannot yield  > 60. MTJ with in-plane magnetization is hopeless !

Perpendicular magnetic recording HDD 500 Gbit/inch2

TEM image of HDD media 7nm

CoPtCrO

 > 80 for the grain size < 10 nm Basic requirements for Gbit-scale Spin-RAM

(1)    /kBT > 60–80 for cell size < 50 nm

(2) MR ratio > 100 – 150% and low RA product

5 2 (3) Switching current density, JC0 = 510 A/cm

(4) Switching speed < 20 ns to replace DRAM < 1 – 3 ns to replace SRAM MRA ratio and RA product required for Gbit Spin-RAM

◆ MR > 100 – 150% is required to attain a high read-out signal (voltage) with a small read-out current.

◆ Low RA is required to satisfy the impedance matching with the pass transistor (CMOS).

The MTJ resistance should be about 10 k.

1 Gbit (F = 65 nm) RA < 30 m2, MR > 100%

5 Gbit (F = 30 nm) RA < 7 m2, MR > 100% 10 Gbit (F = 20 nm) RA < 3.5 m2, MR > 100% How to achieve high MR with perpendicular electrodes ?

(i) The 1 band of perpendicular materials are not fully spin-polarized. (ii) Lattice matching between perp. materials and MgO(001) is not good.

Insertion of CoFeB between MgO and PMA layer

PMA material PMA material CoFe/CoFeB ・Lattice mismatch MgO(001) MgO(001) ・Band mismatch CoFe/CoFeB PMA material PMA material Typical structure of p-MgO-MTJ

Yakushiji, SY et al., Appl. Phys. Express3, 053003 (2010). Yakushiji, SY et al., Appl. Phys. Lett., 97, 232508 (2010).

TbFeCo

CoFe/CoFeB MgO CoFeB/CoFe

Co/Pt

5 nm . Yes ! ? – ?

RA , MR , MR > 100%) 2 ) 2 m m   < 3.5 product ( product RA (

10 Gbit MR ratio (%) ratio MR 10 Gbit Spin-RAM Can we simultaneously attain high MR &high MR low attain we simultaneouslyCan The p-MTJs basically satisfy the requirements for requirements the satisfy basically The p-MTJs The best properties attained with in-plane magnetization

Nagamine, SY et al., Appl. Phys. Lett. 89, 162507 (2006). Maehara, SY et al., Appl. Phys. Express 4, 033002 (2011).

MR = 175% & RA =1.0 ·m2 200 with in-situ annealing of 150 MgO layer w/o in-situ (APEX 2011) annealing of MgO layer (APL 2006) 100

MR = 102%

MR ratioat RT (%) 2 50 & RA =0.9 ·m

0 0.1 1 10 RA ( ·m2 ) Basic requirements for Gbit-scale Spin-RAM

(1)    /kBT > 60–80 for cell size < 50 nm

(2) MR ratio > 100 – 150% and low RA product

5 2 (3) Switching current density, JC0 = 510 A/cm

(4) Switching speed < 20 ns to replace DRAM < 1 – 3 ns to replace SRAM Potential barrier for magnetization switching

In-plane Perpendicular

2e  damp 2 2 2e  damp Ic =  2 therm kB T  2 M S tF Ic =  2 therm kBT  g   g 

STT therm STT

Easy Axis Easy Axis therm STT ~ therm STT ~ 30therm E E

60kBT 60kBT Our latest data (Courtesy of Toshiba)

NEDO – Spintronics Non-Volatile Devices Project (Toshiba, AIST, etc.)

250 MR ratio = 200% at RT 200

150

100 MR ratio(%) 50

Electrode 0 MgO Magnetic field (a.u.) Electrode

30 nm 5 2 Jc0 = 510 A/cm  30 nm MgO-MTJ with perp. magnetization  = 45 Basic requirements for Gbit-scale Spin-RAM

(1)    /kBT > 60–80 for cell size < 50 nm

(2) MR ratio > 100 – 150% and low RA product

5 2 (3) Switching current density, JC0 = 510 A/cm

(4) Switching speed < 20 ns to replace DRAM < 1 – 3 ns to replace SRAM Demonstration of high-speed switching

Kishi, SY et al. IEDM 2008, 12.6. (Toshiba, AIST, etc.)

4 nsec

50 nm 10 nsec MTJ 30 nsec Voltage (a. u.) 01020304050 TEM image time (nsec) Write pulses

4 nsec 10 nsec

R (a. u.) u.) (a. R 30 nsec

CMOS-integrated -1 -0.5 0 0.5 1 Voltage (V) MTJ array Spin-torque writing Summary on Spin-RAM

Perp. In-plane MTJ MTJ

WRITE (Ic)

MR ratio > 100–150% READ (MR & RA) & low RA STABILITY  > 60 – 80 for MTJ size < 50 nm

SPEED < 20 ns writing

ENDURANCE > 1016 write cycles Toshiba – Hynix alliance to commercialize Spin-RAM

http://www.toshiba.co.jp/about/press/2011_07/pr1302.htm Outline

(1) Spintronics

(3) Spin-transfer torque (STT) ・Physics of spin-transfer torque ・Spin-transfer torque MRAM (STT-RAM or Spin-RAM) ・Microwave applications Steady-state precession induced by spin torque

DC current & voltage Steady-state precession Spin torque FM2 Spin (free layer) torque Steady-state precession can be induced by NM adjusting I and H. (ferromag. resonance (FMR) ) FM1 (pinned Giant MR layer) ratio ～100 nm AC current & voltage (FMR frequency)

MgO-MTJ is expected to act as a microwave oscillator and detector.

Microwave power  (MR ratio)2 Microwave functions of MgO-based MTJs

Tulapurkar, SY, Nature 438, 339 (2005). Kubota, SY et al., Nature Phys.4, 37 (2008). Deac, SY et al., Nature Physic 4, 803-809 (2008). Dussaux, SY et al.，Nature Comm. 1, 8 (2010). H. Maehara, SY et al., MMM 2010.

120

100 Power: 2.4 W Line width: 12 MHz V) V) 80  Q factor: 350 W/GHz) W/GHz)

 60

40 PSD ( dc voltage ( dc voltage 20

0 246810 Frequency (GHz) Frequency (GHz) Spin-torque diode effect Spin-torque oscillator (STO) (microwave detection) (microwave emission) Advantages of STO overスピントルク発振器の特徴 conventional microwave oscillators

Klystron Gunn diode MgO-MTJ-based STO oscillator

CoFeB MgO CoFeB

~10 cm ~1 cm 100 nm Large ! Ferromagnetic resonance  No resonance circuit is necessary.

Small (100 nm), cheap, MMIC： low efficiency easily integrated in Silicon LSI

MMIC: Monolithic Microwave IC MR effects Industrial applications MR ratio (RT & low H) Year HDD head AMR effect 1857 MR = 1 - 2 % Inductive head 1985 GMR effect MR = 5 - 15 % : commercialized 1990 MR head : R & D level TMR effect 1995 Al-O barrier MR = 20 - 70 % GMR head 2000 Memory

Giant TMR effect TMR head 2005 MgO(001) barrier MRAM Novel MR = 150 - 600 % MgO-TMR head devices 2010 STT-RAM (Spin-RAM) Microwave