Advanced magnetic tunnel junctions based on CoFeB/MgO interfacial perpendicular anisotropy

Acknowledgement: This work was supported by the project "Research and Development of Ultra-low Power Spintronics-based VLSIs“ under the FIRST Program of JSPS.

SEMATECH 8th International Symposium on Advanced Gate Stack Technology, Session 4: Spin Transfer Torque Memory Bolton Landing, NY 20, October 2011

S. Ikeda1,2 M. Yamanouchi1, H. Sato1, K. Miura3,1,2, K. Mizunuma2, H. Yamamoto3, R. Koizumi2,

H. D. Gan1, S. Kanai2, J. Hayakawa3, F. Matsukura1,2, H. Ohno1,2 1 Center for Spintronics Integrated Systems (CSIS), Tohoku Univ.

2 Laboratory for Nanoelectronics and Spintronics, RIEC, Tohoku Univ. 3 Central Research Laboratory, Hitachi, Ltd.

http://www.csis.tohoku.ac.jp

1

Spintronics devices

2

Nonvolatility High speed operation Virtually unlimited write endurance 1n 10n 100n 1µ 10µ

105

1010

1015

1020

Write time (s)

Flash

FeRAM

PRAM

SRAM/DRAM(Volatile)

MRAM/SPRAM

Num

ber o

f writ

e cy

cles

Magnetoresistive RAMs (MRAMs) SPin transfer torque RAMs (SPRAMs)

Logic-in-memory architecture Reduction of leak current (static power) Reduction of interconnection delay

Memory

Processor core

MTJCMOS

Mochizuki et al., IEICE Transactions on Fundamentals, E88-A (2005) 1408. Matsunaga et al., Appl. Phys. Express, 1 (2008) 091301; ibid., 2 (2009) 023004. Sekikawa et al., IEDM 2008, Suzuki et al., VLSI Technology 2009 Ohno et al., IEDM 2010 Matsunaga et al. VLSI Circuit 2011

(Expectation as low power consumption devices)

MTJWord line

Gate

Bit line

DrainSource

Free

PinnedBarrier

Recording

Reference

1998 2000 2002 2004 2006 2008 2010 2012 2014

Cap

acity

(bits

/chi

p)

Year

MRAM (demo) MRAM (product) SPRAM (i-MTJ)SPRAM (p-MTJ)

1G

1M

10M

100M

100k

10k

1k

100

IBM

Freescale(Motorola)

Sony

Freescale

Samsung

Sony

Toshiba-NEC

Toshiba-NECIBM-Infineon

Everspin

Hitachi-Tohoku Univ.

Cypress

TDK

NTHU-ITRI

ITRI-TSMC-NTHU

Everspin

NECToshiba Hynix-Grandis

MRAM development

2Mb SPRAM ISSCC 2007

Kawahara et al., IEEE J. Solid-State Circuits 43 (2008) 109.

Cross section SEM image

of SPRAM cell

MTJ cells

M3

M2

M1CMOS-Tr.

M4

50 nm

Recording layer

MTJ cell

Lower electrode

pinned layerAnti-ferromagnetic layer

MgObarrier

32Mb SPRAM

VLSI Circuit 2009 Takemura et al., IEEE J. Solid-State Circuits 45 (2010) 869.

Ikeda et al., IEEE Trans. Electron Devices, 54 (2007) 991.

SA & Write

128kbArray

Word D

river/Dec.

Control CircuitsPre-Decoder

Y-Dec.

Magnetic field

Spin transfer torque

3

Nonvolatile circuits for logic-in-memory

Matsunaga, …Hanyu et al., Appl. Phys. Express 1 (2008) 091301.

MTJ-Based Nonvolatile Full Adder

Matsunaga, ... Hanyu et al., Appl. Phys. Express 2 (2009) 023004.

MTJ-Based Nonvolatile Ternary Content- Addressable Memory (TCAM) cell

MTJ-Based Nonvolatile Lookup-Table Circuit Chip

Suzuki, …Hanyu,et al., VLSI Circuit Symposium 2009.

2-kbTCAM

cell array

Column Dec.

Row

Dec

.

Sens

e A

mp.

174 µm

2kb-nonvolatile TCAM chip

Matsunaga, …Hanyu,et al., VLSI Circuit Symposium 2011.

4

Technology issues toward realization of nonvolatile VLISs

High output (TMR ratio>100%)

Low switching current (IC < F μA)

Thermal stability for nonvolatility (E/kBT > 40) Annealing stability in back-end-of-line process (Ta > 350oC)

+ Scalability

To realize nonvolatile VLISs (MRAM and logic-in-memory) using the leading edge technology node, there are still issues to be addressed. ISSUES

• MTJs have to satisfy these requirements with scalability at the same time. • In order to satisfy these requirements, we focus on the MTJs with

perpendicular anisotropy electrodes.

5

Progress of MTJs CoFeB

MgO CoFeB

bcc (001)

rock salt (001)

bcc(001)

0

100

200

300

400

500

600

700

1994 1996 1998 2000 2002 2004 2006 2008 2010

トンネル

磁気抵抗比

(%)

年

AlOx-barrier MTJs

CanonANELVA & AIST

AIST

Tohoku Univ. & Hitachi

604%@RT(1144%@5K)

IBMAIST

AIST

TMR

ratio

(%)

Tohoku Univ.

MgO-barrier MTJs

Year

S. Ikeda et. al., Appl. Phys. Lett. 93, 082508 (2008). Nat. Mat., 9, 721 (2010)

SiO2/Si sub.

Ta(5) Ru(10) Ta(5)

CoFeB(5 ) →

MgO(2.1) CoFeB(6) →

Ta(5) Ru (5) Cr/Au

Cr/Au

SiO2/Si sub.

Ta(5) Ru(10) Ta(5)

CoFeB(1 ) ↑ MgO(0.9)

CoFeB(1.7) ↑ Ta(5) Ru (5)

Cr/Au

In-plane anisotropy pseudo-SV

Perpendicular anisotropy pseudo-SV

6

M-H curves of CoFeB/MgO stack samples with different tCoFeB

-1

0

1

-0.5 0.0 0.5 1.0

-1

0

1

tCoFeB = 2.0 nm

in-plane out-of-plane

M

(T)

tCoFeB = 1.3 nm

µ0H (T)

In-plane anisotropy

Perpendicular anisotropy

Ta=300oC

Ta=300oC

CoFeB(tCoFeB ) MgO(1)

SiO2/Si sub. Ta(5)

Ru(10) Ta(5)

µ0Hs_in-plane = 0.34 T Ms = 1.58 T K = 2.1x105 J/m3

S. Ikeda et al., Nat. Mat. 9 (2010) 721.

Ta=300 oC, 4 kOe, 1h

7

0 1 2-1

0

1

K

⋅ tCo

FeB (

mJ/

m2 )

tCoFeB (nm)

VSMFMR

in-plane

Perpendicular

tCoFeB dependence of KtCoFeB in CoFeB/MgO stack

CoFeB(tCoFeB ) MgO(1)

SiO2/Si sub. Ta(5)

Ru(10) Ta(5)

KtCoFeB = (Kb-MS2/2µ0) tCoFeB+Ki

•From the y-intercept, Ki = 1.3 mJ/m2. •The CoFeB/MgO interfacial anisotropy Ki is dominant in the perpendicular anisotropy because Kb is negligible.

Ta = 300oC

S. Ikeda et al., Nat. Mat. 9 (2010) 721. 8

tMgO dependence of Ms and K

1 monolayer of MgO is ~0.21 nm

More than 3 monolayers of MgO is required to stabilize the perpendicular anisotropy induced by CoFeB-MgO interface.

Saturation magnetization Ms and effective magnetic anisotropy energy density K are evaluated from the out-of-plane M-H curves.

0.0

0.4

0.8

1.2

0.0 0.5 1.0 1.5 2.00.00.51.01.52.0

Ms (

T)

K (1

05 Jm

-3)

tMgO (nm)

tMgO = 0.6 nm

M. Yamanouchi et al., J. Appl. Phys. 109 (2011) 07C712.

0.42 nm

Mg O

9

anti-bonding

bonding

O 2pZ

σ-coupling

23z

dzp2O Fe

Fe 3dz2 (m=0)

K. Nakamura et al., Phys. Rev. B, 81 (2010) 220409. R. Shimabukuro et al., Physica E 42 (2010) 1014.

23z

d : in-plane anisotropy

(m=0)

Possible factor of the perpendicular anisotropy in CoFeB/ MgO stack

• 3dz2 band of Fe is pushed up above the Fermi energy.

• The contribution of 3dz2 orbital of Fe becomes small,

resulting in appearance of perpendicular anisotropy.

10

TMR properties in CoFeB-MgO p-MTJs

-0.2 -0.1 0.0 0.1 0.210

20

30

40

50

R (k

Ω)

µ0H (T)

Ta = 300oC

Clear hysteresis -> top and bottom CoFeB electrodes have perpendicular anisotropy.

SiO2/Si sub.

Ta(5)Ru(10)Ta(5)

CoFeB(0.9 )

MgO(0.9)CoFeB(1.5 )

Ta(5)Ru (5)Cr/Au

Al2O3

V-

V+

(nm)

40 nm

Parallel

Antiparallel

S. Ikeda et al., Nat. Mat. 9 (2010) 721. 11

Junction size dependence of Ic0

Ic0 linearly decreases with junction area S (volume Vrec). We confirmed scalability of switching current for spin transfer torque. 50 nm

SiO2/Si sub.

Ta(5) Ru(10) Ta(5)

CoFeB(0.9 )

MgO(0.9) CoFeB(1.5)

Ta(5) Ru (5) Cr/Au

F = 40 ~ 76nmφ Ta=300oC w/o Hex

0 1000 2000 3000 4000 5000 6000-100

-50

0

50

100

150

200

IC0 (P to AP) IC0 (AP to P)

Critic

al c

urre

nt, I

C0 (µ

A)

Recording layer volume, Vrec (nm3)

76 nm70 nm59 nm

44 nm53 nm

Junction size

= S・tCoFeB

H. Sato et al., Appl. Phys. Lett. 99 (2011) 042501. 12

Junction size dependence of E/kBT

0 1000 2000 3000 4000 5000 60000

20

40

60

80

pulse current experiment pulse field experiment previous report

Ther

mal

sta

bility

fact

or, E

/kBT

Recording layer volume, Vrec (nm3)

76 nm70 nm59 nm

44 nm53 nm

Junction size

E/kBT maintained almost constant values even though the junction diameter was varied from 40 nm to 80. IC0 reduces in proportion to the volume of the recording layer but the E/kBT values are not affected much by the volume down to 40 nm in diameter.

= S・t

H. Sato et al., Appl. Phys. Lett. 99 (2011) 042501. 13

Origin of the junction size dependence of IC0 and E/kBT

Nucleation embryo

Nucleation type magnetization reversal

0 20 40 60 80 1000

20

40

60

80

Effe

ctive

diam

eter

, Dn (

nm)

Junction size in diameter (nm)

No clear reduction in E/kBT with different junction size

Dn showed almost constant values of 40 nm in diameter, which is almost independent of junction size.

Nucleation diameter: Dn

IC0=JC0S=JC0(π(D2-Dn2)/4)+π(Dn/2) 2)∝S

E/kBT=Keffπ(Dn/2) 2/kBT ∝Dn2

Junction area (diameter): S (D)

Nuc

leat

ion

diam

eter

, Dn (

nm)

H. Sato et al., Appl. Phys. Lett. 99 (2011) 042501. 14

Jc0 and E/kBT of MTJs with 40 nm diameter

40 nm Ta(oC) TMR ratio (%)

RA (Ωµm2)

Jc0 (MA/cm2) E/kBT

300 124 18 3.9 43.1 350 113 16 3.8 39.1

w/o Hex

w/o Hex

Ta=300oC

Ta=350oC

S. Ikeda et al., Nat. Mat. 9, (2010) 721. K. Miura et al., MMM 2010, HC-02.

•Jc0 and E/kBT are maintained after annealing at 350oC. •This MTJ system has a back end of the line (BEOL) compatibility.

SiO2/Si sub.

Ta(5)Ru(10)Ta(5)

CoFeB(1.0 )MgO(0.85)CoFeB(1.7)

Ta(5)Ru (5)Cr/Au

Al2O3 Al2O3

15

Comparison of MTJs Type Stack structure

(nm) Size (nm)

MR (%)

RA (Ωμm2)

JC0 (MA/cm2)

IC0 (μA) Δ=E/kBT IC0/Δ Ta (oC) Ref.

i-MTJ CoFeB(2)/Ru(0.65)/CoFeB(1.8) SyF 100x200 >130 ~10 2 ~400 65 ~6.2 300-

350 J. Hayakawa et al., IEEE

T-Magn., 44, 1962 (2008)

p-MTJ L10-

FePt(10)/Fe(t)/Mg(0.4)/ MgO(1.5)/L10-FePt(t)

Blanket 120 (CIPT)

11.8k - - - - 500 M. Yoshizawa et al.,

IEEE T-Magn., 44, 2573 (2008)

p-MTJ L10-

FePt/CoFeB/MgO(1.5)/ CoFeB/Co based

superlattice

Blanket 202 (CIPT)

- - - - - - H. Yoda et al.,

Magnetics Jpn. 5, 184 (2010) [in Japanese].

p-MTJ [Co/Pt]CoFeB/CoFe/MgOCoFe/CoFeB/TbFeCo Blanket 85-97

(CIPT) 4.4-10 - - - - 225 K. Yakushiji et al., APEX

3, 053033 (2010)

p-MTJ [CoFe/Pd]/CoFeB/MgO/CoFeB/[CoFe/Pd]

800x800N

100 (113)

18.7k (20.2k) - - - - 350

(325) K. Mizunuma et al.,

MMM&INTERMAG2010

p-MTJ CoFeB (1)/ TbCoFe (3) 130 φ ~15 4.7 650 107 6.08 - M. Nakayama et al., APL 103, 07A710 (2008)

p-MTJ L10-alloy 50-55 φ - - - 49 56 0.88 - T. Kishi et al., IEDM 2008

p-MTJ Fe based L10 (2)/CoFeB (0.5) - - - - 9 - - -

H.Yoda et al., Magnetics Jpn. 5, 184 (2010) [in

Japanese].

p-MTJ CoFeB/MgO/CoFeB 40 φ 113

(124) 16

(18) 3.8

(3.9) 48

(49) 39

(43) 1.23 (1.14)

350 (300)

S. Ikeda et al., Nat. Mat. 9 (2010) 721.

K. Miura et al., MMM2010,HC-02

These p-MTJ technologies will be a promising building block for nonvolatile VLSIs using spin-transfer torque switching.

16

Issue for p-MTJs

Slide 17

Parallel state Low resistance

Antiparallel state High resistance

Different polarity =Stable

The same polarity =Unstable

Free

Reference

N

S

N

S

N

S

N

S

The thermal stability ∆p=E/kBT of anti-parallel state in p-MTJs becomes low by comparison with parallel state.

Bit Information “0” Bit Information “1”

17

Enhancement of thermal stability ∆P, AP = ∆ [1±Hs/Hc0]2

Free

Reference

Hs

E/kBT

“1” “0” ∆AP ∆P

Hs decreases by employing step structure with large reference layer. ∆AP increases with decreasing diameter of reference layer.

Hs “1” “0”

∆AP ∆P

E/kBT Free

Reference

Stepped structure

Conventional structure

dipole interlayer coupling field

18

Two types of p-MTJs

Slide 19

100 nm

F = 100 nm

CoFeB

CoFeB

MgO

F = 100 nm

300 nm

CoFeB

CoFeB

MgO

100 nm

Stepped structure Conventional structure

F : Feature size

Free layer

Reference layer

K. Miura et al., 2011 VLSI Technology, 11B-3.

1.5 nm 0.9 nm 0.9 nm

1.5 nm 0.9 nm 0.9 nm

19

Hs in two types of p-MTJs

Slide 20

Hs

Stepped structure Conventional structure

Hs

Conventional structure Stepped structure

TMR ratio (%) 100 97 RA (Ωµm2) 13 13

Hs (mT) 22 5

1

2

3

4

-100 -50 0 50 100

Res

ista

nce

(kΩ

)

Magnetic field (mT)

1

2

3

4

-100 -50 0 50 100

Res

ista

nce

(kΩ

)

Magnetic field (mT)

Hs can be reduced by using stepped structure. K. Miura et al., VLSI Technology 2011, 11B-3. 20

∆P and ∆AP in two types of p-MTJs

∆AP in stepped structure increases.

Conventional structure Stepped structure ∆P 71.2 72.9 ∆AP 46.5 70.1

Stepped structure Conventional structure “P”->”AP” “AP”->”P” “P”->”AP” “AP”->”P”

00.20.40.60.8

1

-150 -75 0 75 150

Prob

abili

ty

Magnetic field (mT)

00.20.40.60.8

1

-150 -75 0 75 150

Prob

abili

ty

Magnetic field (mT)

∆P,AP

K. Miura et al., VLSI Technology 2011, 11B-3. 21

Cell area of stepped structure

Cell area of 0.09 µm2 (300nmφ) in the stepped structure corresponds to SRAM cell area at 32 nm technology node. Cell area can be down to 0.04 µm2 (200nmφ) without degrading the retention time over 10 years, which corresponds to SRAM cell area at 20 nm technology node.

0.01

0.1

1

10

0 50 100 150

Cel

l are

a (µ

m2 )

Technology node (nm)

32 nm 300 nm

CoFeB

CoFeB

MgO

F = 100 nm

20 nm

SRAM 100F2 (F: feature size)

F = 100 nm

200 nm

CoFeB

CoFeB

MgO

K. Miura et al., VLSI Technology 2011, 11B-3. 22

Summary We have demonstrated that the critical current Ic0 in CoFeB/MgO

perpendicular anisotropy MTJs (p-MTJs) can be scaled down with decreasing recording layer volume.

The thermal stability factor E/kBT can be maintained at about 40 even though the recording volume was reduced to 40 nm. The magnetization reversal in CoFeB/MgO p-MTJs is dominated by nucleation type magnetization reversal (nucleation diameter ~40 nm).

CoFeB/MgO p-MTJs show the high TMR ratio of more than 100%, high thermal stability at dimension as low as 40 nm diameter and a low switching current of 49 μA at the same time .

CoFeB/MgO p-MTJ with step structure shows the enhancement of thermal stability in antiparallel state, which achieves thermal stability for data retention time over 10 years.

23