1. Thermally Assisted MRAM How does it work ?
Cristian PAPUSOI, Bernard DIENY
SPINTEC, Grenoble, France
Acknowledgments: O.Redon, A.Astier, (LETI), J.P.Nozières (CROCUS Tech.), S.Cardoso, P.Freitas (INESC-MN Portugal)

2. Introduction EDUCATION
1993 – 1998
1987 – 1992
EDUCATION
Ph. D., Faculty of Physics, “Al.I.Cuza” University, Iasi, Romania
Diploma in Physics, Faculty of Physics, “Al.I.Cuza” University, Iasi, Romania
2006 – 2007
2004 – 2006
2003 – 2004
2000 – 2003
1992 – 2000
2006
ACADEMIC AND PROFESSIONAL EXPERIENCE
Postdoctoral Research Fellow, Grenoble, France
Investigation of RAM devices with thermal assisted switching
Postdoctoral Research Fellow, Center for Materials for Information Technology (MINT) Center, University of Alabama, Tuscaloosa, USA
Fabrication and characterization of CPP (Current Perpendicular to the Plane) spin valves
Postdoctoral Research Fellow, RWTH University,2 Physikalisches Institut A, Aachen, Germany
Role of non-magnetic defects inserted in metallic antiferromagnets on exchange bias
Postdoctoral Research Fellow, Information Storage Materials Laboratory, Toyota Technological Institute, Nagoya, Japan 
Thermal stability and recording performance of hard-disk media
Lecturer, Department of Electricity and Physical Electronics, Faculty of Physics, "Alexandru Ioan Cuza" University, 11 Blvd. Carol I, 700 506 Iasi, Romania
AWARDS
Outstanding REU student/postdoc mentor, University of Alabama (11/8/2006)

3. SPINTEC - location
SPINTEC
MINATEC
LETI

4. Main topics of research at SPINTEC
GOAL
Bridge fundamental research and advanced technology in spin electronics
Modeling
Analytical models
Micromagnetism
Finite elements
CAD tools
Basic
phenomena
GMR
TMR
Spin transfer
Functional
materials
Ferro/Antiferro coupling
Materials with perpendicular magnetization
Nanoparticles
Data storage
Patterned media
GMR/TMR sensors
Recording heads
Thermally assisted recording
Spintronics
MRAM
Microwave oscillators
Magnetic logic gates
Created January 2002 as joint CEA/CNRS laboratory affiliated to MINATEC R&D center
27 permanent staff members, 16 PhD students and 7 post-docs

5. Why MRAM ?
• Non-Volatility of FLASH
• Density competitive with DRAM
• Speed competitive with SRAM

6. Why Thermally Assisted MRAM ?
Problems in conventional MRAM
Selectivity –> difficulty in writing a single junction
Scalability –> electromigration in magnetic field lines with decreasing in-plane size
Thermal stability -> reduced life-time of written information
New approaches
1. Thermally assisted MRAM (Spintec Patent + lab. demo)
- good thermal stability ensured by exchange coupling of the storage layer with an Antiferromagnet;
- high selectivity;
- low power consumption during writing at high temperature.
2. Current induced magnetization switching
- linear decrease of power consumption with decreasing junction in-plane area
3. Possibility to integrate 1 and 2

7. Outline TA-MRAM. Definition, structure and principle of operation.
Electric characterization
Regimes of operation. Power of the electric pulse PHP vs. junction temperature TAF.
Exchange bias as a temperature probe. Electric pulse width d vs. junction temperature TAF.
Conclusions

8. Thermally Assisted MRAM (TA-MRAM) – principle
Status 0
OFF 
Writing ON
Heating
Cooling Hw
IHP Iw 
Status 1 R H
Hex
PHP = 0
Hw = 0
T = Room Temperature
Hw < 0
T  Blocking Temperature 1

9. Thermally Assisted MRAM (TA-MRAM) – structure
TMR = 30%
Vbias = 40 mV
250 nm
PtMn (20nm)
CoFe (2.5nm)
Ta (50nm)
Ru (0.8 nm)
CoFe (3nm)
NiFe (3nm)
Al (20nm)
Ta (90nm)
CoFe (1.5nm)
IrMn (5nm)
AlOX (0.5 nm)
Hard mask
(thermal barrier)
Reference layer
Tunnel barrier
Storage layer
450 nm

10. Thermally Assisted MRAM (TA-MRAM) – electric characterization
MTJ
PPB
I MR Tester H
(I>60 mA, |H|<1000 Oe)
Pulse Generator
(d>1.5 ns, UPG<3.8 V)
Switch 1 2
Trigger
OSC
Equivalent electric circuit of the MRAM device (by network analyser)
UAPPL
UMTJ
RMTJ
≈ 800 Ω
4.2 Ω
54 pH
54 pF ~
ZPG = 50 Ω
2UPG
ZOSC = 50 Ω
No amplitude attenuation (UAPPL=UMTJ) for d > 1 ns pulse width
UMTJ = 2UPG – (ZPG+ZOSC)I
I = UOSC / ZOSC
PHP = UMTJ I
What do we measure ?

11. Decrease pulse width d. Set PHP to PHP,min
Thermally Assisted MRAM (TA-MRAM) – measurement procedure
Set pulse width d to dMAX. Set pulse power PHP to PHP, MIN
How do we measure ? d
Time
PHP
HSET
Magnetic field pulse
Electric pulse
Saturate HEX (negative magnetic field pulse -HSET and heating pulse of power PHP, MAX)
Attempt to reverse HEX (positive magnetic field pulse HSET and heating pulse of power PHP)
Increase heating pulse power PHP
Decrease pulse width d. Set PHP to PHP,min
Quasi-Static MR cycle (HEX, HC, R)
HEX<HEX ?
WRT NO
YES
PHP (d)
WRT

12. Stationary regime (TAF=TRT+aPHP)
Thermally Assisted MRAM (TA-MRAM) – operation regimes
1-D model of heat diffusion in the MTJ stack
Thermal barrier 1 (cTB, rTB, dTB)
Magnetic stack (ci, ri, di)
Thermal barrier 2 (cTB, rTB, dTB)
Layer position z
Temperature T
TAF
TAF
PHP
Time
Stationary regime (TAF=TRT+aPHP)
Transient regime
c - specific heat capacity J/(Kg K)
k – thermal conductivity W/(K m)
r - density Kg/m3
d - layer thickness m

13. 120 MHz maximum writing frequency
Thermally Assisted MRAM (TA-MRAM) – transient regime
dPUMP
dPROB t
Time
PPUMP
HSET
Magnetic field pulse
PPROB
Electric pulse
tTR=2.7 ns
120 MHz maximum writing frequency
tTR=2.7 ns

14. Thermally Assisted MRAM (TA-MRAM) – stationary regime
Measurement time per point is 50 ms >> tTR -> always in stationary regime
a=105 K/W
  - measured as a function of temperature
  - measured as a function of PHP and converted into temperature according to
TAF = TRT + a PHP

15. Theory – 1D model of heat diffusion
Thermally Assisted MRAM (TA-MRAM) – consistency of results
Theory – 1D model of heat diffusion
Material r
kg/m3 c*
J/(K kg)
k**
W/(K m)
Ta*** (b)
16327
144
4.3
PtMn
12479
247
4.9
CoFe
8658
446 37 Ru
12370
239
120
AlOx
3900
900 27
IrMn
10181
316
5.7
NiFe
8694
447 Al
2700
904
235
SiO2
2200
730
1.4
rTB = kg/m3
cTB = J/(K kg)
kTB = 4.38 W/(K m) Al
Experiment
(a = 105 K/W, tTR = 2.7 ns)
*for metallic alloys calculated according to Dulong-Petit law
**for metallic alloys calculated according to Widemann-Franz law
***confirmed by electrical resistivity measurements (170 mWxcm) and XRD scan

16. 3-D simulations of heat diffusion in the MTJ stack
Thermally Assisted MRAM (TA-MRAM) – consistency of results
3-D simulations of heat diffusion in the MTJ stack
Electric pulse
(measured write power is used as input)
300 oC
200 oC
100 oC
25 oC
SiO2
electrodes
junction stack Ta
SiO2
231 nm
tTR = 2.3 ns
a = 0.84 x 105 K/W

17. Thermally Assisted MRAM (TA-MRAM) – conclusions
Two temperature regimes of the MTJ evidenced:
transient temperature regime for pulse widths d < 9 ns;
stationary temperature regime for longer pulse widths; in this regime, the relationship between the temperature of the storage layer TAF and the power of the electric pulse PHP is linear: TAF = TRT (K/W) PHP.
Use of thermal barrier layers reduces the electric power density required for writing but also decreases the writing frequency
The writing power density increases with decreasing pulse width from 0.6 mW for d = 1 s up to 6 mW for d = 2 ns; even in the range of pulse widths 1 s - 10 ns, where the storage layer reaches the stationary temperature regime by the end of the electric pulse, the writing power shows a 500 % increase.

18. Thermally Assisted MRAM (TA-MRAM) – exchange bias as temperature probe
AFM layer = collection of non-interacting single domain grains
FM layer = single domain
(-)
(+) D
mAFM q HE S
(-)
(+)

19. Thermally Assisted MRAM (TA-MRAM) – exchange bias as temperature probe
Ni80Fe20/Ir20Mn80 Ferromagnetic/Antiferromagnetic bilayer
TNeel=350 oC, a0=2.7 Å, l=0.33
JINT = 8.11x10-15 erg
KAF = 2.44x105 erg/cm3
R.White et al, J.Appl.Phys. 92, 4828 (2002)
TB = 275 oC
TB = 50 oC
Decreasing the pulse width d may require a considerable increase of the writing temperature !!!

20. Thermally Assisted MRAM (TA-MRAM) – exchange bias as temperature probe
200 nm a
TAF = TRT x103 PHP
KAF = 7.3 x 105 erg/cm3
KF = 1.3 x 104 erg/cm3
JINT = 8.5 x erg
PHP=
() 0.38 mW -> 58 oC
() 0.69 mW -> 85 oC
(▲) 0.92 mW -> 105 oC
(▼) 1.24 mW -> 133 oC
(◄) 1.47 mW -> 153 oC
(►) 1.7 mW -> 173 oC
() 2.24 mW -> 220 oC

21. Thermally Assisted MRAM (TA-MRAM) – exchange bias as temperature probe
HEX = 0 Oe
WRT
 3-D simulation of heat diffusion – experimental write power PHP (d) is used as input
WRT
Exchange bias model - temperature required to set HEX = 0 for heating time equal to the pulse duration d; temperature pulse shape is assumed rectangular.
WRT
Exchange bias model - temperature required to set HEX = 0 for heating time equal to the pulse duration d; temperature pulse shape is calculated by 3-D simulation of heat diffusion.
WRT

22. Conclusion
Writing temperature increases with decreasing pulse width d as a consequence of thermal relaxation in the Antiferromagnetic storage layer and approaches the Neel temperature in the limit d -> 0.
Exemple: writing with 2 ns pulses imply heating at about 300 oC with possible negative effects on the integrity of tunnel barrier and storage layer antiferromagnet.
Solution: decrease the writing temperature by using antiferromagnets of lower Nèel temperature than IrMn (TN ≈ 350 oC) for pinning the storage layer.