1

Giant Magneto-Resistive Switches & Spin Torque Transfer Switches

friendly critic analysis

ERD "Beyond CMOS" Technology Maturity Evaluation Workshop

San Francisco, CaliforniaJuly 12, 2008

Eli YablonovitchUC Berkeley

Electrical Engineering & Computer Sciences Dept.

Si (001) Substrate

Ta 5nm

Ru 50nm

Ta 5nm

NiFe 5nm

Antiferromagnetic MnIr 8nm

CoFe 2nm

Ru 0.8nm

Ferromagnetic CoFeB 3nm

MgO 1.5nm Tunnel Barrier

Ferromagnetic CoFeB 3nm

Ta 5nm

Transpinnor Structure:

Ru 15nm

6:1 Resistance Change inTunnel Magnetoresistive (TMR)

stack[1]

Current Gate

Isignal

Magnetization

Drain

Source

InsulatorCurrent Gate

Drain

Source

Isignal

[1] Ikeda et. al., Japanese Journal of Applied Physics, Vol. 44, No 48, pp. L1442-L1445

BField

BField

Device Area 1μm2

Gate

5μA output5μA input

Complementary Transpinnor Logic

500Ωor

2.275kΩ

2.275kΩor

500Ω

+V +3mV

-V -3mV

Output Power = 1.6*10-8 WTotal Power = 2.5*10-8 W

Efficiency=65%

80

Efficiency for Complementary Transpinnor Circuit

100

90

70

60

50

40

30

20

10

010 100 1000 100001

On/Off Ratio

Efficiency (%)

Best On/Off ratio today, 4.5:1

NAND Gate:

input A

+

output

input B

Transpinnor Logic Example

-

NOR Gate:

+

input A

input Boutput

Transpinnor Logic Example

-

Si (001) Substrate

Ta 5nm

Ru 50nm

Ta 5nm

NiFe 5nm

Antiferromagnetic MnIr 8nm

CoFe 2nm

Ru 0.8nm

Ferromagnetic CoFeB 3nm

MgO 1.5nm Tunnel Barrier

Ferromagnetic CoFeB 3nm

Ta 5nm

Transpinnor Structure:

Ru 15nm

6:1 Resistance Change inTunnel Magnetoresistive (TMR)

stack[1]

Current Gate

Isignal

Magnetization

Drain

Source

InsulatorCurrent Gate

Drain

Source

Isignal

[1] Ikeda et. al., Japanese Journal of Applied Physics, Vol. 44, No 48, pp. L1442-L1445

BField

BField

Device Area 1μm2

Gate

Isignal

InsulatorCurrent Gate

BField

BField

What is the minimum current required for switching?

10nm=10-8m

Ampere's Law: H = J 2r H = I

H needs to be at least H=1 Oersted to switch a GMR deviceequivalent to B=10-4Tesla

(private communication from Stuart Parkin of IBM)(This is equivalent to saying the best relative magnetic permeability is =104

to generate and effective B=1Tesla)

I =2r H = 2r (B/o) = 2r 10-4/(4 10-7) and take r=10nm

I = 10-8 10-4/(2 10-7) Amps

I = 5 Amps are required for switching!This is really pretty good, but required very optimistic assumptions!

l

Repeater Repeater Repeaterl l

a

RC time = (clock period) /2

2R,Resistance

a

l

lorCe,Capacitanc

2

2

2

RC timeRC

a

l

a

loror

cm102F/cm102

sec10

2

periodclock612

10

o

ra

l

4800

a

laspect ratio

of wire

Physics of Wires:

<V2> = 4kT R f

Vsignal = 0.56 milli-Volts

200045

1024800

4800R,Resistance

6

2

nm

cm

aa

l

aa

l

a

lal rr ooCe,Capacitanc

4800

a

laspect ratio

of wire

<I2> = 4(kT/R) f

Isignal = 0.25 Amps

C = r o a 4800

C 7 femto-Farads

Isignal

InsulatorCurrent Gate

BField

BField

What is the minimum current required for switching?

10nm=10-8m

I = 5Amps are required for switching!This is really pretty good, but required very optimistic assumptions!

According to the previous slide, operation at 1micro-Amp implies a good noise margin ~ 48kT.

Operation at 5Amps implies 1200kT per bit function, which is at least 100 better than today's technology,

and might be worth pursuing, but it still falls 25 short

of the practical engineering limit of 48kT.

That was the Giant Magneto-Resistive Effect. What about the Spin Torque effect?

Si (001) Substrate

Ta 5nm

Ru 50nm

Ta 5nm

NiFe 5nm

Antiferromagnetic MnIr 8nm

CoFe 2nm

Ru 0.8nm

Ferromagnetic CoFeB 3nm

MgO 1.5nm Tunnel Barrier

Ferromagnetic CoFeB 3nm Magnetization

Drain

Si (001) Substrate

Ta 5nm

Ru 50nm

Ta 5nm

NiFe 5nm

Antiferromagnetic MnIr 8nm

CoFe 2nm

Ru 0.8nm

Ferromagnetic CoFeB 3nm

MgO 1.5nm Tunnel Barrier

Ferromagnetic CoFeB 3nm

Drain

Ta 5nm

Ru 15nm

Source

Magnetization is changed by literally transferring the electrons!

Take for a minimum domain size, that 1000 electrons have to be transferred.

Current I=1000e- 1.610-19Coul/10-10seconds

I=1.610-6Amps = 1.6 Amps

Slightly better than the GMR case, but not quite to the theoretical goal<1A.

But 1000e- for switching is very optimistic.Further improvements require going slow to keep the current down. Might be interesting at a clock speed <100MHz

Magnetization

Summary:

1. Giant Magneto-Resistive Effect Switch:Better than today's technology, but not quite to the level of theoretical goal.

2. Spin-Torque Switch:Slightly better than GMR Switch, and capable of achieving theoretical goal at slow clock speeds, <100MHz.

Backup Slides:

nano-transformerh

~1eV

A low-voltage technology, or an impedance matching device,needs to be invented/discovered at the Nano-scale:

transistor amplifier with steeper sub-threshold slope

photo-diode

+ ++

-

+VG

MEM's switch

Cryo-ElectronicskT/q~q/C

Cu

Cu

solid electrolyte

Electro-Chemical Switch

giant magneto-resistancespintronics

+

giant magneto-resistance

"spintronics"

These switches are made of metallic components and are of inherently low impedance

+

10μm

1μm

100nm

10nm

Moore's Law

1960 1980 2000

Crit

ical

D

imen

sion

2020 2040 2060Year

Te

chn

olo

gy G

apG

ates only

Gates including w

ires

107

105

102

1

0.1

104

106

10

108

103

Ene

rgy

per

Bit

func

tion

(kT

)

The other , for energy per bit function

Shoorideh and Yablonovitch, UCLA 2006

Transistor Measurements by Robert Chau, Intel

Recommendations:

1. Milli-Volt powering should be regarded as a Goal for future electronic switching devices.

2. There would be both an immediate power benefit, as well as a benefit at the end of the roadmap.

3. Band edge steepness is poorly known, and should be investigated for a number of semiconductors and semi-metals.

4. The full range of technology options should be included.

Moore

You?

!

Transistor

Nano-transformers

High Impedance Magnetically Loaded Transmission Line

C

q

q

kTV

C

kTV

RCRkTV

fRkTV

noise

noise

noise

noise

4

4

14

4

2

2

2

2

What about very short wires?

Johnson Noise:

,4

C

q

q

kTIf

then the signals could be large enough to be efficiently amplified.

kT

qC

2

If The Coulomb Blockade Capacitance.

For wires less than 1m, a conventional transistor amplifier configuration may be adequate.

10 atto-Farads,

mVolt1V

CqqkT4V

C

q

q

kT4V

C

1kT4V

RC

1RkT4V

fRkT4V

Volts10mVolts100

noise

2noise

2noise

2noise

2noise

//

The natural voltage range for wired communication is rather low:

The thermally activated device wants at least one electron at ~1Volt.

The wire wants1000 electrons at 1mVolt each.

(to fulfill the signal-to-noise requirement >1eV of energy)

Voltage Matching Crisis at the nano-scale!

If you ignore it the penalty will be (1Volt/1mVolt)2 = 106

The natural voltage range for a thermally activated switch like transistors is >>kT/q, eg. ~ 40kT/q

or about ~1Volt

In the future, Vdd in digital circuits will drop to 1 milli-Volt,

for communication wires.