Transcript Anton Kapliy MRAM Poster 24 x 48

Magnetoresistive Random Access Memory (MRAM)
Anton Kapliy
University of Chicago
History
•1975 – M. Julierre observes weak TMR effect
under low temperatures and attributes it to
exchange splitting in ferromagnets
•1995 – J.S. Moodera uses new materials (MgO)
to achieve substantial TMR at room temperatures
•2000 – Freescale, IBM, Infineon et al initiate
programs to commerialize the technology
•2006 - present – commercial availability.
Today 1 MB costs > $1, development proceeds
Exchange Splitting in Ferromagnets
•External magnetic field produces Zeeman
splitting in the energies of spin-up and spin-down
electrons.
•In case of ferromagnets, which can maintain
their own B-field, this energy splitting can occur
without any external field. This so-called
exchange splitting can be on the order of an eV,
and explains unequal density of states of
electrons with different spins near Fermi surface:
Tunnel Magnetoresistance (TMR)
Memory Structure and Read Mode
Spin Transfer Switching
Each cell in MRAM is represented by an MTJ. Being in a parallel or
antiparallel state defines binary memory states “0” (low R) and “1” (high R).
Starting around 2006, MRAM research shifted towards the Spin Transfer
Toque technique for changing layer magnetization.
MRAM cells are organized into a 2-d grid such that each MTJ lies at the
intersection of a Word/Digit Line and a Bit Line (see image on the left). This
way we can address any cell of interest.
•Electrons pass through the fixed layer and get spin-polarized along M1
•Electrons next move through the free layer and repolarize along M2
•By angular momentum conservation, they apply a torque on electrons in the
free layer, so that their spin turns in the direction of M1
•With sufficient current, free layer becomes polarized parallel to M1
Read Mode:
•Activate the word line (WL)
•Clamp bit line (BL) to fixed voltage
•Transistor will sink a current (IR)
from collector to emitter
•Compare that current to reference
Total torque on M2 for different spin torques
Note that spin torque is proportional to current
A Magnetic Tunnel Junction (MTJ) consists of a thin insulating
layer (~1 nm, corresponding to ~10 atomic monolayers)
sandwiched between two ferromagnets (~50 angstroms). One
layer has frozen magnetization, and the other is free to change.

Classically, there is no conduction across MTJ when we apply bias voltage.
However, quantum mechanically, electrons can tunnel through the barrier.
To a good approximation, this process can be described in terms of Julierre’s
two-current model:
•Different spins tunnel independently: the probability for an electron to tunnel and
change its spin is a 2nd order effect.
•Electrons near Fermi level account for the majority of tunneling – since they are
the most energetic
•Tunneling probability is proportional to density of states near Fermi level –
according to Fermi’s Golden rule
•If two magnetizations are parallel, majority spins (↓) tunnel to majority states
and account for most of the conduction. Minority spins (↑) tunnel to minority
states and have negligible conduction.
•If two magnetizations are antiparallel, majority spins (↓) tunnel to minority
states, while minority spins (↑) tunnel to majority states. Thus, both channels are
suppressed, resulting in low overall conduction.
Mathematically, we can compute the TMR ratio as follows:
I I
R R
(n  n )
TMR 


I
R
2n n
2 P

 50%
I n n
1 P
U  m  B
I  2n n
Write Mode
The sensing current produced in read mode is not sufficient to change the
magnetization state of free layer. Instead, the state is changed as follows:
•Send a unidirectional (“hard”) current through the digit line.
•Send a bi-directional (“soft”) current through the bit line
•Turn the isolation transistor off so that no current flows through the junction
during write cycle
•Use ferromagnetic material with such coercivity that the switching doesn’t
occur when exposed to B-field from one line only.
Sketch of
state
switching
Reversal gap
between states
vs switch fields
2



2

Hysteresis curve: Remanence and Coercivity



2
2








Device model
from Freescale
2

n n
P
n n




(spin polarization)
Phase space of
switch fields
Since spin transfer method doesn’t require a strong externally-generated Bfield, it promises lower power consumption and better scalability.
Conclusions
MRAM has many benefits over alternative technologies:
•Non-volatile (draws no power when idle)
•No mechanical parts and no wear mechanism
•Low power and high density
•Much faster than Flash memory
However, it still faces many challenges:
•Requires good material uniformity (e.g., for small MTJ resistance variation)
•Close packing reduces switching barrier; prone to thermal auto-switching
•B-field from neighboring locations starts to interfere as size is reduced
•Low adoption, and as a consequence, high prices
References
1. Physics Review, Volume 168, N2 531 (1968) – Spin-disorder scattering and
magnetoresistance of magnetic semiconductors
2. J.M. De Teresa et al, Role of the Barrier in Magnetic Tunnel Junctions
3. JOM-e 52(6) (2000) (J.M. Slaughter et al) – Magnetic Tunnel Junction Materials for
Electronic Applications
4. Freescale Semiconductor – website materials (http://www.freescale.com/)
5. Industrial Embedded Systems (magazine) – Magnetic Tunnel Junction sensor
development for industrial applications
6. H. Kimura et al - A Study of Multiple-Valued MRAM Using Binary MTJ Devices
7. Wikipedia.org (to review many relevant concepts)
8. Plus half-a-dozen other academic and industrial papers/datasheets