Spin Transfer Torque Magnetoresistive Random Access Memory
49
its relative complexity to AMR sensors [16, 17, 18, 21, 22]. These devices have been
used since the mid-1990s to detect the transitions between the magnetic domains,
in which encode the information stored on hard drives. To enable further scaling of
hard disk drives, CPP spin valves are be preferred as the size of a read sensor in a
hard drive must be comparable to the bit size. It is easier to design the read/write
head with a small read sensor if the flow of current is perpendicular to the spin valves
[21, 22]. However, the signal-to-noise ratio (SNR) is still too low to be considered
for RAM applications.
The search for a higher SNR led to the discovery of tunnel magnetoresistance
(TMR) in Fe/Ge–O/Co-junctions by Michel Jullière in 1975 [23]. The TMR effect is
attributed to quantum tunneling through an insulating barrier. The reported change
in resistance was around 14% at 4.2 K, similar to the magnitude of GMR and did not
attract much attention initially. Other tunnel barriers, such as NiO and Gd 2 O 3 , had
also low TMR values at low temperatures [24, 25]. Eventually, room temperature
TMR values of 11.8 to 18% were attained when amorphous AlOx tunnel barrier was
used in conjunction with Fe, Co and CoFe electrodes in 1995 [26, 27], up till 70%
when CoFeB was used in 2004 [28]. The second major breakthrough in achieving
higher SNR occurs when bcc-MgO(100) was used as the tunnel barrier. The highest
reported TMR till date is at 604% at room temperature, or 1144% at 5 K [29], and
can theoretically reach beyond thousands of percent [30, 31, 32, 33, 34].
Although the major conceptual concerns have been addressed, a multitude of
practical designs such as scalability, integration and read/write schemes of MRAM
devices have to be considered too. The first generation of MRAM products available
on the commercial market have the easy axes of the magnetization of the free and
reference layers lying along the plane of film, otherwise known as magnetic materials
with in-plane magnetic anisotropy (IMA) [35]. The MTJ devices were patterned into
an ellipsoid shape to capitalize on shape anisotropy, inducing the magnetization to
preferentially align along the long axis of the ellipsoid, as shown in Fig. 3b. This
poses challenges to scalability of the bit density and the risk of potential anomalous switching due to unintended dipole field formation [36, 37]. In addition, the
initial conception of MRAM requires write lines to generate a local Oersted field
to drive the magnetization along a given orientation. While the MTJ devices can
scale with relative ease, these current carrying wires face the challenges in scaling
of current density and electromigration effects (~10
7 A/cm
2 for copper). Crosstalk
issue may also arise during the scaling of MTJ bits due to the generated Oersted field
unintentionally disturbing neighboring MTJ bits.
To overcome the above issue, toggling and heat-assisted switching have been
proposed and utilized by several companies [22, 38, 39, 40, 41, 42]. Nonetheless,
the added complexity of such switching mechanisms did not increase the overall
attractiveness of MRAM. The discovery of current induced switching, termed spintransfer-torque (STT), revolutionized the design of MRAM. It was discovered that a
pure spin current polarized by the reference layer can impart its angular momentum
to the soft layer, induce magnetization reversal when the minimum threshold is
exceeded. With the read and write currents now passing through the same contact
49
its relative complexity to AMR sensors [16, 17, 18, 21, 22]. These devices have been
used since the mid-1990s to detect the transitions between the magnetic domains,
in which encode the information stored on hard drives. To enable further scaling of
hard disk drives, CPP spin valves are be preferred as the size of a read sensor in a
hard drive must be comparable to the bit size. It is easier to design the read/write
head with a small read sensor if the flow of current is perpendicular to the spin valves
[21, 22]. However, the signal-to-noise ratio (SNR) is still too low to be considered
for RAM applications.
The search for a higher SNR led to the discovery of tunnel magnetoresistance
(TMR) in Fe/Ge–O/Co-junctions by Michel Jullière in 1975 [23]. The TMR effect is
attributed to quantum tunneling through an insulating barrier. The reported change
in resistance was around 14% at 4.2 K, similar to the magnitude of GMR and did not
attract much attention initially. Other tunnel barriers, such as NiO and Gd 2 O 3 , had
also low TMR values at low temperatures [24, 25]. Eventually, room temperature
TMR values of 11.8 to 18% were attained when amorphous AlOx tunnel barrier was
used in conjunction with Fe, Co and CoFe electrodes in 1995 [26, 27], up till 70%
when CoFeB was used in 2004 [28]. The second major breakthrough in achieving
higher SNR occurs when bcc-MgO(100) was used as the tunnel barrier. The highest
reported TMR till date is at 604% at room temperature, or 1144% at 5 K [29], and
can theoretically reach beyond thousands of percent [30, 31, 32, 33, 34].
Although the major conceptual concerns have been addressed, a multitude of
practical designs such as scalability, integration and read/write schemes of MRAM
devices have to be considered too. The first generation of MRAM products available
on the commercial market have the easy axes of the magnetization of the free and
reference layers lying along the plane of film, otherwise known as magnetic materials
with in-plane magnetic anisotropy (IMA) [35]. The MTJ devices were patterned into
an ellipsoid shape to capitalize on shape anisotropy, inducing the magnetization to
preferentially align along the long axis of the ellipsoid, as shown in Fig. 3b. This
poses challenges to scalability of the bit density and the risk of potential anomalous switching due to unintended dipole field formation [36, 37]. In addition, the
initial conception of MRAM requires write lines to generate a local Oersted field
to drive the magnetization along a given orientation. While the MTJ devices can
scale with relative ease, these current carrying wires face the challenges in scaling
of current density and electromigration effects (~10
7 A/cm
2 for copper). Crosstalk
issue may also arise during the scaling of MTJ bits due to the generated Oersted field
unintentionally disturbing neighboring MTJ bits.
To overcome the above issue, toggling and heat-assisted switching have been
proposed and utilized by several companies [22, 38, 39, 40, 41, 42]. Nonetheless,
the added complexity of such switching mechanisms did not increase the overall
attractiveness of MRAM. The discovery of current induced switching, termed spintransfer-torque (STT), revolutionized the design of MRAM. It was discovered that a
pure spin current polarized by the reference layer can impart its angular momentum
to the soft layer, induce magnetization reversal when the minimum threshold is
exceeded. With the read and write currents now passing through the same contact