Title: Development and Characterization of STT-RAM Cells
1Development and Characterization of STT-RAM Cells
Team Members Kang L Wang (PI) - UCLA Pedram
Khalili (PM) UCLA Ken Yang (Investigator) -
UCLA Dejan Markovic (Investigator) - UCLA Hongwen
Jiang (Investigator) - UCLA Yaroslav Tserkovnyak
(Investigator) - UCLA Ilya Krivorotov
(Investigator) - UC Irvine Jian-Ping Wang
(Investigator) - University of Minnesota IBM
Trusted Foundry (Fabrication Vendor) Rep by
Scott Marvenko SVTC/Singulus (Fabrication
Vendor) Rep by Eric Kent, Mike Moore MICRON and
Intel (Supporting and on Advisory Board) Rep by
Gurtej Sandhu Mike Violette (MICRON) Rep by
George Bourianoff, Tahir Ghani, Tanay Karnik
(Intel)
2Outline
- STT-RAM optimization to approaching Phase 1
metrics - Free layer dimensions (area, aspect ratio,
thickness) - Optimal MgO thickness
- Optimal write voltage pulse amplitude and
duration - Energy-efficient switching with non-collinear
polarizer - Measurement techniques
- Thermal stability
- Switching with short voltage pulses
- Recent results and metrics update
- Summary and Outlook
3Where we were 3 month ago
- At the pervious review meeting we were reported
initial results of non-optimized STT-RAM cells.
They showed the following metrics parameters - Write energy per bit 7.5 pJ
- Write time 2.5 ns
- Upper bound on thermal stability
- ? lt 90
- Lower bound on endurance of 105
- Since the review meeting in November 2009, we
have substantially improved the metrics through
STT-RAM device optimization
Data reported in November 2009
Voltage at the sample, Vs
4STT-RAM Optimization Free Layer Dimensions
- critical current for STT-RAM
Fundamental constants and material parameters
- For a given material, to reduce Ic one should
decrease the free layer volume V without
sacrificing thermal stability ?
- Thermal stability
- To decrease volume V while keeping ? constant,
we must increase thickness and decrease width w
5STT-RAM Optimization Barrier Thickness
- - Since energy per write is I2R tw, low RA MgO
also decreases energy per write - Low TMR is signature of pinholes an
lower-voltage dielectric breakdown - MgO thickness with lowest RA that still has high
TMR is needed
RA VS MgO thickness
TMR ratio VS MgO thickness
We found that the optimal MgO thickness for
I-STT-RAM devices is right at the knee in RA and
TRM plots versus MgO thickness.
6STT-RAM Optimization Write Pulse Shaping
For short switching times, switching time versus
pulse duration is well fit by the following
functional dependence
V0 is the (zero-temperature) critical voltage for
switching
This is a signature of quasi-ballistic switching
dominated by angular momentum transfer rather
than temperature
This ?(V) allows us to determine the optimal
write voltage V for minimizing the energy per
write
7STT-RAM Optimization Write Pulse Shaping
Energy per write
Write time
- E(V) has a minimum at V2V0 - Minimum energy
per write is at twice the critical voltage - This
is consistent with our data
8STT-RAM Optimization Write Pulse Shaping
Experimental data
Theory
0.46 pJ
Predicted write energy minimum is experimentally
observed
9STT-RAM Optimization Non-collinear structures
- Micromagnetic simulations show that for collinear
free and fixed layer geometries - there is long incubation time between the
leading edge of the write pulse and the
nanomagnet switching - energy is wasted on excitation of non-uniform
modes -
10STT-RAM Optimization Non-collinear structures
The incubation time results from small initial
spin torque in the collinear geometry (spin
torque ) Polarizer that is non-collinear with the
free layer provides larger spin torque,
accelerates the switching process
Collinear free and fixed layers, simulations
Non-collinear free and fixed layers, simulations
11Materials for non-collinear structures
- We developed materials with perpendicular
anisotropy for - STT-RAM structures with non-collinear
magnetizations
perpendicular filed
12STT-RAM Optimization Non-collinear structures
- - Our measurements of switching of devices with
non-collinear magnetization revealed deep sub-ns
switching. - - This is a salient feature of precessional
switching due to perpendicular polarizer - Pulse shaping is expected to further improve the
energy per write
Device shape and magnetic multilayer optimization
is also expected to significantly improve the
non-collinear device performance compared to this
initial demonstration. This device concept is
very promising for meeting Phase 2 metrics.
13Measurement Techniques of Metrics
- Energy per write and write time
- Switching in response to ns and sub-ns pulses
- Thermal stability measurements
- Thermally activated switching
- Field-assisted switching
- Hard axis hysteresis loop measurement
14Switching by Short Voltage Pulse
Pulse Generator (0.1 10 ns pulse width)
STT-RAM element
Multimeter (resistance measurement)
DMM
- Pulses of variable duration are sent to the
sample - Sample resistance before and after the switching
is measured - - Probability of switching is determined
15Voltage of Short Pulse at the Sample
- Voltage at the sample is a sum of incident and
reflected voltages - Since the sample resistance is much higher than
50 ?, the voltage at the sample is nearly doubled
compared to the incident voltage - We use a pulse generator that absorbs the
reflected pulse without affecting the incident
pulse
Vin
Vs
Vref
Rs
16Thermal Stability Method 1 Thermally-Activated
Switching
- - Switching in response to long, relatively
low-voltage pulses - Switching time histograms are measured and
switching voltage versus pulse duration is
obtained - Extrapolation of the plot of switching voltage
versus pulse duration down to zero voltage,
?(V0) gives the bit lifetime and thermal
stability ?
Fitting to exponential function gives the average
bit life time, ?, at a given voltage
17Method 1 Thermally-Activated Switching
- Applying current-assisted switching, lower bound
on thermal stability ? is determined. - This is only a lower bound due to current noise,
ohmic heating and possible current-induced magnon
excitation
Single switching attempt sequence
Quasi-ballistic switching
Thermally Activated
?63 ts(0 Volt) 31 billion years
Train of 10,000 write/reset pulses
10,000,000 switching attempts
18Method 2 coercivity vs field sweep rate
- Another approach to determining thermal
stability is to measure the sweep rate dependence
of the coercive field. The coercive field under
the sweep time based on the Neel-Arrhenius model
is -
-
Coercivity vs sweep time.
Resistance vs magnetic field at different sweep
time.
19Thermal Stability Method 3 Hard-axis Loop
Estimating the anisotropy field as
Hard Axis Hysteresis Loop
free
free
- - Anisotropy field estimated by this method is
570 Oe - Micromagnetic OOMMF simulations give the
hard-axis saturation field 520 Oe. - The origin of the hard-axis loop asymmetry is not
clear - - Using 520 Oe value, the thermal stability
estimate gives an upper bound on thermal
stability ?
20Current status of STT Cells
As a result of a combination of aforementioned
STT-RAM optimization procedures, device
performance has been substantially improved since
the last review meeting.
Optimized device quasistatic characteristics
21Switching of Improved STT-RAM Cells
In the optimized devices, at the optimal voltage
pulse amplitude, write energy of 0.46 pJ has been
achieved.
?gt55
0.46 pJ
22Two Thermal Stability Measurements of I-STT
Thermally activated switching lower limit for ?
Hard axis saturation upper limit for ?
?55
HK ? 650 Oe as average of HL and HR -gt ? ? 90
Two measurements give the bounds on the thermal
stability 55lt?lt90
HK ? 550 Oe from micromagnetic simulations
23Metrics Update for STT-RAM
Status at previous meeting (Nov 2009) Status (Feb 1, 2010) BAA Targets
Write Energy E7.5 pJ/bit E0.46 pJ/bit E0.25 pJ/bit
Write Speed (tW) 2.5 ns/bit 1.3 ns/bit 5 ns/bit
Cell Size N.A. 0.23 um2 (27F2) 0.24 um2 (lt28F2)
Memory Bit Area (A) 0.01 um2 0.01 um2 0.02 um2
Thermal Stability (?) N.A. 55 lt ? lt 90 60
Endurance gt105 gt107 1x1016
Wafer Yield gt40 gt40 40
24Summary
- We made a significant progress towards optimizing
the performance of STT-RAM cell through device
optimization - Free layer dimensions (area, aspect ratio,
thickness) - Optimal MgO thickness
- Optimal write voltage pulse amplitude and
duration - Energy-efficient switching with non-collinear
polarizer - All Phase 1 metrics are clearly within reach with
modest further optimization. - We are also on the way towards meeting Phase 2
metrics using non-collinear structures.