Recovery dynamics and fast (sub-50ns) read operation with Access Devices for 3D Crosspoint Memory based on Mixed-Ionic-Electronic-Conduction (MIEC),

6-4
Recovery dynamics and fast (sub-50ns) read operation with Access Devices for
3D Crosspoint Memory based on Mixed-Ionic-Electronic-Conduction (MIEC)
G. W. Burr, K. Virwani, R. S. Shenoy, G. Fraczak† , C. T. Rettner, A. Padilla, R. S. King, K. Nguyen, A. N. Bowers,
M. Jurich, M. BrightSky† , E. A. Joseph† , A. J. Kellock, N. Arellano, B. N. Kurdi and K. Gopalakrishnan†
IBM Almaden Research Center, 650 Harry Road, San Jose, CA 95120 († IBM T. J. Watson Research Center, Yorktown Heights, NY 10598)
Tel: (408) 927–1512, Fax: (408) 927–2100, E-mail: [email protected]
Abstract
BEOL-friendly Access Devices (AD) based on Cu-containing
MIEC materials[1-4] are shown to be capable of both maintaining
and moving rapidly between all the roles necessary for 3D crosspoint memory (un-selected, half-selected, selected(read), and selected(write)). Ultra-low leakage is maintained over hours, recovery dynamics after both write (30-50uA) and read (3-6uA) operations are explored, and read operations fast enough for use with
MRAM (sub-50ns) at low voltages are demonstrated.
Keywords: Access device, MIEC, PCM, NVM, RRAM, MRAM
Introduction
3D-stacking of large crosspoint arrays in the BEOL can allow
a nonvolatile memory (NVM) to be as cost-effective as 2D NAND
FLASH (≤4F2 /3). MIEC-based ADs exhibit the large ON/OFF
ratios needed for such arrays (Fig. 1), with high voltage margin
Vm (for which leakage is ≤10nA), ultra-low leakage (<10pA),
high enough write currents (>100uA) even for PCM, and the bidirectional nonlinearity needed for stacking high-density MRAM
and RRAM in the BEOL [1-4]. In addition, we have shown both
scalability to <30nm CD and <15nm thickness [4], and tight distributions and 100% yield in large (512kBit) arrays[3].
In a crosspoint array, each AD must fill multiple roles: holding
ultra-low (∼10pA) leakage at low un-selected bias for hours; maintaining modest (∼10nA) leakage at half-selected bias for seconds;
and occasionally being selected to pass either read-level (3-6uA)
or write-level (30-60uA) currents through its NVM (Fig. 2). The
AD must maintain low leakage over very long periods, yet switch
into a current-passing role and back to low leakage, quickly and
seamlessly. To date, we have shown fast turn-ON speeds (15ns) of
MIEC-based ADs, mostly in the PCM write regime (≥100uA)[4].
In this paper, we show that MIEC-based ADs maintain half- and
un-selected states with ultra-low leakage over hours. We explore
the time to recover low-leakage after both NVM-write (30-50uA)
and NVM-read (3-6uA) operations, demonstrate fast NVM-reads
(<50ns) suitable for MRAM, and show similarly fast turn-ON in
thinner ADs with less overvoltage (thus minimal read disturb).
Stability of low-leakage in MIEC ADs
MIEC-based ADs are integrated on 8” wafers, using sputterdeposition of Cu-containing MIEC material into vias followed by
an optimized CMP process[2] and a confined, non-ionizable TEC.
Small “1T-1AD” device arrays are integrated with 180nm-node
FETs (Fig. 3(a)) [2-4]; “short-loop” devices (Fig. 3(b)) are tested
with Conductive-AFM [3-4], using a proximate bottom contact
with low series resistance made by removing nearby devices.
The long-term, low-bias stability of large 1T-1AD devices
(Vm ∼ 1.05V [2-3]) was tested with VWL ∼ 3.5V (all applied
voltage VDUT appears across the AD). Fig. 4 shows that MIEC
ADs maintain low-leakage over hours of exposure, whether in a
deep (±230mV) or shallow un-select (±350mV) condition (inset).
Repeated exposure to the half-select (±530mV) condition (Fig. 5,
t<70s) has no appreciable effect on any of the leakage conditions,
and even prolonged exposure to higher currents (Fig. 5, t>70s)
induces only subtle effects on subsequent leakage.
Recovery of low-leakage in MIEC ADs
While Figs. 4, 5 show excellent stability, the SMU is too slow
to gauge MIEC AD recovery. With a low VWL , a large first AFG
“over-voltage” pulse can force a rapid turn-on (Fig.6) of a 1T-1AD
T66
978-4-86348-347-7
to a given current level; a second pulse can then probe AD recovery
dynamics under read or lower bias conditions (Figs. 7–11).
Fig. 7 shows that after a strong write pulse (50uA for 1us),
MIEC AD response is in fact affected. For a read following a write
(Fig.7, top), the MIEC AD simply supplies the usual expected current (∼10uA) instantly, without needing overvoltage acceleration.
At much lower bias, however, (Fig. 7, bottom) significantly larger
currents than the expected low leakage can persist. Assumably,
turn-ON involves shifting Cu (slowly under modest bias, rapidly
under overvoltage). Once large currents have been passed, some
Cu may remain “latched” in place, altering device response. However, even brief (1us) exposure to 0V restores much of the original
low-leakage response (Fig.8), and negative voltages accelerate this
significantly (Fig.9). Post-write effects are also greatly suppressed
if write duration is limited (Fig. 10). Despite this, some sort of
explicit post-write recovery sequence is likely to be necessary.
In contrast, recovery after read-level currents is much faster.
Fig.11 shows a 3-pulse sequence, in which two long (4ms) pulses
at 600mV surround a brief (1us) pulse at 2V. Since VWL is significantly lower, the maximum current is limited to 6uA per-device
(to obtain sufficient SNR here, ten devices along a single column
were pulsed in parallel). Before the read pulse and without overvoltage, the current at 600mV slowly builds up to ∼300nA. After
the read pulse, leakage has increased by no more than 50%, and
exposure to 0V (for even 500ns) completely restores the original
leakage characteristics. Fig. 12 quantifies the recovery dynamics
after write and read events.
Role of thickness in MIEC AD turn-ON speed
Since these 1T-1AD MIEC ADs are fairly thick (dmin ∼
75nm), rapid turn-ON requires significant overvoltage (Fig.13(a)).
Despite the slow native response, turn-ON from half-select to 10uA
read currents can be sub-50ns (Fig. 13(b)). However, such large
overvoltages in the presence of bitline capacitance could lead to
read disturbs. “Short loop” devices with proximate BEC contacts
allow testing of thinner MIEC ADs [4]. Fig.14(a) shows that native
turn-ON (without overvoltage) speeds up considerably as MIEC
ADs get thinner, Even modest pulse-shaping in thin MIEC ADs
can readily deliver 100ns read-level (10uA) turn-ON (Fig.14(b)).
Conclusions
BEOL-friendly Access Devices (AD) based on Cu-containing
MIEC materials [1-4] can sustain and move rapidly between unselected, half-selected, selected(read), and selected(write) states.
Ultra-low leakage can be maintained over hours, leakage recovery
after write (30-50uA) operations requires ∼1us (with read (3-6uA)
recovery even faster), read operations can be fast enough for use
with MRAM (sub-50ns), and inherently-fast thin MIEC ADs offer
similar speeds at modest overvoltages (for minimal read disturb).
Acknowledgements
Expert analytical and processing support from T. Topuria, P.
Rice, E. Delenia, L. Krupp, D. Hepner, D. Erpelding and the Microelectronics Research Laboratory (YKT) is gratefully acknowledged.
References
[1] K. Gopalakrishnan et. al., VLSI 2010, T19-4 (2010).
[2] R. Shenoy et. al., VLSI 2011, T5B-1 (2011).
[3] G. W. Burr et. al., VLSI Tech. Symp., T5.4 (2012).
[4] K. Virwani et. al., IEDM Tech. Digest, 2.7 (2012).
2013 Symposium on VLSI Technology Digest of Technical Papers
(300ns)
1uA
100nA
1e-5
0.001
0.0001
0.01
0.1
Current
0.9
0.5
Vm~1.53V
1nA
100pA
a)
10pA
300nA
V1
100nA
via
1pA
DC IV
-1.5
-1.0
-0.5
0
0.5
1.0
30nA
1.5
.99999
.9999
.999999
BEC
via
b)
-1
-0.5
~½Vm + 350mV + …
0
Voltage [V]
0.5
1
1.5
b)
SMU/AFG
dmin
MIEC
SCOPE
196:
100nm
BEC
0us
No overvoltage
Time
1us
2us
3us
4us
25uA
5us
Vc
½Vm (= 520mV)
un-select
0uA
0
25uA
50
… + 750mV
25
Time [hours]
then
0
50
0uA
0
0ns
250ns
500ns
25uA
Time
750s
0uA
1us
Fig. 6 Turn-ON delay—as induced
by RC-delay in establishing a given
voltage across a 1T-1AD, followed
by the response of the MIEC AD to
that voltage—can be greatly reduced
by overvoltage.
40
20
0
10us @ 50uA
40
20
0
3us @ 50uA
40
20
0
1us
0V for 100ns, then
0.9V for 1us
0V for 100ns, then 0.9V for 1us
0V for 100ns, then 0.9V for 1us
@ 50uA
300ns
40
20
0
0V for 100ns, then 0.9V for 1us
@ 50uA
100ns
40
20
0
@ 50uA
-1us
0V for 100ns, then 0.9V for 1us
Time 1us
0us
Fig. 10 Post-write effects can also
be greatly suppressed by reducing the
duration of the 50uA write pulse.
Current
(a)10uA
5uA
0uA
0us
No overvoltage
Time
10us 20us 30us 40us 50us
10uA
Current
5uA
then
100ns
@0V
@ 0.9V
1us
@ 0.9V
5us 10us 0
1us
0ns @0V
0
1us
@0V
1us
@ 0.7V
@ 0.7V
1ms 2ms 0
1us
2us
0ns @0V
2us
100ns
1us
Native response
0
2us
@ 0.7V
0
1us
2us
100ns
@0V
1us
Native response
0
1us
@ 0.5V
@ 0.5V
5ms 10ms 0
1us
Time
2us
@ 0.5V
0
1us
2us
800
600
400
200
0
800
600
400
200
0
800
600
400
200
0
1us
1us
2V for 1us Æ 6uA
then 1us @0V
50ns
25ns
0ns
500ns
1us
1.5us
2us
0.6V for 3us
@ 50uA
0us
1us
2us
3us
4us
2ms
(a)
0.6V for 4ms
Time
dmin =
10uA
4ms
Fig. 11 After a
1us read pulse
at 6uA (perdevice over 10
ADs), leakage
is never more
450nA, or 1.5×
the
expected
static leakage
level that would
build up over
∼1ms anyway.
100ns
0ns
100ns
0ns
100ns
0.9V for 1us
@ 50uA
0us
50
1us
25
Time
2us
500ns at 0V
1us
0.9V for 1us
@ 50uA
0
50
500ns at <0V
1us
25
@ 50uA
0
0.9V for 1us
Leakage Post write (50uA)
current recovery (Fig. 8)
(at 600mV)
3uA
Dotted fits are guide-to-the-eye only
Post read (6uA)
recovery (Fig. 11)
1uA
300nA
0
Static leakage
at 600mV
0
500ns
1us
1.5us
Recovery time at 0V
Fig. 12 Restoration of the static IV characteristics (at 600mV) after write (50uA)
events takes longer than after a read (6uA).
100ns shaped pulse
(max. overvoltage: 350mV)
10uA
Current
36nm
5uA
47nm
0uA
0ns
200ns at <0V
1us
25
(b)
11nm 28nm
5uA
Time
Time
Time
Time
Fig. 13 (a) Despite their slow native response, thick (dmin ∼75 nm) MIEC
1T-1AD devices can be turned ON rapidly with large overvoltage, allowing
(b) the transition from half-select to 10uA read currents to occur in <50ns.
0.9V for 1us
@ 50uA
0
50
0
200ns at 0V
1us
25
0us
1us
2us Time
Fig. 9 Exposure to negative voltages accelerates the recovery after
a 1us write (50uA) pulse.
Time
Current
10uA
5us
50
5uA
0.6V for 4ms
0uA
… + 1.15V
5uA
0.6V for 3us
1us
0.6V for 4ms
(b) 15uA 100ns
Current
300ns @0V
1us @0V
2V for 1us Æ 6uA
then 500ns @0V
0
0.6V for 3us
@ 50uA
0.6V for 4ms
-2ms
100ns @0V
@ 50uA
0.6V for 4ms
0uA
10uA
0uA
1us
2V for 1us Æ 6uA
then 200ns @0V
-4ms
0.6V for 3us
30ns @0V
@ 50uA
5uA
… + 600mV
5uA
@ 50uA
0.6V for 4ms
0.6V for 4ms
800
600
400
200
0
0.6V for 3us
0.6V for 4ms
2V for 1us Æ 6uA
then 100ns @0V
0.6V for 4ms
0ns @0V
1us
Fig. 8 Even brief (1us) exposure to
0V after a 1us write (50uA) pulse
restores much of the original lowleakage response.
2V for 1us Æ 6uA then 50ns @0V
0.6V for 4ms
800
600
400
200
0
… + 300mV
0uA
10uA
15
10
5
0
15
10
5
0
15
10
5
0
15
10
5
0
15
10
5
0
Fig. 7 After a strong write pulse (50uA for
1us), MIEC AD response is affected: devices
remain ON, don’t require overvoltage acceleration to turn back ON, and at lower voltages
where leakage should be undetectable, measurable currents can persist.
Measured current PER DEVICE [nA]
… + 1.0V
1us
@ 50uA
1us
@ 0.9V
50uA
25
0ns @0V
Native response
0
Time [seconds]
Fig. 5 Repeated 2sec exposures to half-select
(±530mV) and even higher bias conditions has
minimal impact on subsequent leakage.
Fig. 4 MIEC ADs maintain low-leakage over
hours of exposure, whether in a deep (±230mV)
or shallow un-select (±350mV) condition.
50uA
… + 350mV
25
@ 50uA
50uA
Measured current
[uA]
50
Current [uA]
0
Vc
Current
½Vm + 120mV
½Vm + 60mV
Measured current [uA]
1us
Current
0
Measured current [uA]
~10pA
un-select
leakage
(across half-selected
ADs of same row)
Proximate
BEC contact
Fig. 3 A Source-Measure Unit (SMU), or an Arbitrary
Function Generator (AFG) and oscilloscope, address either
small “1T-1AD” MIEC arrays [2-4] or “short-loop” MIEC
ADs tested with Conductive-AFM and a nearby BEC contact.
25
Vr
½Vm – 180mV
un-select
½Vm – 300mV
Voltage [V]
MIEC
(DC)
50
(across un-selected
ADs biased NEGATIVE)
~10nA
half-select
leakage
~½Vm
(deep)
TEC
dmin
VWL
Vc
~½Vm – 200mV
Vr
V
Fig. 2 Voltage on a selected NVM+AD affects 3 other sets of ADs: those
in the same row, in the same column (both half-selected), and all others
(un-selected). Each AD must hold ultra-low (∼10pA) leakage for
hours, maintain modest (∼10nA) leakage for seconds, and occasionally
pass either read-level (3-6uA) or write-level (30-60uA) currents.
half-select
(shallow)
C-AFM tip
Vr
~½Vm
(across half-selected
ADs of same column)
Measured current [uA]
a)
uProbe & pad
~3-60uA
Read or
Write
current
+ Required NVM voltage + …
…+ voltage drop
across wiring
(across selected diode)
…but also should not exceed sum of:
via
FET
V
…has to
provide:
ILD
Fig. 1 MIEC-based ADs exhibit the large ON/OFF ratios needed for large crosspoint
arrays, showing (a) high voltage margin Vm (for which leakage stays below 10 nA) [34], high ON current densities [1,4], ultra-low leakage (< 10 pA) [2-4], good scalability
(<30nm CD, <12nm thickness) [4], and (b) tight margins (as well as 100% yield) when
integrated on 8" CMOS wafers in large (512kBit) arrays[3].
Current
VDUT
V across selected cell…
0Volts
TEC
MIEC
10nA
-1.5
Voltage [V]
.999
M2
M1
10nA
.99
Log(I)
pulsed IV
10uA
Current [uA]
1uA
Log(I)
1e-6
100uA
0ns
500ns
0uA
Time
1us
-100ns
0ns
100ns
Time
200ns
Fig. 14 Thinner MIEC ADs are measurably faster, both in
terms of (a) the un-accelerated native turn-ON as well as (b)
when accelerated with modest (+350mV) overvoltage.
2013 Symposium on VLSI Technology Digest of Technical Papers
T67