You are on page 1of 10

Home Search Collections Journals About Contact us My IOPscience

Sub-100 fJ and sub-nanosecond thermally driven threshold switching in niobium oxide

crosspoint nanodevices

This article has been downloaded from IOPscience. Please scroll down to see the full text article.

2012 Nanotechnology 23 215202

(http://iopscience.iop.org/0957-4484/23/21/215202)

View the table of contents for this issue, or go to the journal homepage for more

Download details:
IP Address: 161.45.205.103
The article was downloaded on 25/07/2013 at 19:30

Please note that terms and conditions apply.


IOP PUBLISHING NANOTECHNOLOGY
Nanotechnology 23 (2012) 215202 (9pp) doi:10.1088/0957-4484/23/21/215202

Sub-100 fJ and sub-nanosecond thermally


driven threshold switching in niobium
oxide crosspoint nanodevices
Matthew D Pickett and R Stanley Williams
Hewlett-Packard Laboratories, 1501 Page Mill Road, Palo Alto, CA 94304, USA

E-mail: Matthew.Pickett@HP.com

Received 23 January 2012, in final form 11 April 2012


Published 3 May 2012
Online at stacks.iop.org/Nano/23/215202
Abstract
We built and measured the dynamical current versus time behavior of nanoscale niobium
oxide crosspoint devices which exhibited threshold switching (current-controlled negative
differential resistance). The switching speeds of 110 × 110 nm2 devices were found to be
1tON = 700 ps and 1tOFF = 2.3 ns while the switching energies were of the order of 100 fJ.
We derived a new dynamical model based on the Joule heating rate of a thermally driven
insulator-to-metal phase transition that accurately reproduced the experimental results, and
employed the model to estimate the switching time and energy scaling behavior of such
devices down to the 10 nm scale. These results indicate that threshold switches could be of
practical interest in hybrid CMOS nanoelectronic circuits.
S Online supplementary data available from stacks.iop.org/Nano/23/215202/mmedia
(Some figures may appear in colour only in the online journal)

1. Introduction the sense that its presence is sustained by current flow


but disappears upon the removal of bias. This volatility
For a half century there has been significant fundamental distinguishes threshold switching from ‘memory switching’ in
research interest in the transition metal oxides that exhibit similar oxide two-terminal devices which retain their resistive
a temperature-driven insulator-to-metal phase transition. The state indefinitely without static bias, a distinction outlined in
phenomenon is of practical interest in part because of the detail previously [17].
fact that two-terminal metal–insulator–metal devices based on Due to the utility of the switching effect that characterizes
such materials display current-controlled negative differential CC-NDR, oxide threshold switches could be valuable
resistance [1–13] (CC-NDR). The phenomenon of CC-NDR for a wide variety of circuit applications ranging from
is also known as ‘threshold switching’ owing to the existence signal processing [18] to isolation devices in a memristor
of bistable resistance states under voltage bias as well as crossbar [19–22]. However, widespread use of phase
‘S-NDR’ because of the s-like shape of the current–voltage transition oxides has been limited to date because the energy
curve. Threshold switching in these materials has been and time required to heat the traditional micron-scale device
attributed to a Joule-heating-induced [1–3, 14, 15] filamentary volumes is prohibitive for practical applications. In this work
insulator-to-metallic phase transformation [16, 3] that causes we report the fabrication of nanoscale crosspoint devices
a dramatic drop in device resistance above a bias threshold. and experimental demonstration of their capability, by virtue
This is a positive feedback effect which is driven by the of their small volumes, of switching to their low resistance
fact that once a small part of the device is Joule heated state with a total energy input less than 100 fJ in less than
above the transformation temperature more current flows a nanosecond. To understand this dynamical behavior we
through this region heating it further and expanding the derived a new quantitative model for Joule-heating-induced
region. The formation of the metallic phase is volatile in CC-NDR utilizing the memristive system mathematical

0957-4484/12/215202+09$33.00 1 c 2012 IOP Publishing Ltd Printed in the UK & the USA
Nanotechnology 23 (2012) 215202 M D Pickett and R S Williams

under a 9 nm thick Pt conduction layer. Next, a 20 nm


thick amorphous Nb2 O5 blanket layer was deposited over the
entire substrate at room temperature by reactively sputtering
a metallic Nb target with a gas mixture of 1/5:O2 /Ar at
5 mTorr. The as-grown film composition and structure were
determined by in situ x-ray photoelectron spectroscopy (XPS)
and ex situ x-ray diffraction (XRD), respectively. Finally, a
crossing 110 nm wide, 11 nm thick Pt top electrode was
patterned with a liftoff process perpendicular to the bottom
electrode.
A scanning electron micrograph of a representative
device is included as figure 1(a). Before operation a 6 V, 10 µs
electroforming pulse was required to irreversibly modify the
low-bias resistance of the devices from the 10 G virgin
state to the 1 M operational regime. The forming step
has previously been attributed [12, 27] to a soft breakdown
process that generates a channel of crystalline NbO2 , which
exhibits an insulator-to-metal transition at 1080 K [28],
within the oxide film. This explanation is similar to the
well-documented case of the electroformed TiO2 system, in
which a channel of reduced oxide forms that also exhibits
an insulator-to-metal transition [12, 29–31, 14]. A quasi-DC
current–voltage curve for an electroformed device is included
in figure 1(a), showing the threshold switching characteristic
of these devices. Note that in the experimental configuration
used for performing the quasi-DC voltage sweep, the negative
resistance region of the curve is unstable and there are
consequently no data points for that region. A cross-sectional
schematic of the device during operation is shown in
Figure 1. Threshold switch characteristics and model schematic.
(a) Experimental and modeled current–voltage curve for a figure 1(b), illustrating our biphasic thermal switching model
110 × 110 nm2 niobium oxide threshold switch with an inset SEM for oxide threshold switches outlined in section 3.
image of the crosspoint device. (b) Schematic cross section of the We studied the switching dynamics of the nanodevices by
device model during operation showing an electroformed NbO2 configuring them into a Pearson–Anson circuit [32, 33], a type
channel that consists of two cylindrical phases. (c) Schematic of the
relaxation oscillator circuit used in this work to characterize the
of van der Pol relaxation oscillator [34, 18], with the oxide
thermal transients within the phase transition threshold switch, U, switch acting as the CC-NDR element. Relaxation oscillators
by means of measuring current transients with a 50  input based on CC-NDR elements have been understood since
impedance oscilloscope. the early 1920s when Pearson and Anson first demonstrated
the effect using neon gas discharge tubes as the switching
elements [33, 32]. This circuit, as adapted for this experiment,
framework [23] and compared the model to the experimental is shown schematically in figure 1(c). The CC-NDR element,
results, obtaining excellent agreement. We then utilized the in this case the niobium oxide threshold switch, U, is
model in simulations to predict that a 10 nm radius device connected through the nanowire electrode series resistance,
could switch with an enthalpy input of 10 fJ in less than Re , in parallel with an external capacitor, C. This parallel
100 ps. These results demonstrated that, at the nanoscale, combination is then attached to a DC source with a controlled
Joule-heating-induced CC-NDR devices can be compatible load resistance RL . We note that, although the threshold switch
with transistors in terms of switching speed and energy itself has some intrinsic parallel capacitance, the nanoscale
requirements and thus could be extremely valuable in hybrid size limits the magnitude to below one femtofarad, rendering
integrated circuits [24–26]. it negligible in comparison to the tens of picofarads of external
capacitance used in the experiment.
2. Device fabrication and characterization This classical circuit functions by spontaneously switch-
ing between two stages: in the first stage the switching
In order to experimentally examine the dynamics and element is in the OFF (high resistance) state and the parallel
energetic properties of threshold switching in nanoscale capacitor is charging, and in the second stage the switching
crosspoint devices, we fabricated niobium oxide devices element is in the ON (low resistance) state and the parallel
with the following nanoimprint lithography process. First, capacitor is discharging. Transition between the stages occurs
a 110 nm wide bottom electrode was patterned on a Si when the CC-NDR element is driven to the instability points,
wafer with a 200 nm thick thermally grown oxide. This also referred to as the switching thresholds, which are located
bottom electrode consisted of a 2 nm thick Ti adhesion layer at the two inflection points of the ‘S’ curve. When DC

2
Nanotechnology 23 (2012) 215202 M D Pickett and R S Williams

or material system, the periods of the charging (low


current/baseline) and discharging (high current/pulse) stages
are primarily dictated by the magnitude of the DC voltage,
load resistance and parallel capacitance. Consequently the
oscillation period and its DC voltage dependence provide no
direct information on the switching dynamics or governing
physics of the CC-NDR element itself. Additionally, the
amplitudes of the voltage and current oscillations are dictated
by differences in the inflection points on the quasi-DC
current–voltage curve of the CC-NDR element, since the
relaxation oscillator traverses a ring around the negative
domain of the curve [47, 44], and also contain no information
on the switching dynamics. Since the parallel capacitor must
maintain a continuous voltage which varies much more slowly
than the current during CC-NDR switching, the voltage across
the switch element is roughly constant during the transition
and also provides little information about device dynamics or
physics. Nevertheless, CC-NDR device switching dynamics
may be obtained by monitoring the current during the
transition between stages and the physical model for the
device must therefore reproduce the current dynamics to be
considered accurate. Thus, it is only the current signal at
the transition between charge/discharge stages that provides
a direct measurement of the switch conductance during
switching from ON to OFF or OFF to ON.
As shown schematically in figure 1(c), the current
through the device was measured by using the 50  input
impedance of a LeCroy Wavemaster 20Zi Oscilloscope as
a high bandwidth current digitizer. The performance of
the measurement system was characterized by applying a
1 V, 200 ps rise-time pulse to the circuit instead of a DC
voltage. This voltage was insufficient to switch the device
but demonstrated that the bandwidth of the test setup was
sufficient to measure current signal rise times of ≤200 ps.
Figure 2. Experimental and simulated relaxation oscillator time The motivation for utilizing the relaxation oscillator circuit to
dependences. (a) Five oscillation periods including the raw current characterize the dynamics rather than studying the response of
data at 80 GS s−1 (light blue diamonds), current data time-averaged the device to an external pulse was to avoid complications due
to 10 GS s−1 (dark blue circles), simulation results of the current to reflections of fast (<1 ns) pulse edges caused by impedance
(solid red line) and simulation results of the phase composition state
variable u (dashed orange line, scale on right vertical axis) for the mismatch between the pulse source (50 ) and the CC-NDR
relaxation oscillator. (b) Expanded view of the first rising edge device (>10 k). These reflections are absent when the device
showing a heating transient of 700 ps. (c) Expanded view of the first itself generates the signal and a more conclusive result can
falling edge showing a cooling transient of 2.3 ns. thus be obtained.
The experimental current response of a typical niobium
oxide nanodevice oscillator over five oscillation periods
biased through an appropriately chosen load resistor [35], the is included as figure 2(a). The sequential charging and
device will spontaneously and periodically switch between discharging stages are evident in the low and high current
these stages resulting in a signal which contains current regions of the signal. Expanded views of the rising
pulses that are roughly trapezoidal in shape. A typical current and falling edges are shown as figures 2(b) and (c)
signal is demonstrated in figure 2(a) for our experimental respectively and represent the device conductance during
configuration. the transition process. These results demonstrate that the
The general behavior of the Pearson–Anson circuit OFF-to-ON-transition time is 1tON = 700 ps while the
is independent of the physical phenomena governing the ON-to-OFF-transition time is 1tOFF = 2.3 ns. We tested
CC-NDR element. Indeed, similar relaxation oscillations have more than ten devices for this sample and observed that the
been observed and modeled in a broad variety of two-terminal OFF-to-ON-switching time was always 3–5 times faster than
systems that are known to exhibit CC-NDR including the ON-to-OFF-switching time. An estimate of 100 fJ for
organic materials [36–39], transition metal oxides [1, 40–42, the total energy input during ON-switching can be obtained
13], chalcogenide semiconductors [43, 35, 44], silicon by noting that the voltage across the device was ≈1 V and
nanowires [45], Wigner solids [46] and gas discharge that numerical integration of the measured current over the
bulbs [32, 33]. Regardless of the physical mechanism transition time is ≈100 fC.

3
Nanotechnology 23 (2012) 215202 M D Pickett and R S Williams

The endurance of the threshold switch is of critical outlined above and obeys the thermodynamic requirement that
importance for practical applications, especially as selection CC-NDR results from the growth of conductive filaments that
devices for memory elements. We tested the endurance by run parallel to the current flow [16]. The following derivation
successfully running the oscillator for 5 min at a time which, lays out the expansion of our previously published [14]
at a period of ∼200 ns, yields an endurance value >109 . steady-state model to a fully dynamical model by utilizing an
The experimental results on the magnitude and asym- enthalpy flow approach.
metry of the switching dynamics are significant for several The electrically parallel combination of the coaxial
reasons. Firstly, the asymmetry of the edges indicates that phases yields the following expression for total device
the current rise time is not limited by a parasitic inductance resistance as a function of phase fraction:
in the experimental circuit as this would result in symmetric  −1
ρins L ρins
 
rise and fall times. The order of magnitude of the switching Rch (u) = 1+ − 1 u2 (1)
times also suggests that the switching speed is not limited π rch
2 ρmet
by the lattice dynamics as optically pumped solid–solid where u = rmet /rch is the metallic phase fraction expressed in
phase transformations are known to occur in the femtosecond radial coordinates and all of the model parameters, definitions
time scale in the related material VO2 [48]. Additionally, and values are included as table 1. The dynamical behavior
the direction of the asymmetry, 1tOFF ≈ 31tON , indicates is introduced by noting that the rate of Joule heating in the
that OFF-switching requires an energetic relaxation which device must be equal to the rate at which heat is conducted to
takes substantially longer than the energetic injection during the surroundings plus the rate of change of the total enthalpy
ON-switching. In the metal–oxide–metal device geometry in the channel:
studied here, thermal energy is injected instantaneously in a d1H
distributed manner throughout the device and most of that Piv = 0th (u)1T + (2)
dt
thermal energy is translated directly into enthalpy thus driving
the temperature increase and transformation with little delay. in which Piv is the power dissipated,
If the current is abruptly turned off, however, that enthalpy Piv = i2 Rch (u), (3)
takes some time to be conducted to ambient much like a
capacitor discharging through a resistor. This intuitive picture 0th is the thermal conductance of the insulating shell,
suggests that heating should be faster than cooling in this
1 −1
 
geometry but, more importantly, the thermal model outlined 0th (u) = 2πLκ ln , (4)
below reproduces the asymmetric dynamics quantitatively. u
and a chain rule relates the time derivatives of enthalpy change
3. Analytical memristive model and phase composition,
d1H d1H du
We established a new approach to modeling oxide CC-NDR = . (5)
dt du dt
devices with the goal of obtaining a simple analytical device
The total enthalpy change is determined by the integrated heat
model which can be used to (a) accurately predict the behavior
capacity plus the transformation enthalpy in the following
of these devices in an arbitrary electrical circuit over its entire
manner:
range of operation and (b) understand the impact of different Z rch
device parameters on device behavior. Although previous 1H = 2π Lĉp (T(r) − Tamb )r dr + π rmet
2
L1ĥtrans . (6)
models exist which take a similar approach, these models are 0
either limited to small-signal analysis of narrow regimes of The assumptions outlined above constrain the temperature to
device operation [3], do not directly relate the current–voltage the following:
behavior to the thermal dynamics [49] and are consequently
1T,   0 < r ≤ rmet

difficult to use for circuit simulations, or are limited to the
steady state and do not incorporate device dynamics [14]. T(r) − Tamb = r 1
1T ln , rmet < r ≤ rch .
To meet these goals we employed a memristive rch ln u
system [23] approach which assumes (a) cylindrical (7)
symmetry, (b) constant temperature within the metallic
This functional form for the temperature profile indicates that
core fixed at the transition temperature, (c) ambient
the maximum temperature within the device occurs in the hot
temperature at the exterior of the channel and (d) two-
metallic core which is clamped to the transition temperature
dimensional heat flow along the radial direction. Although
while the minimum temperature in the device occurs at the
these assumptions reduce the thermal environment of the
outer edge which is clamped to ambient temperature. Due to
device to a simple picture, they were previously employed
this result, the metallic phase fraction u is bounded between
for a static model and shown to accurately reproduce
zero and one and the temperature at a specific position is a
the temperature-dependent current–voltage behavior in the
nonlinear function of u.
titanium oxide system [14]. These assumptions yield a Combining equations (6) and (7) yields
biphasic model, shown schematically in figure 1(b), with two
u2 − 1
 
coaxial cylinders: a hot metallic core and cooler insulating
shell. This picture is consistent with the previous models 1H = π rch L ĉp 1T
2
+ 1ĥtr u ,
2
(8)
2 ln u
4
Nanotechnology 23 (2012) 215202 M D Pickett and R S Williams

Table 1. Model parameter definitions, symbols and values.


Model property Symbol Model value (SI) Remarks
Molar heat capacity cp 55 J mol−1 K−1 From [55]a
Molar enthalpy of transformation 1htr 3.4 × 103 J mol −1 From [55]
Molar volume η 21 × 10−6 m3 mol−1 From [56]
Volumetric heat capacity ĉp 2.6 × 106 J m−3 K−1 cp /η
Volumetric enthalpy of transformation 1ĥtr 1.6 × 108 J m−3 1htr /η
Thermal conductivity κ 1.5 W m−1 K−1 Fit to data
Metallic phase electrical resistivity ρmet 1 × 10−4  m From [28]
Insulating phase electrical resistivity ρins 0.7 × 10−3  m From [28]a
Phase transition temperature TMIT 1080 K From [28]
Ambient temperature Tamb 296 K Measured
Heating temperature 1T 784 K TMIT –Tamb
Conduction channel radius rch 30 nm Fit to data
Conduction channel length L 20 nm Measured
Nanowire electrode resistance Re 2700  Measured
External load resistance RL 4200  Measured
Parasitic capacitance Cp 23 pF Measured
DC voltage applied to oscillator VDC 1.8 V Measured
a Indicates that the property is temperature dependent, but an averaged value was used
for analytical simplicity in the model. Literature values and direct measurements were
unavailable for κ and rch , respectively. We obtained values for κ and rch by performing a
best fit between the device model and the experimental static i–v curve of figure 1(a).

and the variation of the enthalpy with respect to u is thus phase fraction state variable approach used here implies that
there is a fundamental mathematical relationship between
1 − u2 + 2u2 ln u
 
d1H threshold switches and memory switches. Both systems
= π L rch2
ĉp 1T + 21 ĥtr u . (9)
du 2u(ln u)2 are governed by a state-dependent current–voltage behavior
The combination of the above equations yields a first order and in both systems the state evolution depends on the
nonlinear dynamical system in which u acts as a state variable dynamically evolving current and/or voltage.
for the biphasic system: This particular memristive model, specifically equa-
tion (11), additionally demonstrates a notable property of
v = f (u, i) = Rch (u)i (10) oxide threshold switches. The physical requirement for
−1
time-continuous enthalpy within the device forces the current

du d1H
= g(u, i) = (Rch (u)i2 − 0th (u)1T). (11) through the device during switching to be continuous in a
dt du
similar manner as would a series inductor. Although the
Here, equation (10) is the quasi-static electron transport thermal energy stored by the oxide in the ON state cannot
relation and equation (11) is the dynamical state evolution be recovered as electrical energy, this inductor-like behavior
relation. This system conforms to the formalism of a contradicts the assumption [34, 47] that a true inductance,
current-controlled memristive system as defined by Chua and
even if just parasitic, is required for an electrical van der Pol
Kang [23] and has similar roots to their electrothermal model
oscillator. An effective inductance in a nanoscale device could
for the negative temperature coefficient thermistor presented
have important utility in circuit designs that employ oxide
in that work. Although the system is memristive and exhibits
threshold switches.
a pinched hysteresis, it only retains ‘memory’ of the ON state
while current is flowing, as the metallic region cools to the
insulating phase rapidly (<3 ns for our devices) after current 4. Simulation of the experimental oscillator
is removed. Indeed, the three examples that Chua and Kang
provide in their original work are all volatile and all exhibit We simulated the experimental relaxation oscillator results
CC-NDR in their DC current–voltage curves. to calibrate and verify the memristive system model. The
The volatility of the phase transformation contrasts time-dependent simulations were performed in LTspice IV
qualitatively with the nonvolatility of metal–oxide–metal with a model that included both the threshold switch and the
memristors based on ionic drift, which can have state experimental test circuit, the details of which are included
retention times measured in units of years [50]. It should as supplementary data (available at stacks.iop.org/Nano/23/
be noted, however, that Driscoll et al demonstrated [51] 215202/mmedia). The device model derived above contains
that nonvolatility can be introduced for insulator–metal phase nine parameters which are either material properties or device
transformation materials, specifically VO2 , by holding the geometries. The parameters that can be found in the literature
device at an ambient temperature that lies within the hysteretic or are known from the device fabrication process are listed in
region of the resistance versus temperature curve. Although table 1 with their sources. However, we did not find literature
the phase transformation is volatile in this work, the metallic values for the thermal conductivity κ and do not have a priori

5
Nanotechnology 23 (2012) 215202 M D Pickett and R S Williams

sweep. The largest departure of the simulated results from


the experimental data is observed in figure 1(a) as the
factor of 3 error between the two ON-switching current
thresholds. We believe that this error primarily arises from the
most aggressive assumption of the model that the electrical
resistance of the insulating phase is linear and temperature
independent. The experimental data demonstrate that the
OFF-state current–voltage curve is indeed nonlinear even well
before the switching threshold and literature data indicates
that the conductivity of NbO2 follows an Arrhenius law [28].
We do not believe that this error is a fundamental flaw of
the model because the primary effect is to introduce some
uncertainty to the modeled internal temperature distribution
rather than the more important circuit parameters like
switching times and energies. Ultimately this assumption is
used to enable the model to remain analytical and tractable for
circuit simulations and, as long as a suitable average electrical
conductivity value is used, results in acceptable static and
dynamical simulations.
Figure 2 shows a comparison of experimental and
simulation results for the relaxation oscillator and demon-
strates that the model quantitatively reproduced the absolute
switching times as well as the asymmetry between OFF to
ON (heating) and ON to OFF (cooling) times (1tOFF /1tON ≈
3). Additionally, the simulation resulted in a total switching
enthalpy of 1Hmax = 65 fJ and a transmitted heat (from
Figure 3. Scaling behavior of the device performance. Switching device to surroundings) during OFF-to-ON-switching of
times and energies were obtained by simulating a relaxation
40 fJ. Thus the total simulated switching energy was 105 fJ,
oscillator and varying only the channel radius or the transition
temperature of the threshold switch. Experimental values for consistent with the experimental estimation of 100 fJ. The
niobium oxide are shown for reference. For these simulations all quantitatively consistent relationship between the experiment
device parameters other than the dependent parameter were kept at and the model for the dynamics of both the electrical and the
their values in table 1. The device switching speed and energy scale thermal properties lends credibility to the assumptions of the
favorably with the device size and transition temperature.
model.

knowledge of the electroformed channel radius rch . In order 5. Device scaling simulation
to obtain these two parameters for simulation we performed
a fit of the model to the experimental current–voltage curve In order to understand device size scaling and the impact
of figure 1(a). The result of fitting κ was 1.5 W m−1 K−1 , of different material choices and operating temperatures,
which is consistent with values of similar sputtered oxides we employed the above model to simulate the effect of
that are typically in the range [52] of 1–5 W m−1 K−1 . varying the channel radius, rch , and the difference between
The result of fitting rch was 30 nm, which is consistent with the transition and ambient temperatures, 1T, on the switching
past observations that the electroforming process in similar times and energies of oxide threshold switches. From each
metal–insulator–metal devices resulted in channels that were simulation, we extracted the heating and cooling times (1tHeat
smaller than the full area of the device [30, 31, 12]. As a and 1tCool ) and the maximum input enthalpy (1Hmax ) during
consequence of the assumptions embodied by the parameter the heating transition. Figure 3(a) demonstrates that the
values used in the analytical model, the fitted values of switching times have a strong dependence on device radius,
the radius and thermal conductivity should be considered as decreasing by an order of magnitude for each 15 nm decrease
effective parameters. Nevertheless, by using the parameter fit in device radius. The switching energies have a nearly linear
to the quasi-static current–voltage measurement of figure 1(a), dependence on device radius, even though the total device
the model reproduced the experimental dynamical behavior of volume decreases parabolically. This apparent inconsistency
figure 2 very well and can therefore be used as a predictive is due to the fact that the metallic phase fraction in the low
tool for circuit design as well as for simulating the parametric resistance state, umax , increases as the device gets smaller (as
behavior of oxide threshold switches. shown in figure 4). Figure 3(b) shows that the primary effect
The results of device and circuit simulations using the of reducing 1T, i.e. choosing a different phase transition
above model and the parameter values of table 1 are included material or increasing the ambient temperature, is to linearly
in figures 1–3. Figure 1(a) shows the quasi-DC characteristics reduce the switching times and energies as well as to decrease
of the model when exposed to a 2 ms triangular current the asymmetry between 1tHeat and 1tCool .

6
Nanotechnology 23 (2012) 215202 M D Pickett and R S Williams

closer to the operating ambient will improve these speed and


energy characteristics. These results show that for devices
with a nanoscale volume, Joule heating is actually a practical
operating mechanism for driving an electronic switch and
could be used in addition to field-effect devices.
In summary, we fabricated and characterized niobium
oxide CC-NDR devices with an effective radius of 30 nm that
switched ON at sub-nanosecond times and ∼100 fJ energies.
We derived an analytical memristive system model based
on Joule heating through an insulator-to-metal transition and
used a SPICE simulation to obtain good agreement with
the experimental measurements of the static and dynamic
behavior of the devices. We applied the model to extrapolate
the performance characteristics and scaling behavior of
Figure 4. Simulated scaling behavior of physical device oxide threshold switches and concluded that 10 nm radius
characteristics. Peak phase boundary velocities and metallic phase
fractions were obtained by simulating a relaxation oscillator and devices should switch in tens of ps with switching energies
varying only the channel radius of the threshold switch. Other of tens of femtojoules. The combination of stackable
device parameters were kept at their values in table 1. Physically process compatibility, attractive switching characteristics and
reasonable values of dr/dt < 1000 m s−1 were obtained until the a simple but accurate physics-based device model enables
model was scaled beyond a 10 nm channel radius, at which the the application of threshold switches in high speed circuits.
velocity exceeds the speed of sound (shown as a horizontal dashed
line). This result is significant because it suggests the practical
applicability of thin-film oxide electronics beyond the realm
of nonvolatile memristors and has significant implications for
In order to evaluate the plausibility of extrapolating the the future of CMOS–nanoelectronics hybrid circuits.
electrothermal model to smaller device sizes, we extracted
the peak phase boundary velocity (s = rch du/dt) from the
simulations reported in figure 3(a). Figure 4 shows that the
Acknowledgments
maximum phase boundary velocities remain under 1 km s−1
We thank Ilan Goldfarb for discussions on deposition
(approximately the speed of sound in an oxide material, shown
techniques and Xuema Li for device fabrication assistance.
by the dashed line in figure 4) for all device radii larger than
We additionally thank John Paul Strachan and J Joshua Yang
5 nm, suggesting that the assumptions inherent to the model
for discussing and commenting on the manuscript.
are tenable for rch = 10 nm. For a channel radius smaller than
10 nm, we expect that the switching speeds will be limited
by the speed of sound and saturate at around 10 ps. Since References
the model was derived assuming bulk material properties, an
[1] Futaki H 1965 A new type semiconductor (critical temperature
additional consideration for extrapolation that requires further resistor) Japan. J. Appl. Phys. 4 28–41
work and understanding is the constraint on thermal gradients [2] Cope R G and Penn A W 1968 High-speed solid-state thermal
based on nanoscale heat flow phenomena [53, 54]. switches based on vanadium dioxide J. Phys. D: Appl.
Despite the favorable dynamic switching energy, the Phys. 1 161
static power dissipation in the ON state is high relative [3] Berglund C N 1969 Thermal filaments in vanadium dioxide
IEEE Trans. Electron. Devices 16 432
to CMOS inverters due to the requirement of maintaining [4] Sovero E, Deakin D, Higgins J A, DeNatale J F and
a static current flow to maintain the filament. This static Pittman S 1990 Fast thin film vanadium dioxide microwave
power consumption is an important consideration in designing switches Technical Digest 1990, 12th Annual Gallium
circuits based on oxide CC-NDR devices. Nevertheless, Arsenide Integrated Circuit (GaAs IC) Symp. 1990
nanoscale oxide threshold switches can enable novel CC- pp 101–3
[5] Sakai J 2008 High-efficiency voltage oscillation in VO2
NDR/CMOS circuits: they are fabricated with a thin-film
planer-type junctions with infinite negative differential
process, they can be integrated into metal interconnect layers, resistance J. Appl. Phys. 103 103708
they can switch in significantly under a nanosecond, and the [6] Ruzmetov D, Gopalakrishnan G, Deng J,
enthalpy input required for switching to the ON state can be Narayanamurti V and Ramanathan S 2009 Electrical
less than 100 fJ. triggering of metal–insulator transition in nanoscale
vanadium oxide junctions J. Appl. Phys. 106 083702
[7] Crunteanu A, Givernaud J, Leroy J, Mardivirin D,
6. Discussion and conclusions Champeaux C, Orlianges J-C, Catherinot A and Blondy P
2010 Voltage- and current-activated metal–insulator
An important conclusion of these simulations is the strongly transition in VO2 -based electrical switches: a lifetime
favorable dependence of the device switching speed and operation analysis Sci. Technol. Adv. Mater. 11 065002
energy on rch and 1T. This behavior is directly due to the fact [8] Chopra K L 1965 Avalanche-induced negative resistance in
thin oxide films J. Appl. Phys. 36 184–7
that the enthalpy of the phase transition during switching is [9] Morris R C, Christopher J E and Coleman R V 1969
very small by virtue of its nanoscale volume. Additionally, the Conduction phenomena in thin layers of iron oxide Phys.
choice of materials that have a phase transition temperature Rev. 184 565–73

7
Nanotechnology 23 (2012) 215202 M D Pickett and R S Williams

[10] Chopra K L 1963 Current-controlled negative resistance in [32] Pearson S O and Anson H S G 1921 The neon tube as a means
thin niobium oxide films Proc. IEEE 51 941–2 of producing intermittent currents Proc. Phys. Soc. Lond.
[11] Geppert D V 1963 A new negative-resistance device Proc. 34 204
IEEE 51 223 [33] Pearson S O and Anson H S G 1921 Demonstration of some
[12] Chudnovskii F A, Odynets L L, Pergament A L and electrical properties of neon-filled lamps Proc. Phys. Soc.
Stefanovich G B 1996 Electroforming and switching in Lond. 34 175
oxides of transition metals: the role of metal–insulator [34] van der Pol B 1926 On relaxation-oscillations Phil. Mag. Ser.
transition in the switching mechanism J. Solid State Chem. 7 2 978–92
122 95–9 [35] Shaw M P and Gastman I J 1971 Circuit controlled current
[13] Kim H T, Kim B J, Choi S, Chae B G, Lee Y W, Driscoll T, instabilities in ‘s-shaped’ negative differential conductivity
Qazilbash M M and Basov D N 2010 Electrical oscillations elements Appl. Phys. Lett. 19 243–5
induced by the metal–insulator transition in VO2 J. Appl. [36] Kishida H, Ito T, Nakamura A, Takaishi S and Yamashita M
Phys. 107 023702 2009 Current oscillation originating from negative
[14] Pickett M D, Borghetti J, Yang J J, Medeiros-Ribeiro G and differential resistance in one-dimensional halogen-bridged
Williams R S 2011 Coexistence of memristance and nickel compounds J. Appl. Phys. 106 016106
negative differential resistance in a nanoscale [37] Kishida H, Ito T, Ito A and Nakamura A 2011
metal–oxide–metal system Adv. Mater. 23 1730–3
Room-temperature current oscillation based on negative
[15] Kim J, Ko C, Frenzel A, Ramanathan S and Hoffman J E 2010
differential resistance in a one-dimensional organic
Nanoscale imaging and control of resistance switching in
charge-transfer complex Appl. Phys. Express 4 031601
VO2 at room temperature Appl. Phys. Lett. 96 213106
[16] Ridley B K 1963 Specific negative resistance in solids Proc. [38] Sawano F, Terasaki I, Mori H, Mori T, Watanabe M, Ikeda N,
Phys. Soc. 82 954 Nogami Y and Noda Y 2005 An organic thyristor Nature
[17] Dearnaley G, Stoneham A M and Morgan D V 1970 Electrical 437 522–4
phenomenain amorphous oxide films Rep. Prog. Phys. [39] Tamura K, Ozawa T, Bando Y, Kawamoto T and Mori T 2010
33 1129–91 Voltage oscillation associated with nonlinear conductivity
[18] van der Pol B and van der Mark J 1927 Frequency in the organic conductor alpha-(BEDT-TTF)2I3 J. Appl.
demultiplication Nature 120 364 Phys. 107 103716
[19] Lee M J et al 2007 Two series oxide resistors applicable to [40] Lalevic B and Shoga M 1981 Relaxation oscillations in NbO2
high speed and high density nonvolatile memory Adv. thinfilm switching devices Thin Solid Films 75 199–204
Mater. 19 3919–23 [41] Hayashi T 1964 The S-type negative resistance phenomena
[20] Chang S H et al 2011 Oxide double-layer nanocross bar for and oscillation in metal–insulator–metal diode Japan. J.
ultra high-density bipolar resistive memory Adv. Mater. Appl. Phys. 3 365
23 4063–7 [42] Gu Q, Falk A, Wu J, Ouyang L and Park H 2007
[21] Liu X, Sadaf S M, Son M, Shin J, Park J, Lee J, Park S and Current-driven phase oscillation and domain-wall
Hwang H 2011 Diode-less bilayer oxide (WOx –NbOx ) propagation in Wx V1 -xO2 nanobeams Nano Lett. 7 363–6
device for cross-point resistive memory applications [43] Ielmini D, Mantegazza D and Lacaita A L 2008
Nanotechnology 22 475702 Voltage-controlled relaxation oscillations in phase-change
[22] Liu X, Sadaf M S, Son M, Park J, Shin J, Lee W, Seo K, memory devices IEEE Electron. Device Lett. 29 568–70
Lee D and Hwang H 2012 Co-occurrence of threshold [44] Lavizzari S, Ielmini D and Lacaita A L 2010 A new transient
switching and memory switching in Pt/NbOx /Pt cells for model for recovery and relaxation oscillations in
cross point memory applications IEEE Electron Device phase-change memories IEEE Trans. Electron. Devices
Lett. 33 236–8 57 1838–45
[23] Chua L and Kang S 1976 Memristive devices and systems [45] Cywar A, Dirisaglik F, Akbulut M, Bakan G, Steen S,
Proc. IEEE 64 209–23 Silva H and Gokirmak A 2011 Scaling of silicon
[24] Borghetti J, Li Z, Straznicky J, Li X, Ohlberg D A A, Wu W, phase-change oscillators IEEE Electron. Device Lett.
Stewart D R and Williams R S 2009 A hybrid nano 32 1486–8
memristor/transistor logic circuit capable of [46] Csáthy G A, Tsui D C, Pfeiffer L N and West K W 2007
self-programming Proc. Natl Acad. Sci. 106 1699–703 A stability and negative differential resistance of the wigner
[25] Xia Q et al 2009 Memristor–CMOS hybrid integrated circuits solid Phys. Rev. Lett. 98 066805
for reconfigurable logic Nano Lett. 9 3640–5 [47] Shaw M P, Grubin H L and Gastman I J 1973 Analysis of an
[26] Türel Ö and Likharev K 2003 Cross nets: possible
inhomogeneous bulk ‘S-shaped’ negative differential
neuromorphic networks based on nanoscale components
conductivity element in a circuit containing reactive
Int. J. Circuit Theory Appl. 31 37–53
elements IEEE Trans. Electron. Devices 20 169–78
[27] Shin S H, Halpern T and Raccah P M 1977 High-speed
high-current field switching of NbO2 J. Appl. Phys. [48] Cavalleri A, Rini M, Chong H H W, Fourmaux S, Glover T E,
48 3150–3 Heimann P A, Kieffer J C and Schoenlein R W 2005
[28] Janninck R F and Whitmore D H 1965 Electrical conductivity Band-selective measurements of electron dynamics in VO2
and thermoelectric power of niobium dioxide J. Phys. using femto second near-edge x-ray absorption Phys. Rev.
Chem. Solids 27 1183–7 Lett. 95 067405
[29] Yang J, Miao F, Pickett M D, Ohlberg D, Stewart D, [49] Zhang Y and Ramanathan S 2011 Analysis of ‘on’ and ‘off’
Lau C and Williams R S 2009 The mechanism of times for thermally driven VO2 metal–insulator transition
electroforming of metal oxide memristive switches nanoscale switching devices Solid State Electron. 62 161–4
Nanotechnology 20 215201 [50] Strukov D and Williams R S 2008 Exponential ionic drift: fast
[30] Kwon D-H et al 2010 Atomic structure of conducting switching and low volatility of thin-film memristors Appl.
nanofilaments in TiO2 resistive switching memory Nature Phys. A 94 515–9
Nanotechnol. 5 148–53 [51] Driscoll T, Kim H T, Chae B G, Di Ventra M and
[31] Strachan J P, Pickett M D, Yang J, Aloni S, Kilcoyne D, Basov D N 2009 Phase-transition driven memristive system
Medeiros-Ribeiro G and Williams R S 2010 Direct Appl. Phys. Lett. 95 043503
identification of the conducting channels in a functioning [52] Lee S M, Cahill D G and Allen T H 1995 Thermal
memristive device Adv. Mater. 22 3573–7 conductivity of sputtered oxide films Phys. Rev. B 52 253

8
Nanotechnology 23 (2012) 215202 M D Pickett and R S Williams

[53] Chen I R and Pop E 2009 Compact thermal model for vertical [55] Jacob K T, Shekhar C, Vinay M and Waseda Y 2010
nanowire phase-change memory cells IEEE Trans. Thermodynamic properties of niobium oxides J. Chem.
Electron. Devices 56 1523–8 Eng. Data 55 4854–63
[54] Ward D R, Corley D A, Tour J M and Natelson D 2011 [56] Magnéli A, Andersson G and Sundkvist G 1955 Note on the
Vibrational and electronic heating in nanoscale junctions crystal structure of niobium dioxide Acta Chem. Scand.
Nature Nanotechnol. 6 33–8 9 1402

You might also like