Current fluctuations and silicon oxide wear"'out in metal . . oxide~
semiconductor tunne~ diodes
K. R. Farmer, R. Sa!etti,a) and R. A. Buhrman
Department ofApplied and Engineering Physics, Cornell University, Ithaca, New York 14853-2501
(Received 11 January 1988; accepted for publication 17 March 1988)
We have studied the behavior of very thin oxide (- 20 A.) metal-oxide-semiconductor tunnel
diodes under high electrical field bias. These devices do not usually experience catastrophic
breakdown, but can be worn out at high fields through the creation of a low barrier tunneling
path. The effective area of the path increases during stress, while the barrier height remains
essentially constant at - I eV. The formation of the path is correlated to the presence of
multilevel switching fluctuations in the diode current. The same complex fluctuations and
excess currents are seen in oxides up to 70 A where the fluctuations show up as noisy
precursors to catastrophic breakdown.
Recently, by studying current fluctuations in very thin
insulator (~20 A) metaI-oxide-semiconductcr CMOS)
tunnel diodes, it has been shown that trap-trap interactions
arise in the Si02 layer at high electric fields and that the
collecti ve action of a group of traps plays an important role
in the device behavior. I In this work we show that catastrophic breakdown does not usuaUy occur in these devices.
Rather, the oxide can be progressively worn out at high
fields by an efIect correlated to the presence of very complex
multilevel fluctuations in the device current. The currentvoltage (1- V) curves of a stressed device show the presence
of a tunneling path that can be well fitted by a Fowler-Nordheim (FN) equation with a barrier height of ~ 1 eV. Higher
levels of field stress increase the effective area of this low
barrier path, but do not significantly change the barrier
height. Furthermore, the same complex fluctuations and
creation of "leakage" paths are seen in thicker oxides (up to
-70 A), where complex fluctuations show up as noisy precursors to catastrophic breakdown.
The devices we have studied are 1-2500 f.im 2 AI.Si0 2 pSi tunnel diodes formed in windows etched through a thick
add oxide. Prior to the final oxidation, the wafers were
cleaned under a class-1 0 hood using a standard RCA process
with a final dip in buffered HF and rinsed in de-ionized water
to a bath resistivity of at least 16 MO em. The tunnel oxide
was grown by a rapid thermal oxidation process,2 resulting
in barrier thicknesses of 16-70 A as determined by ellipsometry, tunnel conductance measurements, and for the
thicker oxides, accumulation capacitance measurements. A
room-temperature curve plotting current versus voltage
(I-V) for a fresh device is shown in Fig. 1 (a). It is in good
agreement with that expected for a nonequilibrium minority-carrier diode with an ~ 18 A ( ± 1 A) oxide. 3 Below
-0.3 V, this characteristic is equivalent to that of an abrupt
n-p diode with an ideality factor equal to 1.3. From -0.3 to
~ 1.1 V, increased conduction through the diode is limited
by the semiconductor as the surface of the silicon moves
from inversion to accumulation. Above the fiatband voltage,
-1.1 Y, conduction is mainly due to the direct tunneling of
,,) Al~o with Centro Studi per Metodi e Dispositivi per Radiotrasmi~sioni,
Consiglio Nazionalc delle Ricerche, via Diotisaivi 2,56100 Pi;;a, Italy.
1749
Appl. Phys. Lett 52 (20), 16 May 1988
electrons from the aluminum to the silir.:ou through a trapezoidal barrier. 4 In thicker oxides FN tunneling arises at
higher fields, exhibiting the usual oscillations due to electron
wave reflections at the Si02 -Si interface. 5
We carry out our experiments using a Hewlett-Packard
414GB picoammeter/dc voltage source and a Wentworth
MP-920 probe station. We stress the diodes by applying a
constant negative voltage to the aluminum electrode. \'fiIe
measure /- V characteristics before and after stress, and monitor and record current fluctuations during stress. The effect
of fields up to ~ ! 5 MV / em in devices with < - 20 A barriers is not usually catastrophic breakdown of the oxide, but
a progressive wear-out as measured by the increased device
conductance. This is shown in Fig. 1(0) where we plot five
J. V's taken on the same device before stress and after stress
at various voltages. In these measurements the stress was
maintained until the gradual increase in conduction saturated. It is important to note that, although the electric field
is quite high, the stress voltage is relatively low; thus electron
energies do not approach those needed for such processes as
FN injection into the oxide conduction band and impact
ionization. The stress-induced excess current in the region
above the flat band voltage can be fitted by the FN formula
1= AV 2 exp( -- B IV) as shown in Fig. I (c), where the accuracy of the fit increases with increased excess current.
Thus the "leakage path" that is created in this process appears to be a region of the oxide that has a significantly
lowered effective tunnel barrier, but is not a path dominated
by hopping or metallic conductivity. If we arbitrarily take
the electron effective mass m" = O.Sm e , the barrier height of
the weakened oxide region as determined from F:N" plots of
Fig. ! (c) is E" ~ 1 eV. We find this to be a very general
result. Incre8.sing the level of stress only increases the effective area of the low barrier region and does not significantly
alter E". In Fig. 1 (b) the effective areas of the increased
conduction regions range from 1.1 X 10- II to 1.5 X 10 - 10
cm~. Stress experiments on numerous other devices, with
oxide thicknesses up 50 A, invariably yield FN-Hke excess
currents with fitted values of Eb within 10% of 1 eY. We
particularly wish to note that a recent experiment with pGlycrystalline silicon gate, 50.4 oxide capacitors has also found
that an excess current path is generated with a barrier height
0003-6951/66/201749-03$01.00
(C) 1988 American Institute of PhySiCS
1749
Downloaded 22 Apr 2011 to 128.84.158.108. Redistribution subject to AIP license or copyright; see http://apl.aip.org/about/rights_and_permissions
0. 106 1(a) V
-4
~
-6
(f)
<:J
L
0..
I
io::X+PX 3 ,
-8
(l)
I
X=V-VFB
E
2.99 Volts
I
« -10
'--"
CJ'l
0
..-l
4.1
-12
ioceqV/nxT, n=1.3
-14
ril
0.0
1.0
0.5
1.5
2.0
2.5
Q..
2:
-3
0.149
L--_ _ _ _ _ _ _ _ _ _ _ _- '
o
O.l
a.. -4
E
3.680
« -5
'--""
I
(c)
1604
V
3.61 Volts
I
I
01
0
3.14 Volts
(!)
~
-.J
V
o
-2
O.l
(n)
L.
E
Volts
"..-....
(f)
0.163
(;)
-6
-7
0.9
1,7
1.3
2.1
2.5
2.9
3,644
-3
..--...
N
>
-4
"-
'"v -5
(f)
()
x
<i>
'-"
Q)
0
-.J
-6
-7
.
~
..
" --~.. "~~~~'.
~.~
~
-8
0.35
0.55
0.75
0.95
1.15
1/V
FIG. 1./- V characteristics of 18-20 A MOS tunnel diodes: (a) Unstressed
device; dashed lines show an exponential n-p diode fit below O.3V and the
Simmons formula fit above the flat band voltage. (b) Unstressed device
(solid line) and increasing excess currents after stresses at 3,3.1, 3.2, and
3.3 V, respectively. (c) Excess currents plotted in a Fowler-Nordheim
plane and their fits.
of 0.9 eV. Other experiments with such thick oxides have
also revealed an increased excess current with stress, but it
was not indicated whether this current had a FN character. 7
Insight into the mechanism for the formation of this low
barrier tunnel path is gained by closely mon.itoring the diode
current while under constant voltage bias. At low to intermediate bias we observe stable, low-frequency, two-level and
multilevel switching fluctuations similar to those previously
reported. 1.8 These fluctuations have been identified as being
due to the slow trapping and emission of electrons at individual and strongly interacting groups of trap sites in the oxide.
At higher biases the stahle switching noise is replaced by
large and very complex switching events which arise at random intervals in time. These events we term "wear-out
events." These events are also composed of dear switching
between discrete levels f as shown in Fig. 2(a) L but they
evolve in time and eventually terminate in a permanent
1750
Appl. Phys. Lett, Vol. 52,
Seconds
32,8
FIG 02. Complex fluctuations in the diode current during stress at (a) 2.'19
V, (b) 3.14 V, and (c) 3.61 V.
,
""
~
"'.
L--_ _ _ _ _ _ _ _ _ _ _ _......I
o
Volts
Noo 20, 16 May 1988
change of diode resistance. As the bias is increased, the time
between such individual events becomes shorter, and eventually many events begin to superimpose (Fig. 2(b) J. Finally, at the highest bias aU that can be observed is a steady and
noisy degradation of the oxide as it is stressed [Fig, 2 (c) 1.
The strong correlation between the ococurrence of wearout events and the creation of the excess current path is
shown in Fig. 3. Here we plot the diode current versus time
for a 20 A device biased at 2.9 V. The current is almost
constant in the first minutes of stress; later, two isolated
events occur; then continuous complex fluctuations arise,
Note that the mean value of the diode current begins to in-
90
I
I
~~
V
80
(fl
v
l.-
v
a.
E 70
0
0
L.
U
~
60
V
2.9 Volts
50
0
4
8
12
16
20
Minutes
FI G. 3. Values of the diode current vs lime during stress at 2.9 V to show the
correlation between the complex fluctuations and the increased current.
Farmer, SaleHi, and Buhrman
1750
Downloaded 22 Apr 2011 to 128.84.158.108. Redistribution subject to AIP license or copyright; see http://apl.aip.org/about/rights_and_permissions
crease at a significant rate just at the onset of the continuous
complex fluctuations. As indicated above, once continuous
fluctuations commence, the oxide current grows rapidly, but
then it tends to saturate over time at a fixed voltage stress.
The values of the saturated excess current are approximately
exponentially dependent on the applied voltage, as can be
seen in Fig. 1 (b). When the excess current in a stressed device saturates, the fluctuations do not cease, indicating that
trap occupancy is still strongly fluctuating, but at this stress
level there is no further degradation of the oxide.
Low-frequency noise in devices before and after stress
has also been measured. Spectra of the current noise power
Sf have a good stationary l/fbehavior, and the values of Sf
at a fixed frequency are proportional to /2. Voltage stress
causes the noise level to increase, indicating an increase in
the number of active but stable traps. For example, the noise
level rose by almost an order of magnitude in one device after
stress at 3.1 V until the excess current saturated.
The interpretation of the wear-out events is that they
occur when a randomly initiated change in occupancy of a
trap or group of traps in an already strained oxide induces an
instability which initiates chaotic filling and emptying of interacting traps.! This process continues until the oxide
reaches a more stable condition where the local strain has
been reduced. The process can result in the breaking of oxide
bonds, leading to the creation of spots through the oxide
which have locally reduced barrier height. This could be the
case if, for example, Si-Si bonds were formed during wearout, as has been suggested. 9
Similar experiments have been performed for thicker
oxides (up to 70 .A..). In these devices the oxide failure is
ultimately due to catastrophic breakdown. But both the formation of excess current and the complex fluctuations are
still present. The complex fluctuations are detectable minutes to milliseconds before breakdown, with this time decreasing sharply with increasing voltage and thickness. Figures 4(a) and 4(b) show two examples of current
fluctuations preceding breakdown in diodes with oxide
thickness of 43 and 69 A, respectively. Figure 4(a) shows
dearly that the complex fluctuations stilI consist of discrete
multilevel switching. Because these noisy phenomena always precede catastrophic breakdown, they are undoubtedly correlated with this destructive event. It is possible that
these oxides can be destroyed when electrons trapped in the
oxide can discharge through the more conductive path that
is created once an instability arises.
We have studied very thin oxide ( ~ 20 A) MOS tunnel
diodes in which catastrophic breakdown does not usually
occur. Rather, the oxide can be progressively worn out at
high fields. The resulting excess current can be described
using the Fowler-Nordheim equation with an - 1 eV barrier
height that remains constant during stress, while the effective area of the low barrier "region" increases. The creation
of this excess current path is completely correlated to the
occurrence of complex multilevel fluctuations in the device
current.
1751
4.80\
.V
5.6 Volts
(a)
I
I
[J)
I
Q)
L
(!)
0..0.80
E
0
16.0
0
g 20.3
V
0
Z
9.0 Volts
(b)
19.9L-________________________
a
Seconds
~
0.95
FIG, 4. Complex multilevel current fluctuations preceding electrical breakdown in MOS diodes with different oxide thicknesses, (a) 43 A and (b) 69
A,
Excess current and complex fluctuations also occur in
thicker oxides, where catastrophic breakdown is the ultimate mode offaHure. The multilevel fluctuations invariably
show up as precursors to this failure, suggesting a correlation between the noisy phenomena and breakdown.
In conclusion, we suggest that a process involving traptrap interactions can provide a low electron energy mechanism which accounts for the wear-out of very thin oxide
barriers and possibly the catastrophic breakdown of thicker
oxides.
We thank B. Soave, J. Nulman, and S. Kugelmass for
their essential contributions to the diode fabrication process.
Th.is research was supported by the Semiconductor Research Corporation, and by the National Science Foundation through the National Nanofabrication Facility and
through the Cornell University Materials Science Center.
'K. R. Farmer, C. T, Rogers, and R. A. Buhrman, Phys, Rev. Lett. 51!, 2255
(1987).
'J. !'>lulman, J. P. Krllsius, and A. Gat, IEEE Electron Device Lett. EDL~6,
20S (1985).
'M. A. Green, F. D. King, and J. Shewchun, Solid-State Electron. 17,551
(1974).
4;. G. Simmons, J. AppL Phys, 34, 238 (1963).
'J. Maserjian, J. Vae. Sci. Techno!' n. 996 (1974).
"T. N. Nguyen, P. Olivo, and B. Ricco, in Proceedings afthe IEEE 1987
International Reliability Physics ::')mposium (Electron Devices and Rdiability Societies of the IEEE, New York, 1(87). p. 66.
7J. Maserjian and N, Zamani, J. App!. Phys. 53, 559 (1982).
"n. Neri, P. Olivo. alld B. Ricco. Appl. Phys. Lett. 51, 2!67 (1987).
"1,. A. Weinberg and T. N. Nguyen, 1. AppL Phys. 61.1947 (1987),
AppL Phys, Lett., Vol. 52, No. 20, 16 May 1988
Farmer, SaleHi, and Buhrman
1751
..•-..•-.... "•••;•.•.•
........n.•_•.":;..>;...... .. •• .-. ••. , ..• ';-;-.".-. " ••.•.•;'"•••••
••••••
...·.:.·.·.·.·.-, •.•.•.•.•.•."'"....... ••-•..•.
Downloaded 22 Apr 2011 to 128.84.158.108. Redistribution subject to AIP license or....."copyright;
see http://apl.aip.org/about/rights_and_permissions
,..v,..·.·.-.-.-.-.~,.-..·.·.~?·."'·.·.·.·.7.-.-
"'7'".~.~.:.~.:.:.;>;
~.v.·.y
.".-.~
>;0; ••••" ........-.-••