Table Of ContentTotal Ionizing Dose Effects in MOS Oxides and Devices
T. R. Oldham', Fellow, IEEE and F. B. McLean,*Fellow,I EEE
I. Introduction
reason, the overall radiation response of a device
The development of military and space or circuit can be extremely complex, sometimes
electronics technology has traditionally been to the point of bewilderment. However, the
heavily influenced by the commercial overall response can be separated into its
semiconductor industry. The development of components, and the components can be studied
MOS technology, and particularly CMOS individually. In fact, this has happened. Many
technology, as dominant commercial different individual investigators have studied
technologies has occurred entirely within the different parts of the radiation response over a
lifetime of the NSREC. For this reason, it is not period of many years, and a reasonable degree of
surprising that the study of radiation interactions understanding has now been achieved. This
with MOS materials, devices and circuits has understanding represents a major
been a major theme of this conference for most accomplishment of the NSREC and the NSREC
of its history. community. Much of this understanding has
been captured elsewhere already-Ma and
The basic radiation problem in a MOS transistor Dressendorfer' edited a major review volume. In
is illustrated in Fig. 1, where Fig. la shows the addition, Oldham*p repared another book, whch
normal operation of a MOSFET. The was intended to update parts of the Ma and
application of an appropriate gate voltage causes Dressendorfer book, to reflect later work. Both
a conducting channel to form between the source volumes discuss the same material to be
and drain, so that current flows when the device presented here, and in far greater detail than is
is turned on. In Fig. lb, the effect of ionizing possible here.
radiation is illustrated. Radiation-induced
trapped charge has built up in the gate oxide, 11. Overview
which causes a shift in the threshold voltage (that
is, a change in the voltage which must be applied We begin with an overview of the time-
to turn the device on). If this shift is large dependent radiation response of MOS systems,
enough, the device cannot be turned off, even at before discussing each of the major physical
zero volts applied, and the device is said to have processes in greater detail. Then we will discuss
failed by going depletion mode. the implications of the radiation response for
testing, prediction, and hardness assurance. We
will also discuss the implications of scaling
(reducing the oxide thickness), and issues
associated with oxide isolation structures and
leakage currents.
(4) RINATDERIAFTAICOEN -ImNDUnCED
WITHIN SI SANWAP
I I
**+&*- ./
,
Figure 1. Schematic of n-channel MOSFET ,-,I --.
dI HOLE
TRAPPlNQ
illustrating radiation-induced charging of the NEAn SIlWO*
INTERFACE
gate oxide: (a) normal operation and (b) post-
irradiation. (2) HOPPINQ TRANSPORT
OF HOLEB THRWQH
LOCALIZED STATES
(1) ELECTRONNOLE PAIR8
QENERATED SI IONlZlNQ IN WLK
In practice, the radiation-induced charging of the RAD1ATK)N
oxide involves several different physical
mechanisms, which take place on very different Figure 2. Schematic energy band diagram for
time scales, with different field dependences and MOS structure, indicating major physical
different temperature dependences. For this processes underlying radiation response.
1. NASA GSFC/QSS Group, Inc.
2. Consultant
Fig. 2 shows a schematic energy band diagram The third process in Fig. 2 and Fig. 3 is that
of a MOS structure, where positive bias is when they reach the Si interface, some fraction
applied to the gate, so that electrons flow toward of the transporting holes fall into relatively deep,
the gate and holes move to the Si substrate. Four long-lived trap states. These trapped holes cause
major physical processes, which contribute to the a remnant negative voltage shift, which can
radiation response of a MOS device, are also persist for hours or even for years. But even
indicated. The most sensitive parts of a MOS these stable trapped holes undergo a gradual
system to radiation are the oxide insulators. annealing, which is illustrated in Fig.3.
When radiation passes through a gate oxide,
electrodhole pairs are created by the deposited The fourth major component of MOS radiation
energy. In SiOz, the electrons are much more response is the radiation-induced buildup of
mobile than the holes3, and they are swept out of interface traps right at the Si/Si02 interface.
the oxide, typically in a picosecond or less. These traps are localized states with energy
However, in that first picosecond, some fraction levels in the Si band-gap. Their occupancy is
of the electrons and holes will recombine. That determined by the Fermi level (or by the applied
fraction will depend greatly on the energy and voltage), giving rise to a voltage-dependent
type of the incident particle. The holes, which threshold shift. Interface traps are highly
escape initial recombination, are relatively dependent on oxide processing, and other
immobile, and remain near their point of variables (applied field and temperature).
generation, where they cause a negative
threshold voltage shift in a MOS transistor. Fig. 3 is schematic in that it does not show real
These processes, electrodhole pair generation data, but it reasonably represents the main
and recombination, together are the first process features of the radiation response of a hardened
depicted in Fig. 2. In Fig. 3, this process n-channel MOS transistor. The range of data,
determines the (maximum) initial threshold from loa s to 10’ s, is enormous, as it has to be
voltage shift. to include qualitatively the four main processes
we have discussed. For the oxide illustrated in
I
Fig. 3, a relatively small fraction of the holes
0 reaching the interface are trapped, which is why
we say it is realistic for a hardened oxide. Many
oxides would trap more charge than is shown
9 here. In addition, the final threshold shift,
Q
including interface traps, is positive (the so-
called rebound or superrecovery effect) here
because the number of negatively charged
interface traps finally exceeds the number of
10- 1.4 10-2 13 13 lr‘ rd le
trapped holes. Not all oxides really have this
TYE rn MDlATlow PlnoE bl
behavior, but it is one of the results, which can
Figure 3. Schematic time-dependent post- be considered “typical.”
irradiation threshold voltage recovery of n-
channel MOSFET, relating major features of the
response to underlying physical processes.
111. Description of Basic Physical
The second process in Fig. 2 is the transport of Processes Underlying the Radiation
the holes to the Si/Si02 interface, which causes Response of MOS Devices
the short-term recovery of the threshold voltage
in Fig. 3. This process is dispersive, meaning Next, we consider these basic physical
that it takes place over many decades in time, mechanisms in more detail, and provide critical
and it is very sensitive to the applied field, references. But for a complete review, the
temperature, oxide thickness, and (to a lesser readers should consult the references.
extent) oxide processing history. This process is
normally over in much less than a second at A. Electron-Hole Pair Generation
room temperature, but it can be many orders of Energy
magnitude slower at low temperature.
The electrodhole pair creation energy, E,, was
determined to be 18+3 eV by Ausman and
McLean? based on experimental data obtained separation between pairs is much greater than the
by Curtis et ai.' This result has been confirmed thermalization distance, the distance between the
independentlyb y including a more hole and the electron. One can treat the
accurate set of measurements and analysis by interaction between the charges of an isolated
Benedetto and Boesch,8w hich established pair, which have a mutual coulomb attraction,
E,=17+1 eV. From this value of E,, one can which undergo drift motion in opposite
calculate the charge pair volume density per rad, directions under the influence of the applied
go= 8.1~10p'a~ir s/cm3-rad. But this initial field, and which have a random diffusion motion
density is quickly reduced by the initial driven by the thermal fluctuations of the system.
recombination process, which we discuss next. But interactions with other pairs can be
neglected. The geminate recombination model
B. Initial Hole Yield was first formulated by Smoluchowski: and later
solved by Onsager," originally for the
The electrons are swept out of the oxide very recombination of electrons and positive ions in
rapidly, in a time on the order of a picosecond, gases. Experimental and theoretical results for
but in that time some fraction of them recombine the geminate process are shown in Fig. 5, where
with the holes. The fraction of holes escaping the theoretical curve was obtained by Ausman,"
recombination, fy(EOx)i,s determined mainly by assuming the average thermalization radius, rr, to
two factors: the magnitude of the electric field, be 5 nm.
which acts to separate the pairs; and the initial
line density of charge pairs created by the
incident radiation. The pair line density is
determined by the linear energy transfer (LET),
and is, therefore, a function of the incident
particle type and energy. The line density is also
inversely proportional to the average separation
distance between electrodhole pairs-obviously,
the closer the average spacing of the pairs, the I
I I I I
more recombination will occur at a given field, a aa 1.m 1.1 La La a.a La 4.a
uILem(Ylk.0
and the less the final yield of holes will be.
Figure 5. Fractional yield as a function of
applied field for Cow gamma rays, 12-MeV
electrons, and geminate model calculation^.^*
-L v 'I* l7
uw€',cA'\r'lu a.i -\- ,i- /' The other case, called columnar recombination,
is illustrated in Fig. 4b, where the separation
(A) GEMINATE
RECOMBINATION between pairs is much less than r,. There are
MOOEL (SEMiNED several electrons closer to any given hole than
ELECTRON-HOLE MIRS)
4i- the electron, which was its original partner, so
(B) COLUMNAR RECOMBINATION the probability of recombination is obviously
MOOEL (OVERLAPPING much greater than in the geminate case. The
ELECTRONHOLE MRS) '
columnar model was originally solved
analytical1 by Jaffe,'* extending earlier work by
Figure 4. Schematic diagrams indicating pair Langevin. More recently, Oldl~am'~h-a's~
separation distances for two recombination presented a more accurate numerical solution of
models: (a) geminate (separate electrodhole the Jaffe equation, which extends the range of
pairs) and (b) columnar (overlapping applicability of the model. Representative
electrodhole pairs). experimental data, for 2 MeV alpha particles, are
presented in Fig. 6,14 along with theoretical
The recombination problem cannot be solved curves. The parameter, b, is the half-diameter
analytically for arbitrary line density, but assumed for the initial Gaussian charge
analytic solutions do exist for the limiting cases, distribution. Recombination results for a
where the pairs are either far apart, or very close variety of incident particles are summarized in
together. These cases are illustrated in Fig. 4. Fig. 7.17 At a field of 1 MV/cm, there is a
Fig. 4a shows the so-called geminate difference of more than an order of magnitude in
recombination model, where the average the yields shown for different particles. Clearly,
recombination is an important effect, which must neither model is strictly applicable, as illustrated
be accounted for, when comparing the effect of in Fig. 8. The original version of this figure was
different radiation sources. published by Oldham18i n 1984, based on
experimental data by Stassinopoulos et
and Brucker et al.”. 23 But additional data points
have been added from time-to-time, as other
experiments were rep~rted.’~T”h~e se later
experiments have generally reported lower yield
(that is, more recombination) than the earlier
experiments. The different experiments have all
been done at different fields, so the geminate
model limit is different in each case, as is now
indicated in Fig. 8. The other key point is that
the LET for a proton and an electron is not really
the same until their energy is about 1000 MeV,
Figure 6. Fractional yield as a function of well above the energy of any incident particle in
applied field for alpha particles incident on SiOz. any of these experiments. In the original version
Solid lines indicate columnar model results for of Fig. 8, the geminate model was indicated
different initial column radii.14’I 6 schematically to apply from 1000 MeV down to
about 150 MeV because it appeared to fit the
As one might expect, there are also a number of data available at that time. But the more recent
intermediate cases of practical interest, where data indicates that the recombination process is
not purely geminate until the proton energy is
I 1 well above 100-200 MeV.
C. Hole Transport
The transport of holes through the oxide layer
has been studied extensively by several
gro~ps,*~an-d~ *it has the following properties:
(1) transport is highly dispersive, taking place
700-k0V PROTONS
over many decades in time following a radiation
pulse. (2) It is universal in nature, meaning that
changes in temperature, field and thickness do
not change the shape or dispersion of the
recovery curves on a log-time plot. Changes in
ELECTRIC FIELD (MVlcm) these variables affect only the time-scale of the
Figure 7. Experimentally measured fractional recovery. (3) The transport is field activated. (4)
hole yield as a function of applied field, for a At temperatures above about 140K, the transport
number of incident particles. is strongly temperature activated, but it is not
temperature activated below about 140K. (5) The
1.0 I hole transport time, or recovery time, has a
strong super-linear power law dependence on
oxide thickness.
The best overall description of the experimental
3
ai
E hole transport data seems to be provided by the
.J
BRUCKER U al. CTRW (continuous-time-random-walk) hopping
e SSTTAUSOSRIN uOW aU. LOS U (1. transport formalism, which was originally
develo ed by Montroll, Weiss, Scher and
others!’” This formalism has been applied to
0.01 I 1I 1I0 101 0 1 hole transport in silicon dioxide by McLeanZ73,0 *
PROTON ENERQV (YEW 31,35-38 and by Hughes.Z8*z9*(3H4o wever, the
multiple trapping 33 also accounts for
Figure 8. Recombination measurements and many of the features of the experimental data.)
calculations for protons incident on SiOz. The specific transfer mechanism seems most
likely to be small polaron hopping of the holes
between localized, shallow trap states having a
random spatial distribution, but having an
average separation of about 1 nm. The term
polaron refers to the situation where the carrier
interacts strongly with the surrounding medium,
creating a lattice distortion in its immediate
vicinity (also referred to as self-trapping). As the
hole hops through the material, it carries the
lattice distortion with it. The strongest evidence Figure 10. Normalized flatband recovery data of
for the polaron hopping mechanism is the Fig. 9 replotted with time scaled to half recovery
transition from thermally activated transport time, showing universal response with respect to
above about 140k, to non-activated transport at temperature. Solid curve is CTRW model result
lower temperatures. This transition is a classic for a=0.25.
signature of polaron h0pping.4~~M' any other CHARGE RELAXATION DATA (T=79K)
features of the hole transport, such as dispersion, 0.0, I I 1 I 1 1
universality, and superlinear thickness
dependence, can be attributed to a wide
distribution of hopping times for the individual
holes.
0
g 0
0.50 A
Representative experimental data are presented
0
in Figs. 9-12. Fig. 9 shows the effect of 0 O0
temperature variation, and Fig. 10 shows the A (4MVlcm) 0
effect of varying the electric field. In both Figs.,
the results cover seven decades in log-time,
following a short radiation pulse. In both Fig. 9
CHARQE, REUXATION DATA (8- = 1MVICM) TIME AFTER PULSE (8)
OD, I
Figure 11 . Normalized flatband voltage recovery
data following pulsed Linac electron beam
0 0 exposure of 96.5 nm oxide capacitor at 80K for
(2474 0 A
tO,,. 0 0 different fields.
, narTtu"rRIEVs0mE~
10' I 1
TIME AFTER FIJLSE (a)
Figure 9. Normalized flatband voltage recovery
following pulsed Linac 12-MeV electron
irradiation of 96.5 nm oxide at different
temperatures.
I I
1
10 2030 5u loo 200
and 10, the flatband voltage shift is plotted as a OXIDE THICKNESS (nm)
function of log-time, normalized to the Figure 12. Recovery time as a function of oxide
calculated shift before any transport occurs. In thickness for etched-back and as-grown oxides.
Fig. 9, the field is 1 MV/cm, and the strong
temperature activation above 140K is apparent. curves from Fig. 9 are replotted for scaled time.
In Fig. 10, all the curves are taken at T=79K. The entire transport process covers 14 decades in
The universality and dispersion of the transport time (!), and all the curves have the same "S"
is better illustrated in Fig. 11 , where all the shape. The time, t1/2,a t which the flatband
voltage reaches 50% recovery, has been used as
the scaling parameter. The solid line is an 69 Both processes can give rise to the linear-
analytical fit of the CTRW model, where the with-log t de endence that has been observed
shape parameter, a,h as the value 0.25. Finally, empirically, but one has to make different
Fig. 12 shows how the hole transit time varies assumptions about the trap energy level
with oxide thickness, as t,”cc, or about b4.T his distribution for the two processes.
oxide thickness dependence arises because the
farther the holes transport, the greater the o [ v K -
probability that some of them will be in states
where the next hop is a difficult one, one that
-1
takes a long time to happen. Then, the farther
the holes go, the slower they move.
D. Deep Hole Trapping and
Annealing
The most complete discussion of hole trapping
(I x 10sr ads)
and annealing is by Oldham2 Deep hole traps
Tm (hr)
near the Si/SiOz interface arise because there is a
transition region where oxidation is not Figure 13. Trapped hole annealing; negative
complete. This region contains excess Si, or bias curve shows that “annealed” holes are not
oxygen vacancies, depending on how one looks really removed.70
at it. The oxidation process was described by
Deal et al.46i n an early review article, which was The study of radiation-induced trapped hole
based on original work done even earlier. annealing has led to new insights about the
Eventually, Feigl et al.47p resented a convincing atomic structure of the oxide trap, which in turn
model for the single oxygen vacancy. Basically, has led to new insights into the structure of
there is one oxygen atom missing from the usual neutral electron traps, which play a critical role
lattice configuration, leaving a weak Si-Si bond, in breakdown studies and in other reliability
where each Si atom is back-bonded to three problems. (For a full discussion, see reference
oxygen atoms. When a positive charge is 2.) One of the key results is illustrated in Fig.
trapped, the Si-Si bond is broken, and the lattice 13, originally reported by Schwank et al.70 An
relaxes. The key point that Feigl added to earlier irradiated sample was annealed under positive
discussions was that the relaxation is bias at 100 C for about one week, and all the
asymmetric-one atom relaxes into a planar trapped positive charge appeared to be removed.
configuration, and the other remains in a But then they applied a negative bias, and about
tetrahedral configuration. The oxygen vacancy half the neutralized positive charge was restored
was also eventually connected to the E’ center, within a day. This result led to the idea that
originally detected by Weeks4*i n a-quartz, but annealing of radiation damage involved a
also later detected in bulk glasses and thermally compensation process. That is, defects were
grown SiOz.49 The correlation of E’ centers, and neutralized without being removed. Although
oxygen vacancies, with radiation-induced other groups quickly confirmed the basic re~ult,’~
trapped holes was first established by Lenahan $7I several years passed before Lelis et al. did a
and Dre~sendorfer.~~ thorough study.72-74O ne of their key results is
shown in Fig. 14, where a hardened oxide is
Oxide trapped holes are relatively stable, but exposed to a short Linac radiation pulse, and
they do undergo a long-term annealing process then subject to a series of alternating positive and
which can extend for hours or even years, with a negative bias annealing steps. Under negative
complex dependence on time, temperature, and bias, a significant amount of neutralized positive
applied field. Generally, trapped hole annealing charge reappears, but there is also a significant
can proceed by either of two processes, tunneling amount of “true” annealing, where the trapped
or thermal excitation. At or near room charge really is removed. Lelis et al. proposed a
temperature, tunneling is the dominant model, illustrated in Fig. 15, to account for their
mechanism, but if the temperature is raised results and results of others. Generally, it had
enough, the thermal process will eventually been assumed that annealing proceeded by an
dominate. Tunneling has been analyzed by electron tunneling
several author^,'^'^ as has the thermal process.s7-
putting an extra electron on a neutral Si atom
instead of a positive Si, required overcoming an
electron-electron repulsion. Lelis et al. pointed
out that adding the extra electron to the positive
Si would require changing a planar configuration
of atoms into a tetrahedral configuration, moving
around atoms in the lattice to change bond
angles, which would require adding energy. The
energy to rearrange the lattice might be greater
than the electron-electron energy, which would
mean stable dipoles would be energetically
favored. But, initially, they lacked the means to
Figure 14. Alternate positive and negative bias quantify this argument, so debate continued for
annealing for capacitor exposed to 4-ps Linac several years.
pulse.
Several other, independent, experiments
to the positively charged Si, neutralizing it, and produced results, whch seemed to support the
reforming the Si-Si bond. Instead, Lelis et al. dipole hypothesis. These included TSC
proposed the electron tunnels to the neutral Si, (thermally stimulated current) measurements,
forming a dipole structure, where the extra first by Shanfield et a1.,60-6a2n d later by
electron can then tunnel back and forth to the Fleetwood et al.63-67I n addition Walters et al.79
substrate in response to bias changes. This concluded that the dipole proposed by Lelis et al.
model is consistent with the ESR work of acted as a neutral electron trap in their injection
Lenahan et and with the electrical results of experiments. Th~ws ork was an important
Schwank and others, explaining a variety of independent confirmation of the dipole
complex results in terms of a single defect, hypothesis, but it was also an important
which was already well-known. The transition extension of it. Basically, they argued that the
from Fig. 15a to 15b was described by Feigl et positive end of the dipole acted as an electron
al. The transition from Fig. 15b to 1% and back trap, so that a second electron could be trapped,
describes the switching reported by Schwank et making the whole complex net negative. The
al. and by Lelis et al. And the transition from reason this result was important was that there is
Fig. 15c back to 15a indicates the true annealing, an enormous body of literature on neutral
which is also observed. electron traps and the critical role they play in
non-radiation-induced reliability problems.
(This literature is too extensive to discuss here-
see reference 2.) But suddenly, the Lelis et al.
dipole hypothesis was critical for explainin all
this other work. And finally, Conley et a1.8’
I l l
Figure 15. Model of hole trapping, permanent 0.5
H+ 01 N1 P1 N2 P2
annealing, and compensation processes. Bias condition
Figure 16. E’ density during alternate positive
The dipole hypothesis by Lelis et al. was and negative bias annealing.
attractive, because it explained many things very
simply, but at first, it was also controversial. It conducted ESR experiments, where they cycled
was criticized by three different group^,^^-^' for charge back and forth by alternating bias, while
different reasons. The biggest problem was that monitoring the E’ signal. Their main result is
shown in Fig. 16. The dipole model was the only
one consistent with this result, so it settled the There have been many conflicting models
debate, at least on the experimental side. Even proposed to describe the process(es) by which
more recently, two theoretical groups have done radiation produces interface traps, and much
quantum mechanical calculations, using different controversy about them. However, a reasonable
mathematical approaches, which have also degree of consensus has finally emerged. All the
indicated that dipoles should be energetically models are consistent with the idea that the
favored under certain conditions.82893 In precursor of the radiation-induced interface trap
addition, Fleetwood et al. have recently extended is a Si atom bonded to three other Si atoms and a
this model to argue that it also accounts for l/f hydrogen atom. When the Si-H bond is broken,
noise results, which they rep~rted.'~ the Si is left with an unpassivated dangling bond,
as an electrically active defect. The process by
We note that, in recent years, there has been which the Si atom is depassivated is where the
much discussion of the role of border traps, models differ. Now it has become clear that the
oxide traps that exchange charge with the Si dominant process is a two stage process
substrate. The proposal to call these traps border involving hopping transport of protons,
traps was made by FleetwoodS5i n 1992. At that originally described by McLean,gO which was
time, the dipole model by Lelis et al. had been in based on a series of experimental studies by his
the literature for four years, and was already well coworker^.^'-^ This work was confirmed and
known. Now, more than ten additional years extended in a series of additional studies by Saks
have passed, and the defect described by Lelis et and his coworkers.'00s'06I n the first stage of this
al. is still the only confirmed border trap, at least process, radiation-induced holes transport
in the Si/Si02 system. Other border trap through the oxide, and free hydrogen, in the form
structures have occasionally been proposed,86 of protons. In the second stage, the protons
but they have not done well in subsequent undergo hopping transport (following the CTRW
experimental tests.8o* formalism described above). When the protons
reach the interface, they react, breaking the SiH
E. Radiation-Induced Interface bonds already there, forming H2 and a tri-valent
Traps Si defect. One of the critical experimental
results is shown in Fig. 17, which shows the
8899 9
Radiation-induced interface states have been
3.0
identified with the so-called Pbor esonance in LINAC. 600 Lrd
ESR studies, by Lenahan and Dressendorfer." L rnl 0-
This center is a tri-valent Si atom, back bonded
to three other Si atoms, with a dangling bond 2.0
extending into the oxide. This defect is
amphoteric, negatively charged above mid-gap,
neutral near mid-gap, and positively charged
below mid-gap. Lenahan and Dressendorfer
showed a very strong correlation between the
build-up of the PM resonance and the buildup of
.
interface traps, as determined by electrical
Tim. (D)
measurements. There is also an extensive
literature suggesting that thls same defect is also Figure 17. Experimental results from field
present as a process-induced interface trap. We switching experiments that support H+ transport
cannot review all this literature here, but other model.
* ''
reviews have already been published.'. (In
2s 89
particular, see the references in Chapter 3 of results of bias switching experiments. For curve
reference 2.) The basic picture, however, is that A, the sample was irradiated under positive bias,
when the oxide is grown, there are still about which was maintained throughout the
1O '3/cm2u npassivated tri-valent Si centers. In experiment, and a large interface trap density
subsequent processing, almost all these centers eventually resulted. For curve B, the bias was
are passivated by reacting with hydrogen. negative during irradiation and hole transport, so
However, they can also be depassivated, either the holes were pushed away from the interface,
by radiation interactions or by other but the bias was switched positive after 1 sec,
environmental stresses. during the proton transport. The final number of
interface states for curves A and B is almost the
same, however. For curve E, the bias is There have also been several models proposed
maintained negative throughout both stages, and where trapped holes are somehow converted to
interface trap production is suppressed
completely. For curves C and D, the bias is
negative during irradiation and hole transport,
..L
but switched positive later than for curve B. In
all cases, bias polarity during the hole generation
and transport made no difference, but positive
bias during the proton transport was necessary to
move the protons to the Si/Si02 interface. The
time scale of the interface trap build-up was
determined by the transport time of the protons.
(For curves B,C, and D, the protons were
0.m1 100 200 !
initially pushed away from the interface, so it
WWrprM FI
took them longer to get there after the proper Figure 18. Isochronal annealing results showing
bias was applied.) McLean also worked out the small neutral hydrogen diffision process and
average hopping distance for protons to be .26 larger H+ transport process for interface trap
nm, which is the average distance between formation.
oxygen atoms. And he determined the activation
energy for the interface trap buildup to be 0.82 interface traps-usually the details of the
eV, which is consistent with proton transport.Io7 conversion process are not specified.' I2-' I6
Saks eventually succeeded in monitoring the These models are not well regarded today. For
motion of the protons directly.lM example, in Fig. 17, curves B and E have the
same hole trapping. For this reason, one might
The two stage, proton transport model is a robust expect these models to predict similar interface
model at this point. It has been confirmed by trap build-ups, contrary to what is shown in the
different groups (McLean and coworkers, and figure. If one studies hole trap removal and
Saks and coworkers), using different test interface trap build-up carehlly, the two
structures (capacitors and transistors, processes have different time dependences,
respectively), different gate technologies (A1 different temperature dependences, and different
active metal and poly-Si, respectively), and bias dependences. They have the same dose
different measurement techniques (C-V analysis dependence, but otherwise seem to be
and charge pumping respectively). Despite all completely independent. Here, we cannot
these variations, this process has always been discuss these things in detail, but fill discussions
the main effect. However, it does not explain appear in Chapter 3 of reference 2, and in
everydung. Boesch'" and Saks'" have reference 88. However, Oldham et a1.88p rovided
identified a second order effect, where a small the most likely explanation for experimental
part of the interface trap build-up seems to results purporting to show trapped hole
correlate with the arrival of transporting holes at conversions. They pointed out that trapped holes
the interface. Presumably, the holes break the that do not undergo a defect transformation can
Si-H bonds instead of protons. Also, Griscom''' account for most of these results-that is,
and Brown"' had also proposed originally, that trapped holes look like interface traps in some
diffision of neutral hydrogen, rather than drift experiments. (The model for these holes is
transport of protons, was the main mechanism discussed in the previous section.) The exchange
for interface trap production. But, Saks et a1."' of charge between trapped holes and the Si
were able to isolate the neutral hydrogen effect, substrate has been extensively studied since then
and showed that it was also small. The key (usually under the name border traps),86'*I7 and
result is shown in Fig. 18, where the interface the idea that such charge exchange takes place is
trap build-up between 120K and 150K is due to no longer considered unusual.
neutral hydrogen, and the build-up above 200K
is due to proton transport. The vertical scale is a Finally, some samples exhibit what has been
log scale here, so the neutral hydrogen process called a latent interface trap build-up, which is
accounts for only a few percent of the total build- illustrated in Fig 19.Ii8 This effect is thought to
up. Griscom and Brown both eventually be due to hydrogen diffusing into the gate oxide
endorsed the McLean model."' region from another part of the structure, perhaps
the field oxide or an encapsulating layer. The however, the final threshold shift is positive,
latency period arises because the hydrogen is about 3.5 V, which is more than enough to fail
diffising from (relatively) far away. From a the device. The effect of raising the temperature
1.2 1 REBOUND EFFECT
0 25%
(1 x 10‘ ndC)
Figure 19. Latent interface trap formation for TIME (kr)
OKI p-channel transistors.’I8 Figure 20. VT, VOT,a nd VITa nnealing; long
term response illustrates rebound effect.”
testing point of view, this is a difficult effect to
account for. There is no trace of it on the time is to anneal the hole traps faster, but the same
scale of most laboratory tests, yet it can final state is reached at room temperature. One
eventually be a large, even dominant, effect, on a could imagine a device that failed due to trapped
time scale of months. Saks et a1.IM subsequently positive charge immediately after irradiation,
reported that the latent build-up is suppressed by that would later work properly for a time as some
a nitride encapsulating layer, which serves as a of the holes annealed, that would then fail again
barrier to hydrogen diffision. Unless one knows later from trapped negative charge as more of the
how the samples are encapsulated, it is not holes annealed. There is very little change in
possible to predict ahead of time whether a latent interface trap density during the annealing
build-up will occur, or not. process at either temperature, so the late time
failure would be due to interface traps that were
IV. Implications for Radiation Testing, there at the end of the irradiation.
Hardness Assurance, and
Prediction To test for rebound, there is now a standard test
method, 1019.4, which calls for a 100 C anneal
We have now completed our review of the basic for 168 hours, which will detect the effect in
physical mechanisms underlying the radiation oxides similar to the one used in Fig. 20. This
response of CMOS devices. Next we consider method represents a compromise, because 100 C
these mechanisms in the context of device and is not a high enough temperature to accelerate
circuit testing, hardness assurance, and the trapped hole annealing in all oxides, but if
prediction. one goes much higher in temperature, the
interface traps may anneal,
A. Rebound or Super-Recovery
The rebound effect is of great practical concern
The rebound effect is illustrated in Fig. 20,” for space environments, because components are
which shows threshold voltage shift (AVT)f or a typically exposed to relatively low dose rates for
MOSFET, along with its components, oxide very long mission lifetimes. There is now strong
trapped charge (AVO,) and interface trapped emphasis on using unhardened commercial
charge (AVIT), during both irradiation and post- technology as much as possible, which is
irradiation annealing. The annealing data is reasonable to consider in space because oxide
shown for two temperatures, 25 C and 125 C. traps may anneal as fast as they are created, or
After irradiation, the threshold shift is less than nearly so. A component that fails at a low dose
IV, but this relatively small shift is obtained by in a laboratory test from oxide trapped charge, as
compensating positive oxide trapped charge with many things do, may work quite well in space,
negatively charged interface traps (in this n- because of the low dose rate and annealing. But
channel device). When the hole traps anneal, this discussion only applies to the positive
