Plasma Charging Damage to gate dielectric

1
Plasma Charging Damage to gate dielectric
-- past, present and future
Kin P. Cheung
Rutgers University
2
Plasma charging damage is electrical stress of
gate dielectric during plasma exposure
Potential difference developed across the MOS
capacitor
Damage device degradation or breakdown
3
Since the first report in 1983, damage evolves
with technology.
High-k
4
Early reports (1983 - 1987)
First report
Y. Yoshida and T. Watanabe, Dry Process Symp.,
1983
Gate oxide 400Å
Damage mode Dead (broken) on arrival.
General characteristics thick gate oxide (gt150Å)
Damage mode oxide breakdown
General observation is that plasma charging tends
to break weak oxides. It high-light the quality
problem in oxide.
If the oxide is not broken, you are okay.
5
Rising period (1987 - 1993)
Plasma front
Migration from batch process to single wafer
process.
Large number of experimental plasma systems.
Higher density plasma introduced
Gate oxide front
Becomes thinner (150Å - 100Å)
Increasing awareness of extrinsic (weak point)
failure.
Result
Many reports based on highly non-uniform plasma.
However, in production systems, plasma is quite
uniform.
Many do not believe plasma charging damage is
serious.
Wisdom if your oxide is good, no need to worry
about plasma charging damage.
6
The main point of confusion damage mode changed
For thick oxide, if it didn't break, you are okay.
For thinner oxide, not broken is not good enough.
During this period, many did not realize that
damage measurement method needs to change.
7
Oxide under stress
The intrinsic breakdown field is a function of
oxide thickness
8
Fowler-Nordheim Tunneling
For thick oxide, very little current can flow
across below breakdown stress.
9
Impact ionization induced current run away
breakdown mode
Once appreciable tunneling is reached, feed back
run away sets in.
For thick oxide, it will either break or nothing
will happen.
This mode dominate until the oxide is too thin to
support the impact ionization.
10
During the rising period, appreciable tunneling
occurs at below breakdown field.
The absence of feed back run away changes the
breakdown mechanism to defect accumulation.
11
New breakdown model
Defects are generated during high-field stress.
It is related to the current that flow across the
oxide.
For a given defect density, there is a
corresponding probability that a linkage between
the two electrodes occurs (breakdown).
These defects can trap charges and lead to device
degradation long before the defect density is
high enough to cause breakdown.
Breakdown is no longer the main mode of failure.
12
Accumulation of defects causes charges to be
trapped in the oxide.
Current decreasing due to electron trapping.
13
Defects in the oxide or its interface cause
device to degrade.
Instead of looking for oxide breakdown, one must
measure transistor parameter degradations.
14
Peak period (1993 - 1999)
Plasma front
Single wafer process has become main stream.
The number of plasma types narrowed down.
High density plasma widely used.
Gate oxide front
Becomes thinner (100Å - 50Å)
Extrinsic (weak point) failure largely eliminated.
Technology front
High aspect ratio windows and line spacing.
Result
New mode of charging damage (electron shading).
Plasma charging damage is found on every plasma
step.
Antenna rule becomes standard design rule.
Device and circuit degradation the main damage
mode.
Plasma damage conference started.
15
Gate oxide becomes so thin that large tunneling
current can be sustained at below breakdown field.
Plasma as a current limited voltage stress source
is widely recognized.
Antenna amplify the stress current.
16
As gate oxide thickness shrink, the voltage
required to achieve significant tunneling
decreases.
The number of plasma process that can sustain
significant stress current increases.
The defect generation rate per tunneling electron
remains roughly constant.
During this period, plasma system improved to
minimize the ability to develop charging voltage,
but the charging current capability remains high
(high density plasma).
The shrinking of feature size leads to electron
shading type of charging damage that improvement
of plasma system would not help.
17
Electron shading charging
e-
ions
Electrons are filtered by the aperture formed by
the high aspect ratio lines ----- Ions can reach
to the bottom of the lines better and thus raise
its potential
FN current
Multiple mode of charging add to each other to
ensure that damage remain severe.
18
Krishnan S et al., IEDM96, pp731.
Damage is severe enough that oxide breakdown
occurs.
19
The down hill period (1999 - 2002)
Plasma front
No major changes except oxide etching largely
changed from high density plasma to medium
density.
Gate oxide front
Electron transport becomes ballistic.
Aggressive use of thinnest possible oxide limited
by breakdown reliability.
Statistic-based scaling effect increasingly
severe.
Reliability as a show-stopper is taken seriously.
General impression
Plasma charging damage is becoming un-important.
20
Since there is no major change in plasma systems,
the continue thinning of gate oxide only lead to
more charging opportunities.
Why would damage go down?
First, the charging current does not go up with
thinner oxide -- plasma charging is current
limited in most cases.
Second, the defect generation rate per tunnel
electron goes down rapidly with stress voltage.
Lower defect density gt less device degradation.
21
At high stress field, it is the energy of the
electron reaching the anode that determines oxide
degradation
Collision with TO phonons
Ballistic transport, electron Energy at anode
equals VG
Electron energy at anode Does not depend on VG
22
Anode-hole injection model of defect generation
As voltage drops below 5V, defect generation rate
drops precipitatively.
23
For the same charging stress, less traps are
created in thinner oxide.
Transistor parameter is no longer a useful
measure for damage.
24
However, charging damage is still there.
The damage mode is once again changed.
H. C. Lin et al., IRPS-98, p312.
Gate oxide breakdown is back as the damage mode.
25
Three factors causes breakdown to become a damage
mode
The density of defects for a given breakdown
probability decreases rapidly with oxide
thickness.
For a given stress voltage, the current is higher
for thinner oxide.
The safe margin of oxide life time decreases
rapidly with thickness.
e.g.
According to the anode-hole injection model, the
life time for 100Å thick gate oxide under 5V
stress will last longer than the life time of the
universe. The life time for gate oxide in current
technology is barely meet the 10 years
specification at 1.2V.
A fundamental different between thin oxide and
thick oxide
For thick oxide, if it didn't break, nothing will
happen.
For thin oxide, even if it didn't break, it could
have been significantly weaken.
26
Initial gate leakage is a popular method for thin
oxide breakdown detection
C. Cismaru et al., P2ID-2001, p48.
1.2V
Tox lt 20Å
0.3V
Inversion
Vt shift has no signal.
Question is charging damage a problem or not?
27
For transistor parameter shifts, the failure
criteria are clear.
For example 10 gm degradation
What is the failure criterion for breakdown?
In the case we just saw, was the charging damage
in pMOS unacceptable? What about nMOS? Was no
failure good enough?
Reliability criterion for gate oxide
0.1cm2 area, lt 0.01 failure after 10 years of
operation at 85ºC.
Oxide thickness is already the thinnest allowed
by reliability.
Not much margin for additional defects created by
charging damage.
Charging damage may cause the oxide to fail
reliability specification.
Charging damage must not exceed the available
margin for reliability.
If N is the critical defect density that lead to
0.01 failure for 0.1cm2 area and if we only have
a 10 margin, we know that charging damage must
generate no more than 10 or the critical defect
density.
How do we know?
28
It is well established that oxide breakdown
follows Weibull statistics
F(t) is the cumulative failure fraction h is the
characteristic (63.2) failure time b is the
Weibull slope
To link one oxide data set to another (with same
oxide thickness)
It can be shown that
and that
gt
When Fi ltlt1
Scaling is very sensitive to Weibull slope.
Experimentally, Weibull slope is a function of
oxide thickness.
29
We now can answer the question.
Start with the reliability specification of A
0.1cm2, F 0.0001
Using
and
Since the test device has an area of 2X10-8cm2,
The reliability specification for the test device
is
In order not to cause the oxide to fail
reliability specification, the failure fraction
for as processed test device must be less than
(10 margin)
If the Weibull slope is close to 1 (most
ultra-thin oxides are), then the failure fraction
must be less then 2 parts in a trillion.
For the pMOS, we know we are in big trouble.
For the nMOS, we have absolutely no idea.
No one has ever measure plasma charging damage
properly for ultra thin oxide. The down hill
period is an illusion.
30
What lies in the future?
Hi-k gate dielectric will be differ from SiO2.
The damage mode most likely will change on us
again.
The challenge is to understand the failure
mechanism.
The lesson from SiO2 is don't jump to conclusion.