on reliability of microelectronic devices
TRANSCRIPT
On Reliability of Microelectronic Devices:An Introductory Lecture on
Negative Bias Temperature Instability
M. A. AlamPurdue UniversityWest Lafayette, IN
Sept. 28, 2005
On Reliability of Microelectronic Devices:
Reliability has always been Important!
A 5000 year old example: Stone vs. Copper tools
A modern example: Honda vs. Yugo
“The car is named Yugo, because it doesn’t …”
Reliability has always been Important!
Microelectronics reliability & viable technologies
Kelly (Bell Labs) to recently-hired Shockley in 1936:
“Instead of mechanical devices, which has annoying maintenance problem, we should look for (reliable) solid state switch ….”
Pauli to his student Peierls, in 1932, on unreliability of Cu2S, Ag2S:
“ Do not work on semiconductors, it is a mess (eine schweinerei); who knows if semiconductor exists at all …..”
Landauer on quantum computing (1992):
“ …. this proposal depends on speculative technology, does not in its current form account all possible sources of noise, unreliability, and manufacturing error, and probably will not work.”
Microelectronics reliability & viabletechnologies
Warranty, Product Recall, and Other Facts of Life
A manufacturer bets the company of the physics of reliability …..
… because the ICs operate in incredibly harsh conditions, turningon and off trillions of timeduring its lifetime ….
… because the lines could open, the source/drain can be shorted, the gateoxide can break ….
Warranty, Product Recall, and Other Facts ofLife
Si-H and SiO2 Bonds
Si-H and SiO2 Bonds
Reliability Issues in Modern Transistors
Initially ….
Drain Voltage
Dra
in C
urre
nt
VG2
VG1GS D
.... a few months later
Broken Si-H bonds: Negative Bias Temperature Instability (NBTI) Hot carrier degradation (HCI)
Broken Si-O bonds: Gate dielectric Breakdown (TDDB)
Reliability Issues in Modern Transistors
Introduction: NBTI defined
GNDVDD
NBTI: Negative Bias Temperature Instability
Gate: GND, Drain: VDD, Source: VDDGate negative with respect to S/D
Other degradation modes: TDDB, HCI, etc.
VDD
Introduction: NBTI defined
VD (volts)
I D (m
A)
before stress
after stress
0 1 2 3 4 0
4
3
2
1
NBTI & Parametric Failure
ΔΣ
Stress Time (sec)
% d
egra
datio
n
101 103 105 107 109
5
10
15
Spec.
War
rant
y
NBTI & Parametric Failure
A Brief History of NBTI
Experiments in late 1960s by Deal and Grove at Fairchild
Role of Si-H bonds and BTI vs. NBTI story (J. Electrochem Soc. 1973;114:266) Came out naturally as PMOS was dominant Important in FAMOS and p-MNOS EEPROMS (Solid State Ckts 1971;6:301)
Theory in late 1970s by Jeppson (JAP, 1977;48:2004)
Generalized Reaction-Diffusion Model Discusses the role of relaxation, bulk traps, ….. Comprehensive study of available experiments
Early 1980s
Issue disappears with NMOS technology and buried channel PMOS
Late 1980s and Early 1990s
Begins to become an issue with dual poly gate, but HCI dominates device reliability
Late 1990s/Early 2000 (Kimizuka, IRPS97;282. Yamamoto, TED99;46:921. Mitani, IEDM02;509)
Voltage scaling reduces HCI and TDDB, but increasing field & temperature reintroduce NBTI concerns for both analog and digital circuits Numerical solution is extensively used for theoretical modeling of NBTI.
A Brief History of NBTI
26 28 30 32 34 36 38 40
100
101
102
TPHY
=36A
Ea=0.49eV
1/kT (eV-1)
D (
arb
. u
nit)
10-3
10-2
10-1
RT
VG=-4.1V
VG=-4.5V
VG=-4.9V
125oC
VG=-3.1V
VG=-3.7V
VG=-4.5V
TPHY
=36A D=1
!NIT
/ [
N1 {
COX
(VG
-VT
)}0.5
] (a
rb.
un
it)
10-3
10-1
101
103
105
107
10-3
10-2
10-1
TPHY
=36AV
G=-4.5V
N1=1
D. t (arb. unit)
RT, T=125oC
T=50oC, T=150
oC
T=90oC
/( ) BEa k T n
ITN t Ae t
!=
n ~ 0.25 Ea ~ 0.5
Exponent and Activation
log( ( )) log( ) (log( ) / )IT BN t n t A Ea k T= + !
Exponent and Activation
A 40-year-old Puzzle …
ln (time)
ln (d
egra
datio
n) T1
T2
Trap Generation
n=0.25 (H diffusion ?) EA=0.5 (H2 diffusion ?)
Before 1980
The exponent is soft (nsat = 0.16) ! NBTI depends on frequency.
ln (time)ln
(deg
rada
tion)
?
Saturation
After 2000
Freq. Dependence
ln (time)
ln (d
egra
datio
n) DC AC
A 40-year-old Puzzle …
A Word about Drawings
Silicon Gate oxide PolySi HSi H
Si H
Si H
Si H
Si H H2
NH
distance
A Word about Drawings
The Reaction-Diffusion Model
Silicon Gate oxide PolySi HSi H
Si H
Si H
Si H
Si H H2
0( ) (0)IT
F IT R H IT
dNk N N k N N
dt= ! !
kF: Si-H dissociation rate const. Creates broken-bond NITkR : Rate of reverse annealing of Si-HN0: Total number of Si-H bonds
2
22
H H IT
H H H
dN d N dND N E
dt dt dt
!µ= + +
NH: Hydrogen densityDH: Hydrogen diffusion coefficientµH: Hydrogen mobility
n=1 n=0
n=1/4
n=1/2
log (time)
log
(NIT
)
NH
distance
The Reaction-Diffusion Model
Meaning of the Parameters
N0 [ ]0
0
/ /
( )
ox F
F F ox
E E E kT
k k p E
e e! " "#
= $
% & % &$ ' (' (
/
0
RE kT
R Rk k e
!"# $= % &/
0
HE kT
H HD D e
!"# $= % &
F
R
k
kSiH h Si H
+ ++ +
oxide
poly
Meaning of the Parameters
A Reformulation of R-D Model
2
22
IT H H
H H H
dN d N dND N E
dt dx dt
!µ= + +
0( ) (0)IT
F IT R H IT
dNk N N k N N
dt= ! !
NH
x
( )H
x t D t=
0( ) ( , )
HD t
IT HN t N x t dx= ! (Neutral)
NH
x
( )H ox
x t E tµ=
0( ) ( , )
H oxE t
IT HN t N x t dx
µ
= ! (Charged)
SiSiSi H
HH
H
H H
H
Si s
ub.
Poly
0 (0)F
H IT
R
k NN N
k
! "#$ %
& '
If trap generation rate is small,and if NIT much smaller than N0, then
( ) ( , , )
0( ) ( , )
H Hx t f D t
IT Hx
N t N x t dxµ=
== !
A Reformulation of R-D Model
NIT with Neutral H Diffusion
0 (0)F
H IT
R
k NN N
k
! "#$ %
& '
NH
x
( )H
x t D t=0
( ) ( , )
1(0)
2
HD t
IT H
H H
N t N x t dx
N D t
=
=
!
10-3
10-2
10-1
RT
VG=-4.1V
VG=-4.5V
VG=-4.9V
125oC
VG=-3.1V
VG=-3.7V
VG=-4.5V
TPHY
=36A D=1
!NIT
/ [
N1 {
COX
(VG
-VT
)}0.5
] (a
rb.
un
it)
10-3
10-1
101
103
105
107
10-3
10-2
10-1
TPHY
=36AV
G=-4.5V
N1=1
D. t (arb. unit)
RT, T=125oC
T=50oC, T=150
oC
T=90oC
SiSiSi H
HH
HH
H
H
Si s
ub.
Poly
Combining these two, we get
10 4( ) ( )
2
F
IT H
R
k NN t D t
k=
n=1/4 even with two sided diffusion
n ∼ 1/4 is a possible signatureof neutral H diffusion
Reproduces results of Jeppson, JAP, 1977.
NIT with Neutral H Diffusion
NIT with Neutral H2 Diffusion
0 (0)F
H IT
R
k NN N
k
! "#$ %
& '
2
2 2
0( ) ( , )
1(0)
2
HD t
IT H
H H
N t N x t dx
N D t
=
=
!
Combining these two, we get
2
10 6( ) ( )
2
F
IT H
R
k NN t D t
k!
( )2
2
2
(0). 2
(0)
H
H
Nconst H H
N=
n ~ 1/6 is a possible signature of neutral H2 diffusion
Small exponent because generation is more difficult.
NH
x
2( )
Hx t D t=
H2
NH
2
SiSiSi H
HH
H2 H2H2
H2
Si s
ub.
Poly
H2
Reproduces results of Chakravarthi, IRPS, 2003.
NIT with Neutral H2 Diffusion
NIT with charged H+ Drift
0 (0)F
H IT
R
k NN N
k
! "#$ %
& '
0( ) ( , )
1(0)
2
H oxE t
IT H
H H ox
N t N x t dx
N E t
µ
µ
=
=
!
Combining these two, we get
( )1
0 2( )2
F
IT H ox
R
k NN t E t
kµ=
NH
x
( )H ox
x t E tµ=
SiSiSi H
H+
H+H+
H+
H+
H+
Si s
ub.
Poly
n ~ 1/2 is a possible signature of charged H diffusion
Rapid removal of H+ by Eox field increase NIT gen. rate.
Reproduces results of Ogawa, PRB, 1995.
NIT with charged H+ Drift
NIT with charge H2+ Drift
n ~ 1/3 is a possible signature of charged H2
+ diffusion
Exponents above 1/3 seldom seen in charge-pumping expt. (uncorrelated to SILC).
0 (0)F
H IT
R
k NN N
k
! "#$ %
& '
20
2
( ) ( , )
1(0)
2
H oxE t
IT H
H H ox
N t N x t dx
N E t
µ
µ
=
=
!
( )2
2
2
(0).
(0)
H
H
Nconst H H H
N
+ += +
Combining these two, we get
( )1
0 3( )2
F
IT H ox
R
k NN t E t
kµ!
NH
2
x
( )H ox
x t E tµ=
SiSiSi H
HH+
H2+
H2+
H2+
H2+
Si s
ub.
Poly
NIT with charge H2+ Drift
Dispersive Diffusion & non-rational nN
H
x
NH
x
*
0( ) ( )2
nF
IT H
R
k NN t D t
k!
0.264-0.2970.33H2+
0.128-0.1440.16H2
0.20-0.250.25H
ndisnideal R-D model predicts n=0.30-0.12
More amorphous oxides for better NBTI
For finite oxides, at very long time all n must be rational (no problem > 10 yrs)
0 0( ) p
HD D t! "
=
NH
x
NH
x
(1 )0 0( )2
n
n pFIT p
R
k N DN t t
k w
!" #$ % &
' (
Shkrob, PRB, 1996; 54:15073
Dispersive Diffusion & non-rational n
26 28 30 32 34 36 38 40
100
101
102
TPHY
=36A
Ea=0.49eV
1/kT (eV-1)
D (
arb
. u
nit)
10-3
10-2
10-1
RT
VG=-4.1V
VG=-4.5V
VG=-4.9V
125oC
VG=-3.1V
VG=-3.7V
VG=-4.5V
TPHY
=36A D=1
!NIT
/ [
N1 {
COX
(VG
-VT
)}0.5
] (a
rb.
un
it)
10-3
10-1
101
103
105
107
10-3
10-2
10-1
TPHY
=36AV
G=-4.5V
N1=1
D. t (arb. unit)
RT, T=125oC
T=50oC, T=150
oC
T=90oC
n ~ 0.25 (H or H2+ diffusion?) Ea ~ 0.5 (H2 diffusion ?)
Puzzle: H or H2 or H2+ ?
/0( ) ( )2
BEa k T nF
IT o
R
k NN t D e t
k
!=
Puzzle: H or H2 or H2+ ?
Clue 1: NBTI Saturation
ln (time)
ln (d
egra
datio
n)
Clue 1: NBTI Saturation
Saturation by Poly Reflection
0 (0)F
H IT
R
k NN N
k
! "#$ %
& '
NH
x
(1)
{ }1
( ) ( ) ( ) (0)2 2
IT H ox H H H
WN t N T D t N W N= + +(2)
( ) ( )
( )
( ) (0) (poly oxH H HH Hpoly
H
N W N N WD D
WD t
!" #= $ %
& '(3)
10-1
100
101
102
103
104
105
10-3
10-2
10-1
T=125OC
TPHY
=36A
EOX
(MV/cm)
7.4
8.8
10.7
VT s
hift (V
)
stress time (s)
122 ( )
(0
( )( ) 2
2
polypolyF ox H
IT ox ox
R
k N T DN t D t T
k D t
! "# + +$ %
$ %& '
Combining, at short time, we get
SiSiSi H
HH
H
H
Poly
H
H
Oxide
{ }1/ 4
( )0( )2
polyFIT H
R
k NN t D t
k!
… but alas, at long time ….
Saturation by Poly Reflection
… but the explanation is wrong
ln (time)
ln (d
egra
datio
n)
NH
distanceKufluoglu, unpublished results
NH
distance
SiSiSi H
HH
Poly
H
H
Oxide
at time t1
at time t2
t1
t2
… but the explanation is wrong
Clue 2: Frequency Dependence
ln (time)
ln (d
egra
datio
n) DC AC
Ming-Fu Li, National University of Singapore, EDL, 2002.
Clue 2: Frequency Dependence
R-D Model at Low Frequencies
stress 1 relax 1 stress 2 relax 2
0 10 20 30 0 10 20 30 0 10 20 30 0 10 20 30 distance into the Oxide (A)
1014
1015
1016
1017
H2
dens
ity [a
.u.]
time (sec)
2 sec
95 sec
1000 3000 4000
450 sec
NIT
1450 s
1002 s
2002 s
3002 s
3450 s
2450 s
R-D Model at Low Frequencies
Analytical Model: Relaxation Phase
0 10 20 30 401014
1015
1016
1017
Distance [A]
H2 c
once
ntra
tion
[a.u
.] Si oxide
NH
0N
H0(0
)
(Dt0)1/2
(0)
0
(*)
1(0)
2
1(0)
2
IT H H
IT H H
N N D
N N D t
!
"
=
=
0( ) (0)IT
F IT R H IT
dNk N N k N N
dt= ! !
(0) (*)
0 0
(0) (*)
2
0 0 0
H H H
IT IT IT
H H
N N N
N N N
AN BN C
= !
= !
+ + =
(0)
0
11
IT IT
x tN N x
x
!
"
# $ # $= % &' ( ' (
+ ) *) *
Analytical Model: Relaxation Phase
Approximate Analytical Models
Time (sec)
abs
(VT)
Shi
ft (m
V)
100 101 102 103 104 10510
20
30
40
50
0.5x104 1x104 1.5x104
linear plotlog plot
VT = a – bt1/4
New modelNumerical
VT = a – bt1/4
New modelNumerical
Approximate Analytical Models
Frequency Dependence Interpreted
Symmetry in R-D model requires frequency-independent degradation
simulationmeas.
10-1 101 103 105 107
Frequency [Hz]
10
20
30
40
50
0
VT
Shi
ft [m
V]
Frequency Dependence Interpreted
Frequency Independence Interpreted
0 20 40 60 80 100Distance into the Oxide [A]
H2
conc
entra
tion
[a.u
.]
100
104
108
1012
1016
1020
High Freq
Low Freq
Low Frequency (1 cycle)
High Frequency (1 cycle)
Frequency Independence Interpreted
R-D model anticipates Frequency Independence!
0 20 40 60 80 100Distance into the Oxide [A]
H2 co
ncen
tratio
n [a
.u.]
100
104
108
1012
1016
1020
High Freq
Low Freq
100/200 cycle
200/400 cycle
Frequency Independence Interpreted
Frequency Independence Interpreted
Measurement: A variable-frequency AC Stress!
1s 10s 100s 1000s Stress time
What we thought we were doing …
S. Rangan, Intel, IEDM 2003.
6s 21s 126s 1031s Stress time
What we were actually doing …
* 5 sec. measurement window (for example).
Measurement: A variable-frequency ACStress!
R-D Simulations for DC NBTI
Actually, n=0.16 at all times (H2 diffusion), measurement delay makes itappear n=0.25 at short times. A 40 year old puzzle finally resolved!
H. Kuflouglu,unpublished results
R-D Simulations for DC NBTI
AC NBTI: Sim. & Measurement
Only DC simulations are fit toexperimental data (VG=2.1V).
AC simulations are inherentlyrelated to DC results
Delay of 6.3 sec is used
AC NBTI: Sim. & Measurement
Conclusive Confirmation
100
101
102
103
10-2
10-1
t-delay
-VG(meas)
-VG(stress)
EOT=21.4AO, Dose=5.54x10
15cm
-2
On-the-fly
delay=50ms
T=50OC
VG=-3.0V (stress)
-VG(meas) / n
3.0V/0.138
1.5V/0.162
1.0V/0.189
stress time (s)
!VT
(V
)
100
101
102
103
Delay
50ms
350ms
Delay Idlin, VG=-3.0V (stress)
VG=-1.5V (measure)
T=150OC
T=50OC
slope:
0.162, 0.189
0.177, 0.211
D. Varghese,IEDM 2005
Conclusive Confirmation
Conclusions
Reliability is, has been, and will be a key consideration for a viable technology.
Over the last fifteen years, reliability engineering (a collection of pragmatic, empirical rules) have gradually evolved into reliability physics (grounded in sound understanding of underlying physics).
In this talk we considered NBTI, similar physical models exist for Hot carrier degradation and Time dependent dielectric breakdown.
The NBTI model provides proper prescription of extrapolating device lifetime. We will consider these extrapolation methods in a separate talk.
Finally, do not underestimate the power of simple models. What we did in 4/5 lines of algebra, is actually equivalent to tens of PRB, JAP, TED, EDL papers over last 30 years.
Conclusions
Collaboration and References
[1] Mahapatra and Alam, IEDM 2002, p. 505.[2] Mahapatra, Kumar, & Alam, IEDM 2003, p. 337.[3] Mahapatra et al. IEDM 2004, p. 105.
[1] Alam, Weir, & Silverman, IWGI 2001, p. 10. [2] Alam, IEDM 2003, p. 346.[3] Kufluoglu & Alam, IEDM 2004, p. 113.
Experiments: S. Mahapatra, S. Kumar, D. Saha, D. Varghese, IIT
Theory: M. Alam, H. Kufluoglu, Purdue University
• For convenience, most of the figures of this talk are taken from these references. I will use other figures to illustrate difference in opinions or to generalize results.
Collaboration and References
Questions & Answers