Académique Documents
Professionnel Documents
Culture Documents
Jason C
C.S.
S Woo
Ion
Ioff
Zin (including Igate)
Parasitics
(G XR t)
Gm (GmXRout)
Noise
Breakdown voltages, Reliability
S D
ND ND
L
NA
Sub
∂φ xi
∫ I D dy = ∫ ∫
L L
q μn Z Fn n( x, y ) dx dy
0 0 ∂y 0
qZ μn |φB |−V n po e β (ψ − βV ) dψ
⇒ ∫ ∫ψ
VD
ID = n dV
L 0 S 2 kT F ( βψ , V , po )
qλ p p po S.M. Sze
Blue: 250nm
Red: 35nm
Taur
For given dielectrics εI and εSi, if (tI, tSi) are both scaled by a
ffactor off α,, Lmin will be scaled byy the same factor α
P. Wong
Jason Woo IWSG2009
2-D potential Profile contour Comparison between 250nm and
35nm DG FETs ((VG = 0.1V,, VD = 0.1V)) ((off state))
Blue: Bulk FET
Red: DG FET
Rfr,ext
Rsur,dp Rac2
Rcsdd Rsur,extt Rac11
Rsp,dp Rsp,ov
Rsp,ext
Lov dy
Rac1 = ∫ Lsilicide ⎛ Ldp ⎞
Lov −Wov μ ac ( y )Cox (Vgs − V fb ,ov ( y ) − Ψs ( y ) − V p ( y )) Rsp,dp = ρ dp / ln⎜ ⎟
Tsilicide ⎜L −L ⎟
Lov −Wov ddy ⎝ dp
p silicide ⎠
Rac 2 = ∫ Lsw − Ldp ⎛ X jext − Wext − Tsilicide Wext − Wdp ⎞
0 μ ac ( y )Cox (Vgs − V fb ,ov ( y ) − Ψs ( y ) − V p ( y )) Rsp,dp = ρ dp ln⎜ +
Lsw
⋅ ⎟
Wext − Wdp ⎜⎝ X jext − Wdp − Tsilicide Lsw − Ldp X jext − Wdp − Tsilicide ⎟⎠
1 Lov −Wov ρ ov ( y )
Rsp ,ov = ∫
tan α1 0 Lov − Wov + Tc / tan α1 − y
dy
( Lsw − Ldp ) 2
R fr ,ext = Total S/D series resistance:
μ fr ,ext (Vgs − V fb ,ext − Ψs ,ext )C fr ,ext
Lsw − Ldp
Rsur ,ext = ρ ext ( N max,ext ) Rsd = 2 Rseries
X p ,ext
1 X jext −Wdp ρext ( x) = 2( Rov + Rext + Rdp + Rcsd )
Rsp,ext =
tan α 2 ∫X jext −Wext x − X p ,ext
dx
25
500 70
S/D Seeries Resisttance [Ωμm
Rcsd 20 Rov
100
10
Rdp
0 0
32 nm 53 nm 70 nm 100 nm 32 nm 53 nm 70 nm 100 nm
Physical Gate Length Physical Gate Length
700 70
S/D Seeries Resisstance [Ωμm
%]
ative Contrribution [%
600 PMOS scaled by ITRS 60 Rcsd PMOS
Rov
500 50
400 Rextt 40
300 30 Rov
200
Rdp 20 Rext
100 Rcsd 10
Rdpp
Rela
0 0
32 nm 53 nm 70 nm 100 nm 32 nm 53 nm 70 nm 100 nm
Physical Gate Length Physical Gate Length
27
Sidewall Gate
Silicide
Nov(y)
Rcsdd Rov Rov:
Rdp Rext
• Nov (y)
• Lov
Next(x)
• Xjext
Lg = 45nm
POLY Lm = 30nm
contact contact
Tox,eq = 0.8nm
0 8nm
Source Drain Xje = 15nm
Xjc = 45nm
Lext,sp+ Lext,ov = Lext
Next = 1~5x1019/cm3
Nd = 3x1020/cm3
Si substrate
•Abrupt
Abrupt junction is assumed
mV/V)
80
A/μm)
650
DIBL (m
600
•Vdd=0.8
08V
Ion (μA
70
550
500 60
450 Ion, Lext,sp = 0nm
•Ioff=10 nA/µm
Ion,L
Lext,sp = 7
7.5nm
5nm 50
400
DIBL,Lext,sp = 0nm
350 40
DIBL,Lext,sp = 7.5nm
300
18 19 20
10 10 10
-3
Extension doping concentration (cm , log)
750
130
1.10
700 10 A/μm 120
Ioff=10nA/
19 -3
C*V/Ion @ Next=5x10 cm
110 1.05 19 -3
650 C*V/Ion @ Next=1x10 cm
100
Normalized C**Vdd/Ion
1.00
600
90
DIBl (mV/V)
m)
Ion (μA/μm
550 80 0.95
70
500 0.90
60
19 -3
450 Ion,SDE=5x10 cm 0.85
19 -3
3 50
Ion,SDE=1x10 cm
400 19
DIBL,SDE=5x10 cm
-3 40 0.80
19 -3 -10 -5 0 5 10 15 20
DIBL,SDE=1x10 cm 30
350 Extension length under sidewall (nm)
-10 -5 0 5 10 15 20
Extension length under sidewall (nm)
95 Source
Drain current (μA/μm)
750
90
DIB
BL (mV/V)
700 85
80
650
75
70
600 19
DIBL @Next=5x10 cm
-3
Ioff=10nA/μm 19
DIBL @Next=1x10 cm
-3 65
550 60
0 5 10 15 20
SDE jjunction depth
p (nm)
( )
500
-9 -8 -7 -6
10 10 10 10
Leakage current (A/μm)
Long channel: device with overlap has higher current drive due to
smaller effective channel length, Ion ∝ Leff-1
Short channel: Similar Ion /Ioff
ff performance due to velocity saturation
and SCE limitation
1.05
with
ith ext.
t overlap
l
1.00 1.1
without ext. overlap
with ext. overlap
1.0
without ext. overlap
0.95
0.9
normalized C**V/I
normalized C**V/I
0.8
0 90
0.90
0.7
0.85 0.6
0.5
0.80 •Lg=500nm
n
04
0.4
0.3
0.75 •Lg=45nm -9
10 10
-8
10
-7
10
-6
0.70
-9 -8 -7 -6
10 10 10 10
leakage current (A/μm)
95 10 A/μm
Ioff=10nA/ Gate
contact Stack
90
750
85
Source
DIBL (mV/V))
IIon (μA/μm)
80
700
75
70
20 -3
DIBL @Nd=3x10 cm 650
65 21 -3
DIBL @Nd=3x10 cm Gate
20 -3
60 Ion @Nd=3x10 cm
21 -3
Stack
Ion @Nd=3x10 cm
55 600 Source
10 20 30 40 50 60
deep S/D junction depth (nm)
•Shallow
Shallow and highly doped Source/Drain region is desired
for high drive current and low DIBL
Jason Woo IWSG2009
MOSFET with optimized
p S/D region
g
POLY
Conventional contact contact
Source Drain
bulk MOSFET
Si substrate
400
Controlling ρc :
2
-77
350 LG = 53
3 nm 10 Midgap silicide →
}
ρc
Ωμm]
low-barrier silicide
300 Rcsd ⇒ ~ 60 % reduction
Resisstance [Ω
100
RovR -9
10
lowering will be an
ext effective way for Rcsd
50 Rdp reduction
0
Co
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 ⎛ qφ ⎞
Schottky Barrier Height, ΦB [eV] ρ c ∝ exp⎜ b ⎟
⎜ N if ⎟
⎝ ⎠
37
Advanced S/D Engineering
g g
300
Graded Junction
Resistancce [Ωμm]
30 f PMOS
for PMOS, and d llower
0 bandgap Si1-xGex layer)
Source/Drain Engineering
38
Annealing Techniques Investigation
1100 700
TSPIKE TSPER
Spike Anneal 1050 600 tSPER
SPER
Temp (C))
on PAI substrates: 1000 500
Temp (C)
Tsper=650C ,700C
Tspike=1050C ,1075C 950 400
tSPER=1min
Ramp Rate=400C/s. 900 300
850 200
-4 -2 0 2 4 6 8 10 0 150 300 450 600
Time (s) Time (s)
750
Teemp (C) 600 tSPER
Max Ramp Rate 600 500 Tspike=850C, 950C
=80C/s.
80C/s 450 400 Max Ramp Rate
300 300
200 =80C/s.
150 100
0 100 200 300 400 0 150 300 450 600 750
Time (s) Time (s)
Diffusion of As during anneal
1x1023
as-implanted As profile
1x1022 Conventional RTA (1075C, 5sec)
m-3)
Concenttration (cm
1x1021
Spike Anneal (1075C, 400C/s)
1x1020
SPER (650C,1min)
1x1019
Ge profile (PAI)
1x1018
1x1017
1x10160.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20
Distance (µm)
SPIKE Annealing MORE SUITABLE THAN CONVENTIONAL RTA
or SPER for controlling
g dopant
p diffusion while maintaining
g an
acceptable degree of activation for arsenic
Diffusion of B during anneal
1x1023
as-implanted B profile
1x1022 Conventional RTA (1075C, 5sec)
m-3)
Concenttration (cm
1x1021
Spike Anneal (1075C, 400C/s)
1x1020
Ge profile (PAI)
1x1019
1x1018
SPER
1x1017 (650C, 1min)
1x1016
0 00
0.00 0 01 0.02
0.01 0 02 0.03
0 03 0.04
0 04 0.05
0 05 0.06
0 06 0.07
0 07 0.08
0 08 0.09
0 09 0.10
0 10 0.11
0 11 0.12
0 12
Distance (µm)
SPIKE Annealing MORE SUITABLE THAN CONVENTIONAL RTA
OR SPER for controlling
g dopant
p diffusion while maintaining
g an
acceptable degree of activation for boron
P-n junctions using spike anneal As=5e14
1.00E+22 Boron_as_implanted
As_as_implanted
1.00E+21
00 Activated Profile
D o pi n g (/cc)
1.00E+20
1.00E+19
1.00E+18
p
1.00E+17 Junction n
1.00E+16
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
depth (um)
SRP results for Spike anneal process (1075C, 400C/s)
performed on PAI samples co-implanted with BF2 and As
(junction in the vertical direction). Very little movement from as
as-
implanted profiles
Gate Oxide Thickness Scaling
• Reducing gate oxide thickness is known to improve:
- short channel effects
- subthreshold characteristics
- drive current and transconductance
- circuit speed
- reduce body effects
- reduce drain induced barrier lowering g
• Oxide scaling also leads to:
- increased gate leakage current, and standby power
- degraded dielectric reliability
- more severe impurity penetration effects
- oxide thickness uniformity control difficulties
- increased gate induced drain leakage (GIDL) and hot electron
effects
ff t
• High dielectric constant (high-K) thin films might provide solution since
physical thickness can be made thicker while maintaining high
capacitance
Criteria Requirements
EOT scalability < 10Å Dielectric constant > 15
Negligible FIBL effect Dielectric constant < 60
L k
Leakage currentt < 1A/cm
1A/ 2 B d
Bandgap > 5eV
5 V
Barrier height>1eV
Thermal stability No silicidation and
reduction
Hysteresis
y <20mV
Interface state density <1011/eVcm2
Mobility >85% of SiO2
Reliability >10 years
S D 25 nm
κ = 3.9
G
19.2 nm
S D 25 nm
κ = 50
Cx / Ctot
0.6
0.4
0.2
L=100nm
0
0 0.2 0.4 0.6 0.8
TK / L (Variation in TK)
• High TK/L ratio ⇒ Less CGSUB, Lesser gate control and more SCEs.
Hi h K ⇒ more fringing.
• High f i i
Jason Woo
Jack Lee IWSG2009
Hih-k Gate Leakage Current
For the
F h same EOT,
EOT high-K
hi h K samples
l exhibit
hibi
much lower leakage current
Jason Woo IWSG2009
Issues: Channel Mobility
μeff E
v= E < Esat
1 + ( E / Esat )
= vsat E > Esat
μeff Esat
i.e., vsat =
2
2 sat
2v
or Esat =
μeff
That is
I D Esat
E( y) =
Esat W μeff C ox ⎣⎡VG′ − V ( y ) ⎦⎤ − I D
⎛ V ⎞
W
= μeff Cox ⎜⎜VG − Dsat ⎟⎟V Dsat
′ 1
L
⎝
2 ⎠
1 +V Dsat / Esat L
Esat LVG′
VDsat =
Esat L + VG′
1
I Dsat = W μeff C ox (VG′ − VDsat ) Esat
2 2
⎛ ′ ⎞
⎜ VG ⎟
= W vsat C ox ⎝ ⎠
VG′ + Esat L
⎛ ′ ⎞2
⎜ VG ⎟
⎝ ⎠
= W vsat C ox
′ 2vsat
V + L
G
μeff
i.e., If EsatL is small, I Dsat ∝VG′ i.e., IDsat to first order ∝(V'G)
Jason Woo
Han Zhao, et. Al. APLIWSG2009
2009
DIGITAL CIRCUIT REQUIREMENTS
High Ion/Ioff ratio
Low intrinsic delay y CgVdd/Ion
Improved DIBL and subthreshold swing (S)
Low parasitic resistances and capacitances
100
00
Tox = 1.5nm
Ids = 100μA/μm
80 Vds = 0.8V
Lg = 150nm
ntrinsic Gaiin
60 Lg = 100nm
Lg = 70nm
Lg = 50nm
40
In
N+poly gate
20 Vth=0.15-0.25V
Vth=0.25-0.40V
0
0 50 100 150 200
fT (GHz)
( V ds – V dsat )
ACLM = ( E s attL g ( 1 + θ Vgtt ) + V gtt) --------------------------------------
V
V -
E s at l (1 + θ V gt ))
2
lt = 3Tox (X dep ⁄ η ) llα
l 3 X j T X dep
, ox
80
8.0
Lg =150nm
7.0 Lg =100nm
Lg =70nm
Lg =50nm
50nm
gm/IIds (V-1)
6.0
Vds=0.8V
50
5.0 Ids=100μ A/μm
Tox =1.5nm
4.0 N+poly gate
Vth=0.15-0.25V
Vth=0.25-0.40V
3.0
0 20 40 60 80 100
Intrinsic Gain
20
2.0 Vds
d =0
0.8V
8V
Ids=100μA/μm
Tox =1.5nm
1.5
1.0
0 50 100 150 200
fT (GHz)
(GH )
100
Empty: Tox =2.0nm Lg=150nm
Filled: Tox =1.5nm Lg=100nm
80 Lg=70nm
Lg=50nm
Inntrinsic gainn
60 Vth=0.15-0.25V
Vth=0.25-0.40V
40
20 Vds = 0.8V
μ μ
Ids = 100μA/μm
0
20.0 70.0 120.0 170.0
fT (GHz)
200
Ids = 30μA
Tox =1.5nm Ids = 60μA
Vds =0.8V Ids = 100μA
150 Ids = 150μA
Intrrinsic Gainn
Ids = 200μA
100
Lg=50nm
50
50
Lg=150nm
0
0 50 100 150 200
fT (GHz)
(ε T C
lt = si si fbb) ⁄ ( η C ox Cbox )
DIBL depends on silicon thickness. It has very weak dependence on channel doping
unlike in the bulk MOSFET
⎛ ΔV ⎞ C ( 0) C
DIVS B = ⎜ ----------th BD--------------------b------------
---⎟ = ------------------
⎝ ΔVds ⎠ Cox ( C BD( 0 ) + Cb )
Ernst, et. al
High Frequency Rout in PDSOI MOSFET
DC Kink Effect:
V t h = Vth0 – γ ( 2Φ B – 2Φ B – Vbs)
Ids and gm expressions modified taking into account the change in threshold voltage
with self-body bias
AC Kink Effect:
Effect of Drain-to-Body coupling on output resistance at high operating frequency
I
dsat
V Acou pled = ----------------------------------------------------------- -
⎛∂Ids ∂V th ∂VBS⎞
⎜ • • ⎟
∂
⎝ th V ∂ V ∂V ⎠
BS ds
I
VAcoupled = --------------------------------------------dsat
---------------------------------------------------
⎛ γ C BD ⎞
⎜ g m • --------------------------------------- • ------------------------------- ⎟
⎝ ( 2 2Φ – V ) C BS + C BD ⎠
b bs
I
dsat
V Acoupled = -----------------------------------------------------------
-
∂I
⎛ ds ∂V th • BS⎞ ∂V
⎜ • ⎟
⎝∂ V th ∂ V BS ∂ Vds ⎠
I
VAcoupled
p = ----
------
------
------
------
------
------
------
------
------
------
------------
----------dsat
------------------------------------------------------------------------------------------
⎛ ⎞
⎜ ⎟
⎜ g • ⎛⎜-----------------------------------------------------
γ Cb ⎞ CBD
- + ------------------------ ⎟ • ------------------------------------------------------------- ⎟
⎜ m ⎝2 2φ – V Cb + Cox ⎠ C ox Cb ⎟
⎜ B bs0 – V bs C + C + ----------------------- ⎟
⎝ BS BD C + C ⎠
ox b
Effect of Channel Length Scaling of Analog Performance
100
Vth=0.25 - 0.35 V Lg =60nm
Lg =100nm
Vth=0.25 – 0.35 V Lg=150nm
80 For Mid ggapp ggate Lg=250nm
nsic Gain FDSOI MOSFET
Black: Bulk
Green: PDSOI
60 Blue: FDSOI
Ids =100μA/μm
Vds =00.8V
8V
Intrin
40 Xj=10nm
TSi =15nm
Tox =1.5nm
f oper=1GHz
20
0
0 50 100 150 200
fT (GHz)
Channel Doping ↑⇒gm ↓ and Rout ↑ in Bulk MOSFETs
Channel Doping ↑⇒gm ↓ in SOI MOSFET; However Rout ↓ in
PDSOI MOSFET while Rout doesn’t improve much in FDSOI
Gain-fT curves show opposite trends in Bulk and SOI
technologies
Jason Woo IWSG2009
Effect of Silicon thickness on Rout in FDSOI
80
NA=1e17cm -3 Vds =0.8V Lg=60nm
Tox=1.5nm
1e5
e5 Lg=100nm
60
%Chaange in Routt
Lg =150nm
KΩ-μm)
40 Lg =250nm
Rout (K
20
0
1e4
1 4 15 20 25 30 35
15 20 25 30 35
Tsi (nm) Tsi (nm)
Long g channel FDSOI MOSFET: Rout improves
p for thin
silicon thickness due to reduction in drain-to-body
coupling
Short channel FDSOI MOSFET: Routt improves with
thinning down silicon film due to reduction in DIBL.
Jason Woo IWSG2009
Excess Low Frequency Noise in
SOI
Gate
Source Drain
1/f
1/f
2qISB
2qIii
Buried Oxide
Gate Noise
DS (mA)
10-10 10-10 0.4
N
ID
10-11 10-11
0
10-12 10-12 0 1.0 2.0
Power SVG (V2/Hz))
PDSOI FDSOI
nMOS (B)
100k
10k PD
PD pMOS (A) nMOS (A)
1k
|VGT | = 0.2 V
100 f0 PD
fkink nMOS (A)
10
-3.0 -2.5 -2.0 -1.5 1.0 1.5 2.0
• Backgate bias?
• DG devices?
• Other
Oth d device
i architectures?
hit t ?
Double Gate devices (thin and low doped Si) show the best performance in terms
of Gain and fT, UTB FETs are worse than Bulk devices for longer
g channel lengths
g
but for lower channel lengths they start showing better performance
DTMOS
h(Vbs) (V)
Lg =1μm
0.2
Vth(0)-Vth
0.1
Lg = 0.14μm
0.0
0.0 0.2 0.4 0.6
Vbs (V)
Change in threshold voltage as a function of body bias for two different gate lengths, Tox =
36Å.
2.5 10-3
V b = 0V NMOSFET NMOS
V b = 0.3V Lg = 0.14 μm Lg = 0.14 μm
V b = 0.6V
0 6V 10-4
2.0 Vds = 1.6V
Ids (A/ μ m)
10-5
mA/μm)
V gs = 1.2V
1.5
10-6
Ids (m
1.0
10-7
V gs = 0.9V Vb = 0V
0.5 10-8 Vb = 0.3V
Vb = 0.6V
0 6V
0.0 10-9
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6
V ds (V) Vgs (V)
Tox = 36Å
Roout (Ω / μ m)
DTMOS
Rin (Ω / μm
10 5
DTMOS
0.2 10 5
Vb =0V
V b=0.3V
=0 3V
m)
V b=0.6V
0.1 DTMOS NMOS 10 4
10 4 Tox = 36 Å
V ds = 0.6V
0 10 3
0 0.5 1.0 1.5 2.0 0 20 40 60 80
Ids ( μA/ μm)
Lg ( μm)
• Th
The ttransconductance
d t iimproves iin th
the DTMOS case, b
butt nott
with static well bias conditions
• Rin decreases with increasing Vgt, increasing loading on
previous stage
Jason Woo IWSG2009
Short Channel Characteristics
NMOS 0.5
0.6 Tox = 25Å Vb=0V
0.4
Vb=0.3V
=0 3V 102
0.5 Vb=0.6V
DTMOS
DIBL ((mV /V )
V th,,lin (V)
0.3
Δ V tth (V)
0.4
0.3 02
0.2
10
0.2 Vb=0V 0.1
Vb=0.3V
01
0.1 Vb=00.6V
6V 00
0.0 NMOS
DTMOS Tox = 36Å
0.0 -0.1 1
0.0 0.2 0.4 0.6 0.8 1.0 0.1 1.0
Lg (μm) Lg (μ m)
• Both the Vth roll-off and DIBL are reduced as the active body
bias increases
• Improvement in SCEs due to the resuced drain electric field
coupling
Jason Woo IWSG2009
INTRINSIC GAIN COMPARISON
1000 Ids same at same Lg
Ids same at same Lg
nsic gain (gm×Rout)
c gain (gm×Rout)
100
100
NMOS
NMOS
Tox =25Å Tox = 36Å
Intrinsic
10
Intrin
10
Vb=0V
Vb=0V
Vb=0.3V Vb=0.3V
Vb=0.6V Vb=0.6V
DTMOS DTMOS
1 1
0.1 1 0.1 1
Lg (μm) Lg (μm)
DTMOS shows the highest improvement (2x), particularly at small channel lengths
due to improved
i gm as well as improved
i Rout (50%)
( 0%) in
i the thick
i Tox devices
i
The Vb = 0.6V case shows improvement upto 50% in the intrinsic gain
Here too Active well bias or DTMOS configuration improves the device
performance
f if the
th original
i i l intrinsic
i t i i device
d i performance
f is
i degraded
d d d by
b severe SCE