perspective on emerging devices and their impact on scaling … · 2010-06-28 · perspective on...
TRANSCRIPT
Perspective on Emerging Devices and their Impact on Scaling Technologies
S. Biesemans IMEC, Kapeldreef 75, B-3001 Leuven, Belgium ([email protected])
Abstract
We review and discuss the latest developments in FinFETs as an emerging device and its impact on scaling technologies.
Introduction Transistor scaling has been the major driving force in the semi-
conductor industry for 40 years. The track has been one of many roadblocks and creative solutions. Most FinFET reports focus on devices, here we discuss technology opportunities and challenges.
The microelectronics industry is constantly seeking ways to scale transistor dimensions in order to allow for more transistors on a single chip, to scale down capacitance and to keep the drive current high while the voltage supply is scaled down to reduce power con-sumption. When scaling conventional planar devices below the 65nm node, short-channel effects (SCE) start to dominate over the device performance. Increased gate control is achieved by reducing the gate insulator thickness, by reducing the depth of S/D regions, and by increasing channel doping. All three strategies have major disadvantages: an increase gate leakage, an increase in the parasitic S/D resistance, a compromised carrier mobility and S/D leakage. Another approach to improve gate control over the channel is by using multiple gate devices [1,2]. It is projected that these 3-D de-vice architectures may start to replace planar MOSFETs for specific applications in the 32nm node and beyond [3].
FinFET process Because the 32nm node does not exist yet, FinFETs are built
with existing higher node processes. In this work, we choose to start from a standard 130nm process with an extension towards a 45/32nm seized 6T-SRAM cell.
Fin patterning A typical FinFET process flow is presented in Fig.1. Most of the
work to date is based on SOI substrates [4,5], although bulk FinFETs [6,7] have been reported. An opportunity of SOI substrates is the ability to work with mesa-type active areas due to the insulating nature of the buried oxide (Box) [8]. The opportunity is that one avoids the need for an STI process and enjoys immunity to latch-up. Several challenges exist. A typical test structure for a FinFET is shown in Fig.2. The various fins are connected through the use of a S/D pad. Despite the fact that they are a convenient way of connect-ing the multiple fins, they require sub-resolution resist openings to be printed. The rounding of the resist opening increases the Wfin as Lfin reduces and can be mitigated by advanced illumination condi-tions and OPC (see Fig.3). It can be avoided by strapping the fins with a local interconnect Metal-0 (see Fig.4), at the expense of extra process complexity. From a lithography point of view, the problem of resist corner rounding is the shifted to a fin line shortening, which is believed to be much more manageable by OPC. A second, related issue is the use of advanced illumination conditions, like dipole [9], and their layout restrictions like forbidden pitches and uni-directionality.
As technologies scale sub-45nm, uni-directionality in layout style may be required to help maintaining the lithography process window. FinFETs provide a great opportunity because by nature they favor single direction, CD and pitch, and provide therefore a regular (like DRAM) style layout to make logic circuits. Fig.5 shows a typical 6T-SRAM active areas pattern with 150nm pitch with and w/o OPC correction [9, 10].
Gate patterning The opportunities provided at the fin level create some chal-
lenges at the gate level, mainly due to the extra topography. A first challenge is that, depending on the fin height (Hfin), the process
window for gate etch may shrink significantly. A typical non-optimized process yields poly residues from micro-masking (see Fig.6). One opportunity is the use of the ‘poly etch-back’ process (see Fig.7) leading to a smoother poly-edge preventing any HM micro masking and hence residues.
A second challenge is the CD uniformity as the gate runs over a fin. Fig.8 (top) shows two examples of logic gates indicating the loss of CD control over narrow fins (Wfin<35nm). The opportunity arises from BARC etch optimization combined with ‘topography-dependent OPC’ leading to a much improved process (Fig.8 bottom). A particular challenge is the simultaneous integration of logic and SRAM style circuits in sub-45nm nodes [9], particularly combined with spacer defined fin patterning.
Metal Gate and Vt tuning A major challenge in the case of any narrow active area device,
including multi-gate devices, it the difficulty to tune the threshold voltage by means of channel doping. Extremely high doping concen-trations are needed for narrow-fin devices, exceeding 5e18cm-3, to have any sizeable effect on VT as shown on Fig.9. This raises con-cerns of VT fluctuations due to random dopant distribution, signifi-cant mobility degradation and junction leakage. In addition, the VT of highly doped fins is sensitive to variations in Wfin (Fig.10).
The opportunity lies in effectively making use of double gate ef-fect to control short channel effects (SCE) combined with tune-able metal gates (MG). Undoped channels maximise mobility, sub-threshold slope and Vt control. Ideally one would use a single n-type (or p-type) metal that is selectively tune-able (implant of C, N, F,…) towards p-type (or n-type). Inserted midgap metals like TaN and TiN have been demonstrated on FinFETs [9,11,12,13] as shown in Fig.10. The use of fully silicided (FUSI) gates in FinFETs has been demon-strated [14] and provides MG action with low Vt’s by use of NiSi phase control [15,16] or alloying.
45/32nm circuit aspects FinFETs have been integrated into various SRAM cell seizes
[17,18] down to 0.274um2 [9] using optical litho (Fig.12). The cell concept has been shown to scale below 0.14um2 [19].
For reasons of cell stability and speed, one needs to design proper beta ratios. Opportunities to modulate the beta ratio are choosing different L’s, selectively etching back the Hfin differently over nFET and pFET regions and selecting the right starting sub-strate surface and notch position.
Recently, the use of channel strain to boost drive currents in bulk and SOI technologies have been demonstrated [20,21]. Initial results on FinFETs have been reported [22], but more work is required to match current state-of-the-art CMOS.
Another opportunity, that requires more in depth study, is the po-tential to optimize the technology for [A/um2] rather than the tradi-tional [A/um]. Very high current densities can be achieved by in-creasing the Hfin that may enable different architectures to be opti-mal (eg, BEOL loaded circuits).
References [1] B.Yu, IEDM 2002, p.251 [2] E. Nowak, IEEE Circ&Dev Mag 2004, p.20[3] E. Nowak, IBM J. R&D Vol.46 2002, p.169 [4] F.L. Yang, IEDM 2002, p.255 [5] B. Doyle, Symp.VLSI Tech. 2003, p.133 [6] T. Park, Symp. VLSI Tech. 2003, p.135 [7] S.M. Kim, Symp. VLSI Tech 2005, p.196 [8] F.L. Yang, Symp. VLSI Tech 2004, p.8 [9] L. Witters, Symp. VLSI Tech. 2005, p.106 [10] A. Nackaerts, IEDM 2004, p.269 [11] K. Henson, Symp. VLSI-TSA Tech. 2005, p.136
[12] N. Collaert, Symp. VLSI Tech. 2005, p.108 [13] C. Jahan, Symp. VLSI Tech. 2005, p.112 [14] A. Kottantharayil, Symp. VLSI Tech. 2005, p.198 [15] M. Terai, Symp. VLSI Tech. 2005, p.68 [16] J. Kittl, Symp. VLSI Tech. 2005, p.72 [17] J-H Yang, IEDM 2003, p.23 [18] T. Park, IEDM 2003, p.27 [19] D. Fried, IEDM 2004, p.261 [20] T. Ghani, IEDM 2003, p.978 [21] C.D. Sheraw, Symp. VLSI Tech. 2005, p.12 [22] P. Verheyen, Symp. VLSI Tech. 2005, p.194
• Mesa-type fin pattering • Surface repair + conditioning • Gate deposition + patterning
o 1.4nm HfO2 o 5nm TiN o 100nm polysilicon o 193nm photo, gate etch
• Extension/HALO implant • HDD Spacer • deep S/D implant • Spike anneal • NiSi formation
Fig. 1 Simplified FinFET process flow
Fig. 2 SEM image of a mesa-type active area Fig. 3 Wfin as function of Lfin
Fig. 4 Two different ways of connecting fins, (top) active areas (bottom) local interconnect M0.
Fig. 5 Fin CD for standard process (a) and OPC optimized process (b)
Fig. 6 FinFET after gate etch, (top) with non-optimised etch process, (bottom) with optimized process (etch back)
Fig. 7 Cartoon showing standard (top) and poly etch-back (bottom) process.
Fig. 8 Gate CD variation over fin topography, (top) examples of standard process in logic area, (bottom left and right) optimized process in logic and 0.3um2 SRAM cell
Fig. 9 2D Device simulation result indicating the sentiv-ity of Vt to channel doping for 3 diffferent device archi-tectures and work function.
Fig. 10 Measured nFET Vtlin as function of Wfin for various channel doping levels.
Fig. 11 TEM showing Fin with inserted TiN / HfO2 gate capped with 100nm amorphous-Si.
Fig. 12 Tilted SEM of a 45nm node SRAM cell with FinFETs
© imec 2005
Perspective on Emerging Devicesand
their Impact on Scaling Technologies
S. Biesemans
SSDM 2006
© imec 2005 2SSDM 2005
OutlineOutline
IntroductionIntroduction
Fin patterning, 6TFin patterning, 6T--SRAM cell fabricationSRAM cell fabrication
Gate patterning, MG and Gate patterning, MG and Vt Vt tuningtuning
ConclusionsConclusions
© imec 2005 3SSDM 2005
How it all got started...1947
The Point contact transistorBell Labs
1958The First IC
Texas Insruments
1961The First Planar IC
Fairchild
© imec 2005 4SSDM 2005
MOSFETPlanar processSiN Passivation
LOCOSIon implant
Plasma Etch
SteppersLDD
Silicide
KrF lithoCMP
Extensions / HALO
Cu metal
Spike RTAArF litho
SiONChannel strainLow-k dielectricCo-implant USJImmersion litho
High-KMetal GateMulti-Gate
SEGmsec anneal
‘70 ‘80 ‘90 ‘00
A Few enabling R&D Breakthroughs
‘60
© imec 2005 5SSDM 2005
Scaling Trends
0
50
100
150
200
250
300
350
400
0 50 100 150 200 250 300 350 400
Technology nodeLg [um]
Technology Node
Lgate Tox
0.5
1
1.5
2
2.5
3
3.5
4
0 50 100 150 200 250 300
Tox [um] MPU
Tox [nm] ITRS
Technology Node
Slower
than I
TRSAccela
ration
Slowdown
Introduction of Channel Strain at 90nm and beyond allows to meetchip performance with less aggressive scale of Lg and Tox
© imec 2005 6SSDM 2005
Power - Performance
Increasing VddDecreasing LgDecreasing Vt
Chip package thermal limit
Fast
par
t
Nom
inal
par
t
Slo
w p
art
Lets have a look at the Power-Performance curve !
© imec 2005 7SSDM 2005
Power - Performance
N-1 NN+1
~25%~25%
After investigation :• Performance ~ CV/I• Power (Iddq) ~ Ioff*V
Power 2x/node Objective of scaling
© imec 2005 8SSDM 2005
Power - Performance
N-1 NN+1
~25%~25%
After investigation :• Performance ~ CV/I => need high Idsat• (Static) Power ~ Ioff*V => need low Ioff
© imec 2005 9SSDM 2005
Power - Performance
Traditionally :High Idsat => Thin Tox
However,Thin Tox limited by gate leakage
New paradigm (2003-beyond) :High Idsat =>High mobility
=>Low RS/D
Whats next ?© imec 2005 10SSDM 2005
Scaling Trends
0
50
100
150
200
250
300
350
400
0 50 100 150 200 250 300 350 400
Technology nodeLg [um]
Technology Node
Lgate Tox
0.5
1
1.5
2
2.5
3
3.5
4
0 50 100 150 200 250 300
Tox [um] MPU
Tox [nm] ITRS
Technology Node
??
© imec 2005 11SSDM 2005
Device Design Parameters
Opportunities :1. RS/D -> Strain, USJ & silicide2. µ -> Strain, orientation, Ge..3. Tinv -> Metal Gate, High-K4. Lg -> Architecture5. Cpar -> SOI, Tgate, Spacer
1 23
4
5
Any new CMOS family member isworking on 1 or more of these.
© imec 2005 12SSDM 2005
Feature Size Scaling (Lg)
Raised S/D
© imec 2005 13SSDM 2005
Device Architecture
Improves:- Short channel effect
double gate effect
- Sub-threshold slopestronger gate-channel coupling
- Mobilitylower channel doping
From Bulk to SOI to Multi-Gate
Bulk MOSFET
Thin Body SOI MOSFET
Double Gate MOSFET
-0.2
-0.1
0
0.1
0.2
0.3
0 0.02 0.04 0.06 0.08 0.1
DG-SOI_5nmDG-SOI_10nmFD-SOI_5nmBulk
Lgate [um]
EMER
ALD
© imec 2005 14SSDM 2005
Device scaling
© imec 2005 15SSDM 2005
Bulk scaling+ strain- high doping levels
Device scaling
© imec 2005 16SSDM 2005
Device Architecture
FinFET (DG) fabrication
© imec 2005 17SSDM 2005
Device Architecture
© imec 2005 18SSDM 2005
Device Architecture
Ring Oscillators
© imec 2005 19SSDM 2005
Co-design of TCS
Improved loop: more internal feedback is needed at the various levels, with the adding of the DEVICES (wires and Xtors) at the tech. level
Design strategies
EnergyDelay
IC High Level model
Feed-back
Feed-back
Technology parameters
Keff
w s p=w+s
Cl
Cp
Cp
Cl
h=p/2
h
h
ρeff
Keff
w s p=w+s
Cl
Cp
Cp
Cl
h=p/2
h
h
ρeff
IC System levelRAM
RAM
Data-path
RAM
RAM
RAM
Data-path
RAM
IC circuit level
Technology levelWires + Xtors
R(f) L(f)
C(f) G(f)
LOAD
R(f) L(f)
C(f) G(f)
LOAD
R(f) L(f)
C(f) G(f)
LOAD
IC System levelRAM
RAM
Data-path
RAM
RAM
RAM
Data-path
RAM
IC circuit level
Technology levelWires + Xtors
R(f) L(f)
C(f) G(f)
LOAD
R(f) L(f)
C(f) G(f)
LOAD
R(f) L(f)
C(f) G(f)
LOAD
Design strategies
EnergyDelay
IC High Level model
Feed-back
Feed-back
Technology parameters
Keff
w s p=w+s
Cl
Cp
Cp
Cl
h=p/2
h
h
ρeff
Keff
w s p=w+s
Cl
Cp
Cp
Cl
h=p/2
h
h
ρeff
Technology parameters
Keff
w s p=w+s
Cl
Cp
Cp
Cl
h=p/2
h
h
ρeff
Keff
w s p=w+s
Cl
Cp
Cp
Cl
h=p/2
h
h
ρeff
IC System levelRAM
RAM
Data-path
RAM
RAM
RAM
Data-path
RAM
IC circuit level
Technology levelWires + Xtors
R(f) L(f)
C(f) G(f)
LOAD
R(f) L(f)
C(f) G(f)
LOAD
R(f) L(f)
C(f) G(f)
LOAD
IC System levelRAM
RAM
Data-path
RAM
RAM
RAM
Data-path
RAM
IC circuit level
Technology levelWires + Xtors
R(f) L(f)
C(f) G(f)
LOAD
R(f) L(f)
C(f) G(f)
LOAD
R(f) L(f)
C(f) G(f)
LOAD
© imec 2005 20SSDM 2005
Conventional scaling
SPICE model
Expectation :Substrate : Si, SOI, Ge, GOI, GaAs, InP,..Gate stack : SiO2, HK, poly, Metal Gate,..Architecture : Bulk, PD-SOI, FD-SOI, finFET,..
Models become more complex capturing thePhysics, but remain abstract for designer
© imec 2005 21SSDM 2005
FinFET benefits
DeviceScaling Lg (improved SCE immunity)Higher IdriveMaintain CV/I improvement 25% per node
Circuit density1 direction1 fin & gate density1 WNo STI isolation issues (stack nFET/pFET closer)
Architectural (high speed)Opportunity to improve V/I wrt bulk (Hfin can be increased)
What is the benefit in case BEOL RC does not scale ?
Process technologyCompatibility with std processing
© imec 2005 22SSDM 2005
© imec 2005 23SSDM 2005 © imec 2005 24SSDM 2005
OutlineOutline
IntroductionIntroduction
Fin patterning, 6TFin patterning, 6T--SRAM cell fabricationSRAM cell fabrication
Gate patterning, MG and Gate patterning, MG and Vt Vt tuningtuning
ConclusionsConclusions
© imec 2005 25SSDM 2005
Process sequence
Mesa-type fin patterning
Sidewall conditioning
Gate deposition + patterning0.4nm chemox1.4nm ALD-HfO25.0nm ALD TiN100nm poly-silicon193nm photogate etch + strip
Extension/HALO
HDD spacer
Deep S/D + spike RTA
Raised S/D epi
NiSiCA/M1
FinFET follows conventionalprocessing, can be donein any Si pilot line.
Process re-useability is higherthan 70%.New steps:
- HK / MG- raised S/D epi(-spacer defined fin)
© imec 2005 26SSDM 2005
Fin patterning
2 approachesResist-defined fin (conventional)Spacer-defined fin
Si finsource drain Si film
buried oxideSi fin
source drain Si film
buried oxide
Box
Fin
HMResist
Recess
Box
Fin
Spacer
Resist-defined Spacer-defined
© imec 2005 27SSDM 2005
Fin patterning
2 approachesResist-defined fin (conventional)Spacer-defined fin
Si finsource drain Si film
buried oxideSi fin
source drain Si film
buried oxide
Box
Fin
HMResist
Recess
Box
Fin
Spacer
Resist-defined Spacer-defined
© imec 2005 28SSDM 2005
Fin patterningResist-defined
Hard MaskTEOS, SiN or SiON, APF (C-based)Used for CD trimmingIssues:
•Removal creates box recess•Too much trimming creates LER
Photo process193nm, attPSM for optimum process control and resolutionCD control
Etch processModified gate etch process (Hfin=30-60nm)
•X-talline Si instead of poly-Si•Stopping on thick Box
Strip processFin conditioning
2nm oxidationH2 anneal @900C
© imec 2005 29SSDM 2005
Fin patterningResist-defined
Bulk CMOS gate etching know-how can beRe-used to etch fins
© imec 2005 30SSDM 2005
Fin patterningResist-defined
CD 10-20nm
Corner roundingH2 anneal @900C
© imec 2005 31SSDM 2005
Fin patterningResist-defined
70 nm35 nm
Fin length
Fin length
Fin width depends on Fin length
© imec 2005 32SSDM 2005
Fin patterningResist-defined
Wfin
Lfin
90
100
110
120
130
140
150
50 100 150 200 250 300 350 400
Fin Length L-eff (nm)
Fin
Wid
th W
(nm
)
conv NA0.63 conv NA0.75 annular NA0.75quasar NA0.63 quasar NA0.75
Diff. in W for L-long to L-short :* NA0.63, conv. = 48.2nm* NA0.75, conv. = 32.9nm* NA0.75, ann. = 38.6nm* NA0.63, quas. = 12.9nm* NA0.75, quas. = 17.7nm
130n
m
90nm
65nm
45nm
32nm
Need for advanced litho settings & OPC
© imec 2005 33SSDM 2005
373.5
101.2
400.4
103.7
351.8
100.1FIN_Lexp
FIN_W
Fin patterningResist-defined
719.5
91.4
226
121.9NO OPC
281.8
99.9
OPC 1
761.9
96.4
OPC 2
270
97.9
737.9
95.5
60
70
80
90
100
110
120
130
200 400 600 800 1000 1200Lexp (nm)
W (nm) NO OPC - NA0.63 sigma 0.89OPC1 - NA0.63 sigma 0.89OPC2 - NA0.63 sigma 0.89
© imec 2005 34SSDM 2005
Fin patterningResist-defined
400.4
103.7
281.8
99.9
761.9
96.4
Dry 193, conventional illumination, No OPC
Wet 193, Quasar illumination, OPC
© imec 2005 35SSDM 2005
Fin patterningResist-defined
Extract more data out of the image : LER
Meas 1Meas 2
LWR Accuracy
0.1
1
10
1 10 100 1000
r (nm)
G(r)
(a.u
.)
σ = 2.6nm
ξ = 35nmα = 0.5
correlation function
[P. Leunissen et al., Proc SPIE 2005]
0
0.2
0.4
0.6
0.8
1
1.2
0 2 4 6 8 10
LWR (nm)
Prob
abili
ty D
ensi
ty (a
.u.)
LS2LS8
© imec 2005 36SSDM 2005
Fin patterning
2 approachesResist-defined fin (conventional)Spacer-defined fin
Si finsource drain Si film
buried oxideSi fin
source drain Si film
buried oxide
Box
Fin
HMResist
Recess
Box
Fin
Spacer
Resist-defined Spacer-defined
© imec 2005 37SSDM 2005
Fin patterningSpacer-defined
Spacer defined patterning looks promising in this phase of device development because:1. Less LER expected
- pattern defined by spacer not by resist2. Fin density is doubled
- relaxes requirements on lithography- or more Idsat per unit area (V/I instead of CV/I)
3. Less CD variation expected over the wafer- spacer material & deposition parameters can be easily optimized (no dopants in place yet)- allows to optimize for the fin etch and surface smoothness (no trimming needed)
© imec 2005 38SSDM 2005
Fin patterningSpacer-defined
Si
BOX
Si
BOX
Si
BOX
BOX
BOX
TEOSnitrideSi
CD=15 +/- 3nm
© imec 2005 39SSDM 2005
Fin patterningSpacer-defined
Si
BOX
resistspacer Need for wider fins !
BOX
Resist based patternSpacer defined pattern
Spacer-defiend fins always come in closed loops© imec 2005 40SSDM 2005
Fin patterning
2 approachesResist definedSpacer defined
Circuit aspects (resist-based)Typical device test layoutSRAM cellLogic library elements (NAND,…)
© imec 2005 41SSDM 2005
Fin patterningResist-defined
Active Area Contact
Gate
Typical device test layout- fins are connected bya fin pad
- 1 contact would be OK
Typical 6T-SRAM layout- stand-alone fins- each fin need a contact
+ fin width independent of fin length- line end shortening (manageable ?)0 need for metal-0 to connect fins and
maintain density (M1 routing)© imec 2005 42SSDM 2005
Fin patterningResist-defined
Optimised process
© imec 2005 43SSDM 2005
Fin patterningResist-defined
0.186um2 cellpitch = 124nmASML /1250i, NA=0.85immersion
© imec 2005 44SSDM 2005
Fin patterningResist-defined
0.134um2 cellpitch = 112nmASML /1400, NA=0.93
Fin process shows excellentscalability
© imec 2005 45SSDM 2005
0.1
1
2000 2001 2002 2003 2004 2005 2006 2007 2008
TSMCIBMNECIFXToshibaSamsungMotorolaIntelSonySTLSI LogicHitachiFujitsuImecTexas Instruments
Year of publication [IEDM, VLSI]
130
9065
SRAM patterningResist-defined
SRAM cell size roadmap
3245
E-beam
0.2
0.3
0.5
2
Optical trend line
0.186um2
0.134um2
© imec 2005 46SSDM 2005
SC45 (22/06/2004)
Cross-section of FIN
SRAM array after poly-Si patterning
SRAM array after spacer etch
SNM = 170mV @ 0.6V VDD
SNM = 240mV @ 1.0V VDD
45nm SRAMX=730nmY=430nm0.314um2
© imec 2005 47SSDM 2005
Cell cross-sections
© imec 2005 48SSDM 2005
Fin patterningResist-defined
Active Area Contact
Gate
Metal-0
Typical logic library element- stand-alone fins- each fin need a contact
Typical device test layout- fins are connected bya fin pad
- 1 contact would be OK
Modified logic library element- stand-alone fins- Metal-0 local interconnect- only 1 contact needed
© imec 2005 49SSDM 2005
OutlineOutline
IntroductionIntroduction
Fin patterning, 6TFin patterning, 6T--SRAM cell fabricationSRAM cell fabrication
Gate patterning, MG and Gate patterning, MG and Vt Vt tuningtuning
ConclusionsConclusions
© imec 2005 50SSDM 2005
Gate patterning
Poly-SiSiONSi-fin
MuGFETBulk FET
© imec 2005 51SSDM 2005
Gate patterning
BOX
POLY
Active Area
TEOS HM
POLY
Active Area
BOX
POLY
Active Area
BOXBOX
Active AreaActive Area
BOX
Active Area
BOX
High topography gate formation
Ex. Micro masking by HM
© imec 2005 52SSDM 2005
Gate patterningEtch-back approach
BOX
POLY
Active Area
BOX
Active Area
POLY
• deposit 200nm poly• anisotropic etchback to 100nm• litho + gate etch
• deposit 100nm poly• anisotropic• idem (same recipes)
+• less over-etch for removal of poly residues • larger process window for hard mask trimming• better CD uniformity • nitride spacer etch: less residual nitride on top of gate-• extra non-uniformity
© imec 2005 53SSDM 2005
Threshold Voltage control
Wfin_design (um)
No fin doping
various fin doping
Line
ar th
resh
old
volta
ge
No fin dopingN+ poly gate
various fin dopingN+/poly gate
Target Vt range
Line
ar th
resh
old
volta
ge
Wfin [um]
0.0350.050
0.0700.100
Undoped fin
Doped fin
© imec 2005 54SSDM 2005
Threshold Voltage control
Wfin_design (um)
No fin doping
various fin doping
Line
ar th
resh
old
volta
ge
No fin dopingN+ poly gate
various fin dopingN+/poly gate
Target Vt range
Line
ar th
resh
old
volta
ge
Wfin [um]
0.0350.050
0.0700.100
Undoped fin
Doped fin
Need MG (midgap +/- 200mV ?)
© imec 2005 55SSDM 2005
Gate patterningMetal Gate
Poly-SiSiONSi-fin
MuGFETBulk FET
Poly-SiTaN, TiN, W, Mo,…SiON, HfO2Si-fin
© imec 2005 56SSDM 2005
poly
TiN
HfO2
TiN
/ HfO
2
Fin
Box
Gate patterningMetal Gate
© imec 2005 57SSDM 2005
PMOS (VDD=-1V)
1.E-11
1.E-10
1.E-09
1.E-08
1.E-07
1.E-06
1.E-05
100 150 200 250 300 350 400 450 500Ion [uA/um]
Ioff
[A/u
m]
poly/SiON bulk results (C065)ARTEMIS targetITRSMUGFET (Wfin=20nm)
Gate patterningMetal Gate
For Narrow FIN (~20nm):Excellent PMOS demonstrated on (HfO2/TiN/poly)Ion uncertainty from in-line Wfin measurements evaluated to be ~ +/- 7%
Poly/SiON IBMTSMC
HITACHIEuropeanpartners
Jg ~ 60A/cm2
Jg=4mA/cm2
Poly/TiN/HfO2_2nm
265uA/um @ 20pA/um, Vdd=1.0V
© imec 2005 58SSDM 2005
NMOS (VDD=1V)
1.E-10
1.E-09
1.E-08
1.E-07
1.E-06
1.E-05
300 400 500 600 700 800 900 1000Ion [uA/um]
Ioff
[A/u
m]
poly/SiON bulk results (C065)ARTEMIS targetITRSMUGFET (Wfin=20nm)MUGFET (Wfin=40nm)HK/MG bulk results
Gate patterningMetal Gate
For Narrow FIN (~20nm):NMOS is fairly weak, in regards to this gate stackMore characterization on going to better quantify drive loss from such narrow FIN (Rext)
Poly/TiN/HfO2_2nm
Europeanpartners
IBMTSMC
HITACHI
Poly/SiON
Poly/SiONJg ~ 60A/cm2
Jg=20mA/cm2
© imec 2005 59SSDM 2005
Mobility improvement
10-11
10-10
10-9
10-8
10-7
10-6
10-5
10-4
200 300 400 500 600 700
referencetensilecompressive
I off [A
/µm
] @ V
gs=-
0.3
V
Ion
[µA/µm] @ Vgs
=0.7 V
Vds
=1 V
15%-30%
NMOS with CESL
10-10
10-9
10-8
10-7
10-6
10-5
10-4
200 300 400 500 600 700
SiGe S/DReferenceI O
FF[A
/µm
] @ V
GS=0
.2V
ION
[µΑ/µm] @ VGS
=-0.8V
25%
10-15% improvement in drive current from implementation
of CESL with strain
implementation of SiGe SD into MUGFET increases 25% PMOS
performance (Rs reduction + compressive strain)
poly
SiGe SiGe
SiGe
-20
-10
0
10
20
30
40
50
0 200 400 600 800 1000
100 nm 800 MPa
100 nm 1.4 GPa
Mob
ility
gai
n ∆µ
/µ [%
]
Lg [nm]
open symbols: top channelclosed symbols: sidewalls
pMOS
nMOS
© imec 2005 60SSDM 2005
Conformal junctions
20nm reduction in Lmin in NMOS and PMOS devices with PLAD
PLAD proved to form conformation junctions*
*) As data
Transistors with PLAD junctions demonstrated
1.E-07
1.E-06
1.E-05
1.E-04
1.E-03
1.E-02
-1 -0.5 0 0.5 1Vg (V)
Id (A
)
NMOSPMOS
Vd=-1V Vd=1V
Vd=-0.1VVd=0.1V
Lpoly=46nm Fin Width=21nm
© imec 2005 61SSDM 2005
OutlineOutline
IntroductionIntroduction
Fin patterning, 6TFin patterning, 6T--SRAM cell fabricationSRAM cell fabrication
Gate patterning, MG and Gate patterning, MG and Vt Vt tuningtuning
ConclusionsConclusions
© imec 2005 62SSDM 2005
Summary
FinFETs can be made with conventional processingShow excellent scalability, devices & circuit densityRemaining issues
Idsat improvementIntroduction of strainJunction engineering w/o defects
© imec 2005
Back up
© imec 2005
Back up
© imec 2005
Back up
© imec 2005
Back up
© imec 2005 67SSDM 2005
Metal gate for 3D-SOI devices
Poly silicon5nm TiN
Poly
TiNSi fins
Gate profileNOTCH
© imec 2005 68SSDM 2005
Gate & Fin patterning
Si fins
Poly-SiSiONSi-fin
MuGFETBulk FET
CDCDU1D Profile1D Notch/FootDefects
CDCDU2D Profile2D Notch / FootDefects
© imec 2005 69SSDM 2005
Metal gate for 3D-SOI devices
Si fins
Active
FieldOxide
Poly
HM
TaN
Active
FieldOxide
Poly
HM
TaN
Poly-SiTaN, TiN, W, Mo,…SiON, HfO2Si-fin
45nm FETBulk FET
+
© imec 2005 70SSDM 2005
Observation : GIDL
Ioff could be reduced by up to 2-3 orders of magnitude if we can fix GIDL
GIDL determines Ioff
*) As data
GIDL in SOI FETs
n+ n+
N-type0V
Vg < 0V
VDD
BOX
h e
GIDL
10-12
10-10
10-8
10-6
10-4
0.01
-0.5 0 0.5 1 1.5
Ids
versus Vg at V
ds=0.02V and V
ds=1.3V, W=20nm
(HfO2/ALD TiN - NMOS)
Lg=1um @Vd=1.3VLg=5um @Vd=1.3VLg=10um @Vd=1.3VLg=1um @Vd=0.02VLg=5um @Vd=0.02VLg=10um @Vd=0.02V
Dra
in c
urre
nt (A
/um
)
Vg (V)
Ioff
GIDL
© imec 2005 71SSDM 2005
HARMONY++, BS20
BS200.186um2 cellPC pitch = 150nm/1250i
Poly
© imec 2005 72SSDM 2005
Fin
Gate
Poly residue
© imec 2005 73SSDM 2005
Poly etch back for smooth topography
BOX
POLY
Active Area
BOX
Active Area
POLY
Etch back
BOX
POLY
Active Area
© imec 2005 74SSDM 2005
© imec 2005 75SSDM 2005
2D Simulation study (Lg=25nm)
1E16 1E17 1E18 1E190
200
400
600
800
1000
1200
DIB
L (m
V/V
)
Bulk FDSOI (TSi) DGFET (WFIN) 15 nm 15 nm
10 nm 10 nm 5 nm 5 nm
Channel doping (cm-3)
1E16 1E17 1E18 1E19
100
200
300
400
500
600
Bulk FDSOI (TSi) DGFET (WFIN) 15 nm 15 nm
10 nm 10 nm 5 nm 5 nm
Sub
thre
shol
d sw
ing
(mV
/dec
.) @
VD
S= 1
V
Channel doping (cm-3)1E16 1E17 1E18 1E19
-0.4
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
ΦM= 4.17 V
ΦM= 4.0 V
ΦM= 4.6 V
Bulk FDSOI DGFET
Thre
shol
d V
olta
ge (V
)
Channel doping (cm-3)
0.00 0.02 0.04 0.06 0.08 0.100
200
400
600
800
1000
DIB
L (m
V/V
)
Bulk FDSOI (TSi) DGFET (WFIN) 10 nm 10 nm
5 nm 5 nm
LG (µm)0.00 0.02 0.04 0.06 0.08 0.10
50
100
150
200
250
300 Bulk FDSOI (TSi) DGFET (WFIN)
10 nm 10 nm 5 nm 5 nm
S SAT
(mV
/dec
.)
LG (µm)0.01 0.02 0.03 0.04 0.05
-0.2
0.0
0.2
0.4
0.6
0.8
Bulk FDSOI (TSi) DGFET (WFIN) 10 nm 10 nm
5 nm 5 nm
V T,LI
N (V
)
LG (µm)
LG= 25 nmWfin/Tsi=15nm
LG= 25 nm LG= 25 nm
© imec 2005 76SSDM 2005
0.1
1
2000 2001 2002 2003 2004 2005 2006 2007 2008
TSMCIBMNECIFXToshibaSamsungMotorolaIntelSonySTLSI LogicHitachiFujitsuImecTexas Instruments
Year of publication [IEDM, VLSI]
130
9065
Status 6T-SRAM cell developmentSRAM cell size roadmap
3245
E-beam
0.2
0.3
0.5
2
Optical trend line
200mm (1100)
300mm(1250i)
(1700)
© imec 2005 77SSDM 2005
SOI MuGFET Device Roadmap
Dual silicideDual silicideNiSiSilicide engineering
W or betterW or betternoneM0
TBITBICESLStrain engineering
LOCOSFETCompound channel
LOCOSFETFinFETArchitecture
SBFETPLAD + laserEPI-style + laserSEG
Implant + spikeSEG
USJ engineering
HK / Dual MGMore WF tuningNotch free MG
(SiON+,HfO2)/OtherMore WF tuningNotch free MG
(SiON,HfO2)/TiNWF tuningNotched MG
Gate engineering
100nm15nm20nm
80nm18nm30nm
60nm25nm45nm
HfinWfinLg
22 nm32 nm45 nm
© imec 2005 78SSDM 2005
Program Status
We have a established baseline process
We have demonstrated 45nm node devices
We have demonstrated 45nm node circuits (SRAM cell)
© imec 2005 79SSDM 2005
Program Status
We have a established baseline processRoute availableDraft Design Rule manual availableV0.1 SPICE model available (HW based)
We have demonstrated 45nm node devices
We have demonstrated 45nm node circuits (SRAM cell)
© imec 2005 80SSDM 2005
Highlights: baseline process
before:
Objective: open up the window for gate etch optimization
Now in POR: polySi etch back
BOX
poly-Si
Active Area
BOX
Active Area
poly-Si
Residue free etch back
BOX
POLY
Active Area
poly-Si residues parallel to active area
© imec 2005 81SSDM 2005
Program Status & Highlights
Nano-scale CD
© imec 2005 82SSDM 2005
Program Status & Highlights
BLUES (preliminary floor plan)XSEM
XS
EM
HPLHPL
TDROMTDRAMDevice arrays
RO
AdderLogic Library SRAM cells
Loco
sfet
ACLV
ACLV
ACLV
ACLV
ACLV
ACLV
ACLVRO
Small circuits
OPC, SEM ESD IPC box
Charac. Charac.DevicesTdaamatching
© imec 2005 83SSDM 2005
Program Status & Highlights
We have a established baseline processRoute availableDraft Design Rule manual availableV0.1 SPICE model available
We have demonstrated 45nm node devicesContain all the gismos : HK/MG/Strain/Conformal USJ/…pFET OK, nFET low on Idsat
We have demonstrated 45nm node circuits (SRAM cell)
© imec 2005 84SSDM 2005
Program Status & Highlights
We have a established baseline processRoute availableDraft Design Rule manual availableV0.1 SPICE model available
We have demonstrated 45nm node deviceswith HK/MG/Strain/Conformal USJ/…pFET OK, nFET low on Idsat
Actions in place to get improved performance
We have demonstrated 45nm node circuits (SRAM cell)
© imec 2005 85SSDM 2005
Short term Plan (2H05)
nFET Idsat improvements Gate process
EOT scaling from 2.1nm towards 1.3nm -> 1.1nm -> …0-notch gate TiN processVt tuning strategy (implant, anneal)Other MG candidates (Mo-based, W-based, WSi,…)
Junction processConformal junctions by Plasma Doping, density scalingLaser annealRaised S/D by SEG
Improved PECVD spacerMobility
High stress CESLDifferent orientationsubstrates
SilicideNiSi -> PtSi, YbSi
1000uA
750uA
© imec 2005 86SSDM 2005
Program Status & Highlights
We have a established baseline processRoute availableDraft Design Rule manual availableV0.1 SPICE model available
We have demonstrated 45nm node deviceswith HK/MG/Strain/Conformal USJ/…pFET OK, nFET low on Idsat
Actions in place to get improved performance
We have demonstrated 45nm node circuits (SRAM cell)Demonstrated 0.274um2 cell with TaN/SiON
In progress 0.245 um2 with TiN/HfO2 : 200mmPlan for 0.180 um2 in place (193i) : 300mm
Printed (optical) Active Area (AF,RX) consistent with 0.13um2
© imec 2005 87SSDM 2005
SC45-SC32 Status
1Q04 2Q04 3Q04 4Q04 1Q05 2Q05 3Q05 4Q05
SC32-1100
SC45-MG
SC45-Logic
SC32-1250i
SC45-MGLogic0.245um2
HARMONY HARMONY+ BLUES28/01/2005
HA++
0.180um2
SC45_Base
0.274um2
0.314um2
© imec 2005 88SSDM 2005
SC45+MG (23/12/2004)
0.2
0.3
0.4
0.50.60.70.80.9
1
2
3
1999 2000 2001 2002 2003 2004 2005 2006
TSMCIBMNECIFXToshibaSamsungMotorolaIntelSonySTLSI LogicHitachiFujitsuImec
Year of publication [IEDM, VLSI]
130
9065
3245
E-beam
Box
Poly-SiInserted-TaxNy
SiONSi-fin
Optical Litho trend line
SC45+MG
SC45
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
10.274 µm2
Node 1 (V)
Nod
e 2
(V)
0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 10
50
100
150
200
250
VDD (V)
SNM
(mV
)
10% VDD
Metal Gate
Poly Gate
Cell: 0.274um2
X=720nmY=380nm0.274um2
© imec 2005 89SSDM 2005
SC32-1100 (15/12/2004)
X=720nmY=340nm0.245um2
Process can be extendedto aggressive pitches on/1100 due to aggressive OPC
Purpose of this work is to prepare the 1250i work
© imec 2005 90SSDM 2005
SC32-1400 (20/1/2005)
Active area printed on /1400 (NA=0.93, 300mm)Active Pitch 112nmCell pitch compatible with 0.134um2 cell
DemoHighest NA possible
© imec 2005
EMERALDprocess learning
© imec 2005 92SSDM 2005
Highlights: Baseline process
FIN scaled down to 20nm without loss in FIN height
Objective: reduce SCE in MUGFET transistor
2.4 nm
15 nm
63 nm
Si- Fin
BOX
Gate Oxide
Poly-Si
2.4 nm
15 nm
63 nm
Si- Fin
BOX
Gate Oxide
Poly-SiResults:
• Better S• Reduced DIBL
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0 0.2 0.4 0.6 0.8 1Lpoly (um)
DIB
L (m
V) Width=1umWidth=60nmWidth=25nm
© imec 2005 93SSDM 2005
Surface smoothening (H2 anneal)
900ºC 5 min (POR)
900ºC 2 min
950ºC 1 min
850ºC 5 min
POR condition is on the edge
looks more rough
© imec 2005 94SSDM 2005
Surface smoothening
H2 anneal vs no H2 annealI of
f[A
/µm
]
Ion [µA/µm]
90 nm10% increase2 orders
© imec 2005 95SSDM 2005
Highlights: baseline process
Physical sputtering+ 2nm dry oxidation Radius 10nm for 2nm SSE
Objective: find a crystal-orientation independent technique of corner rounding
900ºC 2 min
POR: Corner rounding achieved by H2 anneal only
© imec 2005 96SSDM 2005
Highlights: Baseline process
Small CD (14-17nm)CD variations
Objective: Increase FIN density for higher current /area
Spacer defined FIN patterning: feasibility proven for FIN doubling
BOX
Ldummy
© imec 2005 97SSDM 2005
Gate & Fin patterning (193nm resist)
Si fins
Poly-SiSiONSi-fin
MuGFETBulk FET
© imec 2005 98SSDM 2005
Gate & Fin patterning (193nm resist)
Poly resisduesRemaining HM BARC planarisation
Optimised processOptimised process
High topography gate formation
© imec 2005 99SSDM 2005
Metal gate for 3D-SOI devices
Si fins
MuGFET withMG
Active
FieldOxide
Poly
HM
TaN
Active
FieldOxide
Poly
HM
TaN
Poly-SiTaN, TiN, W, MoSiON, HfO2Si-fin
45nm FETBulk FET
+
© imec 2005 100SSDM 2005
Solution: Berkeley approach with Si SEG
Si SEG on S/D after parasitic spacer recess (Hgate > Hfin)
Si SEG
© imec 2005 101SSDM 2005
Source/Drain Resistance … (contd.)
Measurements on fabricated devices are shown Narrow fin MuGFETs show high S/D resistance (RSD)
10 102 103
0
2
4
6
8
Par
asiti
c S
/D R
esis
tanc
e (k
Ohm
)
Fin Width (nm)
© imec 2005 102SSDM 2005
T1.1: SEG for MuGFET — faceting
SEG on MuGFET (no poly present):• 45 nm grown on 60 nm heigh fins• Facets are orientation dependent due to the difference in growth rate• Epi overgrowth
single fin
multiple fins
© imec 2005 103SSDM 2005
Box
Si Sub
Si
Box
Si Sub
Si
Highlights: Conformal junctions
Study on compatibility of different conformal USJ formation techniques with MUGFET structure started
(PLAD, VPD, Solid Source, Epi tip, Schottky barrier….)
VPD
Box
Si Sub
Si
Carrier Gas
Box
Si Sub
Si
Carrier Gas
Solid Source diffusionEpi tip
Doped Epi
Suitability and compatibility of alternative annealing techniques (Flash, Laser) initiated
© imec 2005 104SSDM 2005
Novel Architectures
65nm GP & LP
Novel Architectures
Shottky Barrier FET
Tunnel FET
Bulk MuGFET
The Ultimate USJ
High Ion/Ioff ratio: Low Power applications
If SOI is too expensive
Choices: Open for discussion based on partner feedback
Other structures not yet on radar screen?
LOCOS FETLitho friendly FET
2004 2005 2006 2007 2008 32nm 45nm 22nm
© imec 2005 105SSDM 2005
Poly 170 pitch, 6%AttPSM/1100, 0.75NA dry, CD in resist@BF
69.9072.8072.874.074.0
CQUA300.89/0.65
70.0072.7072.772.271.7
CQUA200.92/0.72
QUAS300.89/0.65
QUAS200.92/0.72
75.574.9XY0573.986.6XY0670.170XY07
73.984.6XY0475.574.9XY0377.277.8XY0277.077.8XY01
ANN0.92/0.72
ANN0.89/0.65
© imec 2005 106SSDM 2005
OPCGap=50nm
© imec 2005 107SSDM 2005
OPCGap=80nm
© imec 2005 108SSDM 2005
STI process with flat topography
© imec 2005 109SSDM 2005
Contact after litho
© imec 2005 110SSDM 2005
© imec 2005 111SSDM 2005
Gate & Fin patterning
Residue ?Left over poly?
© imec 2005 112SSDM 2005
Logic: 30nm
SRAM1: 55nm SRAM2: 51nm SRAM3: 48nm
SRAM1 bridging
SRAM1: 55nm
SRAM3: 46nm SRAM3: 38nm
SRAM2: 30-37nm SRAM3: 31-40nm
Gate patterning
50 nm
After stack etch After silicidation
Fin height reductionin small lines