Transcript The presentation 2010 - Department of Electrical and Computer

Novel dual-Vth independent-gate
FinFET circuits
Masoud Rostami and Kartik Mohanram
Department of Electrical and Computer Engineering
Rice University, Houston, TX
Laboratory for Sub-100nm Design
Department of Electrical and Computer Engineering
Outline






Introduction and motivation
Background
Dual-Vth independent-gate FinFETs
Logic design
Simulation results
Conclusions and future directions
2
Introduction
 Technology scaling
 Process variations
 Leakage power
 Short channel effects
 Planar double-gate FETs and FinFETs
 Compatibility with planar CMOS
 Scalability
 Suppression of short channel effects
 Low parametric variations
 High Ion/Ioff
3
Motivation
 Conventional FinFETs
 Tied-gate devices
 Independent-gate FinFETs
 Removing top oxide
 Electrically isolated, electrostatically coupled gates
 Muttreja, Agarwal, and Jha, ICCD, 2007
 Caciki and Roy, IEEE TED, 2007
 Tawfik and Kursun, Microelectronics Journal, 2009
Motivation
 Conventional FinFETs
 Tied-gate devices
 Independent-gate FinFETs
 Low-power logic gates
 Disabled, reverse-biased back-gate
 Independent-gate logic gates
 Merge parallel transistors, compact logic
Motivation
 Merge series transistors?
 Dual-Vth independent-gate FinFETs
 Device design considerations
Motivation
 Merge series transistors?
 Dual-Vth independent-gate FinFETs
 Device design considerations
 Independent-gate dual-Vth FinFETs
 Logic design opportunities
 Compact logic gates
 New family of logic gates
 Delay+power+area evaluation
Background
 FinFET
 Cross-sectional view
Front Gate
tox
Source
n+
tsi
Channel (undoped)
L
Back Gate
Underlap
Drain
n+
Background
 FinFET models
 University of Florida double-gate (UFDG)
 Physics-based, good agreement with manufactured devices
 Fin height, silicon thickness, S/D doping, underlap, gate
electrode work-function
 Predictive technology model (PTM)
 Modeled as two parallel coupled SOI devices
9
Background
 Finfet threshold voltage VT
 Φms and Cox tuning
 Independent-gate FinFETs
 Electrically isolated, electrostatically coupled
 Vth-dependence model [Colinge 2008]
 Vth-front = Vth-front0 – δ (Vgbs – Vth-back) if Vgbs < Vth-back
 Vth-front = Vth-front0 in all other cases
 Substrate-like effect in planar CMOS
10
Dual-Vth FinFETs
 Φms tuning
 Two additional mask steps
 Cox tuning through tox
 Asymmetric oxides [Masahara, IEEE EDL 2007]
 Device design using UFDG model
Dual-Vth FinFETs
IDS (A)
VDS = 1V
VGS (V)
Dual-Vth FinFETs
 TCAD simulations (2D Sentaurus)
 Same device geometry
2D Sentaurus
UFDG
Dual-Vth FinFETs
UFDG
2D Sentaurus
Dual-Vth FinFETs
 Good Vth separation
 Good noise margin (approx. 0.5Vdd)
 Leakage current in high-Vth device
 pA for low-Vth devices
 nA for high-Vth devices in disabled-gate mode
 Comparable to 32nm CMOS
Dual-Vth logic gates
 Rules for pull-down and pull-up network:
 Parallel structure ↔ series structure
 “Weak” AND-like high-Vth transistor ↔ “strong” ORlike low-Vth transistor
16
Dual-Vth logic gates
Novel dual-Vth logic gates
 Novel logic gates
 Independent back-gate as independent input
 n-input gate with n, n+1, …, 2n inputs
 Example f = (a + b)(c + d)
 n = 2, 12 unique combinations
 Some competitive, some not
 Exponential in n
 Occupancy problem
 Series-parallel graphs
 Functionally equivalent,
electrically different gates
Results
 Technology libraries for n = 3
 Basic library
 Traditional INV, NAND, NOR, AOI, OAI
 Previous work library = Basic library +
 Compact gate with parallel transistors merged
 Low power gates with disabled back-gate
 Merged series = Previous work library +
 Dual-Vth logic gates, with series transistors merged as
appropriate
 Complete library = Merged series library +
 Novel dual-Vth logic gates
Results
 ISCAS and OpenSPARC circuits
 Area (no. of fins), delay, total power
 Improvements: Basic, Previous work, Merged series
 Area savings: 27%, 23%, 12%
 Delay improvement: 7%, 7%, 1%
 Total power: 24%, 21%, 15%
 Static power
 10-100X higher, but net contribution negligible
 Dynamic power
 Dominates
 Improvements with proposed complete library significant
Conclusions
 Dual-Vth FinFET design
 Gate work-function
 Oxide thickness
 UFDG-based search and 2D TCAD validation
 Compact merged series-parallel logic gates
 Novel dual-Vth logic gates
 New opportunities for FinFET-based design
 Leakage power control
 Underlap as a design parameter
Double gate devices
 Reduction in relative strength of gate
 Two gates bring more electrostatic stability
 Double gate devices have:
 Less DIBL, GIDL and leakage power
 Better Ion/Ioff
 Better subthreshold slope
 Fabrication issues (alignment, etc)
22
FinFET
 FinFETs are folded channel MOSFETs
 Easy manufacturing process
 Narrow vertical fin(s) stick up from the surface
[1]
[1] D. Hisamoto, et al, “FinFET—A Self-Aligned Double-Gate MOSFET Scalable to 20nm”,
IEEE Tran on Electron Devices
23
FinFET cross view
 Can you see the underlap?
Front Gate
Tox
Source
n+
Channel-Undoped
Tsi
Drain
p+
Length
Back Gate
 Channel engineering unfeasible
 Different strength for each gates is possible by tuning:
 Work–function
 Oxide thickness
24
Outline




Introduction and motivation
Device characteristics
Circuit innovation with FinFET and results
Future directions and conclusion
25
I-V curves (current vs. drain voltage)
26
 PMOS
 NMOS
Off-current
On-Current
-3
10
-3
10
-4
10
Ids
Ids
Ids
Ids
-4
10
-5
10
Tied-Gate
Vbg=0
Vbg=-.4
Tied-Gate
Vbg=0
Vbg=-.4
-5
-6
10
0
0.1
0.2
0.3
0.4
Vds
Vds
0.5
0.6
0.7
0.8
10
-0.8
-0.7
-0.6
-0.5
-0.4
Vds
Vds
-0.3
-0.2
-0.1
0
Backgate and discrete “width”
 Disabling the backgate:
 An order of magnitude less on-current
 Less static leakage
 Suitable for off-critical paths
 The height cannot be changed across chip
 Stronger devices by adding parallel fins [1]
 Gate sizing will be a discrete problem
 W = n.Hfin n=1,2,…
[1] J. P. Colinge, “FinFET and other multi-gate transistors”, chapter 1 and 7, Springer, 08
27
DC properties
 No dopant in channel:
 No random dopant fluctuations
 No Coulomb scattering => Higher mobility
 Higher concentration of traps
 Higher flicker noise and noise figure
 Due to 3D structure
 Much higher heat transmission resistance
 Danger of thermal runaway [1]
 Performance degrades less in alleviated
temperatures
[1] J. H. Choi, et al, “The Effect of Process Variation on Device Temperature in FinFET circuits”, 2007
28
Analog devices
 The unavoidable underlap
 Big source and drain resistance
 Less gm
 Less FT
 gm/2π(Cgs+Cgd)
 Also due to new fringing capacitances
 Still better gm/gds
 Good for gain of amplifiers
 Not a very good SNR reported in ADCs and LCOs
 Due to high flicker noise and charge trapping
29
Sample RF circuit
 Fast coupling of two independent gates can be
exploited for building a compact low-power mixer
 A mixer for down converting the RF signal
 LO=1.8 GHz
 RF=1.6 GHz
 IF=1.8-1.6 =200Mhz
 Very good THD
FFT of Mixer Output
30
Outline




Introduction and motivation
Device characteristics
Circuit innovation with FinFET and results
Future directions and conclusion
31
Innovative circuit techniques




Disabling the backgate
Merging parallel transistors
Merging series transistors
A novel class of static logic
32
Disabling the backgate
 Disabling backgate increases the threshold voltage
 Less leakage and slower
 Suitable for non-critical paths
 New gate has less Cin, too
 Because the driver; sees less ‘gates’
33
Merging parallel transistors
 If either of the gates is active; we still have a channel
 Works like an OR function [1]
 Suitable for non-critical paths.
 Less static leakage and dynamic power (due to Cin)
 Higher sensitivity to parametric variation
[1] V. Trivedi, et al, “Physics-Based Compact Modeling for Nonclassical CMOS”, 2005.
34
Merging series transistors
 Series transistors can be merged if the transistor acts
like an AND gate [1]
 It has low resistance; only if both of the inputs are
active
 Best design parameters chosen by SPICE simulation
 Oxide thickness and gate work-function tuning
[1] M. Chiang, “High-Density Reduced Stack Logic Circuit Techniques Using Independent-Gate Controlled
Double-Gate Devices”, IEEE Tran On Electron Devices, 2006
35
Why not use these ideas concurrently?
36
The main contributions
 Rules for pull-down and pull-up network:
 Parallel structure=> series structure and vice versa.
 “Weak” type (AND-like) transistor => “strong” type
(OR-like) and vice versa
 No substrate effect in FinFET
 Transistors can be stacked in pull-up or pull-down
network more easily
 Many more complex gates are possible!
37
New gate
 Pull-down Boolean
equation:
 PD = (a+b) * (c+d)
 Pull-up Boolean equation:
 PU = (~a*~b) + (~c*~d)
 PU and PD are
complement
 Static logic
 24 different gates realized
by just four transistors
 Just 3 Boolean
functions with CMOS
38
Results
 All the gates were simulated using UFDG model
 The logical effort [1] model of each model was
calculated
 Technology libraries were constructed
 ISCAS85 benchmarks were mapped
 Addition of the new gates showed
 XX improvement in area
 YY improvement in power
[1] R.F. Sproull and D. Harris, Logical Effort: Designing Fast CMOS Circuits, Morgan
Kaufmann, 1999.
39
Outline




Introduction and motivation
Device characteristics
Circuit innovation with FinFET and results
Future directions and conclusion
40
Conclusions
 FinFET devices has a huge potential for replacing the
planar CMOS technology
 They have:
 Better Ion/Ioff ratios
 Better SCE suppression
 Possibility of a second independent gate
 Innovative circuits were designed exploiting
independent gate of FinFET
 Savings in area and power consumptions observed
41
Future work
 Near term:
 Completing the Synopsys chain
 Using backgate for performance tuning
 Offline/online
 Clustering?
 Long term:
 SRAM
 Issue due to discrete width
 Design centering
 FinFET based RF circuits
 Amplifiers
42