072008_NEMS.Ionescue.Akarvardar

Download Report

Transcript 072008_NEMS.Ionescue.Akarvardar

Nanoelectromechanical Switches
(NEM Relays & NEMFETs)
Dimitrios Tsamados
Adrian Ionescu
EPFL
Kerem Akarvardar
H.-S. Philip Wong
Stanford
Elad Alon
Tsu-Jae King Liu
UC Berkeley
Outline
Part I: NEM Relays
• Motivation
• Device operation
• Logic gate configurations
• State-of-the-art
• Scaling and performance
• Issues
• Conclusion
Part II: NEMFETs
2
Motivation
Motivation
Two key properties unavailable in CMOS
Infinite subthreshold slope
Zero Leakage
Ultra-low VDD
without degrading Ion
Zero static energy
Ultra-low
dynamic energy
Ultra-low energy operation
beyond the capability of CMOS
3
Motivation
Extra
Motivation
• Hysteresis
• Stiction
SRAM and NV Memory
• Low temperature process
3-D integration and
hybrid NEMS/CMOS
Choi et al. IEDM 2007
(UC Berkeley)
Chakraborty et al.
IEEE Trans. Circ. Syst. 2007
(Case Western Reserve U.)
• High temperature operation
• High radiation hardness
Niche applications
• Cheap substrates
Reduced cost
4
air gap
S
G
D
insulating substrate
OFF-state
S
G
ON-state
D
1
2
3
0
Drain Current
limit stop
cantilever beam
Normalized Position
Operation
of the
NEM Relay
Conventional
Device
Structure
& Operation
VGS
infinite
slope
zero
leakage
VGS
Pull-in
Pull-out
voltage, Vpo voltage, Vpi
= VDD,min
5
Operation of theand
NEM
Relay
Logic Implementation
Interconnection
S
D
S
G
MOSFET
VDD
VDD
D
G
NEM RELAY
laterallyactuated
cantilever
VDD
in
out
"p-type"
laterallyactuated
cantilever
"n-type"
CMOS
CNEM
Lee et al. SPIE 2005
(Simon Fraser U. –
Canada)
GND
top view
• CMOS schematics can be used in CNEM logic circuits
• No conductivity difference between the n-type and p-type relays
• Simple layout enabling to small area
6
Operation of
the NEM
Relay
Energy-Reversible
(ER)
CNEM
Logic Gates
out = VDD
GND
VDD
GND
VDD
GND
cantilever
insulator
VDD
in
in = VDD
in = GND
Laterally-actuated
ER CNEM Inverter
out = GND
Elastic potential energy, which
is stored due to beam bending,
is reversibly used for switching
Akarvardar et al. DRC 2008
out = in
Top View
ER principle: Yang et al. MME 2003
Delft University – The Netherlands
7
1
energyreversible
A
B
VDD
ER
ER Relays:
NAND
• Reduced VDD & dynamic energy
Gate
(unless hysteresis is prevented)
• Experimentally demonstrated in RF
switches (VDD = 5 V instead of Vpi = 38 V)
GND
0
0 Vpo
Vpi VGS
(energy-reversible) (conventional)
floating
electrode
VDD,min
Conventional
VDD
Normalized Position
Operation of the
NEM Logic
Relay Gates
Energy-Reversible
CNEM
Pakula et al. (Delft U.) IEEE Sensors 2005
• Any logic function can be realized
out = AB
8
State-of-the-Art
The smallest 2-terminal
switch ever reported:
Jang et al. APL 92, 2008
(KAIST & Samsung – Korea)
Vpo = 8 V Vpi = 13 V
300 nm
• 15 nm gap
• 35 nm beam thickness
• TiN beam, sacrificial poly-Si,
wet etch + critical point dry
• Several hundred cycles
endurance (insulator failure)
zero
leakage
infinite
slope
9
State-of-the-Art
Self-assembled "CNT wafers"
on pre-patterned substrates
Hayamizu et al. Nature Nanotech 3
2008 (AIST – Japan)
>1250
relays
Parallel, scalable, and
reproducible relay fabrication
with > 95% structural yield
~5000
SWNTs
10
Constant-Field Scaling
Scaled
parameters
Resulting
variation
Parameter
Scaling Factor
Gap, thickness, length, width
1/K
Supply voltage
1/K
Spring constant
1/K
Resonance frequency
K
Footprint area
1/K2
Pull-in voltage
1/K
Pull-in delay
1/K
Pull-in energy
1/K3
Electrostatic force
1/K2
Elastic force
1/K2
van der Waals forces
K
• Scaling => smaller and faster relay that dissipates less energy
• vdW Forces tend to become dominant at nanoscale
11
0
0
1
nm
Performance
= 250
nm nm
L =L200
– 300
=
Wsilicon
h = 10 nm
g0 = 10nm
4 nm
D
S
G
• 1 ns switching delay
• 1.5 V supply voltage
• 80 aJ switching energy
• 0.03 μm2 lateral inverter area
=> competitive with CMOS
• Zero leakage
Akarvardar et al. IEDM 2007
scaling
Fvdw ↑
delay ↓
~ 1 ns
stiffer beams
to compensate
for Fvdw
increased VDD ~ 1 V
Negligible Fvdw =>
1 ns @ VDD  150 mV
12
NEM relays vs. Low-Power CMOS
Lg = 45 nm LSTP CMOS (ITRS):
High VT (0.53 V) => Very low leakage (30 pA/μm)
Zero leakage advantage of the NEM relay
would only be apparent in logic circuits with
relatively high device count and low activity
CV/I = 1 ns @ VDD = 425 mV
NEM relays should achieve nanosecond-range
intrinsic delay @ VDD << 425 mV
=> Decrease the vdw forces substantially
• How?
=> And/or operate close to the stiction limit
Akarvardar et al.
submitted to
• Increased sensitivity to device param.
IEDM 2008
13
Issues
1. Contact reliability
• hot switching
• high current density
• high impact velocity
2. Sticking: limits the voltage scaling
3. Packaging: hermetic sealing is required
4. Tunneling: determines the minimum gap
5. Long settling time & tip bouncing: tend to
increase the switching delay
6. Brownian motion: leads to switching errors
14
Conclusion
NEM logic can become an alternative ultra-low
power logic technology if:
• Contacts can be reliably implemented at nanoscale
• Nanosecond range delays can be achieved at a
few 100 mV
Detailed roadmapping and intensive
engineering development are recommended
15
Nano-Electro-Mechanical FETs
Hybrid M/NEMS
Pure M/NEM devices:
Hybrid M/NEM devices:
- micro/nano movable parts
- passive device operation
- micro/nano movable parts
- solid state semiconductor
device involved in operation
Ex: suspended-gate FETs
Ex: suspended nano-beams
Drain
Gate
Source
17
M/NEM-FET: device architectures
In-plane
Move body Move gate
Out-of-plane
Major advantages: new functionality and low power
18
M/NEM-FET abrupt switch
movable
gate
tgap = 220nm
Drain
A
A’
SuspendedGate
Experiment:
Out-of-plane movable gate
10-6
(A)
D
10-8
Pull-in
10-9
A
10-10
Gate up:
• high Vt
• in-series Cgap
-11
10-12
20μm
3
k t gapo
VPullin 
Pull-out
10
Source
Gate down
• low Vt
• Cox
10-7
Drain current, I
• Resonant-Gate FET (Nathanson, 1966)
• Suspended-Gate MOSFET (EPFL: A.M.
Ionescu, ISQED 2001, IEDM 2005, 2006)
• Nano-electro-mechanical FET (UC
Berkeley, T.J. King, IEDM 2005)
• Modeling of SG-FET (Stanford & EPFL:
Akarvardar: IEEE TED 2008, Tsamados: SSE
2008)
0
2
4
6
8
10
12
14
Gate voltage, VG(V)
Applications: power management, low power logic, memory
19
NEMS simulation & modeling
Electrostatic NEMS: mechanical & electrostatic analysis
Source: G. Li et al, Urbana-Champaign.
20
NEM-FET: scaling & simulation
• Multi-physics simulation for hybrid NEM device design
• Coupled FEA: 2D ANSYS-DESSIS for suspended-gate FET
Simulation: 90nm NEM-FET
Suspended
Gate
S
D
21
NEM-FET power management switch
NEM-FET vs. MOSFET
power management switch:
• dynamic VT
• Ioff, Isubthreshold : 102 -104
• sleep area ~ MOSFET
Replaced by NEM-FET
22
Hysteresis: 1T MEM-FET memory
-6
10
1.E-06
Drain
Source
ID (A)
Gate
1
-7
10
1.E-07
Mechanical
pull-in
-8
10
1.E-08
Mechanical
pull-out
-9
10
1.E-09
0
-10
10
1.E-10
0
5
10
15
20
VG (V)
• Electro-mechanical hysteresis: [ Vpull-in – Vpull-out ]
• SG-MOSFET capacitor-less memory feasible
• Hysteresis control! Scaling? Reliability?
23
Size & Voltage operation scaling (1)
• nanogap scaling (Samsung)
tbeam= 20nm
tgap = 20nm
NEM clamp switch with
TiN beam memory cell
array structure for high
density non-volatile
memory application
M.-S. Kim et al., ISDRS 2007
24
Size & Voltage operation scaling (2)
VD=1.2V
SG-FET compliant to ITRS 90nm node:
tox=2nm, L=65nm, channel doping Nch=3×1018cm−3, μ0=278cm2/Vs , airgap g0=5nm, W=400nm, h=10nm, Young modulus, E=170GPa.
25
NEM-FET inverter
• significant power
savings (1-2 decades
reduction) of inverter
peak current
• no leakage power
compared with nanometer scaled CMOS
inverter.
26
Movable/vibrating gate transistor
Laterally (in-plane) vibrating gate
• lateral MOS transistor, detection in drain current
• +4.3dB experimental gain demonstrated compared to capacitive
detection using same structure
LETI-CEA
• e-beam defines gaps (~47nm gap resol.).
• L=10mm, W=165nm, d=120nm
C. Durand et al., IEEE EDL 2008.
27
Double Gate switchable/vibrating body FET
Laterally (in-plane)
movable body:
first demonstration of
+30dB signal improvement
fres=2.4MHz, Q=6’000, Rm=200Ohm
EPFL
D. Grogg, A.M. Ionescu, DRC 2008.
28
Double Gate switchable/vibrating body FET
Experiment
D. Grogg, A.M. Ionescu,
ESSDERC 2008, Confidential
29
Conclusion
• NEM-FET: true hybrid mechanical-solid-state switch with
near-zero point subthreshold swing
• attractive for low-Ioff power management switches,
capacitor-less memory (D-RAM, S-RAM and NVM with
appropriate storage layers) and new analog/RF functionality
(in the resonant-gate configuration).
• fabrication: compatibility of surface micromachining with
CMOS processing.
• Voltage scaling below 1-2V: nanogap technology
• Size: ~as scalable as MOSFET (anchors needed)
• NEMFET does not use mechanical contacts in the path of
current flow: long-term reliability comparable to that of
capacitive RF MEMS switches (>109 cycles), being limited by
oxide charging.
30
MEMS/NEMS application roadmap
More transistors
than MEMS devices
Number of Transistors, NT
increasing power consumption
109
108
Ultra-low-power
NEM-based
Embedded NVM
NEM-FET Power Gating
and Embedded D/SRAM
107
106
CPU’s
OMM 32x32
105
104
Optical Switches
& Aligners
102
101
More MEMS devices
than transistors
Digital Micromirror
Device (DMD)
Adaptive
Optics
ADXL RF MEMS
103
Integrated
Actuator
Systems
NT/NM=1
Early MEMS
(sensing applications)
Ultra-low-power
NEM Relay
Logic
100
100
101
102
103
104
105
106
107
108
109
Number of Mechanical Components, NM
M/NEMS
Integration Levels
Enabled Applications
31
MEMS/NEMS application roadmap
• NEMS Beyond CMOS = low power nano-switch
More than Moore = new functionality
• Key role of NEMS for power savings and new
functionality: future hybrid NEMS-CMOS
• Future role of true hybrid NEM-FET devices:
abrupt switch, memory, resonator, sensing
• Challenges for hybrid NEM-FET:
 additional process control of nanoscale air-gap, thickness and
uniformity of suspended structures, control and uniformity of
mechanical properties
 fabrication: top-down & bottom-up (Si, CNTs)
 wafer-level packaging and reliability
 thermal drift
32
Acknowledgments
EPFL: FP7 IST projects MIMOSA,
MINAMI and NANO-RF
Stanford: DARPA, FCRP C2S2, NSF
Roger T. Howe, David Elata,
J Provine, Roozbeh Parsa,
Kyeongran Yoo, Soogine Chong
UC Berkeley: DARPA, FCRP C2S2
Hei Kam
33