1 Carbon-Based Electronics: Will there be a carbon age to follow the silicon age? Jeffrey Bokor EECS Department UC Berkeley [email protected] Solid State Seminar 9-13-13
Apr 20, 2020
1
Carbon-Based Electronics:Will there be a carbon age to follow the silicon age?
Jeffrey BokorEECS Department
Solid State Seminar
9-13-13
2
Outline
• Review of development of Carbon Nanotube (CNT) transistors (for logic)– Issues, progress, prospects
• Advent of graphene– Recognition of promise of graphene nanoribbons (GNRs) for
logic transistors– Issues, progress, prospects
• Summary of prospects for carbon transistors
3
C60: Birth of carbon Nanotech era
4
Main properties of carbon nanotubes predicted before discovery!
semiconductor
metal
Applied Physics Letters
5
Single-wall carbon nanotubes discovered in carbon ‘soot’ by TEM
Iijima and Ichihashi, Nature (1993) [NEC]
6
CNT Transistor
Tans, et al., Nature (1998) [Dekker group, Delft]
Laser vaporization method for CNT synthesis
7
Catalytic CVD growth of CNTs on a surface
Kong, et al, Nature (1998) [Dai group, Stanford]
Catalyst: Fe(NO3)3⋅9H2O/alumina/methanol suspensionCVD at 1000C with methane
8
-25
-20
-15
-10
-5
I DS (
A)
-0.4 -0.3 -0.2 -0.1 0.0VDS (V)
0.2 V
-0.1 V
-0.4 V
-0.7 V
-1.0 V
-1.3 V
Self-Aligned Ballistic FETs w/High-k
0.5
10-9
10-8
10-7
10-6
10-5
-I DS
(A)
-1.5 -1.0 -0.5 0.0VG (V)
VDS = -0.1,-0.2,-0.3 Vd~1.7 nmL ~ 50 nm
• Pd zero-barrier height contact• > 5 mA/um at Vg = VDS =0.4V
Javey, et al, Nano Lett. (2004) [Dai group, Stanford]
9
High Performance p- and n-FETS
Javey, et al, Nano Lett. (2005) [Dai group, Stanford]
• Doping by adsorption• Lg = 80nm
10
CNT-CMOS Integration Chip
NMOS binary tree 11-bit decoder 2048 back-gated CNT transistors >4000 Si NMOS transistors, 1 m Microlab
baseline process
Tseng, et al, Nano Lett. (2004) [UCB/Stanford, Bokor/Dai groups]
11
Carbon Nanotube + Silicon MOS Integrated Circuit
-15 -10 -5 0 5 10 15
10-8
10-7
10-6
1x10-5
Imin
Ion
Id (A
)
Vgs (V)
0 1 2 3 4 5 6 70
50
100
150
209 Devices
Total: 523 devices
Semiconductingnanotubes only
coun
ts
Log (on/off)
Tseng, et al, Nano Lett. (2004) [UCB/Stanford, Bokor/Dai groups]
12
Direct correlation to diameter variation
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
10-6
1x10-5
Ion
(A)
Diameter (nm)
Ion, Vgs-Vt=-7v
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.010-12
10-11
1x10-10
1x10-9
1x10-8
1x10-7
1x10-6
Measurement Limit
Imin
(A)
Diameter (nm)
1 tube per device
-15 -10 -5 0 5 10 1510-11
1x10-101x10-91x10-81x10-71x10-61x10-5
Id (A
)
Vgs (V)
d=2.9nm d=2.2nm d=1.1nm
Tseng, et al, Nano Lett. (2006) [UCB, Bokor group]
13
Parallel Tube CNTs
• To get large drive, need to stack multiple tubes in parallel with common contacts, gate
• Important for ultimate circuit application
• Do parallel array currents add?• How close can tubes be stacked?
14
10-10
10-8
10-6
10-4
-I DS
(A)
-1 0 1VG (V)
Parallel Array of Self-Aligned Ballistic FETs
-200
-150
-100
-50
I DS
(A
)
-0.6 -0.4 -0.2 0.0VDS (V)
VGS = -0.9 to 0.3 V in 0.2 V steps
D
S
G
D
G
G
G D
S
S
SWNTS
VDS = -0.1,-0.2,-0.3 V
• 1st demonstration of a parallel array• ~200 uA of current for the array of 8 tubes.
Javey, et al.,Nano Lett. (2004)[Stanford, Dai group]
15
CNT Array Density Limited by Screening
Wang, et al. SISPAD (2003) [IBM]
16
CNT Array Transistor Circuit Performance
Jie, et al., ISSSC (2007)[Stanford/USC, Wong/Mitra/Zhou groups]
17
Vision for CNT channel array MOSFETs
• Array of 1D channels, densely packed
• Density 200-250 per m• No metallic tubes• Narrow diameter
distribution
Bulk Dielectric k2
HfO2
LgcLg
pitch
CNTsor
SNWs
Gate Dielectric k1
MetalGate
Substrate
S D
Bulk Dielectric k2
HfO2
LgcLg
pitch
CNTsor
SNWs
Gate Dielectric k1
MetalGate
Substrate
S D
18
A Multiple Growth Strategy to High Densities
Hong, et al, Adv. Mat. (2010) [UIUC, Rogers group]
• Single-crystal quartz growth substrate• “Epitaxial” CNT growth• Layer transfer to Si wafer
19
Density Scaling by Multiple Transfers
as grown: 15/um transferred: 15/um
2X transfer: 29/um 4X transfer: 55/umWang, et al., Nano Res. (2010)[USC, Chou group]
Removal of m-tubes by ‘breakdown’
20
Coat with small molecule film
Induce Joule heating selectively in m‐SWNTsto form trenches by thermocapillarity
Si (p++)
SiO2
S M S M S S
Si (p++)
M MSiO2
O2 plasma etch exposed m‐SWNTs
Remove film and electrodes; build circuits on remaining s‐SWNTs
S S
Si (p++)
SiO2
S S
Si (p++)
SiO2
Selective Removal of m-tubess From Aligned Arrays
J. Rogers group, UIUC
21
Dynamics of Thermocapillary Flow
t= 0 0.1 1 10 100 300 s
2 m
Joule Heating by a SWNT (∆T~5-15C)
Jin, et al., Nat. Nano. (2013) [UIUC, Rogers group]
Heating options:• Gated electrical Joule heating• Selective laser absorption• Selective microwave absorption
22
Solution phase nanotube ‘sorting’/purification
Arnold, et al., Nat. Nano. (2006)[Northwestern, Hersam group]
“Density gradient” centrifugation
23
Electrical results on sorted CNTs
Arnold, et al., Nat. Nano. (2006)[Northwestern, Hersam group]
Percolating network transistor Sorted tube transistor high on/off ratio
24
DNA sequence specific wrapping for sorting
Tu, et al., Nature (2009)[Dupont, Zheng group]
“size exclusion” chromatography
25
Purified Single Chirality (10,5) SWNTs
200nm
(10,5)
StartingHiPco material
Separated SWNTs
(10,5)
Zhang, et al, JACS (2009), [Stanford/Dupont, Dai/Zheng groups]
DNA used: (TTTA)3T
26
FETS with 99% Semiconducting Tubes
10-1410-1310-1210-1110-1010-910-810-710-610-5
-Ids
(A)
-2 -1 0 1 2Vgs(V)
1000mV 500mV 100mV 10mV
100nm
S
D< 2 2 -4 4 -6 > 6
0
1 0
2 0
3 0
4 0
Perc
enta
ge(%
)
lo g (Io n /Io ff)
Mixed
Purelysemiconducting
Average 15 tubes per deviceIon/Ioff >102 : 88%semiconducting tubes: 99% (0.9915 ~ 88%)
Mostly (10,5) SWNTs
Zhang, et al, JACS (2009), [Stanford/Dupont, Dai/Zheng groups]
27
Solution phase array assembly by Langmuir-Blodgett technique
Li, et al. JACS (2007)[Stanford, Dai group]
~80/um~70/um
28
Solution processed CNTs are as good as CVD tubes at nanoscale Lg
Choi, et al., ACS Nano (2013)[UCB, Bokor/Javey groups]
Franklin and Chen,Nat. Nano. (2010)[IBM
CVD tubes
29
Ultimate scaling study
GAA NW
SG AGNR
GAA CNTDG UTB
Also DG AGNR
M. Luisier (Purdue)
30
Device Characteristics:- All: Lg=5nm, VDD=0.5 V, EOT=0.64nm (3.3nm of HfO2 with εR=20)- SG and DG AGNR: width=2.2nm, normalization by width- GAA CNT: diameter=1.58, 1.0, and 0.6 nm, normalization by diameter - GAA and -NW: Si, diameter=3nm, transport=<110>, 1% uniaxial strain- DG UTB: Si, body=3nm,, transport=<110>, 1% uniaxial strain
Simulation Approach:- Same quantum transport simulator for all devices based on Non-equilibrium
Green’s Functions (NEGF) formalism with atomistic resolution of simulation domain and finite element method for Poisson equation
- Bandstructure model: single-pz for carbon and sp3d5s* for silicon (tight-binding)- Ballistic limit of transport (no electron-phonon scattering nor interface
roughness taken into account)- Intrinsic device performances (no contact series resistances included)- No gate leakage currents included- No structure optimization for any of the selected devices
Simulation parameters and assumptions
31
Id-Vgs at Vds=0.5V in carbon-based Devices
SiO2HfO2
AGNR width: 2.2nm / CNT diameter: 1.58nm / Band Gap Eg=0.56 eV
• Same EOT gives very different electrostatic gate-channel coupling
EOT=0.64nm EOT=0.64nm
M. Luisier (Purdue)
32
Gate Dielectric Effect In Carbon-Based Devices
Comparison of Conduction Band Edge and Spectral Current in Single-Gate AGNR with 0.64nm SiO2 (εR=3.9) and 3.3nm HfO2 (εR=20) => same EOT=0.64nm
OFF- stateSiO2 HfO2
• Effective channel length is longer for the thicker HfO2• Barrier widens and tunneling current drops
33
Extreme (sub-10 nm S-D Tunneling regime) d=1.58 nm CNT FETs
Transfer Characteristics Sub-threshold swing
Vds=0.5V
• Bandgap 0.56 eV GAA- CNT (d=1.58 nm) scales poorly
5nm≤Lg≤12nm
Vds=0.5V
d=1.58 nm
3.3nm HfO2 EOT=0.64nm
M. Luisier, et al., IEDM (2011)[Purdue/MIT/UCB, Lundstrom/Antoniadis/Bokor groups]
34
Transfer Characteristics Sub-threshold Slope
Gate-length trend for 1 nm CNTs
• Bandgap 0.8 eV GAA- CNT (d=1.0 nm) scales better
M. Luisier, et al., IEDM (2011)[Purdue/MIT/UCB, Lundstrom/Antoniadis/Bokor groups]
35
Transfer Characteristics
Id-Vgs at Vds=0.5V in CNT FETs with d=0.6nm and 5≤Lg≤12 nm
Gate-length trend for 0.6 nm CNTs
Sub-threshold Slope
• Bandgap 1.4 eV GAA- CNT (d=0.6 nm) scales well
M. Luisier, et al., IEDM (2011)[Purdue/MIT/UCB, Lundstrom/Antoniadis/Bokor groups]
36
• CNT with d=0.6nm and NW with d=3nm have same band gap Eg=1.4eV
• CNT with d=1.0nm has band gap Eg=0.817eV
Comparison of different channel materials
Id-Vgs at Vds=0.5V in CNT, NW, and UTB Devices3.3nm HfO2, EOT=0.64nm
• Bandgap 0.8 eV GAA-CNT (d=1.0 nm) scales poorly• Bandgap 1.4 eV GAA-CNT (d=0.6 nm) scales well• Si NW (d=3 nm) scales very well due to high-mass and band-gap
37
9 nm CNT transistor
2012
(5 nm)
(20 nm)
38
Monolithic 3D CNT Circuits!
Hai, et al., IEDM (2010)[Stanford, Mitra/Wong groups]
39
Graphene
Forms of graphene Graphene resistivity
Geim and Novoselov, Nat. Mat. (2007) [Manchester]
40
Bandgap Prediction for Graphene Nanoribbons
Son, et al., PRL (2006)[UCB, Louie group]
41
Bandgap Measurements of Etched GNRs
Han, et al., PRL (2007) [Columbia, Kim group]
42
GNR formation by unzipping CNTs
Jiao, Nat. Nano. (2010) [Stanford, Dai group]
43
0.5
0.4
0.3
0.2
0.1
0.0E
g(eV
)50403020100
W (nm)
100
101
102
103
104
105
106
107
I on/
I off
50403020100W (nm)
)/exp( /II offon TkE Bg
nmWeVEg
8.0
Li, et al. Science (2008)[Dai group]
All sub-10nm GNRs are semiconductingIon currents few uA
GNR Bandgap vs. width
44
Bottom-up Synthesized GNRs
Cai et al. Nature (2010) [EMPA (Switzerland), Fasel group]
• Atomically perfect edges!
• 7 C atoms wide• W = 0.74 nm!
45
Aligned Growth – Bandstructure Measured
Ruffieux, et al. ACS Nano (2012) [Fasel group]
m* = 0.21 meEg = 2.3 eV
~2 nm pitch!
46
Synthesized GNR Transferred to SiO2
Bennett, et al., unpublished [UCB, Bokor/Crommie/Fischer groups]
47
Synthesized GNR Transistor Results
Bennett, et al., unpublished [UCB, Bokor/Crommie/Fischer groups]
48
Wider GNRs Synthesized with 1.4 eV Gap
Chen, et al., ACS Nano (2013)[UCB, Fischer/Crommie groups]
49
Single-Molecule Heterostructures
50
Summary/Outlook
• CNT and GNR both promising candidates for CMOS channel material for 8 nm gate length
• Why?• High drive at low V• Good scalability• 3D layer stacking:
10 layers = 3 nodes on roadmap!
• More work needed:• Purified chirality for tubes• Longer, wider GNRs• Dense aligned arrays• Low resistance contacts
• GSR opportunities in my group