NCN 1 Neophytos Neophytou Advisory Committee Chairs: Mark Lundstrom, Gerhard Klimeck Members: Ashraful Alam, Ahmed Sameh Network for Computational

1

NCN NCN

www.nanoHUB.orgwww.nanoHUB.org

Neophytos Neophytou

Advisory Committee Chairs: Mark Lundstrom, Gerhard Klimeck Members: Ashraful Alam, Ahmed Sameh

Network for Computational Nanotechnology Purdue UniversityWest Lafayette, Indiana USA

Quantum and atomistic effects in nano-electronic devices

Ph.D. Thesis Defense, May 22nd, 2008

2


NCN NCN

Introduction – Device trend

Robert Chau (Intel), 2004Robert Chau (Intel), 2004

Device Challenges:1) Atoms are countable2) Strain3) Material /potential variations on nanoscale4) Crystal orientation5) III-V, Ge, InGaAs

Electronic structure features:1) Strong quantization2) Band coupling3) Non-parabolicities4) Quantum mechanics

Design Challenges1) Low dimensionality2) Parameter fluctuations3) Scalable – last for 2 generations

3


NCN NCN

Si MOSFET alternatives? CNTs

Issues: Chirality Metallic vs. Semiconducting Alignment

CNT (Delft group-1998)

Top gate+High-k(Javey et. al.) BTBT(IBM) Oscillator (IBM)

4


NCN NCN

Si MOSFET alternatives? NWs

Singapore group (IEDM 2006)

Samuelson groupEDL 2006

Still based on Si, so might have easier integration Gate all around for better electrostatics

Scattering in 1D – surface roughness? Bandstructure effects?

Samsung (APL 2008)D=8nmL=22nm

5


NCN NCN

Si MOSFET alternatives? III-Vs

Kim et. al. 200660nm InGaAs

Kim et. al. 200740nm InAs

Freescale IEDM 2007

Intel EDL 2008

High mobility, high speed, close to the ballistic limit, but low DOS (DOS Bottleneck) Lower VD, low dissipation - tunneling/leakage? Large series resistance

6


NCN NCN

Open questions

How will these devices perform at the scaling limit? What parameters control their performance? Low dimensionality: Sensitivity to defects and fluctuations?

Are they advantageous to the Si MOSFET?

Modeling tools used here: Quantum transport – NEGF (CNT, III-V HEMT) 3D (CNTs) Atomistic bandstructure (NWs)

7


NCN NCN

Why atomistic is needed –motivation

Valley splitting Band coupling

Valence band anisotropy Warped bands

m* valid m* NOT valid

8


NCN NCN

Motivation for TB

NN sp3d5s*-SO

The bulk bandstructure(from Anisur Rahman’s thesis)

[100] [110] [111]

Based on Localized Atomic OrbitalsSuitable for: Structure deformations, strain Material variations, heterostructures Surface truncation Potential variations: treated easily

Needs a large set of fitting parameters Computationally expensive

9


NCN NCN

Outline

1) 1D channel sensitivity to atomistic defects

2) Bandstructure effects in nanowires:

Self consistent model for NWs

Electron transport

Hole transport

3) III-V HEMT devices

4) Conclusions – Future work

10


NCN NCN

Defects in 1D channels

vacancy

Neophytou APL 2006, APL 2007, JCE 2007

Gate

Gate

Insulator: 4nm HfO2 (k=16)S D

22.5 nmDoping: ND = 109 /m


25 nmIntrinsic

Gate

Gate

Insulator: 4nm HfO2 (k=16)S D



25 nmIntrinsic

1) NEGF2) 3D electrostatics3) Atomistic TB

CNTFET on nanoHUB.org

ID: ~27% reduction

Dangling bonds in NWs

CB

A

CB

A

CB

A

CB

A

CB

A

CB

A

CB

A

CB

A

ID: ~30% reduction

charged impurity

11


NCN NCN

Outline




Electron transport

Hole transport



12


NCN NCN

The self-consistent model for NWs

Bandstructure

(states)

+k-k

EF1 - qVDS

Uscf

EF1

E(k)

EC(x)Semiclassical

Ballistic Transport

+k-k

EF1 - qVDS

Uscf

EF1

E(k)

EC(x)Semiclassical

Ballistic Transport

Transport(state filling - charge)

oxide

gate

CHANNEL

oxide

gate

CHANNEL

Poisson

Charge Potential

Simple model but provides physical insight

13


NCN NCN

Why need a SC model? ~0.5nm

Neophytou SISPAD 2007

Charge Variations

CS

Ec changes

Ev changes

14


NCN NCN

Numerical issues of the SC model

Hamiltonian size – (dep. on wire size and orientation)3nm x 3nm with SO: 4k x 4k - 9k x 9k12m x 12nm with SO: 55k x 55k

eigenvalue problem150 k-points, 60 eigenvalues

Parallelized per bias point: Vd is constant Vg is varied

Timing (per bias point):3nm device: a few hours 12nm devices: 1 – 2 days

15


NCN NCN

Outline




Electron transport

Hole transport



16


NCN NCN

NMOS [100], [110], [111] wire comparison

Cox

CS

VG

ψs

GND

Same capacitance/ charge in all wires [110], [100], then [111] on performance

OX Stot

OX S

C CC

C C

Neophytou TED 2008

CS: - CQ (30%)-potential/charge variations

17


NCN NCN

Masses change with quantization

C

B’

AA’

C’

B

C

B’

AA’

C’

B

mlmtmtC

mtmlmtB

mtmtmlA

mz *my *mx *

mlmtmtC

mtmlmtB

mtmtmlA

mz *my *mx *

zx

y

NW mass is controlled by

quantization of the 6 ellipsoids

[100], [111] wire masses increase [110] mass decreases

Neophytou TED 2008

18


NCN NCN

Non-parabolicity and anisotropy in the dispersion

kx

Neophytou TED 2008

[010]

[110]

19


NCN NCN

Outline




Electron transport

Hole transport



20


NCN NCN

Ek for holes in 6nm wires

Corner effects – electrostatics Directionality in the charge - bandstructure

High gate bias

Neophytou TED 2008

(100

)(1

-10)

(010)

(11-

2)

(1-10)

(001)

Energy surfaces

[100]

[110]

[111]

21


NCN NCN

Anisotropy implications on the device performance

Kobayashi et. al.JAP 103, 2008

[110] side variations, do not affect the device – VT, Ion [100] side variations, affect the device

22


NCN NCN

Outline




Electron transport

Hole transport



23


NCN NCN

Motivation: del Alamo group HEMTs

Source Drain

InGaAs/InAlAs

InAlAs(11 nm)

InGaAs(MQW)

InAlAsBuffer

InP(6 nm)

tins

Lside

Typical IDS vs. VDS

Reference: Dae-Hyun Kim et al. IEDM 2006

• how close to ballistic limit? • role of mobility• degradation of gm at high VG

Typical Gm vs. VGS

24


NCN NCN

Approach

Simulation:

• 2D Poisson in the cross section • NEGF in the channel and upper buffer layer (ballistic)• Include Rs to fit low VDS conductance to experiment• Bulk material masses (In0.7Ga0.3As)• Adjust ΦB to achieve the experimental VT

• Parallelization: One VG set per CPU (constant VD) 2 hours per bias point – 20 hours per I-V

δ-dop. n++

Gate

L_side2.1e12/cm2 10e12/cm2

3nm

sou

rce

dra

in

L_side2.1e12/cm210e12/cm2

n+

60nm40nm 60nm

15nmInGaAs

InAlAs

500nmInAlaAs

n+ n++δ-dop. n++

Gate

L_side2.1e12/cm2 10e12/cm2

3nm

sou

rce

dra

in

L_side2.1e12/cm210e12/cm2

n+

60nm40nm 60nm

15nmInGaAs

InAlAs

500nmInAlaAs

n+ n++

25


NCN NCN

Series resistance and “Ballistic” mobility

2

170 - m( )

~ 170 - 450 cm /V-s

DSCH

DS B ins G T

B

V LR

I W C V V

(Depending on the Tins)

ballistic simulation

measured(Tins = 3 nm, L = 60 nm)

ballistic + Rs = 400 -m

“ballistic mobility:”

26


NCN NCN

LG = 60 nm vs. Tins

Tins=3nm Tins=11nmTins=7nm

1) Except for high VG, all results can be explained as a ballistic FET with series R

2) Series resistance increases as Tins decreases

27


NCN NCN

Source limits

2) Barrier collapses

Gm rolls off in the ballistic model too.

1) OFF state

3) Gate loses control

28


NCN NCN

Charge

CS degrades Cins by 2.5 x

Q=Cins(VG-VT)

29


NCN NCN

Velocity

1) Non-parabolicity degrades the velocity by ~10%

Velocity is low:

Due to quantum mechanical reflections and tunneling

v ~ 2.7

v~ 4

v~ 3.6

30


NCN NCN

Outline



Self consistent model

Electron transport

Hole transport



31


NCN NCN

Conclusions (1)

1) 1D channel are sensitivity to single atomistic defects:

Vacancy, charged impurities, dangling bonds

2) Transport in nanowires:

Non-parabolicity, anisotropy causes mass variations, charge distribution variations

EMA cannot be used in general

NMOS: [110], [100] perform better, [111] worse

PMOS: [111], [110] perform better, [100] worse

32


NCN NCN

Conclusions (2)

3) III-V HEMT devices:

Ballistic channel + RSD

Low charge and velocity, low “apparent” mobility

Importance of the source design

33


NCN NCN

General concluding comments

Low dimensional devices:

Importance of CS, CQ, that degrade COX

But variations in parameters that influence C do not affect the device

Velocity is important

But, low mass tunnels more so the velocity can be reduced

Device is mostly controlled by external parameters rather than the channel (RSD, parasitics)

34


NCN NCN

Future work

Identifying the ultimate MOSFET: Perform appropriate comparisons between Si MOSFETs and UTB, NW devices at the scaling limit. Power, speed, Ion, Ion/Ioff, leakage, parasitics Device to circuit level Optimal strain and wafer/transport orientation, material, Is it different at each technology node? Which device for which application – identify appropriate use

Modeling for nanoscale devices: Contacts in low dimensional devices: DD + Ballistic NEGF, dephasing mechanisms (equilibrium or not?) Schottky barriers Inexpensive treatment and use of complex bandstructures, and distortions. (zone unfolding). NEGF + TB – real or mode space for PMOS in NWs and UTB

35


NCN NCN

AcknowledgementsProf. Mark LundstromProf. Gerhard KlimeckProf. Ashraful AlamProf. Ahmed Sameh

Jin Guo, Siyu Koswatta, M.P. Anantram (CNT)Diego Kienle, Eric Polizzi, Shaikh Ahmed (CNTFET)Anisur Rahman, Jing Wang, Mathieu Luisier (Bandstructure)Abhijeet Paul (generalized poisson for SC model)Titash Rakshit (HEMT)

Yang Liu (UTB work)Gengchiau Liang, Dmitri Nikonov (Graphene)

All others in EE350

NCN for the computational resourcesCheryl Haines

36


NCN NCN

BACKUP

Neophytou APL 2006

37


NCN NCN

Explanations for the Ek - transport

[100] subbands [110] subbands

Quantization surfaces-Structural

-Electrostatic

Neophytou TED 2008

38


NCN NCN

Anisotropy implications on the device performance

Neophytou, in preparation

[100]

[1-10]

(i)

(ii)

(iii)

(vii)

(vi)

(v)

(iv)

(110)

(100)

BA

39


NCN NCN

Figure 1 – The different quantizations of the different surfaces

(110) surface(100) surface

40


NCN NCN

Figure 2 – Current surface for variation of the dimensions

[100]

[1-10]

(i)

(ii)

(iii)

(vii)

(vi)

(v)

(iv)

(110)

(100)

BA

41


NCN NCN

Tins / gate length dependence

- Low DOS degrades charge

- At high VG the measured current deviates from the ballistic limit.

- As LG decreases, ID approaches the ballistic limit

42


NCN NCN

3nm wire dispersions in different orientations

3nm-[100] 3nm-[110] 3nm-[111]

Mass at Γ: 0.27 (0.19) Degeneracy : 4 Excited states shift down

Mass at Γ: 0.16 (0.19) Degeneracy : 2 Valley Splitting

Mass: 0.46 (0.43) Degeneracy: 6

OFF-Γ

43


NCN NCN

Approach - Bandstructure

In0.7Ga0.3As – undistortedm*=0.048m0 (matches DOS up to 0.2eV) (account for non-parabolicity)L valleys are too high

44


NCN NCN

Transconductance degradation

1) Cannot be explained by series resistance

2) Possibly scattering at high VG (not loss of confinement or upper valleys

3) Is there a ballistic mechanism that can

explain this?

Documents

NCN 1 Neophytos Neophytou Advisory Committee Chairs: Mark Lundstrom, Gerhard Klimeck Members: Ashraful Alam, Ahmed Sameh Network for Computational