Feng Yuan Ping (???)

1
First Principles Studies on High-k Oxides and
Their Interfaces with Silicon and Metal Gate

• Feng Yuan Ping (???)
• Department of Physics
• National University of Singapore
• phyfyp_at_nus.edu.sg

2
(No Transcript)
3
www.mrs.org.sg
4
Outline

• Introduction
• Oxygen vacancy in HfO2 and La2Hf2O7
• Tuning of metal work function at metal gate and
  high-k oxide interface
• Properties of high-k oxide and Si interface
• Conclusion

5
CMOS Scaling
ITRS roadmap shows the expected reduction in
device dimensions
6
Why High-k oxides ?
Gate
CB Si

• 1.2 nm (5 atomic layers) physical SiO2 in
  production of 90 nm logic technology node 0.8 nm
  physical SiO2 in research of transistors with 15
  nm physical Lg
• Gate leakage is increasing with reducing physical
  SiO2 thickness. SiO2 layers lt1.6 nm have high
  leakage current due to direct tunneling. Not
  insulating
• SiO2 running out of atoms for further scaling.
  Will eventually need high-K
  

Rober Chau, Intel
7
Choice of High K Oxide
8
Growth of ZrO2 on Si Interface
Wang et al. APL 78, 1604 (2001) Wang Ong, APL
80, 2541 (2002)
9
Problems with High K oxides
Among other problems, oxide has too many charge
traps, and the threshold voltage (Vth) shifts
from CMOS standards.
10
Dynamic Charge Trapping
Power law shift!
Oxygen vacancy?
Negative-U traps?
Time evolution of threshold voltage Vth under
static and dynamic stresses of different
frequencies, for (a) n-MOSFET, and (b) p-MOSFET.
The Vth evolution has a power law dependence on
stress time. C. Shen, H. Y.Yu, X. P. Wang, M. F.
Li, Y.-C. Yeo, D. S. H. Chan, K. L. Bera, and D.
L. Kwong, International Reliability Physics
Symposium Proceedings 2004, 601.
11
Hydrogen in HfO2
Formation energies for (a) interstitial H and H2
molecules, and (b) the VO-H complex. J. Kang et
al., APL, 84, 3894 (2004).
12
Bulk HfO2
P21/c Monoclinic
Fm3m Cubic
P42/nmc Tetragonal
J. Kang, E.-C. Lee and K. J. Chang, PRB, 68,
054106 (2003)
13
Cubic HfO2
Vasp Cutoff energy 495 eV GGA Eg 3.68 eV
(direct) (Exp gap 5.8 eV)
Valence band O 2p Conduction band Hf d
Peacock and Robertson, JAP (2002)
14
Computational Details

• DFT, planewave, pseudopotential method (vasp)
• 2s and 2p electrons of O, 5d and 6s electrons of
  Hf are treated as valence electrons.
• Cut off energy 495 eV
• 80 atom supercell (3x3x3 primitive cells)
• Uniform background charge for charged vacancy

15
Supercell
16
Total Energy
Charge State Energy (eV)
V-- 13.73
V- 7.02
V0 0.00
V -6.20
V -13.35
17
Energetics
Negative-U Property!
18
Charge Trapping Mechanism
Negative bias for p-MOSFET Holes are injected to
HK V0 ? V (meta-stable) ? V
Positive bias for n-MOSFET Electrons are injected
to HK V0 ? V- (meta-stable) ? V--
In both cases, when the gate bias is removed, no
charges are injected to HK, all charges in the O
traps will be de-trapped, the gate dielectric
remains neutral
19
Frequency Dependence of Vth
Experimental and simulation results for n-MOSFET
20
Formation Energy
A. S. Foster, et al. PRB 65, 174117
(2002) Formation energy for neutral vacancy 9.36
eV (O3) 9.34 eV (O4) Present calculation 9.33
eV (relative to O atom)
21
Band Structures
V0
22
Band Structures
AC plane
V-2
23
Relaxation of NN Hf atoms
?2
?1
V
V
?
(b)
(a)
C2v Mode
Breathing Mode
24
Relaxation of NN Hf Atoms
Charge State Breathing Mode C2v Mode C2v Mode
Charge State ? (Å) ?1 (Å) ?2 (Å)
V-- 0.14 0.11 -0.006
V- 0.07 0.06 0.002
V0 0.03 ? ?
V -0.08 ? ?
V -0.16 ? ?
25
Effect of Lanthanum
X. P. Wang et al. VLSI2006
Charge trapping induced Vth shift under constant
voltage stress for HfO2, HfLaO with 15 and 50
La gate dielectric NMOSFETs.
26
Effect of La
The formation energies of oxygen vacancies at
varies sites in monoclinic HfO2 and pyrochlore
HfLaO, calculated by ab initio total energy
calculations.
27
Summary

• Oxygen vacancy in HfO2 has negative-U property.
  It is energetically favors trapping two electrons
  or two holes.
• Oxygen vacancy is a main source of charge
  trapping in HfO2 and the origin for frequency
  dependence of dynamic charge trapping in HfO2 MOS
  transistors.
• Large lattice relaxation for charged vacancies,
  due to strong electron-lattice interaction.
• Oxygen vacancy has higher formation energy at Td
  site in La2Hf2O7.

28
Gate Material

• Currently polycrystalline silicon (poly-Si) gate
  electrode is used.
• Problems
• high gate resistance
• boron penetration
• Fermi level pinning
• poor compatibility with high-? gate dielectrics
• increase of EOT due to gate depletion
• Need metal gate!
• Eliminates the gate depletion problem
• Eliminates boron penetration problem
• Reduces the gate sheet resistance
• Generally more compatible with alternative gate
  dielectric or high-permittivity (high-k) gate
  dielectric materials than poly-Si.
• The urgent need for alternative gate dielectrics
  to suppress excessive transistor gate leakage and
  power consumption could speed up the introduction
  of metal gates in complementary metal oxide
  semiconductor (CMOS) transistors.

29
Issues

• The integration of metal gate with high-? gate
  dielectric requires the metal effective work
  functions to be within 0.1 eV of the Si valence-
  and conduction-band edges for positive- (PMOS)
  and negative-channel metal-oxide-semiconductor
  (NMOS) devices, respectively.
• However, to find two metals with suitable work
  functions and to integrate them with current
  semiconductor technology remains a challenge.

30
Work Function of Metals
Work function of several elemental metals in
vacuum, on a scale ranging from the positions of
the conduction band to the valence band of
silicon.
Metal work functions are generally dependent on
the crystal orientation and on the underlying
gate dielectric.
31

• Can we tune the metal workfunction?

32
Tuning of Workfunction?
Ni-m-ZrO2
Ni
ZrO2
m Au, Pt, Ni, Ru, Mo, Al, V, Zr and W (for half
monolayer) m Ni, V, and Al (for one monolayer)
33
Bulk ZrO2
Very small lattice mismatch (lt2)
34
Models

• Supercells for the Ni-m-ZrO2 interfaces,
• The interface is formed using c-ZrO2(001) and fcc
  Ni(001) surfaces.

1. with one monolayer metal m (mNi, V, and Al).
2. with half monolayer metal m (mAu, Pt, Ni, Ru,
   Mo, Al, V, Zr and W)

35
Computational Details

• DFT, planewave, pseudopotential method (vasp)
• Ultrasoft pseudopotential GGA
• Cut off energy 350 eV
• K points 8x8x1
• In plane lattice constants constrained to that of
  c-ZrO2
• Electronic energy was minimized using a fairly
  robust mixture of the blocked Davidson and
  RMM-DIIS algorithm. Conjugate gradient method for
  ionic relaxation

36
Density of States
Spin resolved and atomic site-projected density
of states (PDOS) for (a) Ni-Pt-ZrO2 interface and
(b) Ni-Al-ZrO2 interface, with half monolayer of
metal insertion. The PDOS for the Ni in the bulk
region (Ni-bulk), interface metal m (Pt or Al),
interface oxygen (O-Int.), and oxygen in the bulk
region (O-bulk) are shown.
37
Schottky Barrier Heights
38
p-type Schottky Barrier Height

• p-type SBH is obtained using the bulk plus
  lineup procedure, using the average
  electrostatic potential at the core (Vcore) of
  ions in the bulk region as reference energy
  DEb the difference between the Fermi energy of
  Ni and the energy of the valence band maximum
  (VBM) of the oxide, each measured relative to
  Vcore of the corresponding bulk ions, DV is the
  lineup of Vcore through the interface.
• DEb is adjusted by quasiparticle and spin-orbital
  corrections (0.29 eV for Ni, 1.23 eV to the
  valence-band maximum of ZrO2, ? overall
  correction of 0.94 eV).

39
Vcore
Average electrostatic potential at the cores
(Vcore) of Ni (filled dark circle) and Zr (open
circle) as a function of the distance from the
interface for Ni-m-ZrO2 interfaces (m Au, Ru,
Ti) with half monolayer metal insertion. Breaks
were introduced in the vertical axis (Vcore)
between - 41 eV and -36 eV.
40
n-type Schottky Barrier Height

• where Eg is the energy gap of the dielectric
• The experimental band gap of 5.80 eV was used.
• The SBH can also be estimated directly from the
  difference between the Fermi energy and the
  energy corresponding to the top of the valence
  band given in the PDOS of oxygen in the bulk
  region. Results obtained using the two methods
  are in good agreement (within 0.1 0.2 eV).

41
Results
m ? ? WF Qm p-SBH n-SBH
Au Pt Ni Ru Mo Al V Zr Ti W Ni V Al 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 1 1 1 5.77 5.6 4.40 4.5 3.9 3.23 3.6 3.64 3.45 4.40 4.40 3.6 3.23 5.1 5.65 5.15 4.71 4.6 4.28 4.3 4.05 4.33 4.55 5.15 4.3 4.28 0.16 0.16 0.37 0.27 0.51 1.06 0.69 1.01 0.80 0.15 0.24 0.44 0.63 1.20 1.98 3.06 3.06 3.44 3.64 3.73 3.86 3.87 4.02 2.19 3.17 4.00 4.60 3.82 2.74 2.74 2.36 2.16 2.07 1.94 1.93 1.78 3.61 2.63 1.80
42
SBH Tunability
Range of tuning 2.8 eV!
43
n-type Schottky Barrier Height
n-SBHs of Ni-m-ZrO2 interfaces are shown as a
function of electronegativity (Mulliken scale) of
m. The straight line is a least-squares fit to
data points shown in filled squares (Al and W
were not included).
44
Workfunction of Ni(001) with m
Work functions of Ni(001) with half monolayer of
metal m coverage are shown as a function of
electronegativity (Mulliken scale) of m. The
straight line is a least-squares fit to data
points shown in filled squares.
45
Mechanism?

• Contribution from the tails of the metallic wave
  functions which tunnel into the oxide band gaps
  or metal induced gap sates can be ruled out, due
  to short delay length (0.9Å) which is nearly
  independent of the interlayer metal.
• Interface dipole can contribute significantly to
  band alignment between the metal and oxide.
• Ionic m-O bonds
• Charged metal layer and its image

46
Gap States
Penetration of electronic density of the gap
states into the ZrO2 of Ni-m-ZrO2 interfaces.
Position of the surface oxygen is set to z 0 Å.
47
Interface bonding dependent SBH experimental
evidence (in-situ XPS)
Afanas'ev et al. JAP 91, 3079 (2002).
48
Interface bonding dependent SBH experimental
evidence (in-situ XPS)
49
Summary

• A scheme for tuning the Schottky barrier height
  or workfunction of metal gate high-k dielectric
  interface was proposed and has been
  experimentally confirmed.
• By including a monolayer or half monolayer of
  transition metal between the metal gate and
  high-k dielectric, a tunability as wide as 2.8 eV
  can be achieved.
• There exists a linear correlationship between the
  Schottky barrier heights / workfunction and the
  electronegativity
• Preliminary experimental results with mAl agree
  with prediction.

50
Acknowledgement

• Y F Dong Physics Department, NUS
• Y Y Sun Physics Department, NUS
• S J Wang Institute of Materials Research
  Engineering
• A Huan Institute of Materials Research
  Engineering
• M F Li Dept of Electrical Computer
  Engineering, NUS Institute of Microelectronics

51

• Thank you!