You are on page 1of 28

Analytical modeling of nano MOSFETs in the quasi ballistic regime : beyond the drift diffusion approximation

R. Clerc, G. Ghibaudo
IMEP-LAHC Grenoble Institute of Technology France

P. Palestri, L. Selmi
DIEGM University of Udine Italy

1 MOS-AK Workshop 2010

Transport in Nano MOSFET (L < 20 nm) challenges Compact Modeling
Since the pioneering work of Natori (1994) and Lundstrom (1996), the quasi ballistic regime of transport in nano MOSFETs has been extensively investigated. However, existing compact model are still based on Drift Diffusion and saturation velocity concepts « Sophisticated computer simulations using techniques such as full band Monte Carlo and full quantum transport approaches are being used to explore the physics of the ultimate MOSFET, but circuit models continue to be based on concepts and approaches developed in the 1960's. »
M. Lundstrom, Int. SOI Conference, 2006

Let us examine the applicability of Quasi Ballistic theories to compact modeling

2 MOS-AK Workshop 2010

Outline The “orthodox” Lundstrom theory Beyond Lundstrom’s theory Ballisticity extraction from experiments 3 MOS-AK Workshop 2010 .

The “orthodox” Lundstrom theory 4 MOS-AK Workshop 2010 .

Concept of Ballistic Limit Energy Flux of carrier Source (In quasi equilibrium) Gate Drain (In quasi equilibrium) Transistor in diffusive regime Scattering I d ( y) = W Q i µ eff d E Fn ( y) dy Virtual Source Vd Channel L Energy Flux of carrier Source (In quasi equilibrium) Gate Drain (In quasi equilibrium) Transistor in ballistic regime Net current I d BAL ≈ W Q i Vinj Performance are no longer limited by transport along the channel. but by injection at the source end Virtual Source Vd Channel L 5 MOS-AK Workshop 2010 .

Rev. P. vol 123 pp. P. Rahman and M. L. Lundstrom. TED. Assuming an Inversion Charge Qi at the virtual source controlled by the gate : (like in the Natori’s model) Id (1 − r ) − exp ( − eVd / kT) (1 − r) = Q i v th W (1 + r ) + exp ( − eVd / kT) (1 − r) I d Sat 1 − r = Q i v th W 1+ r I d Lin e Vd Q = (1 − r ) i v th W 2 kT A. McKelvey. R.Concept of Backscattering Coefficient Using Mc Kelvey flux theory of Transport : js+ Perfect non degenerated reservoir of carriers “Source like” thermalisation (1 – rs) js+ rs js+ (1– rd) jd- rd jdjdx=L thermalisation Perfect non degenerated reservoir of carriers “Drain like” Distance x=0 J. 49 p 481 . Phys.489 (2002) 6 MOS-AK Workshop 2010 . Longini and T. « A compact scattering model for the nanoscale double-gate MOSFET ». S. « Alternative approach to the solution of added carrier transport problems in semiconductors ». vol. 51-57 (1961). Brody.. IEEE.

Lundstrom Z. Ren.Concept of Backscattering Coefficient rLF = Back Scattering Coefficient at Low Field (Ohmic Regime) assuming a constant isotropic mean free path : rLF = L L+λ (using Boltzmann or Fermi Statistics) M. « Fundamentals of Carrier Transport ». Cambridge university press.133 (2002) Drain 7 MOS-AK Workshop 2010 . second edition. 293 (2001) rHF = Back Scattering Coefficient at High Field (Saturation Regime) Gate F+ kT rHF = LkT L L kT L kT + λ empirical rLF = L L+λ Source F- intuited as a generalization of rLF M. Lundstrom IEEE EDL 22 p. this is consistent with the Drift Diffusion model : I d Lin e Vd Q = (1 − r ) i v th W 2 kT L >>λ I d Lin DD 1 = µ eff Qi Vd W L if λ= 2 µ eff kT v th e M. 2000 When L >>λ. IEEE TED 49 p. Lundstrom.

this is consistent with the Drift Diffusion model : Id Sat BAL Id Lin BAL (Vg − VT ) 2 I d Sat DD 1 = µ eff C ox W L 2 if Id Lin λ= 2 µ eff kT v th e (same expression than in ohmic regime) 10 λ and L kT = L 100 1000 10000 2 kT / e 2 kT / e =L Vd SAT Vg − VT 1 1 10 Channel Length L (nm) which is in fact equal to the LkT layer calculated from the Drift Diffusion potential profile ! µ = 200 cm2V-1s-1 Vinj = 1. 8 MOS-AK Workshop 2010 .Backscattering Coefficient Modeling In the MOSFET model.45×1013 cm-2 • In ohmic regime. S. Rhew Journal of Computational Electronics.2 × 105 m/s Ninv = 1. Quasi ballistic transport occurs when L ∼ λ • In saturation. 1 pp 481 . vol. Quasi ballistic transport occurs when LkT ∼ λ i. H. Lundstrom and J. we have : I d Sat 1 − r = Q i v th W 1+ r with rHF = L kT L kT + λ Drain Current (µA/µm) Id Sat 1000 Id Lin DD 100 Id Sat DD When L >>λ.489 (2002).e when L ∼ 10 λ M.

L. Clerc.Validation by Monte Carlo Simulation rHF = L kT L kT + λ ?? λ0 = 2 kT µ ev th ?? rHF extracted is in fact a function of LkT 20 λ extracted is in fact proportional to L kT L kT + λ Dev Extracted λDev (nm) rMC = λ0 = 2 kT µ ev th DG 10 nm 10 Strained Bulk Bulk 0 0 λ λ= 0 2 DG 4 nm IEDM 2006 « Multi Subband Monte Carlo investigation of the mean free path and of the kT layer in degenerated quasi ballistic MOSFETs » P. Esseni. R. L. Selmi 100 200 300 400 MC low field mobility (cm2V-1s-1) 500 9 MOS-AK Workshop 2010 . D. Palestri. Lucci.

Acoustic Ph.2 0.5 1 0.1 0 r= λ= L kT L kT + λ 1 2 µ eff ( E eff ) kT/e × 2 v th Bulk DG 12 10 MOS-AK Workshop 2010 . Proc. 166 Analytical Model 2. Roughness Tsi Fluctuation DG 12 0. Optical Surf.3 0.5 Back Scattering r 0.al. Essderc 2006 p. Ponton et.4 0.Application to device performance modeling Strain Silicon : double advantage in QB devices Multi Subband Monte Carlo Simulations Injection Velocity 107cm/s D.5 0 Bulk Ph.5 2 1.

Limits of Lundstrom’s Model rHF L kT = L kT + λ • is only an empirical formula : where does it come from ? • How to evaluate LkT ? (V(x) is not known) (In particular the impact of velocity overshoot on V(x). and thus LkT is not self consistently taken into account) λ= 2 kT/e µ v th • OK in low field • But in high field rather equal to λ=α 2 kT/e µ v th α≈ 1 2 with Vth and rHF are valid in non degenerated inversion layers • How to generalize LkT and λ in degenerated inversion layer ? (An increase of Vinj may also degrade r : is subband engineering a viable strategy ?) 11 MOS-AK Workshop 2010 .

Palestri.Beyond Lundstrom’s Model R. Selmi « On the Physical Understanding of the kT-Layer Concept in Quasi-Ballistic Regime of Transport in Nanoscale Devices » IEEE TED 53. P. L. p 1634 – 1640 (2006) 12 MOS-AK Workshop 2010 . Clerc.

v) F1 Collision Integral (relaxation length approximation ) : x F3 x + ∆x ∞ f-(x.v) C A → B = ∆x ∫ 0 F4 B F5 f(x.1D flux conservation in the relaxation length approximation 1D Balance Equation in the phase space vx A F1 + F3 = F2 + C A → B ∞ F1 = F2 ∫ 0 f(x. v x ) Φ + (x) − Φ + (x) LE dv x = ∆x λ /v x λ Balance Equation for the flux of carrier along the device + + d Φ− − e F Φ ( x ) − Φ LE ( x ) = −f(x. v x ) v x dv x = Φ + ( x ) f+(x.0 ) − dx m λ Assumption on the shape of f(x. v x ) − f LE ( x.vx) needed 13 MOS-AK Workshop 2010 .

6 L=λ/5 0.1 1 0.The quasi ballistic Drift Diffusion Distribution Function (a.2 0.u) Assuming local Maxwellian distribution function (at each point) : This work Drift Diffusion Approximation m m v2 ± f ( x . v) = 2 n (x) exp (− ) 2 π kT 2 kT ± Similar to Drift Diffusion -4 -2 0 2 4 velocity v/vth BackScattering Coefficient r 1 L = 50 λ 0.01 rLF = Drain Voltage Vd (V) L λ+L rHF = L kT λ + L kT 14 MOS-AK Workshop 2010 .8 L L+λ Application to the backscattering coefficient calculation: 0.4 This model L kT L kT + λ r= L kT (1 − β ) λ + L kT (1 − β ) β = exp( − L / LkT) This equation includes both : 0.

if we derive the current flux : Φ = q (n + − n − ) v th = − λ x v th n = n+ + n− d n λ x v th − 2n − Fx dx kT / q dn − n µ ' Fx dx µ'= µ 1 + µ Fx / v th Using : Φ = (n + − n − ) v th = − D' Except for the boundary conditions. it is equivalent to Drift Diffusion.. Wang. including saturation velocity ! Idem for H. Dig. in IEDM Tech. Gildenblat. 125–128 (2002) 15 MOS-AK Workshop 2010 . “Scattering matrix based compact MOSFET model”.The quasi ballistic Drift Diffusion We thus have now a new formalism which includes : Thermal velocity limitation at the source (Ballistic limit) Same backscattering coefficient in high and low field (Ballistic mobility) With no need to calculate LkT … An alternative to Drift Diffusion ? Using this approach. G. pp.

u.) Backscattered Electron Ballistic Electron MB Velocity 16 MOS-AK Workshop 2010 . v) = 2 n (x) exp (− ) 2 π kT 2 kT ± A more suitable approximated distribution function (at each point) : From MC simulations 2 U(x) m Distribution Function (a.Non Thermal Approach What is wrong with quasi ballistic Drift Diffusion ? m m v2 ± f ( x .

Backscattering Coefficient r 2 λth = 18 nm Backscattering Coefficient r 2 λth = 18 nm 0.01 1 kT layer length LkT (nm) 10 100 0. L) = ∫ 0 U (u )   2λ(u ) (1 + ) du  2λ kT    −1 1 Linear Pot. 1 Parabolic Pot. L) L kT / L A(L kT .01 1 10 100 kT layer length LkT (nm) 17 MOS-AK Workshop 2010 .1 2 λth = 64 nm 2 λth = 36 nm MC This model Thermal 0.1 2 λth = 36 nm 2 λth = 64 nm MC This model Thermal 0. L) 2 λ + L kT A (L kT .Non Thermal Approach Backscattering formula : r= L kT A(L kT .

LkT = 2 nm 2 λth = 36 nm V+ V Symbols : MC Lines : Model Velocity (105 m/s) 2.5 2 1.5 1 0.5 1 0.5 V V− V− vth V thermal 0 5 10 15 20 vth V thermal 2 4 6 8 10 12 14 Distance (nm) Distance (nm) 18 MOS-AK Workshop 2010 .5 Symbols : MC Lines : Model V+ With real drain (emitting) : Linear Pot.5 0 0 Velocity (105 m/s) LkT = 2 nm 2 λth = 36 nm 2 1.Non Thermal Approach Velocity profile in high field in a non self consistent linear potential profile With absorbing drain : 3 Linear Pot. 2.

Ballisticity extraction from experiments 19 MOS-AK Workshop 2010 .

Ballisticity extraction from experiments Is there any experimental confirmation of the quasi ballistic nature of transport ? Not clear yet ! Let us show some results : • let us consider only the ohmic regime (more simple : you do not need to know the charge at the virtual source) • a standard low field mobility extraction has been performed on : 65 like nm Bulk and undoped Fully Depleted SOI technology featuring physical gate length down to 40 nm • according quasi ballistic theory : Id Lin e Vd 1 Q = (1 − r) i v th = µ app Q iVd W 2 kT L µ app = L µ dd L+λ Apparent mobility should decrease with L in the quasi ballistic regime 20 MOS-AK Workshop 2010 .

01 0.1 1 10 Undoped FD-SOI nMOS Low Field Mobility µexp (cm2/Vs) 200 180 160 140 120 100 80 0.2nm) polysilicon gate.01 0.Ballisticity extraction from experiments Bulk nMOS Low Field Mobility µexp (cm2/Vs) 700 600 500 µdd 400 300 µexp 200 0.5 nm of HfSiON dielectric Raised source and drain 21 MOS-AK Workshop 2010 . 2. SiON gate oxide (CET=2.1 1 10 µdd µexp Channel length L (µm) Channel length L (µm) Bulk 65nm CMOS technology doped channel (≈1017/cm3) with halos. Undoped Fully depleted SOI Body thickness of 10 nm Metal gate TiN.

9 0.85 0.Ballisticity extraction from experiments r = 1− µ app µ bal The backscattering coefficient has been extracted using Bulk nMOS Backscattering coefficient r 1 0.9 Bulk nMOS 0.7 0.1 1 10 FD-SOI nMOS 0.95 0.85 0.8 0.1 1 10 Channel length L (µm) Channel length L (µm) Extracted value of r are very large … Neutral defects ? 22 MOS-AK Workshop 2010 .7 Undoped FD-SOI nMOS Backscattering coefficient r 1 rLF ~ 0.6 0.01 rLF ~ 0.01 0.

Conclusions 23 MOS-AK Workshop 2010 .

Conclusions The Lundstrom backscattering theory provides a powerful guideline to analyse qualitatively device performance in the quasi ballistic regime However. experiments shows a more complex picture : we are not sure yet to operate in the quasi ballistic regime (neutral defects ?) 24 MOS-AK Workshop 2010 . It is not a compact model (How to compute LkT ?) If you try to generalize this approach along the channel → quasi ballistic drift diffusion → which is similar to drift diffusion with saturation velocity in high field regime You need to account for the ballistic distribution of carriers However.

Thank you for your attention ! 25 MOS-AK Workshop 2010 .

u) Ballistic limit Ballistic enhancement factor Drift Diffusion Analytical Model (Velocity Saturation) Mean Free Path 1 10 100 Drift Diffusion 1/L 1000 10000 Channel Length (nm) 26 MOS-AK Workshop 2010 .More Exact Quasi Ballistic Model Drain Current (a.

45×1013 cm-2 Undoped UTB µ = 200 cm2V-1s-1 Vinj = 1.3 × 105 m/s Ninv = 1.Application to device performance enhancement Device Optimisation in the Quasi Ballistic Regime 3500 2 possible strategies to improve Ion : 3000 Ion Current (µA/µm) Strained Undoped UTB Undoped UTB Bulk DD vsat=105 m/s 2500 2000 1500 1000 500 0 1 10 100 1000 10000 • improving λ.45×1013 cm-2 Strained Undoped UTB µ = 370 cm2V-1s-1 Vinj = 1.2 × 105 m/s Ninv = 1. which mean improving µ = effective field mobility like in pure Drift Diffusion model ! • improving Vinj (subband engineering) by DOS reduction Still no clear experimental evidence Channel Length (nm) BULK µ = 130 cm2V-1s-1 Vinj = 1.2 × 105 m/s Ninv = 1.45×1013 cm-2 27 MOS-AK Workshop 2010 .

regardless of the channel length and of the magnitude of the electric field reasonable approximation Only close to the virtual source Or along the channel when many collisions are involved 28 MOS-AK Workshop 2010 .Re-investigation of the kT layer Concept rHF L kT = L kT + λ has been derived using Quasi Ballistic Drift Diffusion. key assumptions on which this formula is based : • Boltzmann statistics in the contacts • Non Self consistent potential • use of the relaxation length approximation with a constant λ λ should be energy dependent (especially at high field) • the population of backscattered carriers (f−) has an equilibrium Maxwellian distribution : at each point x.