Date

2003-12-03

Description

Comment: 27 pages, 13 figures, submitted to IEEE Trans. Electron. Dev.,

corrected typos, conclusions clarified, less meaningful figures suppressed

corrected typos, conclusions clarified, less meaningful figures suppressed

The scattering effects are studied in nanometer-scaled double-gate MOSFET,

using Monte Carlo simulation. The non-equilibrium transport in the channel is

analyzed with the help of the spectroscopy of the number of scatterings

experienced by electrons. We show that the number of ballistic electrons at the

drain end, even in terms of flux, is not the only relevant characteristic of

ballistic transport. Then the drive current in the 15 nm-long channel

transistor generations should be very close to the value obtained in the

ballistic limit even if all electrons are not ballistic. Additionally, most

back-scattering events which deteriorates the ON current, take place in the

first half of the channel and in particular in the first low field region.

However, the contribution of the second half of the channel can not be

considered as negligible in any studied case i.e. for a channel length below 25

nm. Furthermore, the contribution of the second half of the channel tends to be

more important as the channel length is reduced. So, in ultra short channel

transistors, it becomes very difficult to extract a region of the channel which

itself determine the drive current Ion.

using Monte Carlo simulation. The non-equilibrium transport in the channel is

analyzed with the help of the spectroscopy of the number of scatterings

experienced by electrons. We show that the number of ballistic electrons at the

drain end, even in terms of flux, is not the only relevant characteristic of

ballistic transport. Then the drive current in the 15 nm-long channel

transistor generations should be very close to the value obtained in the

ballistic limit even if all electrons are not ballistic. Additionally, most

back-scattering events which deteriorates the ON current, take place in the

first half of the channel and in particular in the first low field region.

However, the contribution of the second half of the channel can not be

considered as negligible in any studied case i.e. for a channel length below 25

nm. Furthermore, the contribution of the second half of the channel tends to be

more important as the channel length is reduced. So, in ultra short channel

transistors, it becomes very difficult to extract a region of the channel which

itself determine the drive current Ion.

Type

Identifier

doi:10.1109/TED.2004.829904

IEEE Trans. Electron Devices 51 (7), 1148-1155 (2004)

Database

Link to record

Show preview

Hide preview

1

On the ballistic transport in nanometer-scaled double-gate MOSFET

Jérôme Saint Martin, Arnaud Bournel, Philippe Dollfus Institut d'Electronique Fondamentale, CNRS UMR 8622, Université Paris Sud, Bât. 220,

F-91405 Orsay cedex, France

Corresponding author

Jérôme Saint Martin Institut d'Electronique Fondamentale CNRS UMR 8622, Université Paris Sud, Bât. 220 F-91405 Orsay cedex, France Tel.: 33 1 69 15 40 37, Fax: 33 1 69 15 40 20, E-mail: stmartin@ief.u-psud.fr

Abstract

The scattering effects are studied in nanometer-scaled double-gate MOSFET, using Monte Carlo

simulation. The non-equilibrium transport in the channel is analyzed with the help of the

spectroscopy of the number of scatterings experienced by electrons. We show that the number of

ballistic electrons at the drain-end, even in terms of flux, is not the only relevant characteristic of

ballistic transport. Then the drive current in the 15 nm-long channel transistor generations should be

very close to the value obtained in the ballistic limit even if all electrons are not ballistic.

Additionally, most back-scattering events which deteriorates the ON current, take place in the first

half of the channel and in particular in the first low field region. However, the contribution of the

second half of the channel can not be considered as negligible in any studied case i.e. for a channel

length below 25 nm. Furthermore, the contribution of the second half of the channel tends to be

more important as the channel length is reduced. So, in ultra short channel transistors, it becomes

very difficult to extract a region of the channel which itself determine the drive current Ion.

Keywords

Silicon, SOI, MOS devices, MOSFETs, nanotechnology, semiconductor device modeling, Monte

Carlo methods, backscattering, ballistic transport.

2

1 Introduction

Double-gate MOSFET architecture (DGMOS) is a potential solution to overcome short channel

effects in the 65 nm ITRS node [1], that is for physical gate lengths smaller than 25 nm. In such

nano-transistors where the channel length is comparable to the electron mean free path, non

stationary [2] or even ballistic [3] transport is probably of great importance regarding the device

performance. This question is however rather controversial from a theoretical point of view.

According to Natori prediction [4], non stationary transport in ultra-small devices and statistical

fluctuation of random scattering events undergone by charge carriers in the channel should lead to

dramatic time fluctuations of drive current. According to Monte Carlo simulation results, small

variations in the number and the position of doping atoms in the channel of a 50 nm bulk MOSFET

significantly influence the transport properties and the drain current [5]. Ballistic transport in

undoped channel may be the solution to limit these types of fluctuations.

Lundstrom and coworkers have studied in detail the influence of ballistic transport in such

devices on the drain current [6],[7]. They developed models describing nano-DGMOS operating in

the ballistic or quasi-ballistic limit. These models are based on the concept of thermionic injection

from source-end into the channel. According to such approach, the electron velocity at the

source-end, and thus the drive current, should be limited by the "source-to-channel" energy barrier

and by back-scatterings in the low-field region, i.e. in the vicinity of the barrier.

Svizhenko and Anantram have also investigated the role of scatterings in nanometer-scaled

DGMOS [8]. Using a Green-function approach, they show that scatterings at both source- and

drain-end influence significantly the drain current.

More recently, Mouis and Barraud [9] have used the Monte Carlo simulator MONACO,

developed in our group, to discuss the evolution of the velocity distribution along the channel of a

DGMOS. Their work puts into evidence that some electrons can be reflected towards the channel

after having experienced Coulomb interactions in the highly doped drain region i.e. far from the low

field region.

To analyze in more details the transport in nanometer-scaled DGMOS, we study in this paper the

behavior of electrons injected from the source as a function of the number of scattering events

undergone in the channel. We have introduced in the device Monte Carlo simulator a procedure

which allows us to get accurate information on the number of scattering events experienced by each

electron during its travel between the source-end and the drain-end of the channel. We can also

extract at many positions in the channel the number of back-scattered electrons and the velocity

distributions of electrons having undergone either 0, 1, 2, … or N scattering events. We thus have

3

relevant information to investigate in detail in this paper the notion of ballisticity [6] and to discuss

the actual influence of both ballistic and backscattering effects on the device characteristics and

performance.

The paper is organized as follows. The simulated devices and the Monte Carlo model are

presented in Section 2 and the results are described and discussed in section 3. First, we analyze the

velocity spectra in the channel by highlighting the ballistic and quasi-ballistic phenomena, which

completes the work initiated in Ref.[9]. Then the notion of intrinsic ballisticity is defined and its

influence on drive current is evaluated by tuning the scatterings intensity in the channel. Finally, we

investigate the evolution of backscattering coefficient along the channel and, by separating the

channel in two parts with different scattering properties, we analyze the impact of the region where

scattering events take place on the I-V characteristics.

2 Model and simulated devices

The simulated DGMOS devices, described in Fig. 1, have an effective channel length Lch equal

to 10 nm, 15 nm, 25 nm or 50 nm and a gate lengths LG equal to 10, 25, 25 and 50 nm respectively.

However, the study is focused on 15 nm-and 25 nm-effective channel lengths. The SiO2 gate oxide

Tox and Si body thicknesses TSi are equal to 1.2 nm and 10 nm, respectively. The doping density is

ND = 5 × 1019 cm-3 in N+ S-D regions and NA = 2 × 1015 cm-3 in the body (P type). The N+ doping level is relatively low, and may induce a significant series resistance, but it is a realistic value

considering the difficulty of electrically activating dopants in such thin body. The N+/P junctions

are assumed to be abrupt. The work function of the gate material is 4.46 eV to achieve the

theoretical threshold voltage VT of 0.2 V. The power supply voltage VDD is fixed at 0.7 V to abide

by the 2007 ITRS specification.

A classical particle Monte Carlo algorithm is self-consistently coupled with a 2D Poisson solver.

The Poisson's equation is solved at each time step equal to 0.1 fs with standard boundary conditions

[10]. The number of simulated particles is typically 50000. The scattering mechanisms included in

the simulation are phonon scattering, impurity scattering and surface roughness scattering. The

acoustic intra-valley phonon scattering is treated as an elastic process and the intervalley phonon

transitions, consisting of three f-type and three g-type processes, are considered via either zeroth-

order or first-order transition matrix in agreement with selection rules [11]. The phonon coupling

constants given in Ref.[10] are used. The impurity scattering rate is derived from the screened

Coulomb potential with the momentum-dependent screening length given in Ref.[12]. The surface

roughness scattering is treated with an empirical combination of diffusive and specular reflections

4

which correctly reproduces the experimental universal mobility curve [13],[14]. Unless otherwise

indicated we have considered in this work a fraction of diffusive reflections Cdiff equal to 0.14.

The originality of the present work, in which quantum confinement effects are not included, lies

in the spectroscopy of the number of scattering events undergone by electrons crossing the active

part of the channel (or any predefined part of the device) and in the detailed study of the velocity

spectra of different electron groups. This makes possible the careful analysis of scattering effects.

Practically, a scattering counting region is predefined with an entrance surface and an exit surface.

Typically, the entrance surface is defined either at the source/channel junction or at the position of

the top of the gate-induced potential barrier, and the exit surface is placed at the channel/drain

junction. Each electron entering the counting region by the entrance surface is flagged and while it

remains inside this region the number of scattering events undergone is recorded. At the exit surface

we can thus separate electrons into different groups, corresponding to ballistic electrons,

once-scattered electrons, twice-scattered electrons, … N-times scattered electrons, respectively. Of

course these groups can be enumerated. Additionally, for each of these groups the energy and

velocity spectra are recorded. Such spectra may be also obtained at intermediate surfaces defined all

along the counting region, i.e. the channel. It should be noted that these spectra only include

electrons coming from the entrance surface and do not consider the electrons entering the counting

region by the exit surface. The information about back-scattered electrons is obtained separately: if

an electron is back-scattered after having crossed the counting region and re-enter this region by the

exit surface, its scattering-story is still recorded and the velocity spectra of such electrons are

recorded too. Thus, by possibly changing the place of the exit surface, we can have very detailed

information on all carriers participating in the source-drain current. It is very useful to quantify and

analyze the effects of ballistic, nearly-ballistic and back-scattered electrons.

3 Results

The variations of drain current ID as a function of source-drain voltage VDS obtained in the

15 nm-, 25 nm- and 50 nm-long transistors for VGS = VDD are shown in Fig. 2. For the 15 nm-long

channel, the Ion current at VDS = VGS = VDD = 0.7 V is 2140 µA/µm. Such a high value is related to

the aggressive scaling of the gate length, to the double-gate architecture and to the relatively thick

body film (Tsi = 10 nm). The drain conductance, and more generally the short channel effects

(SCE), are however rather strong in the 15 nm-long channel, as a consequence of the non optimized

body thickness. The 25 nm-long channel, better designed, is less sensitive to SCE while exhibiting

Ion-value (1600 µA/µm) still higher than the ITRS Roadmap specification that is 900 µA/µm. We

now examine the transport in these devices biased in the "on-state", i.e. with VGS = VDS = VDD.

5

3.1 Velocity spectra analysis

The conduction band profile and the corresponding velocity spectra evolution of electrons along

the 15 nm-long channel have been plotted respectively in Fig. 3 and Fig. 4. The Fig. 4 represents the

velocity spectra evolution calculated in the first sheets of cells under the front oxide, i.e. between

y = 0 and y = 1 nm (see Fig. 1). The transport is stationary in the highly doped source well, as

illustrated by the quasi-Gaussian shape of the velocity distribution in the middle of this region.

Other spectra are taken in the channel and, in contrast to the results shown in Ref.[9], they only

concern electrons injected by the source into the channel, excluding those which have been injected

or re-injected from the drain. These spectra are very different from stationary ones. In the vicinity of

the top of the barrier (x = 2 nm), the spectrum is quite similar to a hemi–Gaussian distribution.

However electrons with a negative velocity have not completely disappeared, which is the signature

of backscattered electrons. Thus, the regime of pure thermionic injection in the channel is not

completely reached in this structure. At a position x = 4 nm after the source/channel junction, two

separate distributions appear in the velocity spectrum. Two peaks can be distinguished, each of

them corresponding to electrons with either a transverse or a longitudinal conductive effective mass

along the source-to-drain direction. For greater distance x, we observe the propagation of these two

distributions along the channel, each one at its own velocity. The electron density decreases as the

average velocity increases in accordance with the current conservation and the distributions become

narrower and narrower.

At the drain-end of the 15 nm-long device, the velocity spectrum of electrons injected from the

source calculated at the Si/SiO2 interface looks like two very well-defined peaks as shown in Fig. 5.

According to energy conservation between the barrier region and the drain-end of the channel, it is

easily demonstrated that the peak velocity values in Fig. 5 correspond to ballistic electrons flowing

from source to drain with a transverse (mt = 0.19 m0 where m0 is the free electron mass) or a

longitudinal (ml = 0.916 m0) effective mass.

To complete the description of electron populations present in the channel, the spectrum of

electrons injected from the drain at x = 15 nm is also plotted in dotted line in Fig. 5. As confirmed

by a velocity spectroscopy distinguishing electrons with transverse mass from those with

longitudinal mass (not shown), this ‘drain injected’ distribution is made up of the sum of two nearly

Gaussian distributions. The widest corresponds to electrons with a transverse mass and the thinnest

to electrons with a longitudinal mass. Even if they are the most numerous at the drain-end, their net

contribution to the current at the drain-end is only 0.9%.

6

Besides, the number of “drain-injected” electrons rapidly decreases near the source. Indeed, at

x = 15, 12, 4 and 1 nm, the part of “drain injected” electrons represents respectively 72.3%, 42.3%,

5.6% and 1.7 % of the electrons. At the source-end (x = 1 nm) 6.3% of electrons which have a

negative velocity have been injected from the highly-doped drain region.

Those first analyses of velocity evolution along the channel indicate clearly that the transport is

completely out of equilibrium. Furthermore, the velocity spectrum of “source-injected” electrons at

the drain-end (Fig. 5) suggests that ballistic electrons have an important role in the transport for

nanometer-scaled DGMOS devices. In this connection, we present a new kind of analysis which

allows dissecting the velocity spectra presented in Fig. 5. In this purpose, we have defined a

counting region by an entrance surface and an exit surface located at the source/channel and

channel/drain junctions, respectively. This study does not consider “drain-injected” electrons.

The Fig. 6 represents the velocity distributions calculated at the drain-end of the 15 nm-long

channel device for electrons flowing from the source by undergoing 2, 1 or 0 scattering events in

the counting region. Contrary to the spectra presented above (i.e. in Fig. 4 and Fig. 5) and calculated

at the Si/SiO2 interface, these new distributions are calculated on the full body thickness, i.e.

between y = 0 and y = Tsi (cf. Fig. 1). A 2D effect is clearly observed in Fig. 6: the ballistic peaks

are wider than those presented in Fig. 5 as they result from the sum of peaks with different maxima.

Indeed, as shown in Fig. 3, the potential drop varies as a function of the channel depth, i.e. along the

y-axis (cf. Fig. 1), which makes the ballistic peak position dependent on y. For instance, according

to the potential drop plotted in Fig. 3, the transverse mass peak velocity is equal to 6.5 × 107 cm/s at

y = 0 nm and to 7.0 × 107 cm/s at y = Tsi/2. We also verify that the distribution tails observed in Fig.

5 are caused by electrons which have undergone interactions during their channel crossing.

Moreover, the shapes of the quasi-ballistic velocity spectra, i.e. the spectra obtained for electrons

which have undergone 1 or 2 interactions, is similar to that of a ballistic spectrum. This suggests

that devices can drive on-current Ion very close to the limit value obtained for a ballistic channel,

even if all electrons are not purely ballistic at the drain-end.

The Fig. 7 represents the same kind of velocity distributions for “source injected” electrons but

calculated near the top of the injecting barrier, i.e. at x = 3 nm from the source-end. This figure

allows us to verify that only ballistic electrons are purely thermionically injected since their velocity

spectrum looks like a strict hemi-Gaussian distribution. Additionally, the contribution of electrons

which have undergone more than two scatterings in the channel is negligible (not shown). At

VDS = 0.7 V, the backscattered population is essentially formed by once– and twice–scattered

electrons.

7

3.2 Influence of the ballisticity on drive current

To get a more quantitative insight into ballistic and/or quasi-ballistic transport, we have

calculated the number of scattering events Nscatt experienced by each carrier crossing the counting

region, i.e. the channel. The resulting electron distribution is plotted in Fig. 8 as a function of Nscatt

for DGMOS of different effective channel lengths: 15 nm, 25 nm and 50 nm. For comparison, we

have also indicated, in dotted line the similar distribution obtained at the drain-end for a

conventional 50 nm bulk MOSFET with the following characteristics: a single gate, a channel

length Lch = 50 nm, an oxide thickness Tox = 1.2 nm, a junction depth Xj = 20 nm and a highly

doped channel NA = 1018 at/cm3. The latter distribution is a bell-curve with a maximum centered on

Nscatt = 5, which corresponds to the ratio Lch/λeff where Lch is the effective channel length and λeff is

defined as an effective mean free path. On the contrary, in all lightly-doped DGMOS the group of

ballistic electrons is the most populated, but it forms the majority only in the 15 nm-long channel.

This indicates that the effective mean free path λeff is smaller than the channel length in all

simulated DGMOS. However, the fraction of electrons decreases as Nscatt (the number of

scatterings) increases and the distribution spreads out when Lch increases. For the 15 nm-long

channel, the curve is a pure exponential function. In both 50 nm-long channel devices (DGMOS

and bulk) the interaction spectrum tends to a more stationary-like one.

Now, we define the intrinsic ballisticity Bint as the percentage of electrons which are ballistic at

the drain-end (Nscatt = 0 in Fig. 8) [15]. The line in Fig. 9 is an interpolated curve obtained by

linking the intrinsic ballisticity Bint for the 3 different DGMOS with channel effective length

Lch equal to 15 nm, 25 nm and 50 nm and by assuming Bint = 100% for Lch = 0. The smooth curve

obtained from these three DGMOS seems particularly relevant because for Lch = 10 nm it gives the

same result as the complete Monte Carlo simulation of a 10 nm-long device (closed circle). From

such a semi classical Monte Carlo approach, one can thus estimate that ballistic electrons should be

largely predominant (i.e. with an intrinsic ballisticity Bint greater than 90%) only for channel lengths

smaller than about 3 nm. However, quantum transport effects have to be considered for

investigating Lch-values smaller than 10 nm [7], which may modify this prediction.

To determine how the intrinsic ballisticity Bint at the drain-end is related to the on-current Ion, we

have artificially modified it from 0 as in full stationary transport to 100% as in pure ballistic

transport. The quantity Ion_bal stands for the on-current obtained for a ballistic channel, that is

without any phonon or roughness effects. Then, we study the effective ballisticity Beff, defined as

Beff = Ion/Ion_bal [6], as a function of the intrinsic ballisticity Bint at the drain-end.

8

To this end, we have varied the oxide roughness coefficient Cdiff from 0 to 1 and introduced a

phonon scattering coefficient Kph in the phonon scattering rates: all standard values are multiplied

by the coefficient Kph varying from 0 for a ballistic channel, to 20 for a very resistive channel. The

evolution of Beff as a function of Bint for a given Cdiff and for different phonon scattering

coefficients, is plotted in dotted line in Fig. 10. The results obtained for Cdiff = 1 (respectively

Cdiff = 0.14) and for Kph = 0, 1, 2, 5, 10 and 20 (respectively Kph = 0, 0.5, 1, 2), are indicated with

open squares (respectively closed triangles). The results obtained for various roughness coefficients

Cdiff and for a given phonon coefficient Kph are shown in solid lines: on the one hand for Kph = 1 and

Cdiff = 0, 0.07, 0.14, 0.21, 0.4, 0.7 and 1 (open circles), and, on the other hand, for Kph = 0 with

Cdiff = 0, 0.14, 0.5, and 1 (closed squares).

These results show an overall view of the effect of each kind of interaction. For an intrinsic

ballisticity Bint greater than 20% the effect of each type of interaction (phonon or roughness

scattering) yields a linear behavior Beff(Bint) but with a slope depending on the type of interaction.

Below this limit, when the transport is more stationary, the Beff(Bint) relation is no more simply

linear. Thus there is neither an equality nor a unique linear relation between the effective ballisticity

Beff and the intrinsic ballisticity Bint. The effective ballisticity Beff (in terms of current) alone does

not provide enough information to quantify accurately the intrinsic ballisticity Bint. However, there

is no denying that these two quantities are strongly correlated. Besides, we notice that the intrinsic

ballisticity Bint is always overestimated by the effective ballisticity Beff: for instance Bint = 52% and

Beff = 84% for the standard DGMOS (Kph = 1 and Cdiff = 0.14).

3.3 Back-scattering localization and its effect on the current

To investigate further the effects of scattering, a flux approach may be used [6]. We propose to

identify the part of the channel giving the highest contribution to back-scattering effects which are

known to degrade the drive current. So, a control volume is defined, as shown in Fig. 11, by an

entrance surface and an exit surface that can be moved along the channel. By calculating, the

different fluxes: ΦI+, ΦO+, ΦI1- and ΦI2-, the relevant back-scattering coefficients may be extracted.

Fluxes ΦI+ and ΦO+, oriented as indicated in Fig. 11, are the ingoing flux at the entrance surface and

the outgoing flux at the exit surface, respectively. The flux ΦI1- represents the flux of electrons

which have entered the control volume by the entrance surface and which have crossed back the

entrance surface without having crossed the exit surface. The flux ΦI2- represents the flux of

electrons which have entered the control volume by the entrance surface and which have crossed

back the entrance surface after having crossed the exit surface.

9

The back-scattering coefficient R(x) of the control volume between the entrance surface located

at the top of the barrier and an exit surface located at the distance x from the source end is equal to

ΦI1-(x)/ΦI+. The evolution of R(x) obtained by moving the exit surface along the channel from the

top of the barrier to the drain–end is plotted in Fig. 12 for 15 nm- and 25 nm-long devices. The

maximum back-scattering coefficient Rmax for the whole structure including the drain contact can be

calculated as Rmax = max[R(x)] = [ΦI1-(x) + ΦI2-(x)]/ΦI+.

First, we remark that at the drain-end of the 25 nm–long channel the backscattering coefficient

R(Lch) is slightly lower than Rmax: Rmax = 1.035 × R(Lch). Moreover, the difference between Rmax and R(Lch) strongly increases when Lch decreases to 15 nm: Rmax = 1.195 × R(Lch). This difference, due to electrons back-scattered in the drain, may be more accurately estimated by taking into

account other scattering mechanisms, as for instance, the short range electron-electron interactions.

A monotonous variation of R(x) along the channel is observed. The increase of R(x) is much

more important in the first half part of the channel as the back-scattering events take place mainly

there. However this is less true when the effective channel length Lch decreases:

R(x = Lch/2) = 0.91 × R(Lch) for Lch = 25 nm, while R(x = Lch/2) = 0.81 × R(Lch) for Lch = 15 nm. In the first part of the channel the increase of R(x) is rather uniform, just a bit faster in the

vicinity of the top of the barrier. Thus, it is difficult to accurately define in the first channel half a

region which has a predominant impact in terms of back-scattering.

It has been suggested in Ref. [6] that the backscattering coefficient R can be estimated at the

position x in the channel where the potential drops by kT/q. Our results show that this definition

leads to a significant underestimation of R. For instance, consider the position x0 where the

potential drops by at least 50 meV on the full channel thickness. We have then x0 = 9.5 nm for

Lch = 25 nm and x0 = 6.3 nm for Lch = 15 nm. At this position x0, the backscattering coefficient is

only R(x0) = 0.86 × R (Lch) = 0.78 × Rmax. for Lch = 25 nm. This underestimation is more

pronounced for Lch = 15 nm: R(x0) = 0.73 × R(Lch) = 0.61 × Rmax.

This trend confirms that reducing Lch gives more importance to the second half of the channel in

terms of contribution to backscattering. Indeed, the number of scattering in ultra-short channel is

not important enough (cf. Fig. 8) to prevent electrons scattered in the second half of the channel

from being backscattered to the source–end [8].

To analyze the influence on drain current of scattering in the different parts of the channel, new

simulations have been performed by changing the scattering properties along the channel. Three

new 25 nm-long DGMOS called ‘bal-bal’, ‘bal-sta’ and ‘sta-bal’ have been simulated. In these

devices, the channel is divided in 2 equal parts. The term ‘sta’ stands for the channel–half with

10

standard scattering properties of Si doped to 2×1015 at/cm3: standard phonon scattering coefficient

Kph = 1 (see Sec. 3.2) and roughness coefficient Cdiff = 0.14. The term ‘bal’ stands for the ballistic

channel–half (without any scattering). So, for instance, ‘sta-bal’ is a DGMOS with a standard first

channel–half and a ballistic second channel–half.

The inset of Fig. 13 shows on both linear and logarithmic scales the evolution of drain current ID

as a function of gate voltage VGS at VDS = VDD. First we notice that all devices have the same

threshold voltage VT ≈ 0.3 V. Moreover, the subthreshold behavior is not degraded by SCE, the

subthreshold slope being equal to 70 mV/dec. The characteristics only differ in the

transconductance gm = GSVhighGS

D V I∂∂ above the threshold voltage VT and, as a consequence, in the

on-current Ion.

The variations of drain current ID as a function of source-drain voltage VDS obtained in these

DGMOS at VGS = 0.7 V are shown in solid lines in Fig. 13. In linear regime, the resistance

Ron = DSVlowD

DS I

V∂∂ extracted from Fig. 13 simply follows the Ohm’s law: Ron is determined by the

average channel conductivity and has the same value for ’bal-sat’ and ‘sta-bal’. It is greater for the

standard channel and lower for the ballistic channel. In the saturation regime, with nearly the same

saturation drain voltage VDSsat for all devices, the following values of output conductance

0.7GS D

D V V DS

IG V = ∂= ∂ are extracted at VGS = 0.7 V: GD_BB = 360 µS/µm for ‘bal-bal’,

GD_BS = 340 µS/µm for ‘bal-sta’, GD_SB = 270 µS/µm for ‘sta-bal’, GD_SS = 190 µS/µm for ‘standard’. We observe that the more ballistic the channel is, in particular in its first half, the greater

the conductance is. Thus, we deduce that the lack of scattering degrades the saturation behavior. It

should be mentioned that similar results have been obtained for bulk MOSFET of higher channel

length, i.e. 50 nm and 180 nm (not shown). Low Ron, i.e. high channel conductivity, is beneficial for

the on-current Ion, and a weak conductivity, in particular the first channel–half, is suitable to get a

low output conductance GD.

Besides, if each channel part had the same influence on drive current, ‘sta-bal’ and ‘bal-sta’ on–

current would be equal to the average on-current between ‘standard’ and ‘bal-bal’, i.e. equal to

1980 µA/µm. As the on-current Ion is equal to 2000 µA/µm for ‘bal-sta’ and 1850 µA/µm for ‘sta-bal’, we deduce that the ballistic channel part location, i.e. the scattering localization, has an

influence on the on-current.

This result is not fully inconsistent with the conventional belief that the first channel-half has a

greater importance than the second channel-half in terms of backscattering coefficient. However, in

11

contrast with this belief, our study reveals that the second part of the channel has a significant

impact on device operation and performance. Indeed Ion is 15% greater for ‘sta-bal’ than for

‘standard’ and it is 14% greater for ‘bal-bal’ than ‘bal-sta’. It should be mentioned that the intrinsic

ballisticity is higher for ‘sta-bal’ (61%) than for ‘bal-sta’ (53%), the resulting Ion is higher for the

latter device. Moreover the relative difference between ‘sta-bal’ and ‘bal-sta’ is low:

Ion(bal-sta) - Ion(sta-bal) = 4.4% × (Ion(standard) - Ion(bal-bal)). At last, the evolution of the

backscattering coefficient along the channel (Fig14) suggests that this impact of the second

channel–half should be still stronger for smaller channel length.

4 Conclusion

We have investigated in detail the velocity spectra of the electrons present in the channel as a

function of their origin: “source-” and “drain-injected” and of the number of the experienced

scatterings. We show that the transport in the 15 nm-long channel is neither purely ballistic nor

purely thermionic.

Nevertheless, ballistic electrons are of great importance in nanometer-scaled double-gate

MOSFET. In nano-DGMOS, the ballistic limit (intrinsic ballisticity Bint = 100% i.e. 100% of

electrons injected from the source are ballistic at the drain-end) is far to be reached for channel

lengths larger than 10 nm: the intrinsic ballisticity is about 50% in 15 nm-long channel. However,

the drive current Ion is closer to the value Ion_bal obtained with a pure ballistic channel: the ratio

Ion/Ion_bal, i.e. the effective ballisticity Beff, is more than 80%. Then the number of ballistic electrons,

even in terms of flux as we defined the intrinsic ballisticity Bint, is not the only relevant

characteristic of ballistic transport. There even if the "ballistic limit” is still “a mere pipe dream”

considering the numerous types of significant interactions in nano-scaled structures [16], the

on-current in next transistor generations should be very close to this limit.

Even if our results are not fully inconsistent with the conventional belief that most back-

scattering take place in the first half of the channel, they show that the role of the second half of the

channel cannot be considered as negligible for a channel length lower than 25 nm. Furthermore the

contribution of the second half to back-scatterings tends to be more and more important as the

channel length is reduced. As a consequence, it becomes more and more difficult to extract a

particularly significant region of the channel which would determine the value of the drive current

Ion.

This work is supported by the French RMNT project CMOS-D-ALI and we thank Emile

Grémion for his contribution.

12

References

[1] International Technology Roadmap for Semiconductors: 2002 Update, http://public.itrs.net/

[2] T. Mizuno and R. Ohba, “Experimental study of nonstationary electron transport in sub- 0.1 µm metal-oxide-silicon devices: Velocity overshoot and its degradation mechanism”, J. Appl. Phys., vol. 82, pp. 5235–5240, 1997

[3] S. E. Laux and M. V. Fischetti, "Monte Carlo simulation of submicrometer Si n-MOSFET’s at 77 and 300 K”, IEEE Electron Dev. Lett., vol. 9, pp 467–469, 1988

[4] K. Natori, "Ballistic metal-oxide-semiconductor field effect transistor", J. Appl. Phys., vol. 76, pp. 4879–4890, 1994

[5] P. Dollfus, A. Bournel, S. Galdin, S. Barraud, and P. Hesto, "Effect of discrete impurities on electron transport in ultra-short MOSFET using 3D Monte Carlo simulation", cond-mat/0310718, to be published in IEEE Trans. Electron Devices

[6] M. Lundstrom and Z. Ren, "Essential physics of carrier transport in nanoscale MOSFETs", IEEE Trans. Electron. Dev., vol. 49, pp. 133-141, 2002

[7] J.-H. Rhew, Z. Ren, and M. S. Lundstrom, "A numerical study of ballistic transport in a nanoscale MOSFET", Solid-State Electron., vol. 46, pp. 1899-1906, 2002

[8] A. Svizhenko and M. P. Anantram, "Role of scattering in nanotransistors", IEEE Trans. Electron Devices., vol. 50, pp. 1459-1466, 2003

[9] M. Mouis and S. Barraud, "Velocity distribution of electrons along the channel of nanoscale MOS transistors", in Proc. ESSDERC'2003 (IEEE), 2003, pp. 147-150

[10] P. Dollfus, "Si/Si1-xGex heterostructures: electron transport and field-effect transistor operation using Monte Carlo simulation", J. Appl. Phys., vol. 82, pp. 3911-3916, 1997

[11] T. Yamada, J. R. Zhou, H. Miyata, and D. K. Ferry, "In-plane transport properties of Si/Si1-xGex structure and its FET performance by computer simulation", IEEE Trans. Electron Devices, vol. 41, pp. 1513-1522, 1994

[12] N. Takimoto, "On the screening of impurity potential by conduction electrons", J. Phys. Soc. Jpn., vol. 14, pp. 1142-1158, 1959

[13] E. Sangiorgi and M. R. Pinto, "A semi-empirical model of surface scattering for Monte Carlo simulation of silicon n-MOSFET", IEEE Trans. Electron Devices, vol. 39, pp. 356-361, 1992

[14] S. Keith, F. M. Bufler, and B. Meinerzhagen, "Full-band Monte Carlo device simulation of an 0.1µm n-channel MOSFET in strained silicon material", in Proc. ESSDERC'97 (Editions Frontières, Paris), 1997, pp. 200-203

[15] It must be highlighted that the intrinsic ballisticity is a percentage in terms of positive current flux and not of electron density.

[16] M. V. Fischetti, "Scaling the MOSFETs to the limit: A physicist's perspective", Journal of Computational Electronics, vol. 2, pp. 73-79, 2003

13

Figure captions

Fig. 1: Schematic of DGMOS structures.

Fig. 2: Drain current ID versus drain voltage VDS at VGS = 0.7 V in DGMOS of different channel

lengths Lch. Crosses: Lch = 50 nm, closed circles: Lch = 25 nm, and open squares:

Lch = 15 nm.

Fig. 3: Conduction band versus distance x along the S-D direction in the on-state

(VGS = VDS = VDD) for the 15 nm-long transistor: Lch = 15 nm and LG = 25 nm for different

distances y from the gate: y = 0 nm for the Si/SiO2 interface and y = TSi / 2 for the body

center. The dotted lines indicate the S-D gate overlap, the x-axis origin corresponds to the

position of the source/channel junction.

Fig. 4: Velocity vx distributions calculated in the N+ source region and at different positions into

the 15 nm-long channel. For x>0, only electrons injected from the source are considered.

Fig. 5: Velocity vx distributions of electrons flowing from the source (solid lines) and electrons

flowing from the drain (dotted lined) calculated at the drain-end (x = 15 nm) of the

15 nm-long channel.

Fig. 6: Velocity vx distributions calculated at the drain-end (x = 15 nm) of the 15 nm-long channel

for electrons flowing from the source and undergoing either 2, 1 or 0 scattering events.

Fig. 7: Velocity vx distributions calculated at x = 3 nm from the source-end of the 15 nm-long

channel for electrons flowing from the source, and undergoing either 2, 1 or 0 scattering

events.

Fig. 8: Fraction of electrons flowing from S to D versus the number of scattering events Nscatt undergone during the channel crossing. Conventional bulk MOSFET (Single gate,

Lch = 50 nm) distribution: dotted line. DGMOS distributions: solid lines. Crosses:

Lch = 50 nm, closed circles: Lch = 25 nm, and open squares: Lch = 15 nm

Fig. 9: Intrinsic Ballisticity Bint at the drain-end versus channel effective length Lch.

Fig. 10: Effective ballisticity Beff = Ion / Ion_bal versus intrinsic ballisticity Bint in the 15 nm-long

device. Results are obtained by varying the intensity of the oxide roughness Cdiff or the

phonon scattering coefficient Kph.For Cdiff = 1 (respectively Cdiff = 0.14) and Kph = 0, 1, 2,

5, 10 and 20 (respectively Kph = 0, 0.5, 1, 2): open squares (respectively closed triangles).

For Kph = 1 (respectively Kph = 0) and Cdiff: 0, 0.07, 0.14, 0.21, 0.4, 0.7 and 1 (respectively

Cdiff = 0, 0.14, 0.5, and 1): open circles(respectively closed squares).

14

Fig. 11: Fluxes schematic and conduction band versus distance x along the S-D direction. Vertical

dotted lines indicate source/channel and channel/drain junctions.

Fig. 12: Back-scattering coefficient R(x) versus S-D distance x in the channel. Closed circles:

Lch = 25 nm and open squares: Lch = 15 nm.

Fig. 13: ID versus VGS at VDS = 0.7 V for different devices: ‘standard’ (closed circles), 'sta-bal’

(open diamonds), ‘bal-sta’ (open circles), and ’bal-bal' (closed squares). Inset: ID versus

VGS at VDS = 0.7 V

15

Tox = 1.2 nm

TSi = 10 nm ND = 5 × 1019 cm-3

G

S D

Lch

LG

Tox = 1.2 nm SiO2

G

ΦMG = 4.46 eV VDD = 0.7 V

ND =

5 × 1019 cm-3 NA =

2 × 1015 cm-3

SiO2

x

y

0

Fig. 1

Saint Martin et al.

16

0

500

1000

1500

2000

2500

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

D ra

in c

ur re

nt I D

(µ A

/µ m

)

Source-drain voltage VDS (V)

VGS = 0.7 V

50 nm

15 nm

25 nm

Fig. 2

Saint Martin et al.

17

-0.7

-0.6

-0.5

-0.4

-0.3

-0.2

-0.1

0

-20 -10 0 10 20 30 40

C on

du ct

io n

ba nd

(e V

)

Distance x along S-D (nm)

Si/SiO 2 interface

Body center

V DS

= V GS

= 0.7 V

Fig. 3

Saint Martin et al.

18

0

5 106 1 107

1.5 107 2 107

2.5 107 3 107

3.5 107

-6 107 -4 107 -2 107 0 2 107 4 107 6 107

N um

be r o

f e le

ct ro

ns (a

.u .)

Velocity vx along S-D (cm/s)

VDS = VGS = 0.7 VSource (x = -10 nm)

x = 1 nm 2 nm

4 nm 12 nm

Fig. 4

Saint Martin et al.

19

0

1 106

2 106

3 106

4 106

5 106

-8 107 -4 107 0 4 107 8 107

VDS = VGS = 0.7 V

Drain injected

Source injected

N um

be r o

f e le

ct ro

ns (a

.u .)

Velocity vx along S-D (cm/s)

x = 15 nm

Fig. 5

Saint Martin et al.

20

5000

1 104

1.5 104

2 104

2.5 104

0 2 107 4 107 6 107 8 107

N um

be r o

f e le

ct ro

ns (a

.u .)

Velocity vx along S-D (cm/s)

Ballistic

1 scatt.

2 scatt.

VDS = VGS = 0.7 V

x = 15 nm

Fig. 6

Saint Martin et al.

21

0

1 104

2 104

3 104

4 104

5 104

6 104

-6 107 -4 107 -2 107 0 2 107 4 107 6 107

N um

be r o

f e le

ct ro

ns (a

.u .)

Velocity vx along S-D (cm/s)

Ballistic

1 scatt.

2 scatt.

x = 3 nm VDS = VGS = 0.7 V

Fig. 7

Saint Martin et al.

22

1

10

0 5 10 15Fr ac

tio n

of S

-D e

le ct

ro ns

(% )

Number of scattering events Nscatt

VDS = VGS = 0.7 V

15 nm 25 nm 50 nm

SG BULK 50 nm

Fig. 8

Saint Martin et al.

23

0

20

40

60

80

100

0 10 20 30 40 50I nt

rin si

c ba

lli st

ic ity

B in

t ( %

)

Channel effective length Lch (nm)

VDS = VGS = 0.7 V

Fig. 9

Saint Martin et al.

24

30

40

50

60

70

80

90

100

0 20 40 60 80 100

B ef

f = I o

n / I o

n_ ba

lli st

ic (%

)

Intrinsic ballisticity Bint (%)

Standard channel Cdiff = 0,..,1

Channel without phonon, Cdiff = 0,..,1Cdiff = 1

Cdiff = 0.14

Fig. 10

Saint Martin et al.

25

Fig. 11

Saint Martin et al.

Entrance Surface

Exit Surface

I +Φ O+Φ

2I −Φ

1I −Φ

-0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1

0 0.1

0 10 20

C on

du ct

io n

ba nd

(e V

)

Distance x along S-D (nm)

V DS

= V GS

= 0.7 V

Entrance Surface

Exit Surface

I +Φ O+Φ

2I −Φ

1I −Φ

-0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1

0 0.1

0 10 20

C on

du ct

io n

ba nd

(e V

)

Distance x along S-D (nm)

V DS

= V GS

= 0.7 V

26

0

5

10

15

20

25

0 10 20 30 40 50

B ac

k- sc

at te

rin g

co ef

fic ie

nt R

(% )

Distance x along S-D (nm)

25nm

15nm

Fig. 12

Saint Martin et al.

27

0

500

1000

1500

2000

2500

0 0.2 0.4 0.6 0.8 1 1.2

D ra

in c

ur re

nt I D

(µ A

/µ m

)

Source-drain voltage VDS (V)

VGS = 0.7 V Bal-sta

Standard Sta-bal

Bal-bal

0

500

1000

1500

2000

2500

3000

10-2

10-1

100

101

102

103

104

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

I D (µ

A /µ

m )

VGS (V)

VDS = 0.7 V

Fig. 13

Saint Martin et al.

On the ballistic transport in nanometer-scaled double-gate MOSFET

Jérôme Saint Martin, Arnaud Bournel, Philippe Dollfus Institut d'Electronique Fondamentale, CNRS UMR 8622, Université Paris Sud, Bât. 220,

F-91405 Orsay cedex, France

Corresponding author

Jérôme Saint Martin Institut d'Electronique Fondamentale CNRS UMR 8622, Université Paris Sud, Bât. 220 F-91405 Orsay cedex, France Tel.: 33 1 69 15 40 37, Fax: 33 1 69 15 40 20, E-mail: stmartin@ief.u-psud.fr

Abstract

The scattering effects are studied in nanometer-scaled double-gate MOSFET, using Monte Carlo

simulation. The non-equilibrium transport in the channel is analyzed with the help of the

spectroscopy of the number of scatterings experienced by electrons. We show that the number of

ballistic electrons at the drain-end, even in terms of flux, is not the only relevant characteristic of

ballistic transport. Then the drive current in the 15 nm-long channel transistor generations should be

very close to the value obtained in the ballistic limit even if all electrons are not ballistic.

Additionally, most back-scattering events which deteriorates the ON current, take place in the first

half of the channel and in particular in the first low field region. However, the contribution of the

second half of the channel can not be considered as negligible in any studied case i.e. for a channel

length below 25 nm. Furthermore, the contribution of the second half of the channel tends to be

more important as the channel length is reduced. So, in ultra short channel transistors, it becomes

very difficult to extract a region of the channel which itself determine the drive current Ion.

Keywords

Silicon, SOI, MOS devices, MOSFETs, nanotechnology, semiconductor device modeling, Monte

Carlo methods, backscattering, ballistic transport.

2

1 Introduction

Double-gate MOSFET architecture (DGMOS) is a potential solution to overcome short channel

effects in the 65 nm ITRS node [1], that is for physical gate lengths smaller than 25 nm. In such

nano-transistors where the channel length is comparable to the electron mean free path, non

stationary [2] or even ballistic [3] transport is probably of great importance regarding the device

performance. This question is however rather controversial from a theoretical point of view.

According to Natori prediction [4], non stationary transport in ultra-small devices and statistical

fluctuation of random scattering events undergone by charge carriers in the channel should lead to

dramatic time fluctuations of drive current. According to Monte Carlo simulation results, small

variations in the number and the position of doping atoms in the channel of a 50 nm bulk MOSFET

significantly influence the transport properties and the drain current [5]. Ballistic transport in

undoped channel may be the solution to limit these types of fluctuations.

Lundstrom and coworkers have studied in detail the influence of ballistic transport in such

devices on the drain current [6],[7]. They developed models describing nano-DGMOS operating in

the ballistic or quasi-ballistic limit. These models are based on the concept of thermionic injection

from source-end into the channel. According to such approach, the electron velocity at the

source-end, and thus the drive current, should be limited by the "source-to-channel" energy barrier

and by back-scatterings in the low-field region, i.e. in the vicinity of the barrier.

Svizhenko and Anantram have also investigated the role of scatterings in nanometer-scaled

DGMOS [8]. Using a Green-function approach, they show that scatterings at both source- and

drain-end influence significantly the drain current.

More recently, Mouis and Barraud [9] have used the Monte Carlo simulator MONACO,

developed in our group, to discuss the evolution of the velocity distribution along the channel of a

DGMOS. Their work puts into evidence that some electrons can be reflected towards the channel

after having experienced Coulomb interactions in the highly doped drain region i.e. far from the low

field region.

To analyze in more details the transport in nanometer-scaled DGMOS, we study in this paper the

behavior of electrons injected from the source as a function of the number of scattering events

undergone in the channel. We have introduced in the device Monte Carlo simulator a procedure

which allows us to get accurate information on the number of scattering events experienced by each

electron during its travel between the source-end and the drain-end of the channel. We can also

extract at many positions in the channel the number of back-scattered electrons and the velocity

distributions of electrons having undergone either 0, 1, 2, … or N scattering events. We thus have

3

relevant information to investigate in detail in this paper the notion of ballisticity [6] and to discuss

the actual influence of both ballistic and backscattering effects on the device characteristics and

performance.

The paper is organized as follows. The simulated devices and the Monte Carlo model are

presented in Section 2 and the results are described and discussed in section 3. First, we analyze the

velocity spectra in the channel by highlighting the ballistic and quasi-ballistic phenomena, which

completes the work initiated in Ref.[9]. Then the notion of intrinsic ballisticity is defined and its

influence on drive current is evaluated by tuning the scatterings intensity in the channel. Finally, we

investigate the evolution of backscattering coefficient along the channel and, by separating the

channel in two parts with different scattering properties, we analyze the impact of the region where

scattering events take place on the I-V characteristics.

2 Model and simulated devices

The simulated DGMOS devices, described in Fig. 1, have an effective channel length Lch equal

to 10 nm, 15 nm, 25 nm or 50 nm and a gate lengths LG equal to 10, 25, 25 and 50 nm respectively.

However, the study is focused on 15 nm-and 25 nm-effective channel lengths. The SiO2 gate oxide

Tox and Si body thicknesses TSi are equal to 1.2 nm and 10 nm, respectively. The doping density is

ND = 5 × 1019 cm-3 in N+ S-D regions and NA = 2 × 1015 cm-3 in the body (P type). The N+ doping level is relatively low, and may induce a significant series resistance, but it is a realistic value

considering the difficulty of electrically activating dopants in such thin body. The N+/P junctions

are assumed to be abrupt. The work function of the gate material is 4.46 eV to achieve the

theoretical threshold voltage VT of 0.2 V. The power supply voltage VDD is fixed at 0.7 V to abide

by the 2007 ITRS specification.

A classical particle Monte Carlo algorithm is self-consistently coupled with a 2D Poisson solver.

The Poisson's equation is solved at each time step equal to 0.1 fs with standard boundary conditions

[10]. The number of simulated particles is typically 50000. The scattering mechanisms included in

the simulation are phonon scattering, impurity scattering and surface roughness scattering. The

acoustic intra-valley phonon scattering is treated as an elastic process and the intervalley phonon

transitions, consisting of three f-type and three g-type processes, are considered via either zeroth-

order or first-order transition matrix in agreement with selection rules [11]. The phonon coupling

constants given in Ref.[10] are used. The impurity scattering rate is derived from the screened

Coulomb potential with the momentum-dependent screening length given in Ref.[12]. The surface

roughness scattering is treated with an empirical combination of diffusive and specular reflections

4

which correctly reproduces the experimental universal mobility curve [13],[14]. Unless otherwise

indicated we have considered in this work a fraction of diffusive reflections Cdiff equal to 0.14.

The originality of the present work, in which quantum confinement effects are not included, lies

in the spectroscopy of the number of scattering events undergone by electrons crossing the active

part of the channel (or any predefined part of the device) and in the detailed study of the velocity

spectra of different electron groups. This makes possible the careful analysis of scattering effects.

Practically, a scattering counting region is predefined with an entrance surface and an exit surface.

Typically, the entrance surface is defined either at the source/channel junction or at the position of

the top of the gate-induced potential barrier, and the exit surface is placed at the channel/drain

junction. Each electron entering the counting region by the entrance surface is flagged and while it

remains inside this region the number of scattering events undergone is recorded. At the exit surface

we can thus separate electrons into different groups, corresponding to ballistic electrons,

once-scattered electrons, twice-scattered electrons, … N-times scattered electrons, respectively. Of

course these groups can be enumerated. Additionally, for each of these groups the energy and

velocity spectra are recorded. Such spectra may be also obtained at intermediate surfaces defined all

along the counting region, i.e. the channel. It should be noted that these spectra only include

electrons coming from the entrance surface and do not consider the electrons entering the counting

region by the exit surface. The information about back-scattered electrons is obtained separately: if

an electron is back-scattered after having crossed the counting region and re-enter this region by the

exit surface, its scattering-story is still recorded and the velocity spectra of such electrons are

recorded too. Thus, by possibly changing the place of the exit surface, we can have very detailed

information on all carriers participating in the source-drain current. It is very useful to quantify and

analyze the effects of ballistic, nearly-ballistic and back-scattered electrons.

3 Results

The variations of drain current ID as a function of source-drain voltage VDS obtained in the

15 nm-, 25 nm- and 50 nm-long transistors for VGS = VDD are shown in Fig. 2. For the 15 nm-long

channel, the Ion current at VDS = VGS = VDD = 0.7 V is 2140 µA/µm. Such a high value is related to

the aggressive scaling of the gate length, to the double-gate architecture and to the relatively thick

body film (Tsi = 10 nm). The drain conductance, and more generally the short channel effects

(SCE), are however rather strong in the 15 nm-long channel, as a consequence of the non optimized

body thickness. The 25 nm-long channel, better designed, is less sensitive to SCE while exhibiting

Ion-value (1600 µA/µm) still higher than the ITRS Roadmap specification that is 900 µA/µm. We

now examine the transport in these devices biased in the "on-state", i.e. with VGS = VDS = VDD.

5

3.1 Velocity spectra analysis

The conduction band profile and the corresponding velocity spectra evolution of electrons along

the 15 nm-long channel have been plotted respectively in Fig. 3 and Fig. 4. The Fig. 4 represents the

velocity spectra evolution calculated in the first sheets of cells under the front oxide, i.e. between

y = 0 and y = 1 nm (see Fig. 1). The transport is stationary in the highly doped source well, as

illustrated by the quasi-Gaussian shape of the velocity distribution in the middle of this region.

Other spectra are taken in the channel and, in contrast to the results shown in Ref.[9], they only

concern electrons injected by the source into the channel, excluding those which have been injected

or re-injected from the drain. These spectra are very different from stationary ones. In the vicinity of

the top of the barrier (x = 2 nm), the spectrum is quite similar to a hemi–Gaussian distribution.

However electrons with a negative velocity have not completely disappeared, which is the signature

of backscattered electrons. Thus, the regime of pure thermionic injection in the channel is not

completely reached in this structure. At a position x = 4 nm after the source/channel junction, two

separate distributions appear in the velocity spectrum. Two peaks can be distinguished, each of

them corresponding to electrons with either a transverse or a longitudinal conductive effective mass

along the source-to-drain direction. For greater distance x, we observe the propagation of these two

distributions along the channel, each one at its own velocity. The electron density decreases as the

average velocity increases in accordance with the current conservation and the distributions become

narrower and narrower.

At the drain-end of the 15 nm-long device, the velocity spectrum of electrons injected from the

source calculated at the Si/SiO2 interface looks like two very well-defined peaks as shown in Fig. 5.

According to energy conservation between the barrier region and the drain-end of the channel, it is

easily demonstrated that the peak velocity values in Fig. 5 correspond to ballistic electrons flowing

from source to drain with a transverse (mt = 0.19 m0 where m0 is the free electron mass) or a

longitudinal (ml = 0.916 m0) effective mass.

To complete the description of electron populations present in the channel, the spectrum of

electrons injected from the drain at x = 15 nm is also plotted in dotted line in Fig. 5. As confirmed

by a velocity spectroscopy distinguishing electrons with transverse mass from those with

longitudinal mass (not shown), this ‘drain injected’ distribution is made up of the sum of two nearly

Gaussian distributions. The widest corresponds to electrons with a transverse mass and the thinnest

to electrons with a longitudinal mass. Even if they are the most numerous at the drain-end, their net

contribution to the current at the drain-end is only 0.9%.

6

Besides, the number of “drain-injected” electrons rapidly decreases near the source. Indeed, at

x = 15, 12, 4 and 1 nm, the part of “drain injected” electrons represents respectively 72.3%, 42.3%,

5.6% and 1.7 % of the electrons. At the source-end (x = 1 nm) 6.3% of electrons which have a

negative velocity have been injected from the highly-doped drain region.

Those first analyses of velocity evolution along the channel indicate clearly that the transport is

completely out of equilibrium. Furthermore, the velocity spectrum of “source-injected” electrons at

the drain-end (Fig. 5) suggests that ballistic electrons have an important role in the transport for

nanometer-scaled DGMOS devices. In this connection, we present a new kind of analysis which

allows dissecting the velocity spectra presented in Fig. 5. In this purpose, we have defined a

counting region by an entrance surface and an exit surface located at the source/channel and

channel/drain junctions, respectively. This study does not consider “drain-injected” electrons.

The Fig. 6 represents the velocity distributions calculated at the drain-end of the 15 nm-long

channel device for electrons flowing from the source by undergoing 2, 1 or 0 scattering events in

the counting region. Contrary to the spectra presented above (i.e. in Fig. 4 and Fig. 5) and calculated

at the Si/SiO2 interface, these new distributions are calculated on the full body thickness, i.e.

between y = 0 and y = Tsi (cf. Fig. 1). A 2D effect is clearly observed in Fig. 6: the ballistic peaks

are wider than those presented in Fig. 5 as they result from the sum of peaks with different maxima.

Indeed, as shown in Fig. 3, the potential drop varies as a function of the channel depth, i.e. along the

y-axis (cf. Fig. 1), which makes the ballistic peak position dependent on y. For instance, according

to the potential drop plotted in Fig. 3, the transverse mass peak velocity is equal to 6.5 × 107 cm/s at

y = 0 nm and to 7.0 × 107 cm/s at y = Tsi/2. We also verify that the distribution tails observed in Fig.

5 are caused by electrons which have undergone interactions during their channel crossing.

Moreover, the shapes of the quasi-ballistic velocity spectra, i.e. the spectra obtained for electrons

which have undergone 1 or 2 interactions, is similar to that of a ballistic spectrum. This suggests

that devices can drive on-current Ion very close to the limit value obtained for a ballistic channel,

even if all electrons are not purely ballistic at the drain-end.

The Fig. 7 represents the same kind of velocity distributions for “source injected” electrons but

calculated near the top of the injecting barrier, i.e. at x = 3 nm from the source-end. This figure

allows us to verify that only ballistic electrons are purely thermionically injected since their velocity

spectrum looks like a strict hemi-Gaussian distribution. Additionally, the contribution of electrons

which have undergone more than two scatterings in the channel is negligible (not shown). At

VDS = 0.7 V, the backscattered population is essentially formed by once– and twice–scattered

electrons.

7

3.2 Influence of the ballisticity on drive current

To get a more quantitative insight into ballistic and/or quasi-ballistic transport, we have

calculated the number of scattering events Nscatt experienced by each carrier crossing the counting

region, i.e. the channel. The resulting electron distribution is plotted in Fig. 8 as a function of Nscatt

for DGMOS of different effective channel lengths: 15 nm, 25 nm and 50 nm. For comparison, we

have also indicated, in dotted line the similar distribution obtained at the drain-end for a

conventional 50 nm bulk MOSFET with the following characteristics: a single gate, a channel

length Lch = 50 nm, an oxide thickness Tox = 1.2 nm, a junction depth Xj = 20 nm and a highly

doped channel NA = 1018 at/cm3. The latter distribution is a bell-curve with a maximum centered on

Nscatt = 5, which corresponds to the ratio Lch/λeff where Lch is the effective channel length and λeff is

defined as an effective mean free path. On the contrary, in all lightly-doped DGMOS the group of

ballistic electrons is the most populated, but it forms the majority only in the 15 nm-long channel.

This indicates that the effective mean free path λeff is smaller than the channel length in all

simulated DGMOS. However, the fraction of electrons decreases as Nscatt (the number of

scatterings) increases and the distribution spreads out when Lch increases. For the 15 nm-long

channel, the curve is a pure exponential function. In both 50 nm-long channel devices (DGMOS

and bulk) the interaction spectrum tends to a more stationary-like one.

Now, we define the intrinsic ballisticity Bint as the percentage of electrons which are ballistic at

the drain-end (Nscatt = 0 in Fig. 8) [15]. The line in Fig. 9 is an interpolated curve obtained by

linking the intrinsic ballisticity Bint for the 3 different DGMOS with channel effective length

Lch equal to 15 nm, 25 nm and 50 nm and by assuming Bint = 100% for Lch = 0. The smooth curve

obtained from these three DGMOS seems particularly relevant because for Lch = 10 nm it gives the

same result as the complete Monte Carlo simulation of a 10 nm-long device (closed circle). From

such a semi classical Monte Carlo approach, one can thus estimate that ballistic electrons should be

largely predominant (i.e. with an intrinsic ballisticity Bint greater than 90%) only for channel lengths

smaller than about 3 nm. However, quantum transport effects have to be considered for

investigating Lch-values smaller than 10 nm [7], which may modify this prediction.

To determine how the intrinsic ballisticity Bint at the drain-end is related to the on-current Ion, we

have artificially modified it from 0 as in full stationary transport to 100% as in pure ballistic

transport. The quantity Ion_bal stands for the on-current obtained for a ballistic channel, that is

without any phonon or roughness effects. Then, we study the effective ballisticity Beff, defined as

Beff = Ion/Ion_bal [6], as a function of the intrinsic ballisticity Bint at the drain-end.

8

To this end, we have varied the oxide roughness coefficient Cdiff from 0 to 1 and introduced a

phonon scattering coefficient Kph in the phonon scattering rates: all standard values are multiplied

by the coefficient Kph varying from 0 for a ballistic channel, to 20 for a very resistive channel. The

evolution of Beff as a function of Bint for a given Cdiff and for different phonon scattering

coefficients, is plotted in dotted line in Fig. 10. The results obtained for Cdiff = 1 (respectively

Cdiff = 0.14) and for Kph = 0, 1, 2, 5, 10 and 20 (respectively Kph = 0, 0.5, 1, 2), are indicated with

open squares (respectively closed triangles). The results obtained for various roughness coefficients

Cdiff and for a given phonon coefficient Kph are shown in solid lines: on the one hand for Kph = 1 and

Cdiff = 0, 0.07, 0.14, 0.21, 0.4, 0.7 and 1 (open circles), and, on the other hand, for Kph = 0 with

Cdiff = 0, 0.14, 0.5, and 1 (closed squares).

These results show an overall view of the effect of each kind of interaction. For an intrinsic

ballisticity Bint greater than 20% the effect of each type of interaction (phonon or roughness

scattering) yields a linear behavior Beff(Bint) but with a slope depending on the type of interaction.

Below this limit, when the transport is more stationary, the Beff(Bint) relation is no more simply

linear. Thus there is neither an equality nor a unique linear relation between the effective ballisticity

Beff and the intrinsic ballisticity Bint. The effective ballisticity Beff (in terms of current) alone does

not provide enough information to quantify accurately the intrinsic ballisticity Bint. However, there

is no denying that these two quantities are strongly correlated. Besides, we notice that the intrinsic

ballisticity Bint is always overestimated by the effective ballisticity Beff: for instance Bint = 52% and

Beff = 84% for the standard DGMOS (Kph = 1 and Cdiff = 0.14).

3.3 Back-scattering localization and its effect on the current

To investigate further the effects of scattering, a flux approach may be used [6]. We propose to

identify the part of the channel giving the highest contribution to back-scattering effects which are

known to degrade the drive current. So, a control volume is defined, as shown in Fig. 11, by an

entrance surface and an exit surface that can be moved along the channel. By calculating, the

different fluxes: ΦI+, ΦO+, ΦI1- and ΦI2-, the relevant back-scattering coefficients may be extracted.

Fluxes ΦI+ and ΦO+, oriented as indicated in Fig. 11, are the ingoing flux at the entrance surface and

the outgoing flux at the exit surface, respectively. The flux ΦI1- represents the flux of electrons

which have entered the control volume by the entrance surface and which have crossed back the

entrance surface without having crossed the exit surface. The flux ΦI2- represents the flux of

electrons which have entered the control volume by the entrance surface and which have crossed

back the entrance surface after having crossed the exit surface.

9

The back-scattering coefficient R(x) of the control volume between the entrance surface located

at the top of the barrier and an exit surface located at the distance x from the source end is equal to

ΦI1-(x)/ΦI+. The evolution of R(x) obtained by moving the exit surface along the channel from the

top of the barrier to the drain–end is plotted in Fig. 12 for 15 nm- and 25 nm-long devices. The

maximum back-scattering coefficient Rmax for the whole structure including the drain contact can be

calculated as Rmax = max[R(x)] = [ΦI1-(x) + ΦI2-(x)]/ΦI+.

First, we remark that at the drain-end of the 25 nm–long channel the backscattering coefficient

R(Lch) is slightly lower than Rmax: Rmax = 1.035 × R(Lch). Moreover, the difference between Rmax and R(Lch) strongly increases when Lch decreases to 15 nm: Rmax = 1.195 × R(Lch). This difference, due to electrons back-scattered in the drain, may be more accurately estimated by taking into

account other scattering mechanisms, as for instance, the short range electron-electron interactions.

A monotonous variation of R(x) along the channel is observed. The increase of R(x) is much

more important in the first half part of the channel as the back-scattering events take place mainly

there. However this is less true when the effective channel length Lch decreases:

R(x = Lch/2) = 0.91 × R(Lch) for Lch = 25 nm, while R(x = Lch/2) = 0.81 × R(Lch) for Lch = 15 nm. In the first part of the channel the increase of R(x) is rather uniform, just a bit faster in the

vicinity of the top of the barrier. Thus, it is difficult to accurately define in the first channel half a

region which has a predominant impact in terms of back-scattering.

It has been suggested in Ref. [6] that the backscattering coefficient R can be estimated at the

position x in the channel where the potential drops by kT/q. Our results show that this definition

leads to a significant underestimation of R. For instance, consider the position x0 where the

potential drops by at least 50 meV on the full channel thickness. We have then x0 = 9.5 nm for

Lch = 25 nm and x0 = 6.3 nm for Lch = 15 nm. At this position x0, the backscattering coefficient is

only R(x0) = 0.86 × R (Lch) = 0.78 × Rmax. for Lch = 25 nm. This underestimation is more

pronounced for Lch = 15 nm: R(x0) = 0.73 × R(Lch) = 0.61 × Rmax.

This trend confirms that reducing Lch gives more importance to the second half of the channel in

terms of contribution to backscattering. Indeed, the number of scattering in ultra-short channel is

not important enough (cf. Fig. 8) to prevent electrons scattered in the second half of the channel

from being backscattered to the source–end [8].

To analyze the influence on drain current of scattering in the different parts of the channel, new

simulations have been performed by changing the scattering properties along the channel. Three

new 25 nm-long DGMOS called ‘bal-bal’, ‘bal-sta’ and ‘sta-bal’ have been simulated. In these

devices, the channel is divided in 2 equal parts. The term ‘sta’ stands for the channel–half with

10

standard scattering properties of Si doped to 2×1015 at/cm3: standard phonon scattering coefficient

Kph = 1 (see Sec. 3.2) and roughness coefficient Cdiff = 0.14. The term ‘bal’ stands for the ballistic

channel–half (without any scattering). So, for instance, ‘sta-bal’ is a DGMOS with a standard first

channel–half and a ballistic second channel–half.

The inset of Fig. 13 shows on both linear and logarithmic scales the evolution of drain current ID

as a function of gate voltage VGS at VDS = VDD. First we notice that all devices have the same

threshold voltage VT ≈ 0.3 V. Moreover, the subthreshold behavior is not degraded by SCE, the

subthreshold slope being equal to 70 mV/dec. The characteristics only differ in the

transconductance gm = GSVhighGS

D V I∂∂ above the threshold voltage VT and, as a consequence, in the

on-current Ion.

The variations of drain current ID as a function of source-drain voltage VDS obtained in these

DGMOS at VGS = 0.7 V are shown in solid lines in Fig. 13. In linear regime, the resistance

Ron = DSVlowD

DS I

V∂∂ extracted from Fig. 13 simply follows the Ohm’s law: Ron is determined by the

average channel conductivity and has the same value for ’bal-sat’ and ‘sta-bal’. It is greater for the

standard channel and lower for the ballistic channel. In the saturation regime, with nearly the same

saturation drain voltage VDSsat for all devices, the following values of output conductance

0.7GS D

D V V DS

IG V = ∂= ∂ are extracted at VGS = 0.7 V: GD_BB = 360 µS/µm for ‘bal-bal’,

GD_BS = 340 µS/µm for ‘bal-sta’, GD_SB = 270 µS/µm for ‘sta-bal’, GD_SS = 190 µS/µm for ‘standard’. We observe that the more ballistic the channel is, in particular in its first half, the greater

the conductance is. Thus, we deduce that the lack of scattering degrades the saturation behavior. It

should be mentioned that similar results have been obtained for bulk MOSFET of higher channel

length, i.e. 50 nm and 180 nm (not shown). Low Ron, i.e. high channel conductivity, is beneficial for

the on-current Ion, and a weak conductivity, in particular the first channel–half, is suitable to get a

low output conductance GD.

Besides, if each channel part had the same influence on drive current, ‘sta-bal’ and ‘bal-sta’ on–

current would be equal to the average on-current between ‘standard’ and ‘bal-bal’, i.e. equal to

1980 µA/µm. As the on-current Ion is equal to 2000 µA/µm for ‘bal-sta’ and 1850 µA/µm for ‘sta-bal’, we deduce that the ballistic channel part location, i.e. the scattering localization, has an

influence on the on-current.

This result is not fully inconsistent with the conventional belief that the first channel-half has a

greater importance than the second channel-half in terms of backscattering coefficient. However, in

11

contrast with this belief, our study reveals that the second part of the channel has a significant

impact on device operation and performance. Indeed Ion is 15% greater for ‘sta-bal’ than for

‘standard’ and it is 14% greater for ‘bal-bal’ than ‘bal-sta’. It should be mentioned that the intrinsic

ballisticity is higher for ‘sta-bal’ (61%) than for ‘bal-sta’ (53%), the resulting Ion is higher for the

latter device. Moreover the relative difference between ‘sta-bal’ and ‘bal-sta’ is low:

Ion(bal-sta) - Ion(sta-bal) = 4.4% × (Ion(standard) - Ion(bal-bal)). At last, the evolution of the

backscattering coefficient along the channel (Fig14) suggests that this impact of the second

channel–half should be still stronger for smaller channel length.

4 Conclusion

We have investigated in detail the velocity spectra of the electrons present in the channel as a

function of their origin: “source-” and “drain-injected” and of the number of the experienced

scatterings. We show that the transport in the 15 nm-long channel is neither purely ballistic nor

purely thermionic.

Nevertheless, ballistic electrons are of great importance in nanometer-scaled double-gate

MOSFET. In nano-DGMOS, the ballistic limit (intrinsic ballisticity Bint = 100% i.e. 100% of

electrons injected from the source are ballistic at the drain-end) is far to be reached for channel

lengths larger than 10 nm: the intrinsic ballisticity is about 50% in 15 nm-long channel. However,

the drive current Ion is closer to the value Ion_bal obtained with a pure ballistic channel: the ratio

Ion/Ion_bal, i.e. the effective ballisticity Beff, is more than 80%. Then the number of ballistic electrons,

even in terms of flux as we defined the intrinsic ballisticity Bint, is not the only relevant

characteristic of ballistic transport. There even if the "ballistic limit” is still “a mere pipe dream”

considering the numerous types of significant interactions in nano-scaled structures [16], the

on-current in next transistor generations should be very close to this limit.

Even if our results are not fully inconsistent with the conventional belief that most back-

scattering take place in the first half of the channel, they show that the role of the second half of the

channel cannot be considered as negligible for a channel length lower than 25 nm. Furthermore the

contribution of the second half to back-scatterings tends to be more and more important as the

channel length is reduced. As a consequence, it becomes more and more difficult to extract a

particularly significant region of the channel which would determine the value of the drive current

Ion.

This work is supported by the French RMNT project CMOS-D-ALI and we thank Emile

Grémion for his contribution.

12

References

[1] International Technology Roadmap for Semiconductors: 2002 Update, http://public.itrs.net/

[2] T. Mizuno and R. Ohba, “Experimental study of nonstationary electron transport in sub- 0.1 µm metal-oxide-silicon devices: Velocity overshoot and its degradation mechanism”, J. Appl. Phys., vol. 82, pp. 5235–5240, 1997

[3] S. E. Laux and M. V. Fischetti, "Monte Carlo simulation of submicrometer Si n-MOSFET’s at 77 and 300 K”, IEEE Electron Dev. Lett., vol. 9, pp 467–469, 1988

[4] K. Natori, "Ballistic metal-oxide-semiconductor field effect transistor", J. Appl. Phys., vol. 76, pp. 4879–4890, 1994

[5] P. Dollfus, A. Bournel, S. Galdin, S. Barraud, and P. Hesto, "Effect of discrete impurities on electron transport in ultra-short MOSFET using 3D Monte Carlo simulation", cond-mat/0310718, to be published in IEEE Trans. Electron Devices

[6] M. Lundstrom and Z. Ren, "Essential physics of carrier transport in nanoscale MOSFETs", IEEE Trans. Electron. Dev., vol. 49, pp. 133-141, 2002

[7] J.-H. Rhew, Z. Ren, and M. S. Lundstrom, "A numerical study of ballistic transport in a nanoscale MOSFET", Solid-State Electron., vol. 46, pp. 1899-1906, 2002

[8] A. Svizhenko and M. P. Anantram, "Role of scattering in nanotransistors", IEEE Trans. Electron Devices., vol. 50, pp. 1459-1466, 2003

[9] M. Mouis and S. Barraud, "Velocity distribution of electrons along the channel of nanoscale MOS transistors", in Proc. ESSDERC'2003 (IEEE), 2003, pp. 147-150

[10] P. Dollfus, "Si/Si1-xGex heterostructures: electron transport and field-effect transistor operation using Monte Carlo simulation", J. Appl. Phys., vol. 82, pp. 3911-3916, 1997

[11] T. Yamada, J. R. Zhou, H. Miyata, and D. K. Ferry, "In-plane transport properties of Si/Si1-xGex structure and its FET performance by computer simulation", IEEE Trans. Electron Devices, vol. 41, pp. 1513-1522, 1994

[12] N. Takimoto, "On the screening of impurity potential by conduction electrons", J. Phys. Soc. Jpn., vol. 14, pp. 1142-1158, 1959

[13] E. Sangiorgi and M. R. Pinto, "A semi-empirical model of surface scattering for Monte Carlo simulation of silicon n-MOSFET", IEEE Trans. Electron Devices, vol. 39, pp. 356-361, 1992

[14] S. Keith, F. M. Bufler, and B. Meinerzhagen, "Full-band Monte Carlo device simulation of an 0.1µm n-channel MOSFET in strained silicon material", in Proc. ESSDERC'97 (Editions Frontières, Paris), 1997, pp. 200-203

[15] It must be highlighted that the intrinsic ballisticity is a percentage in terms of positive current flux and not of electron density.

[16] M. V. Fischetti, "Scaling the MOSFETs to the limit: A physicist's perspective", Journal of Computational Electronics, vol. 2, pp. 73-79, 2003

13

Figure captions

Fig. 1: Schematic of DGMOS structures.

Fig. 2: Drain current ID versus drain voltage VDS at VGS = 0.7 V in DGMOS of different channel

lengths Lch. Crosses: Lch = 50 nm, closed circles: Lch = 25 nm, and open squares:

Lch = 15 nm.

Fig. 3: Conduction band versus distance x along the S-D direction in the on-state

(VGS = VDS = VDD) for the 15 nm-long transistor: Lch = 15 nm and LG = 25 nm for different

distances y from the gate: y = 0 nm for the Si/SiO2 interface and y = TSi / 2 for the body

center. The dotted lines indicate the S-D gate overlap, the x-axis origin corresponds to the

position of the source/channel junction.

Fig. 4: Velocity vx distributions calculated in the N+ source region and at different positions into

the 15 nm-long channel. For x>0, only electrons injected from the source are considered.

Fig. 5: Velocity vx distributions of electrons flowing from the source (solid lines) and electrons

flowing from the drain (dotted lined) calculated at the drain-end (x = 15 nm) of the

15 nm-long channel.

Fig. 6: Velocity vx distributions calculated at the drain-end (x = 15 nm) of the 15 nm-long channel

for electrons flowing from the source and undergoing either 2, 1 or 0 scattering events.

Fig. 7: Velocity vx distributions calculated at x = 3 nm from the source-end of the 15 nm-long

channel for electrons flowing from the source, and undergoing either 2, 1 or 0 scattering

events.

Fig. 8: Fraction of electrons flowing from S to D versus the number of scattering events Nscatt undergone during the channel crossing. Conventional bulk MOSFET (Single gate,

Lch = 50 nm) distribution: dotted line. DGMOS distributions: solid lines. Crosses:

Lch = 50 nm, closed circles: Lch = 25 nm, and open squares: Lch = 15 nm

Fig. 9: Intrinsic Ballisticity Bint at the drain-end versus channel effective length Lch.

Fig. 10: Effective ballisticity Beff = Ion / Ion_bal versus intrinsic ballisticity Bint in the 15 nm-long

device. Results are obtained by varying the intensity of the oxide roughness Cdiff or the

phonon scattering coefficient Kph.For Cdiff = 1 (respectively Cdiff = 0.14) and Kph = 0, 1, 2,

5, 10 and 20 (respectively Kph = 0, 0.5, 1, 2): open squares (respectively closed triangles).

For Kph = 1 (respectively Kph = 0) and Cdiff: 0, 0.07, 0.14, 0.21, 0.4, 0.7 and 1 (respectively

Cdiff = 0, 0.14, 0.5, and 1): open circles(respectively closed squares).

14

Fig. 11: Fluxes schematic and conduction band versus distance x along the S-D direction. Vertical

dotted lines indicate source/channel and channel/drain junctions.

Fig. 12: Back-scattering coefficient R(x) versus S-D distance x in the channel. Closed circles:

Lch = 25 nm and open squares: Lch = 15 nm.

Fig. 13: ID versus VGS at VDS = 0.7 V for different devices: ‘standard’ (closed circles), 'sta-bal’

(open diamonds), ‘bal-sta’ (open circles), and ’bal-bal' (closed squares). Inset: ID versus

VGS at VDS = 0.7 V

15

Tox = 1.2 nm

TSi = 10 nm ND = 5 × 1019 cm-3

G

S D

Lch

LG

Tox = 1.2 nm SiO2

G

ΦMG = 4.46 eV VDD = 0.7 V

ND =

5 × 1019 cm-3 NA =

2 × 1015 cm-3

SiO2

x

y

0

Fig. 1

Saint Martin et al.

16

0

500

1000

1500

2000

2500

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

D ra

in c

ur re

nt I D

(µ A

/µ m

)

Source-drain voltage VDS (V)

VGS = 0.7 V

50 nm

15 nm

25 nm

Fig. 2

Saint Martin et al.

17

-0.7

-0.6

-0.5

-0.4

-0.3

-0.2

-0.1

0

-20 -10 0 10 20 30 40

C on

du ct

io n

ba nd

(e V

)

Distance x along S-D (nm)

Si/SiO 2 interface

Body center

V DS

= V GS

= 0.7 V

Fig. 3

Saint Martin et al.

18

0

5 106 1 107

1.5 107 2 107

2.5 107 3 107

3.5 107

-6 107 -4 107 -2 107 0 2 107 4 107 6 107

N um

be r o

f e le

ct ro

ns (a

.u .)

Velocity vx along S-D (cm/s)

VDS = VGS = 0.7 VSource (x = -10 nm)

x = 1 nm 2 nm

4 nm 12 nm

Fig. 4

Saint Martin et al.

19

0

1 106

2 106

3 106

4 106

5 106

-8 107 -4 107 0 4 107 8 107

VDS = VGS = 0.7 V

Drain injected

Source injected

N um

be r o

f e le

ct ro

ns (a

.u .)

Velocity vx along S-D (cm/s)

x = 15 nm

Fig. 5

Saint Martin et al.

20

5000

1 104

1.5 104

2 104

2.5 104

0 2 107 4 107 6 107 8 107

N um

be r o

f e le

ct ro

ns (a

.u .)

Velocity vx along S-D (cm/s)

Ballistic

1 scatt.

2 scatt.

VDS = VGS = 0.7 V

x = 15 nm

Fig. 6

Saint Martin et al.

21

0

1 104

2 104

3 104

4 104

5 104

6 104

-6 107 -4 107 -2 107 0 2 107 4 107 6 107

N um

be r o

f e le

ct ro

ns (a

.u .)

Velocity vx along S-D (cm/s)

Ballistic

1 scatt.

2 scatt.

x = 3 nm VDS = VGS = 0.7 V

Fig. 7

Saint Martin et al.

22

1

10

0 5 10 15Fr ac

tio n

of S

-D e

le ct

ro ns

(% )

Number of scattering events Nscatt

VDS = VGS = 0.7 V

15 nm 25 nm 50 nm

SG BULK 50 nm

Fig. 8

Saint Martin et al.

23

0

20

40

60

80

100

0 10 20 30 40 50I nt

rin si

c ba

lli st

ic ity

B in

t ( %

)

Channel effective length Lch (nm)

VDS = VGS = 0.7 V

Fig. 9

Saint Martin et al.

24

30

40

50

60

70

80

90

100

0 20 40 60 80 100

B ef

f = I o

n / I o

n_ ba

lli st

ic (%

)

Intrinsic ballisticity Bint (%)

Standard channel Cdiff = 0,..,1

Channel without phonon, Cdiff = 0,..,1Cdiff = 1

Cdiff = 0.14

Fig. 10

Saint Martin et al.

25

Fig. 11

Saint Martin et al.

Entrance Surface

Exit Surface

I +Φ O+Φ

2I −Φ

1I −Φ

-0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1

0 0.1

0 10 20

C on

du ct

io n

ba nd

(e V

)

Distance x along S-D (nm)

V DS

= V GS

= 0.7 V

Entrance Surface

Exit Surface

I +Φ O+Φ

2I −Φ

1I −Φ

-0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1

0 0.1

0 10 20

C on

du ct

io n

ba nd

(e V

)

Distance x along S-D (nm)

V DS

= V GS

= 0.7 V

26

0

5

10

15

20

25

0 10 20 30 40 50

B ac

k- sc

at te

rin g

co ef

fic ie

nt R

(% )

Distance x along S-D (nm)

25nm

15nm

Fig. 12

Saint Martin et al.

27

0

500

1000

1500

2000

2500

0 0.2 0.4 0.6 0.8 1 1.2

D ra

in c

ur re

nt I D

(µ A

/µ m

)

Source-drain voltage VDS (V)

VGS = 0.7 V Bal-sta

Standard Sta-bal

Bal-bal

0

500

1000

1500

2000

2500

3000

10-2

10-1

100

101

102

103

104

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

I D (µ

A /µ

m )

VGS (V)

VDS = 0.7 V

Fig. 13

Saint Martin et al.

Comments