IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. ED-25, NO.

3, MARCH 1978

ACKNOWLEDGMENT
The authors would like to thank F. Hasegawa and A.
Higashisaka for developing the power GaAs Schottkybarrier gate FETs and K. Ayaki for his encouragement
and guidance throughout this study.

REFERENCES
[ l ] C. A. Liechti, Microwave field-effect transistors--1976, IEEE
Trans. Microwave Theory Tech.,vol. MTT-24, pp.279-299, June
1976.
[2] W. Shockley, A unipolar field-effect transistors,Proc. IRE, vol.
40, pp. 1365-1376, Nov. 1952.
[3] P. Wolf, Microwave properties of Schottky-barrier field-effect
transistors, IBM J. Res. Develop., vol.14, pp. 125-141, Mar.
1970.
[4] K. E. Drangeid and R. Sommerhalder, Dynamic performanceof
Schottky-barrier field-effect transistors, IBM J . Res. Develop.,
vol. 14, pp.82-94, Mar. 1970.
[5] A. B. Grebene andS. K. Ghandhi, General theoryof pinched opSolid-state Electron.,vol. 12,
eration of the junction-gate FET,
pp. 573-589, July 1969.
of the junc[6] D. Mo and H. Yanai, Current-voltage characteristics
tion-gate field-effect transistor with field-dependent mobility,
IEEE Trans. Electron Devices,
vol. ED-17,pp. 577-586,Aug.
1970.
[7] H. Statz, H. Haus, andR. A. Pucel, Noise characteristics
of gallium
arsenide field effect transistors,IEEE Trans. Electron Devices,
vol. ED-21, pp. 549-562, Sept. 1974.
[SI K. Lehovec and R. S. Miller, Field-distributionin junction fieldeffect transistors at large drain voltages, IEEE Trans. Electron
Devices, vol. ED-22, pp. 273-281, May 1975.
[9] T. Sugeta, M. Ido, and M. Uchida, Design theoryof a GaAs-SBG.

337

FET with considerationof the electron velocitysaturated region,

Trans. Inst. Electronics Commun. Eng. [Link]. C, vol. J59-C,
pp. 107-114, Feb. 1976.
small signal
[IO] P. L. Hower and N. G. Bechtel, Current saturation and
characteristics of GaAs field-effecttransistors, IEEETrans.
Electron Devices, vol. ED-20. pp. 213-220, Mar. 1973.
[ l l ] D. P. Kennedy andR. R. OBrien,Computer aided two-dimensional
analysis of the junction field-effect transistor,IBM J. Res. Develop., vol. 14, pp.95-116, Mar. 1970.
[12] B. Himsworth, A two dimensional analysis of Gallium Arsenide
junction field-effect transistors with long and short channels,
Solid-state Electron.,vol. 15, pp.1353-1361, Dec. 1972.
of junction gate
[13] K. Yamaguchi andH. Kodera, Drain conductance
FETs in the hot
electron range,IEEE Trans. Electron Devices,
vol. ED-23, pp.545-554, June 1976.
[I41 K. Yamaguchi, S. Asai, and H. Kodera, Two-dimensional numerical
analysis of the stability criteria for GaAs FETs, IEEE Trans.
Electron Devices, vol. ED-23, pp. 1283-1290, Dec. 1976.
FET model fordc, ac, and
[I51 M. Reiser, A two dimensional numerical
large signal analysis,IEEE Trans. Electron Devices,vol. ED-20,
pp. 35-45, Jan. 1973.
[IS] G. D. Alley and H. E. Talley, A theoretical studyof the high-frequency performance of a Schottky-barrier field-effect transistor
fabricated on a high resistivity substrate,IEEE Trans. Microwave
Theory Tech.,vol. MTT-22, pp.183-189, Mar. 1974.
[I71 J. Sone and Y. Takayama, Analysisof field distributionsof GaAs
MESFET at large drain voltages, Electron. Lett., vol. 12, pp.
622-624, Nov. 1976.
[18] S. M. Sze, Physics of Semiconductor Devices. NewYork and
London: Wiley,1969.
[I91 A. Higashisaka,[Link], Y. Takayama,and F. Hasegawa,
Microwave power GaAs FET, presented a t the 1977 National
Convention of the Institute of Electronics and Communications
Engineers of Japan.
[20] G. D. Vendelin and M. Omori, Try CAD for accurateGaAs MESFET models, Microwaues, pp. 58-70, June 1975.

Subthreshold Conduction inMOSFETs

GEOFFREY W. TAYLOR, MEMBER, IEEE

Abstract-The dependence of channel currentin subthreshold

operation upon drain, gate, and substrate voltages is formulated
in terms of a simple model. The basic results are consistent with
earlier approaches for long-channel devices. For short-channel
devices, the variation of current with drain voltage up to the
punch-through voltage is accurately described. The threshold
voltage of a short-channel device as a function
of applied voltages
follows as natural
a
result of the derivation. Results are presented
which confirm the theory over a wide range of drain and gate
voltages. Withthe application of substrate bias it is concluded from
the data and the theory that two-dimensional effects can cause
dramatic increases in the
drain conductance.

Manuscript received March 7,1977; revised November 3,1977.

The authoris with Texas Instruments,
Inc., Dallas,TX 75222.

LIST OF

SYMBOLS

Gate to source
voltage.
Punch-through voltage.
Flat-band voltage.
Substrate tosource voltage.
Drain tosource voltage.
Built-in voltage of the source to
substrate diode.
Surface potential (referred to the
equilibrium bulk conduction band
edge).
Surface potential (referred to the
substrate).

0018-9383/78/0300-0337$00.75 0 1978 IEEE

IEEE TRANSAC'IQNS 3 N ELECTRON DEVICES, VOL. ED-23,

Intrinsic concentration.
Doping concentration in thebulk.
Effective charge density at the
surface.
Effective surface state density.
Electron density in the channel
at the
source.
Electron density normal to
the
surface.
Electron density parallel to the
surface.
Change in electron density.
Electron concentration in the source.
Arbitrary energy level in the energy
gap.
Dielectric constant of silicon.
Electronic charge.
Oxide capacitance.
Temperature.
Thermal velocity.
Effective spread of charge a t surface.
Distance normal to surface.
Distance parallel to surface.
Channel widthof MOS device.
Channel lengthof MOS device.
Various lengths used to partition
charge in the MOSchannel.
Width of drain depletion region a t
the surface.
Width of source depletion region a t
the surface.
Surface density of free charge.
Electron mobility.
Electron diffusion constant.
Diffusion component of current
density.
Drift component of current density.
Current density.
Total device current.
Trapped charge concentration.
Electron Fermi level.
Electric-field variable.
Voltage variable.
Conduction band edge.
Position in channelof equal current
components.
Electron diffusion length.
Punch-through current(surface
component).
Punch-through current (bulk
component).
Length associated withfringing
fields.
Length of transition from diffusion to
drift.
Average value of cps.
Bulk charge of silicon for a longchannel device.

IC
i'

NU. o,

lyln.w--_

Depletion constant.
Short-channel
factor.

I. INTRODUCTION
HE EXPERIMENTAL andtheoretical characteriza-

r rtion of the MOS transistor hasbeen well established

for conduction above the conventional threshold voltage
and for moderately long-channel devices [1]-[3]. However,
many of the currentapplications of MOS and related [Link] depend critically upon current flowing in the device
in the normally off or subthreshold region of operation.
The interestin this mode of conduction arises not only in
determining design limitations on packing densities and
refresh times in high-density dynamic circuits
but also as
the principal mechanism inthe design of critical circuits,
e.g., the control of threshold voltages for inputs toa CCD
analog signal processor or of blooming currents in CCD
imagers.
Several articles have appeared inthe literaturewhich
[4]-[8] (sometimes
dealwithsubthresholdconduction
referred to as the weak inversioncondition) and almostall
of the work has considered only the variation of drain
current with gate voltage. In long-channel devices, the
variation of channel currentwith drain voltage is of little
concern because the characteristics are virtually constant
with drain voltage. However, as the channel length is reduced in many practical applications, the drain current
exhibits asignificant dependence onthe drainvoltage. As
the drain and source depletion regions extend into the
channel, avery rapid rise of current is observed as the device enters the punch-through state. The
mechanisms of
conduction in the voltage range up to andincluding the
punch-through point have received little attention and
generally are notwell understood.
In this paper, we will formulate the problem of subthreshold conduction on the same assumptions as those
of previous authors, but using a more direct approach. The
results are shown to be consistent with the former work
and also to include the effects of drain and substrate
voltage in a straightforward manner. Furthermore, the
occurrence of punch-through in short -channel devices
arises naturally from the theoretical derivation. All of the
features of the theoreticalmodel are verified by a close
comparison with experimental data.

11. THEORY
The currentis derived as a functionof drain voltage for
both small and large drain bias over a wide range of gate
and substratebias.
A. Surface Potential Relations
1 ) Dependence of Surface Potentialupon Gate, Drain,
and Substrate Bias: The approach taken here will be
similar in some respects to a former derivation of the

CONDUCTION
dOR: SUBTHRESHOLD
c

xd

F---Ilves

ns

/&-E&v&

NSS (SURFACE
STATES)

E2

G S

339

IN MOSFETS

SOURCE

POTENTIAL
(POSITION OF FERMILEVEL
IN SOURCE R E G I O N 1

nb

Fig. 1. Energy diagram for the MOS channel with a substrate bias applied, in region 11.

tial would be expected to be a continuous function of y

along the channel. The assumption we will make is that we
can still define a unique surface potentialwhich itself is
used to determine the fractionof charge associated with
the source, drain, and gateelectrodes. The charge in the
channel is assumed to be divided into regions I, 11,and I11
as depictedin Fig. 2(a) by the chain-dottedboundaries. In
I and I11 it is assumed that all field lines terminate on the
source and drain, respectively, and hence one-dimensional
I1
solutions in the y direction may be used. Similarily, in
it is assumed that all field lines terminate on the gate
electrode, allowing a one-dimensional solution in the x
direction and yielding a single value of cps, The boundaries
between I, 111, and I1 are the loci y ( x ) determined (for
vertical junctions)by the condition

K-ly2 = & (1

,
\\

- )

(la)

xd

or (for a circular radiusof curvature) by the Condition

is the depletion width in11(see Fig. l),N A is the bulk acceptor

doping for n-channel devices, and & = & VBSis
(b)
the
surface
potential referred to thebulk conduction band
Fig. 2. Cross-sectional viewof MOS device showing the division of
charge between the source, gate, and drain electrodes for a shortedge. With a substratebias applied, therefore, & may take
channel device (a) before [Link] occurrence of punch-through, (b) after
negative values but & will always be positive. The locus
punch-through in the bulk occurred.
has
( l a ) is astraight line and thelocus (lb) is approximately
so and hence we may use the trapezoidal representation
in Fig.2. To be consistent with the regional approximation,
threshold voltage in short-channel IGFETs[9]. The sur- it is assumed that thelength of the gate effective in terface potential +s will be measuredwith respect to thebulk minating thebulk electric field corresponds to thewidth
conduction band edge for VBS= 0 as shown in Fig. 1and of 11 (the distanceLz) plus the effects of fringing fields.
we will concern ourselves only with depletion conditions Relating the totalcharge on the gate electrode to the total
in the range of surface potentials between flat-band con- charge in I1 we have
ditions (& = 0) and the stronginversion condition (& =
2 4 ~ where
)
4~ is the bulk Fermi potential. In the accu- VGS - VFBmulation condition, where it has been well established that
$J~
varies logarithmically with the gate voltage, the subthreshold current becomes exceedingly small and very
difficult to measure, and hence of little interest. For >
Z+F the device is in the onor above threshold condition
and herealso & varies logarithmically withthe gatevolt- where 6 accounts for the effects of fringing at theoverlap
age. Hence the observable subthreshold currents are de- of the source and drain and L2 (see Fig.2(a)) is the distance
termined by the normal depletion range of surface po- along the surface between the boundaries of I and 111.
tentials.
Using the abrupt depletion approximation
in I and 111, it
In a long-channel device, a one-dimensional charge
is expressed
analysis may be usedto relate the charge on the gate to the
L2 = L - dK(Vbi - 4 s ) - d K ( v b i + VOS - 4s). (3)
charge in the silicon substrate. However, ina short-channel
device, a significant fractionof the electric-field linesas- Similarily, L1 and L3 are the widths of I1 and 111,respecsociated with the gate charge may be terminated on the tively, at the depthof Xd below the surface and may be
source and drainjunctions. Therefore, the surface poten- determined by geometric analysisto be
IVBS

340

IEEE "FtANSACTIONS
DEVICES,
ELECTRON
ON

VOL. ED-25, NO.

I //

where rj is the radius of curvature of the junctions (see Fig.

2(a)) andalso corresponds to theoverlap of the gate on h e
source and drain. In
( 2 ) ,Nss is a uniformdensity of suditce
states (acceptor-like) extending
over an energy range E1
to E2 in theenergy gap asshown in theenergy diagramof
Fig. 1. Substituting (4)and (3) into ( 2 ) yields

3, MARCH

/'

Fig. 3. Variation of surface potential with gate voltage for a short- and
long-channeldevice showingthe effect of drain and substrate
voltages.
The parameters VFB= 0, N A = 6 X lOI5 cm-3, and to, = 500 8, were
used.

where

,.

f L + 26 - dK(Vbi

- 6s) - dK(Vbi + VDS - &)]

(54

Since theeffect of fringing fields at thesource and drain

as determined by the
size of
6 = to, ln

[2-J

and

F=
is small (to, is the oxide thickness) becausegenerally t o K
<< L , then itwill not be included in the subsequent derivation. It will be noted, however, that when Lz << L, then
(7d)
LSand 6 become comparable and the importance
of 6 may
become significant. For the simplified
case of source and
drain junctions with vertical
sides (or the equivalent
con. Because F is a weak function of 6 an iteration may be made
using (7a,d) to obtain a more accurate value of &. The
dition, rj >> d K ( & VBs)),(5) reduces to theform
results of such calculations are shown plotted
Fig.in3 for
VGS- VFB- &
various combinations of gate lengths and device biases.
The curves show that for a long-channel device, the qjS
versus VGcurve is independent of drain bias but that in
a short-channel device a drain voltage dependence will
exist. In addition, the
influence of the drainvoltage onthe
versus VGcurve is considerably stronger when a sub(6:) strate bias is applied.
When the substratebias has beenincreased to thevalue
For large values of L , the expression in the braces re- where L' = 0, then (6) yields the value
duces to 2 and thefamiliar result is obtained. For smaller
values of L , a solution for all ranges of qjs cannot be
[Link], we are generally interested in $s
only over a small range, say 0.35 < dS < 0.7. Hence the
term dK(vbi - &) will be small compared toL and the
-- qNA J K ( A + VBS) (8)
term t / K (Vbi VDS- $2will depend very weakly onL
2CO
and so we may use an average value $ 0.5. Solving (6)
yields
where VBS is determined by the condition of punch-

$s

= VG

+ Vo - v'(VG + V o ) z - V s + 2VOvBs

(7a)

through in the bulk(L' = 0) which is

TAYLOR SUBTHRESHOLDCONDUCTIONINMOSFETS

or, solving for

34 1

VBS,

In thiscase $s is givenby (7a), with(6) changed according

to (8),and is no longer a functionof substrate bias because
punch-through of the source and drain depletionregions
has occurred in the bulk. As indicated in Fig. 2(b), the gate
now controls a triangular section of charge which increases
in size as the gate bias is increased,but which decreases in
size as the drain bias is increased. The area of the triangle
decreases for drain voltages in excessof that which satisfies
(9) because the potential barrier at the apex of the triangle
is reduced as VDSis increased. The surface potential may
be described then by (8) with 7 ~replaced
s
by $D(VDS)
where C$D is the drain-voltage-dependent energy barrier
Fig. 4. Variation of threshold voltage with channel length in a shortin the bulk in the punch-through condition. The details
channel MOSFET for various substrate and drainbias [Link]
parameters used were VFB= -0.7, N A = 1 X 10l6,rj = 0.5 pm, and to,
of the conduction process in this potential configuration
= 500 A.
will be discussed elsewhere [12].
2) Determination of Threshold Voltage: The threshold
condition is achieved when C$s = 2$F, which follows directly from (6) and is (in theabsence of surface states)

(loa)

It will be noted that this condition could also be stated

using the more complicated form ( 5 ) and in this case if we
assume vbi = 2$F, and VDS= 0, we obtain a result identical to thatof Y au [9], which is
V T = VFB+ 2 $ ~
1

+ci
L - 5Ld 2 K

NUMERICAL S3LilTION
APPROXIMATE SOLUTICN

(2d:

vBs)

+ 11

(lob)

where QB is the bulk silicon charge. Using (loa) and the

same conditions (i.e., deep junctions andzero drain voltage) we obtain

LiUl

Fig. 5. Variation of threshold voltage with oxide thickness and channel

length andcomparison withnumerical solutions. The
calculations were
made forN A = 5.6 X 10 ~ r n - ~VFB
, = 0 , and rj = 0.5 pm.

combined with (9a)to give the rather simple form

The threshold voltage is also a functionof drain voltage

according to
Since a current will flow in thedevice
in this
and the current will be a function of vDs due
to channel length modulation effects, then a particular
currentat a givenvoltage cannotbe
used as a measure of
VT. The voltage is measured most easily by the extrapolation to zero of the 6versus VGcurve of the device in
qNA
VT = v m 2 4 +~__ d i K ( 2 4 ~ VBS). (11) thesaturation region above threshold.
2CO
The threshold
function
isaplotted
as
of channel length
Since, generallyvbi z 2C$F, then for VDS= 0, (11)may be in Fig. 4 for two values of drain voltage and threedifferent

As the substratebias is increased, punch-throughin the

bulk of the device will occur as described. For greater
substrate biases, the charge in the silicon in 11will remain
reach a constant
unchanged and the
value which is obtained from ( 8 ) as

342

JEEE rRANSA(TI0NS ON ELECTRONDEVICES,

VOL. ED-25, NO.

3 , MARCH 1978

device for two different values of drain voltage. The experimental points are taken from device
a
having the indicated [Link] fitbetween the data and theory
is excellent exceptfor the deviation nearV S S= 10 V and
VDS = 10 V. In this region of bias it is expected that
punch-through effects are occurring and thetheory would
require modifications.
3 ) Other Assumptions: An important resultof previous
work [4]-[8] is that the channel current in subthreshold
conduction is a diffusioncurrent in contrast to the
above
threshold case in which drift is the dominantmechanism.
For gate voltages such that the surface is in normal depletion, the dominantspace charges in the channelregion
are theionized acceptor ions. Hence, variationsin electron
density at thesurface havenegligible effect uponthe surface potential. Eventhe presence of a uniform densityof
surface states causes only a slight variation
of surface potential
along
the
channel.
As
a
result,
the
field
along the
2
4
LIUI
channel, dt$,/dy, is very small and the subthreshold drift
Fig. 6. Variation of threshold voltage with substrate bias for several
current becomes much smaller than the subthreshold
substrate doping concentrations and
two drain biases. A substrate bias
of 1V was assumed and the parametersVFB= 0, to, = 500 A, and rj
diffusion current. This resultis consistent withthe basic
= 0.5 Nm were used.
approximation used herein for the treatmentof the subthreshold [Link] was the case in the determination
of t$*, the device will bedivided into thesame three regions
as shown by the energy diagram in Fig. 8. Region I, in
which all field lines are assumed parallel to thesurface,
extends from 0 < y < ys at thesurface, and region 111in the
same fashion extends fromy c < y < L. In region I1 ( y , <
y < y c ) ,the assumption of field lines only inthe x direction
with a current flowing in the device is the gradual channel
,approximation commonly used in the derivation of on
state characteristics. It will be noted in Fig. 8 that the
effect of the drainvoltage isto move the boundary yc in the
direction of the source and also to change the value of the
barrier between 1and I1 because of the two-dimensional
sharing of charge between the gate and the drain as pre,I
,2
,3
viously
discussed. In Fig. 8(b) the energy diagramis shown
Jvss-Jvv2)
between the source and drainaway fromthe surface at the
Fig. 7. Dependence of the threshold voltage upon substrate voltage for
edge of the gate depletionregion (along the line L in Fig.
long- and short-channel devices and two values
of drain [Link]
both theory and experiment N A = 8 X
VF, = 0, and rj = 4 X lob5 2(a)) for short-channel
devices.
urn.
Most of the discussion to datehas considered the subthreshold conduction only for the case of long-channel
values of substrate bias. It is apparent that both the draindevices or small sourceto drainvoltages. In thesederivaand [Link] bias have a strong effect in reducing the tions, the analysis ignores the effect of recombination in
threshold voltage. Fig. 5 shows similar behavior for two the channel and
follows the generalized theory introduced
[IO]in which the concept of a quasi-Fermi
different oxide thicknesses and compares these with points by Sah and Pao
calculated from a two-dimensional numerical calculation. level and the normalized field function F ( V,k,Vp) are
The agreement of the approximate model
is quite good for employed where V,E,V, are the electrostatic potential, the
L 5 2 pm. For L 5 2 pm, punch-through has occurred in electron quasi-Fermi potential, andthe equilibrium pothe bulk for the case of VDS= 10 V, resulting in the dis- tential in the bulk all normalizedto kT/q. Although this
crepancies shown. Fig.6 shows the resultsof the model for approach is rigorously correct, it does not provide the
various substrate dopings withdifferent drainvoltages and reader with much physical insight
or an interpretationof
a fixed value of 1V on the [Link] expected, the effect the approximations which are used to obtainan analytical
diminishes for higher substrate doping levels. Finally, Fig. form. In this paper, the subthreshold currentis derived
7 displays the dependence
of the threshold on the square from a somewhat more direct approach. The theory deroot of the substratebias for along- and a short-channel scribes more accurately the region of conduction in the

TAYLOR:SUBTHRESHOLDCONDUCTION

IN MOSFET'S

343

statement of quasi-equilibrium itmay be relatedto conditions in the substrate and thus

n ( y , ) = n, = n(xd)eq@JkT

(13)

where 4, is determined from(7) and n ( x d ) is the electron

concentration at theedge of the gate depletion region, as
noted in Fig. 1 and explained below. Alternatively, (13)
may be written as a statementof quasi-equilibrium between the source diodeand the channel as
n ( y , ) = n, = nSDe-q(Vbi-4s)/kT
(14)
(bl

where n S D is the concentrationof electrons in thesource

diode. It willbe noted that n ( x d ) is smaller than the
equilibrium minority carrier density,
npo,in the substrate
bulk whena substratebias is applied. This mustbe the case
subthreshold regime for larger drain voltages which cor- since the effect of substrate bias is to reduce the electron
responds to theabove-threshold saturation region. In this concentration from npo to n ( x d ) a t all points on the periphery of the source diode depletion region, due to the
region, for short-length devices, the channel-length modpresence of a finitereverse-biased current flowing fromthe
ulation effects normally encountered in the "on" state, and
substrate to the
source. Sincethe gate depletionregion is
two-dimensional effects in the channelbecome significant
coupled to thesource depletionregion via the channel, it
in the subthreshold state. For short-channeldevices the
follows that n is also reduced to n ( x d ) along the edge of the
phenomenon of punch-through occurs before avalanche
gate depletionregion.
breakdown in the drainjunction. The voltage and current
When the drainbias is increased from zero, region I11
at the onset of punch-through are derived as a natural
increases and region I1 decreases in width as y c moves
extension of the subthreshold theory.
towards the source. This situationis depicted in Fig. 8(a).
effect of drain
Also region I is reduced in widthdue to the
B. Dependence of Subthreshold Current upon Drain
bias
on
the
surface
potential
(cf.
(7)). However the electron
Voltage
density at y s is maintained at thevalue of (14) due to the
For the MOS device in the subthreshold region of op- quasi-equilibrium condition at thesource-channel junceration there are differentmodes of conduction, each of
tion. The drain-voltage-dependentc$s is taken as constant
which determine the channel currentfor a particularrange
in I1 in keeping withour basic assumptionthat region I1
of gate and drainvoltage (Vcs and V D S )The
. variation of
is free of a y component of field until punch-throughoccurrent asa function of drain voltage is of particular in- curs.
terest as the length of the channel is reduced. For very
For voltages of the order of a few k T l q , the free-electron
small drain voltages the channel diffusion current is of the
density at the channel to draindiode boundary is described
order of the thermalequilibrium drift currentflowing from by Boltzmann statistics andwe may say
the drain into the channel. Hence the current
varies exponentially withthe drainvoltage in the same fashion as
in a reverse-biased junctionat low drain voltages. As the
drain voltage is increased, the depletion region associated
This boundary conditionis identical tothat used for the
with the drain(Region I11 in Fig. 8(b)) extends towards the
minority carrier density in a p-n junction (incase,
this n,
source and the current exhibits
a significant dependence is the equilibrium minority carrier density on the p side)
on drain voltage. This behavior is considered in subsection
when a reverse bias is applied, which in this case is the
11-B1. If the channel length is small,enough punchdrain to channel junction.
through of the source and drain depletion
regions (region
As discussed in Section11-A,the currentis a diffusion
I and region 111) occurs at a voltage VDS= Vptbefore av- current and thus the distribution
of free carriersalong the
alanche breakdown has occurred at the drain. The conchannel is given by the solution of the familiar continuity
ductioncharacteristic at punch-through is derived in
equation inregions of small electric fields
subsection 11-B2 wherea novel approach is introduced to
remove the anomaly of an infinite punch-through current.
1 ) Drain Voltages Below Punch-Through,VDs < V p t : subject to theboundary conditions (14) and (15).An inFor VDS= 0, the potentialprofile is indicated in Fig. 8(a). teresting observation that can be made fromthe descripThe free electron density inthe channel at thesource is
tion of the conduction both physically [ l l ] and matheequal to its valueeverywherein the channel; from a
MOS transistor in the
matically is that conduction in the
Fig. 8. MOS device and associated channel potential in the subthreshold
region (a) at the
surface, (b) below the surface.

844

IEEE TRAKSACTIONS ON ELECTRON DEVICES, VOL. ED-25, NO. 3, MARCH

1978

subthreshold region is identical to that of the bipo1a.r The subthreshold currentis then determinedat theedge
transistor in the on state. The
role of the base of the bi- of the draindepletion region to be
polar and thegate or substrate
of the MOST, however, are
not totallycomparable. The effect of recombination in the
channel region isincluded by the second term in (16)where
L, is the diffusion length of electrons in the channel given
Substituting ( 2 2 ) and ( 2 1 ) into ( 2 3 ) we obtain
by

L, = V m

(1711

and D is the diffusion constant of electrons inthe channel.

The surface lifetime 7 is constant with position in the
channel, because of the constantsurface potential andis
given by

tanh

( - E- )

where 6, is determined by VG according to (7).

where a d is determined as follows. To find the total current
For the typical surface state densities encountered in
it is necessary to consider the total charge available for practical MOS devicesL, > 10 pm; hence, for deviceswith
conduction in the surface depletion region. By integrating L < 10 pm the term
in the x direction over the surface depletion region the total
charge Q, is obtained as

If we make use of ( 2 5 ) and, in addition,make the subwhere ? ( x ) is measured in the same way as & (see Fig. l), stitutions in ( 2 4 ) of the relations
and this integral is performed approximately in the apN A N p b = n;e94F/kT
(26)
pendix to yield1
from the substratebulk, and
nsDe-9Vbi/kT = nie-9@~/kT
(27)
Qn N 4
n,.
(20)

d y(g+. .
2q2N*

1)

at thesource to channel boundarywe obtain

A simple interpretation [7] of (10) is that thefree charge

available for conduction is confined to a distancefrom the
surface of
tkT

xd =

(21)

2q2N,4

($+

1)

This resultis very close to thatobtained by Barron,

using
the concept of a quasi-Fermilevel and the standard intewhich is constant because 3, is consideredconstant.
gration that is used to obtain the characteristics in the
Hence, we are assuming that we can represent the free ubove-threshold region. The approach used here is equally
charge in the x direction by an average value over a dis- a u rigorous in that thesame final results are
attained and
tance f d . We are also assuming that thesurface states in- yet the derivation relates more directly to thephysics of
fluence all of this charge and thus behave as a volume the conduction mechanism and indicates more clearly the
density of defects with a densityN , , / f d .
physical limitations of the approximations that are inFrom the solution of ( 1 6 ) )the free-carrier density vari- troduced in order to obtain a close-form solution. Furation, An = n, - n along the channel at thesurface is given thermore, we have taken account of the location of the
drain and source depletion
region to channel boundaries,
by
a necessary consideration in the evaluation of the subthreshold current.
For small drain voltages (several k T / q in size) the value
o f 6, is constant and the drain
voltage variation of ( 2 8 ) is
exponential in nature in correspondence with previous
findings. For voltages VDS2 3 k T / q , this exponential
variation rapidly disappears and the variationof Yd and
l Actually inthe region ys < y < L1 (see Fig.2) the integral (19) extends
p , become dominant. In this range of voltage ( 2 8 ) may be
from 0 to x where x is given by(1).The result will not be changed, how,mitten,using ( 3 ) and (7a) as
ever, because of the dominance of the surface contribution.

TAYLOR: IN
SUBTHRESHOLD CONDUCTION

345

MOSFETS

of drift anddiffusion currents throughout the

device; that
is
dV
dx

where

Io =

qZD~dnie-q~F/kTe-qVBs/kT

- Z/K(vbi - 6 ) - d K ( v b i + VDS- 4)

(29b)

dn
qDdx

qpn-.

This condition is not valid [12] for reverse-bias voltages

2 3 k T l q because of the dominance of the driftcomponent
in the depleted
region. By using a methodof regional apy c and the
proximations, the boundary conditions at
channel currentat theonset of punch-through canbe determined more accurately
as follows.
A device in which punch-through can be achievedwill
have a channel sufficiently short that thecondition (25)
applies. In thiscase, recombination in the channel
is un(6) simplifies to
important and the continuity equation

and F is given by (7d).It will be noted that(29) shows a

substantial dependenceon drain voltage through thefactor
F. Of particular interest is the presence of the crossproduct
of the drain and substrate
voltages; hence as the substrate
bias is increased, the dependence of the currentupon drain in region 11,where
voltage will also be expected to increase. In general, the
J = I/Z$d.
(324
dependence of the current on drain voltage will be a
function of oxide thickness (through C,) and the substrate In thework that follows it will be notedthat J is anegative number because the currentis negativein sign. Using
doping (throughK ) .
2 ) The Approacht o Punch-Through: As indicated in the boundary condition(14) with (32) yields
Fig. 8, the position yc = L - Y d moves towards thesource
J
n=n,+(y-y,)-.
as the drain depletion
region expands. Thislocation corqD
responds t o the pinchoff point normally referred to in
In
region
I11
it
is
assumed
that
the
electric field is deabove-threshold operation. However, unlike the oncase,
termined
by
the
solution
of
Poissons
equation in the y
the channel potential a t pinchoff, VDSAT, is approximately
direction.
The
negative
space
charge
in
region I11 consists
zero and the total appliedvoltage is taken upacross the
of
the
ionized
donors
and
the
charge
in
trapsa t the surface
depleted region. It will be noted from (18) that asYd in(surface
states)
which
have
been
filled
due
to thepresence
creases with increasing drain voltage the channel current
of
a
finite
current.
The
charge
in
surface
states
need only
will increase. If the channelis sufficiently short that the
be
considered
if
the
surface
state
density
is
comparable
to
condition
the acceptor density andalso if the quasi-Fermilevel for
L-y,-yd=O
(30) electrons inregion I1 is positioned between E1 and E2 as
T

can be achieved without the occurrence of avalanche shown in Fig. [Link] equation at thesurface in region
breakdown in the draindepletion region, then we have the I11 is then expressed
condition of punch-through between the source and drain.
Using the punch-through condition (30) in (28) yields the
anomalous result of an infinite punch-through current.
The inaccuracies in (28) are now discussed and modifica- where
tions introduced to
provide a continuouscurrent from the
NA=NA+p
(33b)
subthreshold to the punch-through condition.
and Epn is the position of the quasi-Fermilevel in region
For voltages larger than several k T l q the boundary
condition (15) at thedrain is no longer valid.
This situation [Link] the electron density in this region is decreasing
is identical to thatwhich exists at theedge of the depletion as the drainis approached, thenEpn is a function of y.
region in a reverse-biased p-n [Link] both the p-n However, it will be shownthat n is not a strong function
MOS device (for voltages much less than of y and since Epn a In ( n )then we will use an average
junction and the
punch-through) the current [see (28)] remains approxi- constant value of p for this purpose as indicated
in (33b).
mately correct because the terme-qvDslkT becomes much The boundary conditionto be used with(33) is
less than 1 and hence has almostno effect on the resulting
6, = 0, Y = y c
(34)
(15) is
value of current. However, the boundary condition
which is in accord withthe assumption that the potential
based uponthe fact that thermal equilibrium conditions
continue to apply to the junction under
reverse bias which in thechannel is constant in region I1 (i.e., they component
is expressed mathematically by the approximate balance
of field is neglected in 11).From (34) and (33), the
electric

1RANSACTIONS
IEEE

field in region 111is

ON ELECTRON
DEVICES,

VOL. ED-25, NO.

3, MARCH 1978

continuity of the electron density and its gradient yat=

9 or
Throughout mostof region 111the currentis a drift current and
(as we will show) and thus
q p n & = J.

(36)

Using (36) and (3.9, the free-electron densityis

n. =

-J

From (41), (32), (32b), (38), and (37) we obtain the results
(3 7)

Differentiation of (37) gives the gradientof free chargs

in I11 to be
d-n-

(38.1

Inspection of (37) supports our assumption that n ( y ) is

not astrong function of y except for values of y close to ye
and hence that p is approximately a constantin I11 since
(from 33)

where

Nt is a density of traps as definedby (39), andN c is the

effective density of states in the conduction band. Also, the
is calculated from (38) to
diffusion current in 111,JDIFF,
be

and so it is clear from (35) and (40)that except for y N- yc

the diffusion component is negligible [Link], close
to y c , (36) and (37) are no longer valid and an adjustment
must be made in the
boundary betweenI1 and I11 as faras
the free carrier density is [Link] (34) it is evident
that the currentis entirely carried bydiffusion a t yc. As
G increases for y > yc, the diffusion component decreases
and the drift component
increases. A t some position j i =
y e A, the drift anddiffusion components become equal.
To locate this position we will make the approximation
that the currentis entirely carried by
diffusion for y < 9
and by drift for y > [Link] 37, the drift anddiffusion components are equal and the
value of A is determined by the

and

It will benoted that (42) is identical to theextrinsic Debye

length. Then, asone would expect intuitively, the transition fromdiffusion to driftoccurs over one Debyelength.
Since the total voltage is dropped across the depletion
region according to (34) then, expressing y c - y s as in (3)
the channel currentis obtained from (43) and (3) to be

It will be noted from (28) that the exponential term

may be included in the numerator of (37) to make thisform
valid for the total range
of drain voltage.
A t the punch-through condition (at the surface), the
punch-through voltage, Vpt,is determined by the solution
of (3) for the conditionL2 = 0 and the current is determined bythe substitutionof the same relation into (37) to
yield
qZDnie-qF/kTeq~pt/kT
Ipt =
(45)

2d35q

where &,t is the value of 4s when punch-through occurs

,at the surface.2
For the corresponding condition of punch-through in
the bulk of the device, the punch-through voltage, V p t S ,
is given by the solution of (9a) for VDS,and the corresponding currentmay be determined in a like manner to
lot!

IptB

q Z D ( r . - xd)nie-qF/kTe-qiiBs/kT

(46)

In a practical case, punch-through in the bulk of the device, avalanche

in tLle drain region, or turn-on of the channel would usually occur before
pur ch-through at thesurface.

347

TAYLOR: SUBTHRESHOLD CONDUCTION IN MOSFET'S

It will be appreciated that this current

will be very small
compared t o the other componentsflowing, since the energy barrier to carrierinjection may be quite large (with
substrate bias applied) asindicated in Fig. 8(b). However,
the current that
flows in this region of the device for voltages greater than V p tis~a punch-through current since
it is at this voltage that the mechanism of carrier flow
changes to that characteristic of punch-through.
For anyposition x between the bulkedge of the depletion region associated with I1 and thesurface, the current
a t punch-through will be given by
(45) with the value of &
(see (44))replaced by 4 ~the
, value of potential at theapex
of the trianglein Fig. 2(b). It will also be appreciated that
the punch-throughvoltage for any position x between the
bulk and the surface
will be a functionof gate voltage.
At this point we are now in a position to evaluate the
electric field in region I1 and toverify, in a self-consistent
way, that the drift current
is indeed negligible compared
to thediffusion component. From(2) we may calculate the
electric field in I1 approximately as
(47)
In calculating (47), we haveused G, = A & / A y , and
q N s s E l in (7) has been replaced by q n ( y ) z d . We have
neglected the square-root term and then
A& is obtained
by evaluating (7)a t each end of 11. Using (44) the ratioof
the drift to the
diffusion term in I1 may be expressed approximately as
JDRIFT
5-

JDIFF

and this term<<1until ns N

4' n s f d
hTCo
NA.

111. EXPERIMENTAL

Fig. 9. Variation of subthreshold current with drain voltage for several

different gate lengths. The thin dashedlines are theoretical curves.

TABLE I
n-Channel Process Sequence
field implant with SisN4/SiOz masking
source-drain-gate region
selective field oxidation
nitride etch
gate threshold adjustment implant (boron)
energy = 120 keV
dose = 1.75 X 10"
polysilicon deposition and etch
source-drain predeposition withPH3 as source
CVD oxide of 5000 A
drive-in anneal in0 2 a t 105OOC
open contacts
AI/Si metallization and cut

A. Device Structure and Fabrication

The measurements were performed on devices having

W = 7.5 pmand L = 10.0,7.5, or 5.0 pm as definedby the

photomask. The IGFET's were fabricated on (100)p-type,
10 s2 cm siliconsubstrates. Thefield region wasimplanted
with boron using Si3N4as a mask and followed by selective
field oxidation. The transistorshave polysilicon gates and
self-aligned drain and source regions. The process sequence is listedin Table I.

B. Measurements
The currentis measured inthe source lead as shown
Fig. 2 so that the channel current alone is obtained. A
number of devices were compared and several different
curves were obtained as shown in Fig. 9. One currentvoltage curve chosen to be well within the subthreshold
region is displayed for each device. Device 1 (nominal
length = 10.0 pm)has a resultant
gate length of L = 9.0 pm
and shows avalanche breakdowna t VDS= 26 V. Device 2
has aneffective length of 6.5 pm and shows punch-through
or avalanche at VDSN 26 V. Devices 3-4 (nominal length

7.5 ym) and device 5-6 (nominally 5.0 pm) show considerablevariation in theirsubthresholdcharacteristics.
I-V charDevices 3,4, and5 show distinct
a
break in the
acteristics a t 12,8, and 4V, respectively, which is associated with the approach to punch-through as explained
later on. Device 6 is probably in the punch-throughcondition for all drain voltages. It is evident from thesecurves
that for a short-channeldevice, the amountof etching of
the polysilicon gate and the extent
of side diffusion of the
source and drain diodes under the gate can causevariations
inin gate length which can have dramaticeffects on the I-V
characteristics in the subthreshold region. Using the effective gate lengthsof 9.0,6.5, 5.0,4.5, and4.0 pm a theoretical curve is provided for devices using the value Of f l A
as discussed later for device 3. Device 6 cannot be compared with the theory because of the influence of the
punch-through current.
Device 3 was chosen for more completecharacterization.
A family of channel current versus drain voltage curves
with gate bias as a parameteris displayed in Fig. 10 and

34,8

IEEETRAXSACTIONSONELECTRONDEVICES,

0.2

0.4

3, MARCH 1978

VOL. ED-25, NO.

0.6

OB

t0

12

1.4

x (U)
Fig. 11. Calculated substrate doping profile for a processed device.

16

12

20

24

28

VDS IVOllS)

Fig. 10. Variation of subthreshold current with drain voltage for VBS
= 0 having gate voltage as a parameter showing the transition
to the
punch-through condition. The theoretical curves are the thin solid
lines.

compared with the theoretical result (44). The theory is

quite successful in matching the data,
even to the extent
of predicting the distinct
changes in slope observed onall
curves in the vicinity of 18 V. On closerexamination of (7d)
and (29) it is found that dFldVDs becomes more rapid
when the condition

v k ( v b i -I-VDS- I$)

VGS(VI

L - 2-

Fig. 12. Variation of subthreshold current with gate voltage having drain
voltage as a parameter; the exponential dependence on gate voltage
[Link]
points.

is achieved. This condition implies that punch-through at

the surface will soon occur. Calculationsshow that
punch-through at thesurface will occurat -26 V whereas
punch-through in the bulk
will occur a t -21 V. This curfit the data in
Fig. 12,
rent is not observable, however, since it is dominated by values of substrate bias. In order to
the subthreshold surface component. In the fitting
of the a distribution of surface stateswas used withthe value of
data, avalue of mobility of y = 600 cm2/V s, determined Nss = 5 X 10 cm-2 eV-l. Using these parameters andVFB
from the on-stateconductivity, was used to determine the == -1.4 V the datain Fig. 12 can befit very closely and it
electron diffusion constant. A value of N A = NA = 2.0 X is difficult to detect significantdifferences between the
1015 wasused as anaverage value chosen
to represent the theory and the experiment in the exponential region of
doping profile shown in Fig. 11which was computed for current (no attempt was made here to fit the nonlinear
Howthe channel implant and
processing used. The discrepan- regions which correspond to the on conduction).
cies which appear as a displacementof the current with ever, the data in Fig. 10 show in more detail the inadegate bias for VGS= -0.23 V and -0.136 V are discussed quacies of the assumptionof a uniformdensity, [Link]
below in connection withFig. [Link] calculating the theo-gate voltages VGS= -0.23 and -0.136, the predicted
6 (cf. current is smaller by about 60 percent from the measured
retical curve, the effects of fringing as represented by
(5b)) were included where Lz is calculated according to value. This resultsuggests that the defect density
or the
11)
is
decreasing
for
increasing
energy
doping
level
(cf.
Fig.
(3).
The corresponding channel currentversus gate bias with in theenergy gap. For additionalincreases in gatevoltage
[Link]),the fitbetween theory and experiment
drain bias as the parameter
is illustrated in Fig. 12 for two (-0.001,

4-/y k

112

VDS( V )

349

MOSFET'S

TAYLOR: SUBTHRESHOLDCONDUCTIONIN

16

210

'

Fig. 13. Drain voltage dependence of the subthreshold current with a

substrate bias applied. The theoretical predictions are thin
solid
lines.

is again very [Link] reason for this result

could bethat
the defect density hasbegun to increase withenergy for
higher energies in the gap, but this point
is not clear.
It is interesting to note that
for the VBS= 0 data, there
is a variation in current of about afactor of 2 for the curves
at drainvoltages of 2 and 6 V but amuch larger change for
VDS= 14 V. This behavior agrees with the model of a
greater variation of current with drain voltage for VOS>
1 2 V. For VBS= - 1V, the movement of the curves with
drain voltage reflects the decrease in drain conductance
with substrate bias.
In both the drain and gate voltage measurements the
lowest currents which can be conveniently measured are
in the picoampere range, as shown. In most applications,
these are the smallest currents of interest because the
leakage currents of the drain diode or the source diode
(with substrate bias applied) begin to dominate at these
levels. A t these levels, the surface potentialis already at
0.45 eV which puts the Fermilevel at thesurface a t 0.66
eV. Hence, the variation of surface potential in these
measurements always corresponds to movementsof the
Fermi level in the upperhalf of the energy gap.
The applicationof substrate bias has a dramaticeffect
on the [Link] -1 V is applied as illustratedin Fig.
13, the gradientof the drain currentversus drain voltage
increases markedly. The theory continues to follow the
data well although thevalue of RAhad tobe changed to
1.5 X 1015to achieve better agreement. This behavior isnot
surprising, since small changes in N A can cause noticeable
changes in shape. The theorywas derived for a constant
value of N A ;as the substratebias is increased the average
value of N A is expected to decrease because of the charge
profile (cf. Fig. 11).
In Fig. 12, the substratebias has theeffect of shifting the

Fig. 14. Drain voltage dependence of subthresholdcurrentafter

Troutman [7]. For the theoretical predictions (circles) the parameters
tox= 600 A, L = 2.1 Fm, N A = 1 X 10l6~ m - and
~ , ion-implanted source
and drain junctions
were assumed.

transconductance curves to higher gate voltages, a result

we would expect because a larger gate bias is then required
for the same value of Cps. As previously mentioned, the
drain bias now has a greater effect onthe transconductance
curves which is also due to thedecreasing resistance (increased slope of the curves in Fig. 13).
As a final verification of the theory, observations of other
workers [7] are shown in Fig. 14 together with the results
predicted here (see (29)). The correlation of the theory and
data over the complete rangeof bias voltages is very good.
In thiscase the punch-throughis confined to thebulk as
was described because of its independence of gate voltage.

IV. CONCLUSIONS
We have developed a simple model for subthreshold
conductionbased on one-dimensional concepts. It is
compatible with existing theories and deals for the first
time with theeffect of drain voltage on the current up to
the punch-through condition. An analytical framework has
been presented which allows the prediction of surface
potential, thresholdvoltages, and subthreshold currents
in short channel FET'sas a function of applied voltages
without theneed of empirical constants.
The agreementbetween the model and the datais excellent when the effect of surface states are included.
However, the inclusion of surface statesin the interpretation is not simply the use of an arbitrary parameter to
allow one to match experimental [Link] of the
exponential dependence of current upon surface potential,
the small effects of trapped surface or bulk charge upon
the surface potential resultin large changes in the current.
This situationis incontrast to the
behavior afterturn-on
in which the trappedcharge is effectively constant for in-

380

IEEE 1RANSACTIONS ON ELECTRON

DEVICES,

VOL. ED-25, NO. 3, MARCH 1978

ueasing voltages and so accounts for only an adjustment

in the threshold voltage. As a result, in comparison to
conventional designprocedures, circuit design with dev. ces
operating inthe subthreshold region has the disadvantsge
of being more sensitiveto device processing.
ACKNOWLEDGMENT
APPENDIX

The author wishes to thank D. Giles of Honeywell's

S.S.E.C. for his conscientious efforts inthe measurement
of these devices.

The integral

REFERENCES

is written

The above representation of G in the depletion a])proximation is valid for 4 > kT/q. T o include the limiting
case of flat-band conditions, the term under the square
root is increased by hT/q so that thelower limit of may
extend to zero. The dominant term in the integral is the
exponential and hence the integral may be performed
approximately by integrating only the exponentialterrn
and evaluating the square-root term at the appropriate
limits. Since only the upper limit has
a significant contri
bution. then

H. K. J. Ihantola and J. L. Moll, Solid-state Electron., vol. 7, p. 423,

1964.
S.R. Hofstein and F. P. Heiman, Proc. I E E E , vol. 51, p. 1190,
1963.
C. T. Sah, IEEE Trans. Electron Devices,
vol. ED-11, p. 324,
1964.
M. B. Barron, Solid-State Electron.,vol. 15, p. 293, 1972.
R. J. Van Overstraeten, G. J. Declerck, and P. A. Muls, IEEE Trans.
Electron Deuices, vol. ED-22, p. 282, 1975.
R. R. Troutmanand S. N. Chakravarti, IEEE Trans. Circuit
Theory, vol. CT-20, p. 659, 1973.
R. R. Troutman, I E R E J. Solid-state Circuits, vol. SC-9, p. 55,
1974.
R. A. Stuart and W. Eccleston, Electron. Lett., vol. 8, p. 225,
1972.
L. D. Yau, Solid-state Electron., vol. 17, p. 1059,1974.
H.C. Pao and c. T. Sah, Solid-state Electron., vol. 9, p.927,
1966.
Johnson, RCA Reo., pp. 80-94, Mar. 1973.
112j G. W. Taylor, tobe published.