Physics of the ferroelectric nonvolatile memory field effect transistor

S. L. Miller and P. J. McWhorter

Citation: Journal of Applied Physics 72, 5999 (1992); doi: 10.1063/1.351910

View online: http://dx.doi.org/10.1063/1.351910
View Table of Contents: http://aip.scitation.org/toc/jap/72/12
Published by the American Institute of Physics

Articles you may be interested in

Ferroelectric thin films: Review of materials, properties, and applications
Journal of Applied Physics 100, 051606 (2006); 10.1063/1.2336999

Device modeling of ferroelectric capacitors

Journal of Applied Physics 68, 6463 (1998); 10.1063/1.346845

Depolarizing field and “real” hysteresis loops in nanometer-scale ferroelectric films

Applied Physics Letters 89, 253108 (2006); 10.1063/1.2408650

Depolarization fields in thin ferroelectric films

Journal of Applied Physics 44, 3379 (2003); 10.1063/1.1662770

On device design for steep-slope negative-capacitance field-effect-transistor operating at sub-0.2V supply

voltage with ferroelectric HfO2 thin film
AIP Advances 6, 025113 (2016); 10.1063/1.4942427

Ferroelectric Zr0.5Hf0.5O2 thin films for nonvolatile memory applications

Applied Physics Letters 99, 112901 (2011); 10.1063/1.3636417
Physics of the ferroelectric nonvolatile memory field effect transistor
S. L. Miller and P. J. McWhorter
Sandia National Laboratories, Albuquerque, New Mexico 87185
(Received 29 June 1992; accepted for publication 9 September 1992)
The operation of the ferroelectric nonvolatile memory field effect transistor is theoretically
examined extensively for the first time. The ferroelectric transistor device properties are derived
by combining the silicon charge-sheet model of metal-oxide-semiconductor field-effect transistor
device operation with Maxwell’s first equation which describes the properties of the ferroelectric
film. The model we present describes ferroelectric transistor 1-V and C-V behavior when
time-dependent voltages are applied which result in hysteresis due to ferroelectric switching. The
theoretical results provide unique insight into the effects of geometrical and material parameters
on the electrical properties of the transistor. These parameters include the ferroelectric
spontaneous and remanent polarization, the coercive field, and dielectric layer thicknesses. We
have found that the conventional concept of threshold voltage is no longer useful, and that
increasing the spontaneous polarization has only a minor impact on memory operation due to
reverse dipole switching of the ferroelectric layer. The application of the model to optimize
design and fabrication parameters is illustrated with a virtual prototyping example. The model
is also used to develop a practical testing methodology for this unique device.

I. INTRODUCTION material parameters can significantly accelerate technology

development. The successful design of nonvolatile memo-
The hysteresis in the polarization of ferroelectric ma- ries implementing the ferroelectric transistor requires a
terials makes possible their use as electrically switchable thorough understanding of how electrical performance is
nonvolatile data storage elements.lb5 The more commonly related to design parameters. To characterize this relatively
implemented ferroelectric data storage element is a capac- unfamiliar device, new electrical characterization tech-
itor consisting of a thin ferroelectric film sandwiched be- niques must be developed which measure the fundamental
tween two conductive electrodes.677 The direction of the device properties. In order to rapidly optimize design and
ferroelectric polarization vector is set by a voltage pulse to fabrication parameters, and develop relevant characteriza-
the capacitor. The stored bit is read by applying another tion techniques, we need to quantitatively model the elec-
voltage pulse and determining whether or not the polariza- trical performance as a function of these parameters. This
tion switched direction. This method of nonvolatile mem- is a capability that has not been previously developed.
ory implementation suffers from the fact that data are de- To address this modeling need, we have derived and
stroyed during the reading process. Consequently the bit solved the equations governing the operation of the ferro-
must be reprogrammed after each read, resulting in the electric field effect transistor. The derivation of the model
accumulation of large numbers of read/write cycles. Dur- is presented in Sec. II. Numerical analysis techniques used
ing the read process data storage is volatile, i.e., data could to solve the resulting equations in a practical and efficient
be lost if power is lost during the read. The ferroelectric manner are described in Sec. III. Theoretical results are
memory field-effect transistor ( FEMFET) 8-’’ data storage presented and discussed in Sec. IV for a range of relevant
element offers a significant advantage over the ferroelectric device parameters. We focus on the basic (often nonintu-
capacitor approach: reading the data does not require the itive) device properties of this novel structure. In Sec. V,
memory to be reprogrammed. an example of technology optimization through the appli-
The ferroelectric memory field effect transistor consists cation of virtual prototyping is given. A testing methodol-
of a field-effect transistor (FET) whose gate dielectric is ogy, based on the model, is also presented and discussed.
comprised of a ferroelectric material, or a stack of dielec- Finally, the paper is summarized in Sec. VI.
trics with one ferroelectric layer.s-” The application of a
voltage pulse to the gate sets the direction of the polariza- II. THEORY
tion, and hence the value of the drain current of the tran-
sistor. The electrically programmable drain current can be A. Approach
used to define two logic states, much like other semicon- To develop the ferroelectric transistor model, we com-
ductor nonvolatile memories,12 such as the silicon-nitride- bine the switching physics of ferroelectric capacitors with
oxide-semiconductor (SNOS) nonvolatile memory transis- the silicon physics of metal-oxide-semiconductor (MOS)
tor.13 FETs. Brews developed a successful model of MOSFET
A comprehensive, quantitative model of ferroelectric transistor device operation14”* based on the charge-
transistor operation has not been previously developed, sheet16”” approximation. This approach assumes that the
and is now needed for several reasons. Understanding the inversion layer of the silicon can be approximated by a
dependence of the electrical properties on fabrication and conducting plane of zero thickness. In Brews’ model, there

5999 J. Appl. Phys.72(12),15 December1992 0021-8979/92/245999-12$04.00 @ 1992 American institute of Physics 5999
 20 -
Ferroelectric 72
5 IO-

Id(t) g o-

v, s
B -10 -

drain / ~ s

B -20 -
N,
@ms I
Vb
ELECTRIC FIELD (kV/cm)

FIG. 1. Cross-sectional schematic of the FEMFET device being modeled.

The gate dielectric stack consists of a switching ferroelectric sandwiched FIG. 2. The dipole switching contribution to the ferroelectric polariza-
between two nonswitching dielectric layers. tion is a function of the history of the electric field. The polarization must
lie on or within the saturated hysteresis loop.

are two contributions to the current: drift (governed by the

lateral electric field, which is determined by the drain- directly applied to a p-channel device by making the stan-
source bias) and diffusion (governed by the carrier con- dard sign transformations.
centration, which is a function of the quasi-Fermi poten- The ferroelectric dielectric layer is assumed to have a
tial). Miller et al. is*19 have developed a model that dipole polarization that is a function of the history of the
accurately describes ferroelectric switching in capacitor electric field internal to the ferroelectric. The basic hyster-
structures by solving Maxwell’s first equation in the dielec- etic nature of the ferroelectric polarization is illustrated in
tric stack. We combine these two approaches to derive the Fig. 2, where the switched polarization (the linear contri-
equations governing the operation of the ferroelectric tran- bution to the polarization is omitted in this figure) is plot-
sistor. ted as a function of the ferroelectric field. Minor hysteresis
The transistor drain current is physically determined loops, along with the saturated loop, are shown. We note
by the quasi-Fermi potential at the surface of the silicon that the polarization must lie on or within the saturated
along the length of the channel. The quasi-Fermi potential polarization hysteresis loop.
is related to the electrostatic potential at the silicon surface,
which in turn is related to the applied gate voltage; it is a
function of the properties of the dielectric layers on top of C. Capacitor equations
the silicon, and of the silicon itself. Our approach now is to
first derive the equations governing the polarization, field, The silicon physics of a MOS device (capacitor or
and silicon surface potential as a function of gate voltage transistor) is independent of the dielectric structure that
for a ferroelectric capacitor on silicon. Using these results, exists above the surface of the silicon. The structure above
we then solve the basic transistor equations to obtain an the silicon simply determines the relationship between the
expression for the drain current. applied gate-to-substrate voltage Vgb and the silicon sur-
face potential 4, For most MOS devices, once this rela-
tionship is determined, the device characteristics can be
B. Definitions
expressed directly in terms of V86’The same is true for a
Before proceeding with the derivation of the relevant ferroelectric capacitor or transistor, except that the rela-
equations, we define the physical structure being modeled, tionship between Y86 and (p, depends on the electrical his-
along with the associated terminology. tory of the device. In this section, we derive the equations
The structure being modeled is shown in Fig. 1. (The relating the silicon surface potential to the gate-to-
definition of the physical parameters are given in the Ap- substrate voltage as a function of the properties of the
pendix.) A stack of three dielectrics exists between the dielectric layers on top of the silicon.
silicon and conductive gate. The central dielectric consists The electrostatic equations for the capacitor structure
of a switching ferroelectric dielectric layer, while the outer illustrated in Fig. 1 are derived starting with Maxwell’s
two layers are nonswitching dielectrics. The thickness and equation
linear dielectric constants for each layer are independent.
For this derivation we assume that there is no space charge
V-D =p, (1)
in the dielectric layers; free charge exists only on the gate where p is the free charge density and D, the displacement
and in the silicon. This capacitor structure, with the addi- vector, is given by D =e,,E+ P,,. (Vector quantities are
tion of the source and drain regions, comprise the FEM- signified by bold symbols, and are taken to be positive in
FET. We derive the equations for an n-channel device (p- the direction pointing downward from the top electrode, as
type substrate, n-type source and drain). The results can be shown in Fig. 1.) The total polarization is the sum of the

6000 J. Appl. Phys., Vol. 72, No. 12, 15 December 1992 S. L. Miller and P. J. McWhorter 6000
linear contribution and the contribution resulting from the lateral electric field resulting from the source-drain
switching dipoles. Thus bias. The other contribution is diffusion of carriers due to
the carrier concentration gradient. The relative contribu-
D=E,,E+E,,xE+P~=E~E+P~, (2)
tions of the two mechanisms to the total current may vary
where E= 1 +x is the linear dielectric constant for a given along the length of the channel. However, the total current
dielectric layer, and Pd, the contribution to the polarization itself (steady state) clearly must be independent of posi-
due to switching dipoles, is nonzero only in the ferroelec- tion, otherwise charge would build up in the channel. The
tric layer. Since Pd is a function of the electric field in the carrier density automatically adjusts to result in the same
ferroelectric layer (as well as the field history), it is written total current. This variation of carrier density with channel
as PJ E2). Using Eqs. ( 1) and (2) in conjunction with the position is physically explained in terms of an adjustment
definition E= -VI$, where 4 is the electrostatic potential, of the Fermi level. Specifically, the conventional Fermi
and recalling that the dielectric layers contain no trapped level is replaced by a quasi-Fermi level that varies along the
charge, we arrive at length of the channel, resulting in a total current that is
4
-pd(Ed
~04, (3)
constant.
Taking into account the mechanisms of drift and dif-
fusion, the current can be expressed in terms of the quasi-
where Fermi potential asl’

(4)

and the field is given by where the derivative is along the length of the channel, and
x is the distance from the source.
E2= - [as+pd(E2) 1 The goal now is to find an expression for the inversion
652 * charge density IV1 and the derivative of the quasi-Fermi
The silicon charge a, is given as a function of the silicon potential dd,/dx.
surface potential c$, by”
2. Inversion charge N,
a,(A) = -SGN(&) fi(e&lfiL~J [ (e-B4s+P#s- 1) The determination of the inversion charge layer is fa-
+ (nJNJ2(P+34s- 1) Y2, (6) cilitated by making a simple observation. The inversion
charge layer is typically a very thin sheet of charge located
where the bulk Debye length LB is given by at the silicon surface. Thus, the inversion layer is approx-
LJy= ( Eo$@qNJ 1’2. (7) imated by a sheet of charge. This approximation eliminates
the need to obtain complex solutions to the electrostatic
Equations (3), (5), and (6) together completely describe equations in the silicon.
the capacitor operation of the device illustrated in Fig. 1, The total silicon charge density is just the sum of the
when the function Pd(E2) is known. These equations are charge in the inversion layer -qN, and the charge in the
valid for accumulation, depletion, and inversion. The solu- depletion layer -qN,w, i.e.,
tion to this set of equations will be addressed later.
a,= -qNr-qN,w, (9)
D. Transistor equations: General formulation where the depletion width w is given by
Having derived the equations relating the silicon sur- w = JzLB( pep,, 1’2. (10)
face potential to the applied voltage, we now proceed to
derive the transistor equations. This derivation follows Eliminating w from Eqs. (9) and (lo), one obtains the
closely that given by Brewst4*t5 for the nonferroelectric desired result
MOSFET. Since the polarization is related to the silicon c stack es
surface potential and charge [see Eq. (3)], Brews’ equa- q&=7 -4~w2-~ (11)
tions are rewritten to properly take into account the effects ( stack

of the hysteretic ferroelectric polarization. The equations where

are derived below for n-channel transistors; application to
0= & 6J’%&~stack). (12)
p-channel devices can be made with the standard sign
changes.
3. Quasi-Fermi potential &,,
1. Basic current equation
A suitable expression for the quasi-Fermi potential is
When the surface potential is sufficiently large to form needed to evaluate Eq. (8) for the current. The equation
an inversion region at the surface of the silicon, an electri- relating the silicon charge and surface potential can be
cal channel connecting the source and drain is formed. The expressed in terms of the quasi-Fermi potential by”
current that flows through this channel consists of two
contributions, each caused by fundamentally different as= -OF [ p$+ (~)‘@L l)&%+bp] 1’2. (13)
mechanisms. One contribution is drift of the carriers due to

6001 J. Appt. Phys., Vol. 72, No. 12, 15 December 1992 S. L. Miller and P. J. McWhorter 6001
Equation ( 13) is strictly valid only for the range of surface This expression may be simplified considerably by making
potentials resulting in depletion and strong inversion. The the following observations. The first two terms of Eq. ( 17)
complete expression that is also valid for accumulation is are simply PqN/Cstack Thus the third term of Eq. (17)
not necessary for the analysis of transistor operation, since contributes significantly to the current only when N1 is
negligible current flows in accumulation. Equation ( 13) is small, i.e., near pinch-off. Consequently, we replace the
inverted to obtain the quasi-Fermi potential third term of Eq. (17) by its value near pinch-off. Using
Eq. ( 15) to do this, IQ. ( 17) becomes
( -fld~c,tack)2-fl#s
* (14)
Idx=
The 1 in the denominator can typically be dropped since
exp(&,)>l when not in accumulation.
It can be inferred from Eq. ( 14) that there is a value of (18)
@$, that cannot be exceeded. As the surface potential &$,
increases, the numerator in the log term approaches zero, The current is obtained by integrating the above ex-
and hence /3+F,, diverges to positive infinity. The value of 4, pression along the length of the channel, from the source to
at which $Fn diverges is called &, and is given by the drain. We designate the effective channel length by L,
and the silicon surface potential at the source and drain by
-&(&at>~&ack= (Pdsat)1’2* (15)
+d) and 4sL, respectively. Since the current is constant
In I$. ( 15) we indicate the explicit dependence of the along the channel, the left-hand side of Eq. ( 18) becomes
charge a, on the surface potential. For fixed gate and IL. All terms on the right-hand side of Eq. (18) can be
source voltages, increasing the drain voltage will increase integrated directly except those that contain a,. We recall
the silicon surface potential at the drain. This potential that o, is a function of @, and is given by [see Eq. (3)]
asymptotically approaches &; this is the condition that
results in pinch-off of the channel. The term pinch-off is
used because at this point the inversion layer vanishes [see -‘;““)
stack
=&y-&b,+P,(E2)
2 . (19)
Eqs. (11) and (15)], and the channel is “pinched-off.” The
value of the drain voltage V,,, that results in saturation of If the drain-to-source bias V,, is kept sufficiently low (typ-
the surface potential at the drain (pinch-off) is defined by ically < -0.5 V), the FE polarization will not vary signif-
V,, = I$,,~ - I&,, where & is the surface potential at the icantly along the length of the channel, and Pd( E2) can be
source end of the channel. treated as a constant for the purpose of integration. Inte-
Finally, a useful expression for d/3 $F,,/dx is obtained. grating Eq. ( 18) using Eq. ( 19), we obtain
This is done by taking the derivative of Eq. ( 14). The
result is
I= -$‘+ [ (1+&++z f%E,))(@$s~-&)
l- [2&d(~~stack)2] (~~~s@~s)
&Fn=( 1+ (-&d~cstack)
2
+‘#%

The terms on the right-hand side of Eq. (8) have now

been defined [see Eqs. ( 11) and ( 16)]. These quantities
+a[. wsLY2- pAilY21 *
1 (20)

The values of the surface potentials 4,L and 4a are

cannot be evaluated directly to obtain the current. How- determined using the boundary conditions15
ever, if Eq. (8) is integrated from the source to the drain,
the current can be determined. +F&ource) =‘#‘F+ vbs (21)
and
4. Determining the current #Fn(drain)=&+ VbYt- vds. (22)
Now that the inversion charge N1 and the derivative of
The potentials #a and & are then given [combining Eq.
the quasi-Fermi potential d&$&dx have been determined,
( 14) with Eqs. (21) and (22)] by solving the transcenden-
we proceed to evaluate Eq. (8) to obtain the current. Sub-
tal equations
stituting Eqs. ( 11) and ( 16) into Eq. (8)) we obtain
- w@stack -@s l/2
I=
p2 c,,, - ‘(“‘)

+
a(1 - [2~q/(~~,,,,k)21
[ ( -&duCstack)
(%J-~~~~~s)~
+ (8h)1’21
-fvK+ (23)

(17) and

6002 J. Appl. Phys., Vol. 72, No. 12, 15 December 1992 S. L. Miller and P. J. McWhorter 6002
 Wz TABLE I. FEMFET deviceparameters.
Pv.b+Gpd(EZ)
Parameter Value Unit

p* 1.0 &/cm2
(24) p/p, 0.8
EC 100 kV/cm
The ferroelectric dipole polarization in Eqs. (20), (23), 4 300 A
and (24) is evaluated at the source end of the channel. EI 3.9
Brews’ original raUlt iS recovered by Setting Pd=O. 4 1000 A
4 10
In order to solve Eqs. (20), (23 ), and (24) for the 10 .A
4
drain current, explicit definitions of the dipole polarization 63
Pd(E2) must be given. We discuss this in the following NC 3;ik cmm3
section, after first making a few comments regarding the P 0.08 m’/(V s)
treatment of interfacial charge. W 8 w
L 4 Pm
We note that the derivation presented here is in a form -1.15 V
4,
that allows the quantitative treatment of interface states at vd 0.1 V
the silicon surface, the occupancy of which is a function of V* 0 V
the difference between the surface potential and quasi- vb 0 V
Fermi potential. The inclusion of interfacial charge results
in slight modifications to Eqs. (3) and (5); these changes
are then carried through the remainder of the analysis.
Care must be taken when integrating the equation analo- The equations have thus been given that govern the
gous to Fq. ( 17) since both the surface potential and quasi- ferroelectric transistor drain current as a function of the
Fermi potential are functions of position along the channel applied voltage history, geometry, and material properties.
length. The detailed treatment of interface states is omitted In the following section we discuss some issues associated
with the numerical evaluation of the above equations.
in the present paper.

III. NUMERICAL ANALYSIS TECHNIQUE

E. Dipole polarization Since the polarization (and hence the surface potential
The value of the dipole polarization depends on the and current) depend on the history of the electric field in
previous history of the ferroelectric electric field. Thus the the ferroelectric, the drain current is calculated as a func-
polarization is determined by integrating dP,(E,)/dE, tion of the applied gate voltage by integrating the polariza-
from some specified initial condition. Since the form of tion from known initial conditions. The procedure to ac-
dPd(E)/dE (we drop the subscript 2 for notational conve- complish this is outlined below.
nience) has been extensively discussed previously for arbi- A. initial condition computations
trary field histories, 18V19
the results are simply stated here.
The saturated polarization hysteresis loop is defined by ( 1) Define the constants required for the computa-
tions. These include P,, P, E. T, N,, p, W, L, d,, d2, d3, el,
P&(E)=P&.anh[ (E-E,)/2S], (25) E2, E3, v, vb, and v,.
where

S=Ec[ ln( :Tig)]-‘, (26)

and the + superscript signifies the positive-going branch of

the loop. The negative-going branch of the loop is given by
P;,(E) = -P&( -E). (27)
The derivative of the polarization, whether or not the
polarization lies on or within the saturated hysteresis loop,
is given by

(28)

where
0
I’=l-tanh[ (E)1’2], (29) time
and c= + 1 when dE/dt > 0 and ,$= - 1 when dE/dt < 0. FIG. 3. The time-dependent voltage Vg applied to the transistor.

6003 J. Appl. Phys., Vol. 72, No. 12, 15 December 1992 S. L. Miller and P. J. McWhotter 6003
 (2) Define the initial conditions, i.e., the initial gate solve Eqs. (3), (5), (6), and (28) must be developed. The
voltage Ve and initial polarization Pa Using these values, approach we pursue is to compute the mth value of the
the silicon surface potential +S and silicon charge density a, relevant quantities, and numerically integrate. In order to
are calculated by simultaneously solving Eqs. (3) and (6). avoid confusion, we drop the subscript 2 on the ferroelec-
The initial value of the ferroelectric field is then calculated tric electric field. We now introduce a new subscript, m,
using Eq. (5). The derivative dP,(E)/dE is also evaluated which signifies that the given quantity is evaluated at the
[using Eq. (28)]. Note that the initial choice of Ve and Pd mth integration increment.
must satisfy the physical requirement that the polarization First, select an mth value for 4, Calculate the mth
lie on or within the saturated hysteresis loop. value of a,( 4,) from Eq. (6). Write the dipole polarization
(3) Finally, the initial current is computed. If 4, < +p as
then the channel is off and the drain current is effectively
zero. If #$> 4p then the current is computed as follows.
Calculate fl+d) and p+SL using Eqs. (23) and (24). Then
substitute these values into Eq. (20) to give the drain cur- PcAEm)
=P~&,+I) + (-%--Em-,) -& [P&3 11~~~~
,
rent. The value of Pd used to evaluate Eqs. (23), (24)) and (30)
(20) is that used in step 2 above.

B. Numerical integration
where the vertical bar signifies that the derivative is eval-
To perform the integration from the initial conditions uated at the field E,,+ i. Using Eq. (5) to eliminate E,, E~J.
to a new set of conditions, a technique to simultaneously (30) becomes

(31)

Using Eq. (3 1) in conjunction with Eq. (28), compute the maximum and minimum field for each curve are different
mth value of the dipole polarization. Next, calculate Y86 in magnitude due to the non-zero flatband voltage.
from Eq. (3). Using the above results, solve Eqs. (23) and The ferroelectric polarization Pd is shown as a function
(24) for the silicon surface potential at the source and of the applied gate voltage Vg in Fig. 5 for several values of
drain end of the channel. Finally, evaluate Eq. (20) to PF Note that at Vg=O, the polarization is less than the
obtain the mth value of the drain current. remanent polarization P, This nonintuitive result occurs
Though the numerical implementation of the equa- because the ferroelectric field reverses direction when the
tions to obtain the drain current as a function of an arbi- gate voltage is reduced from a saturating value back to 0 V.
trary time-dependent gate voltage may appear somewhat This field-reversal results in the additional polarization re-
complex, it is actually quite straightforward. duction observed in Fig. 5. It also results in the fact that
the polarization at zero gate bias is relatively insensitive to
IV. MODEL RESULTS the spontaneous polarization when it is increased beyond
The behavior of a FEMFET device is now investigated -0.5 ,uC/cm2. The nature of the electrostatic boundary
using the equations derived above. The parameters used to conditions that must be satisfied at the dielectric layer in-
perform the analyses are given in Table I, unless otherwise terfaces is the physical origin of this behavior. For com-
specified. The initial conditions are ?‘a= Vfb and Pd=O. pleteness, we show the dependence of the ferroelectric field
The applied voltage history is shown in Fig. 3. To maintain on the applied gate voltage in Fig. 6. Note that the direc-
clarity in the following figures, only the quantities associ- tion of the path taken around the hysteresis loop is opposite
ated with the portion of the voltage history indicated by to that of the polarization shown in Figs. 4 and 5. We also
the heavy line in Fig. 3 are shown. point out that the changing depletion layer capacitance of
For reference, we first show the “standard” polariza- the silicon is the origin of the “kink” that disrupts the
tion hysteresis loop, i.e., the contribution to the ferroelec- symmetry in Figs. 5 and 6.
tric polarization due to the switching dipoles as a function The electric field in the non-switching layers (or the
of the ferroelectric field. This is plotted in Fig. 4 for several silicon) can become quite large if the ratio of the dielectric
values of P, while maintaining a constant ratio P/P, The constant of the ferroelectric layer to that of the nonswitch-
arrows indicate the direction of the hysteresis path in this ing dielectric layers (or the silicon) is not close to unity.
and the following figures. Though the amplitude of the We illustrate this by plotting the field in the bottom non-
applied signal is the same for all three cases, the maximum switching layer as a function of the gate voltage in Fig. 7.
ferroelectric field decreases with increasing P,. In fact, for The rate at which the field increases with voltage scales
P,= 1.0 &/cm2, the hysteresis loop is unsaturated. The with P, during switching, and is independent of P, after

6004 J. Appl. Phys., Vol. 72, No. 12, 15 D&ember 1992 S. L. Miller and P. J. McWhorter 6004
 0.8
0.4

0.2

z
g 0
V
N
w -0.2

-0.8l’ ’ s ” “I ”
-500 -250 0 250 500 -12 -8 -4 0 4 8 12

E, (kV/cm) vg(V)
FIG. 4. The ferroelectric polarization P,+as a function of the ferroelectric FIG. 6. The direction of the hysteresis of the ferroelectric field vs the gate
field is hysteretic due to the applied signal shown in Fig. 3. voltage is oppositethat of the polarization, resulting in field reversal when
the gate bias is reduced to 0.

switching is complete. This rapid increase in the field dur-

ing switching is a direct consequence of the boundary con- and surface potential. The silicon surface potential (with
ditions associated with this stacked dielectric structure. no drain bias) is shown in Fig. 8 as a function of gate
The polarization of the FE layer reduces the electric field voltage for several values of Pr (Note that increasing P, by
internal to the FE film, and hence increases the field in the a factor of four has only a minor impact on the gate voltage
dielectric layers external to the FE film. Spontaneous po- at which the transition from accumulation to inversion
larization values that are not very small can result in fields occurs.) The occurrence of ferroelectric switching not only
exterior to the ferroelectric that can easily become compa- results in hysteresis of the d,( Vg) curve, but also results in
rable to the breakdown field for such films, and can cer- a change in shape relative to that of a non-switching de-
tainly result in parasitic effects such as charge injection or vice. In particular, when the gate voltage passes through
transport through the films. the range where switching occurs (see Fig. 5), the slope
Since the transistor drain current is a function of the dt&/dV, increases due to the changing polarization. These
silicon surface potential, the I-V characteristic of the tran- results significantly impact our intuition regarding such
sistor depends on the relationship between the gate voltage concepts as threshold voltage.

0.8
3

0.4 2

E 1
Ni
0 O Y
2
is O
ci?
-0.4 w” -1

-2

-0.8
, * 1 * * * * ” a ’ 1 ’ ” I”” a ‘3
-12 -8 -4 0 4 8 12 -3

vg09 -12 -8 -4

vgW)
0 4 8 12

FIG. 5. The hysteresis of the polarization Pd as a function of gate voltage

k’s demonstrates the reverse switching due to field reversal in the ferro- FIG. 7. The field in the non-switching dielectric layers is amplified by the
electric. The “kink” in the curves disrupting the symmetry is due to the switching polarization of the ferroelectric layer. Depending on the geom-
changing capacitance of the silicon. etry, it can easily reach values comparable to the breakdown field.

6005 J. Appl. Phys., Vol. 72, No. 12, 15 December 1992 S. L. Miller and P. J. McWhorter 6005
 0.25
0.8

-12 -8 -4 0 4 8 12

vg W) 120 ,

FIG. 8. The silicon surface potential as a function of the gate voltage is

hysteretic, resulting in distorted transistor Z-V characteristics exhibiting 80 -
hysteresis.

For an “ideal” MOS transistor, the threshold voltage

can be defined as the gate voltage that results in the silicon
surface potential at the source being equal to 2+P The
subthreshold region corresponds to gate biases resulting in ‘fl Ps = 0.0 yC/cm’

+F < +S< 2#F The #,( Vg) hysteresis curves in Fig. 8 are
those for saturating (in the sense that the polarization sat- -2 0 2 4 6 a 10 12
urates) or nearly saturating gate biases. Minor hysteresis vg (V)
loops can take any path within the corresponding saturated
loop. Consequently, the notion of threshold voltage is com- FIG. 9. Transistor Z-V curves exhibit hysteresis due to ferroelectric
pletely ambiguous; the gate voltage that corresponds to switching. The same data are plotted in both (a) and (b); the only
$,=21$~ depends on the voltage history. Moreover, all as- difference is current scale. The apparent “threshold” voltage shift is gov-
pects of the transistor I-V curve (e.g., subthreshold swing erned by EO while P, governs the current offset at large voltages.
and k’) depend on the voltage history due to the fact that
the ferroelectric polarization depends on the internal field
history. of P,. The boundary conditions are such that the onset of
Transistor I-V curves for several values of P, are strong inversion corresponds to a relatively small value of
shown in Fig. 9; Fig. 9(b) is identical to 9(a) except for ferroelectric polarization. This polarization is relatively in-
the scale on which the current is plotted. The amplitude of dependent of P, (as is clearly seen in Fig. 5) as long as P,
the gate bias is sufficient to saturate the polarization (see is larger than some very small critical value (which de-
Fig. 4). The “subthreshold slopes” of the switching tran- pends on the geometry of the device). The “apparent”
sistors are greater than that for the nonswitching transistor threshold voltage shift, rather than being dependent on the
( PS=O); this occurs because of the greater sensitivity of the spontaneous polarization, is directly proportional to the co-
surface potential to the gate voltage during the switching ercive field E,
process. In addition, the maximum transistor current is The dependence of the flatband (or equivalently,
greater due to the larger fields resulting from the switched “threshold voltage”) shift on E, is clearly illustrated by the
dipoles (see the discussion associated with Fig. 7). C-V curves in Fig. 10. Recall that the C-V curves only
Figure 9(b) highlights some additional striking prop- sample the range in surface potential corresponding to the
erties of FEMFET hysteresis. The apparent slope of the formation of the depletion layer, i.e., - 0 < 4, < - 24,+ This
“on” I-V curves has two components; the larger slope re- corresponds to a very limited range of the FE polarization
sults when dipole switching is still occurring, while the Pd, as can be inferred from Figs. 8 and 5. As noted earlier,
smaller slope results when switching is complete. Note that the gate biases corresponding to this range of polarization
for voltages large enough for switching to be complete, the are relatively independent of the spontaneous polarization
slope is the same as that of a nonswitching transistor, and P, To summarize, the apparent flatband shift of the C-V
the actual current is proportional to the spontaneous po- hysteresis curves is proportional to the coercive field. The
larization P,. In addition, the “apparent” threshold voltage shape of the C-V hysteresis curves is governed by the de-
shift associated with the hysteresis is relatively independent pendence of the FE polarization on gate voltage for a range

6006 J. Appl. Phys., Vol. 72, No. 12, 15 December 1992 S. L. Miller and P. J. McWhorter 6006
 Ps = 1 .o j.lc/cm*

0.9 ,ZOlarizatiin Swi,ChlnQ field

decreases with thickness
2
0 0.8
3
0.7

0.6

0 0.5 1 1.5
d2 W-d

vgw FIG. 11. Following a positive voltage pulse, the transistor drain current
is significant only for a finite range of ferroelectric film thicknesses. This
virtual prototyping approach can be used to rapidly optimize a number of
FIG. 10. Capacitor C-Y curves exhibit hysteresis; the flatband voltage design and fabrication parameters.
shift is proportional to E, The shape of the C-Y curves results from the
ferroelectric polarization sweeping through only a small range in polar-
ization.

cantly reduce both the time and expense of performing

in polarization where this dependence is a very weak func- technology optimization, as well as reduce the number of
tion of the spontaneous polarization.
design iterations.
We illustrate a virtual prototyping application using
V. DISCUSSION the above FEMFET model. A fabrication parameter that
The above analyses reveal a number of unique and impacts the electrical performance of a FEMFET memory
nonintuitive aspects of the behavior of FEMFET devices device is the thickness of the ferroelectric layer. In our
that are not observed with either fixed-threshold FET de- virtual prototyping example, we determine the optimum
vices or other types of conventional silicon-based nonvol- ferroelectric layer thickness to maximize the transistor
atile memories. It is not possible to measure a “conven- drain current after a programming voltage pulse to the
tional” C&-threshold transistor I-V curve. The act of gate. The physical parameters associated with the device of
performing the measurement changes the ferroelectric po- interest are those listed in Table I. For the sake of illustra-
larization, and the resulting curve depends on the previous tion, we assume that the polarization Pd is zero immedi-
electrical history. The concept of a well-defined threshold ately before the programming voltage pulse. A positive 10
voltage does not exist; the gate voltage resulting in #s=2r$F V pulse is then applied to the gate, after which the gate-
depends on the voltage path taken due to the changing to-substrate voltage is maintained at 0 V. Immediately af-
ferroelectric polarization. In addition, the “subthreshold” ter the pulse, the polarization will be different from zero,
portion of the I-Y curve is always steeper than that of a resulting in a change in the transistor drain current. The
similar nonswitching FET. Specifically, as the gate voltage desired information is the value of the drain current (im-
is increased (decreased), the transistor turns on (off) mediately after the pulse) as a function of the ferroelectric
faster due to the ferroelectric switching; this is why the 1-V layer thickness.
hysteresis occurs in the counter-clockwise direction (see The conventional way to determine this information is
Fig. 9). to develop the appropriate fabrication processes, fabricate
These types of behavior present some interesting prob- devices with different ferroelectric thicknesses, and then
lems regarding FEMFET technology development and de- perform electrical measurements. Instead, we perform vir-
vice characterization. Since much of our standard intuition tual prototyping to determine the same information within
and characterization approaches are not applicable to minutes.
FEMFET devices, we show how the model developed The results of the virtual prototyping analysis are
above can facilitate the design and electrical characteriza- shown in Fig. 11, where the drain current at zero gate bias
tion of these devices. is plotted as a function of the ferroelectric film thickness.
Note that the transistor is on after the pulse for only a
A. Virtual prototyping application
limited range of thicknesses. For large thicknesses, the field
Virtual prototyping is the application of predictive resulting from the applied voltage is not sufficient to switch
models to answer questions that would normally be ad- the polarization, and the transistor remains off (as it was
dressed by designing, fabricating, and testing devices (i.e., with zero polarization). For slightly thinner films, the field
physical prototyping ). Virtual prototyping can signifi- is sufficient to switch the polarization. However, when the

6007 J. Appl. Phys., Vol. 72, No. 12, 15 December 1992 S. L. Miller and P. J. McWhorter 6007
gate voltage is reduced to zero after the pulse, reverse
switching of the dipoles occurs due to the field reversal in
the ferroelectric film (see Fig. 5 and the associated discus-
sion). Consequently, the transistor current after the pulse
is reduced to a value that is less than it was during the
pulse. For very thin films, the current again drops to zero.
This is because the impact of the ferroelectric polarization
on the silicon surface potential is roughly proportional to
the film thickness [see Eq. (3)]. We note that after negative
gate pulses, the transistor is turned off.
Though we have considered a single application of vir-
tual prototyping here, it is clear that the FEMFET device
model may be used to rapidly optimize a number of design
and fabrication parameters associated with the FEMFET 0.2 0.3

technology, e.g., film thicknesses, device areas, operating Pd (pC/cm2)

voltage, bit line capacitance, and FE film properties such as
E. P,, P, and E. FIG. 12. The computed current vs polarization curve can be used to infer
the ferroelcctric polarization from drain current measurements at zero
8. Testing methodology gate bias. The polarization is thus not disturbed during the measurement
process.
There are several physical properties of the ferroelec-
tric film that are relevant to the successful design, fabrica-
tion, and delivery of FEMFET devices. These include the
polarization versus field hysteresis loop (impacts design device parameters such as channel doping, channel length
parameters and material selection), the relaxation of the and width, film thicknesses, and dielectric constants must
polarization with time (determines data retention), and be known or determined using other standard techniques.
the change in these properties with repeated reprogram- Data retention is governed by the rate at which the
ming (endurance). The polarization vs field hysteresis loop polarization relaxes with time after the programming
of FE capacitors is typically obtained using a Sawyer- pulse. To determine the polarization as a function of time
Tower circuit;” unfortunately, the standard analysis typi- after a positive gate pulse, the drain current at zero gate
cally performed is not valid due to the highly nonlinear bias is measured as a function of time. Note that this mea-
relationship between the applied voltage and ferroelectric surement process does not disturb the polarization pro-
electric field. Data retention in conventional silicon-based vided the drain bias is kept low. To determine the time
nonvolatile memories is typically characterized by measur- dependence of the polarization, the drain current as a func-
ing the change in threshold voltage with time; this cannot tion of the polarization (and at zero gate bias), is com-
be done for FEMFET devices since changing the gate volt- puted using the previously determined physical parame-
age changes the “threshold voltage.” Characterization of ters. Figure 12 illustrates the results of such a computation
endurance simply involves measuring the above properties, for the parameters listed in Table I. Using Fig. 12, a given
for example, as a function of accumulated reprogramming current value is mapped directly to the ferroelectric film
cycles. Clearly, a new approach must be taken to electri- polarization. We note that this technique permits polariza-
cally characterize FEMFET devices. tion measurements for a limited range of values due to the
The development of a testing methodology must be fact that the transistor must have a measurable drain cur-
performed with extreme care due to the unique properties rent at zero gate bias. To measure the decay of polarization
of the FEMFET device. It requires a quantitative under- having a different range, one may perform the same mea-
standing of the relationship between measured quantities surements on an identical n-channel transistor having a
(such as currents and voltages) and basic device parame- threshold-adjust implant, or test a p-channel transistor fol-
ters in order to electrically characterize the relevant phys- lowing a negative gate pulse.
ical properties. We discuss below how the FEMFET device Using the above approach, the basic ferroelectric pa-
model is used to develop a relevant testing methodology. rameters may be determined after various numbers of ac-
The first step is to determine the basic ferroelectric cumulated reprogramming cycles. In this way, the endur-
hysteresis loop parameters P,, P, and EC We recall from ance of the ferroelectric film is characterized.
Fig. 9(b) that these parameters are directly related to the The testing methodology described above may be di-
saturated I-V hysteresis loop. The model parameters are rectly extended using the FEMFET device model to de-
determined by adjusting them so that the model 1-V curve velop additional characterization techniques.
matches the experimental I-V data; EC is adjusted to match
the apparent “threshold voltage” shift, P, is adjusted to
give the correct parallel offset for the saturated portion of VI. SUMMARY
the 1-V curve, the dielectric constant e2 is adjusted to give
the correct saturated slope, and P, is adjusted to give the The device properties of a ferroelectric memory field
correct curvature during switching. To do this, the other effect transistor (FEMFET) have been theoretically inves-

6008 J. Appl. Phys., Vol. 72, No. 12, 15 December 1992 S. L. Miller and P. J. McWhorter 6008
tigated. This was done by solving Maxwell’s first equation strates two characteristic slopes. At high voltages when
for the stacked dielectric structure in conjunction with the switching is complete, the slope is governed by the same
charge-sheet approximation to the silicon inversion layer. parameters as a normal nonswitching FET (but the cur-
The resulting equations allow the determination of the rent itself is higher by an amount proportional to the spon-
transistor drain current, ferroelectric polarization, and di- taneous polarization). At lower voltages the slope is larger,
electric layer fields as functions of arbitrary time- and is impacted by the shape of the ferroelectric P-E hys-
dependent gate voltages. It was found that the polarization teresis curve. Having two different slopes renders the nor-
at zero gate bias (which is relevant to device applications) mal concept of k’ somewhat ambiguous. Hence, we have
is relatively insensitive to the spontaneous polarization due quantitatively shown that most aspects of FEMFET device
to reverse switching, a result that can significantly impact operation differ significantly from those of standard MOS-
the development of the FEMFET technology. This reverse FET operation.
switching is caused by field reversal in the ferroelectric Because of these differences, the FEMFET device
layer; boundary conditions derived from Maxwell’s first model was applied to develop a viable testing methodology
equation for a ferroelectric film on silicon require this to that allows the experimental determination of relevant
occur. It was quantitatively demonstrated that the fields in physical properties of the ferroelectric film. Use of the
nonswitching films comprising the dielectric stack can eas- model was essential because quantitative relationships be-
ily become comparable to the breakdown field for such tween measured quantities and physical device properties
films, another result that impacts FEMFET technology are required.
development. Depending on the properties of the film, par- The application of the FEMFET device model to per-
asitic effects such as charge injection could easily result. form technology optimization was illustrated with an ex-
The FEMFET I-V curves were extensively investi- ample of virtual prototyping. The drain current as a func-
gated. The I-V hysteresis occurs in a counter-clockwise tion of the ferroelectric film thickness was computed for a
direction (this is opposite to that of other silicon-based specific programming pulse. It was found that the transis-
nonvolatile memories such as floating gate or SNOS) due tor is in a high current state after the pulse for only a
to polarization switching. The well-defined features nor- narrow range of thicknesses, allowing the optimum choice
mally associated with fixed-threshold FET I-V curves are of ferroelectric layer thickness without costly fabrication
dependent on the gate voltage history, and are thus ambig- iterations.
uous. For example, the concept of a well-defined threshold
voltage is no longer relevant since the gate voltage at which ACKNOWLEDGMENTS
+,=24F is history dependent. Also, the “subthreshold
slope” is steeper than that of an equivalent fixed-threshold We thank D. Lampe at Westinghouse Electric Corpo-
device due to dipole switching. The coercive field governs ration for stimulating discussions. We are also grateful to
the apparent threshold voltage shift of saturated I-V B. Filter and T. Meisenheimer for critically reviewing this
curves. The spontaneous polarization plays only a minor manuscript.
role in determining the apparent threshold voltage shift This work performed at Sandia National Laboratories
(provided it is larger than some critical value), contrary to supported by the U.S. Department of Energy under Con-
normal intuition and contrary to the way such data is com- tract No. DE-AC04-76DPOO789.
monly interpreted. In the triode region of operation (often
referred to as the linear region), the I-V curve demon- APPENDIX
The physical parameter and symbol definitions are
given below, along with their associated MKS units.
Silicon di, d,, d3, = dielectric layer thicknesses (m) .
Surface el, Q, e3, =linear dielectric constant (unitless).
tc l ,=linear dielectric constant of the silicon (unitless).
eo=permittivity of free space -8.85X 10-l’ (F/m).
P,=remanent polarization [C/m’], i.e., the ferroelec-
tric dipole polarization occurring at zero electric field on a
saturated hysteresis loop.
P,=spontaneous polarization [C/m*], i.e., the maxi-
E
Fn mum ferroelectric dipole polarization occurring on a satu-
rated hysteresis loop.
&=coercive field (V/m), i.e., the electric field at
which the ferroelectric dipole polarization is zero on a sat-
urated hysteresis loop.
V,=applied transistor or capacitor gate voltage (V).
V,= applied transistor drain voltage (V).
V,=applied transistor source voltage (V).
FIG. 13. A band diagram illustrating the definitions of the relevant po- Vb= applied transistor or capacitor substrate voltage
tentials in the silicon. 09.

6009 J. Appl. Phys., Vol. 72, No. 12, 15 December 1992 S. L. Miller and P. J. McWhorter 6009
 v,,= v,- v, 4~=[Ei(~0)-E&q=P-‘log(N,/n~)=Fermi poten-
V,,= - ( V,- V,). The sign of V,, is chosen so that V,, tial (V).
is a positive number when the source/substrate diode is #F,, = [Ei( 00 ) - EFn]/q = quasi-Fermi potential (V).
reversed biased. n=n,exp[(E,,-Ei)/k~=n,exp[P(~-~Fn)]=carrier
vgs= vg- v, concentration ( mm3).
Vgb= Va- V, = I&+ V,, In order to account for con-
tributions to the flatband voltage not explicitly addressed
in the following derivation, such as the gate-to-substrate ‘J. Moll and Y. Tarui, IEEE Trans. Electron Devices ED-IO, 338
work function difference 4, or fixed trapped dielectric (1963).
charge, VBbshould be replaced in the equations that follow *R. Zuleeg and H. H. Wieder, Solid-State Electron. 9, 657 ( 1966).
‘P. M. Heyman and G. H. Heilmeier, Proc. IEEE 54, 842 (1966).
with Vgb- Vu, where Vu, is the additional contribution to ‘J. C. Crawford and F. L. English, IEEE Trans. Electron Devices ED-
the flatband voltage. For example, Vn, = 4,, when no other 16, 525 (1969).
charges are present, and Vgb+ Vgb- Vfb= Vgb-~ms ‘C. A. Araujo, L. D. McMillan, B. M. Melnick, J. D. Cuchiaro, and J.
W=electrical channel width (m). F. Scott, Ferroelectrics 104, 241 ( 1990).
6J. F. Scott and C. A. Araujo, Science 246, 1400 (1989); in Molecular
L = electrical channel length (m). Electronics, edited by M. Borisson (World Scientific, Singapore, 1986),
,u=channel mobility [m2/(V s)]. pp. 206-214.
ni=intrinsic carrier concentration ( mm3). ‘J. T. Evans and R. Womack, IEEE J. Solid-State Circuits 23, 1171
N, = substrate doping concentration (assumed to be (1988).
8S. Y. Wu, IEEE Trans. Electron Devices ED-21, 499 (1974).
uniform) [mm3]. 9K. Sugibuchi, Y. Kurogi, and N. Endo, J. Appl. Phys. 46,2877 (1975).
n =free carrier concentration ( me3). ‘OS. Y. Wu, Ferroelectrics 11, 379 (1976).
q=charge unit - 1.6~ lo-i9 (C). ‘ID. R. Lampe, D. A. Adams, S. Sinharoy, and H. Buhay, Presented at
T= absolute temperature (K) . the 4th International Symposium on Integrated Ferroelectrics, Mon-
terey CA, 1992.
k=Boltzmann’s constant - 1.38~ 1O-123(J/K). “Y. Nishi and H. Iizuka, in Applied Solid State Science, edited by D.
8=q/(kT) (V-l). Kahng (Academic, New York, 1981), Suppl. 2A, pp. 121-251.
Figure 13 contains a band diagram illustrating the fol- “S. L. Miller, P. J. McWhorter, T. A. Dellin, and G. T. Zimmerman, J.
Appl. Phys. 67, 7115 (1990); S. L. Miller and P. J. McWhorter, ibid.
lowing definitions: 70, 4569 (1991).
Ei= intrinsic level (J) . 14J. R. Brews, Solid-State Electron. 21, 345 (1977).
EF= Fermi level (J) . I5 J . R . Brews, in Applied Solid State Science, edited by D. Kahng (Aca-
EFn=electron quasi-Fermi level (J). In a capacitor, demic, New York, 1981), Suppl. 2A, pp. 11-120.
16J. A. Guerst, Solid-State Electron. 9, 129 ( 1966).
this is typically equal to the Fermi level. “H. W. Loeb, R. Andrew, and W. Love, Electron. Lett. 4, 352 ( 1968).
N,= inversion layer charge density ( mW2). “S. L. Miller, R. D. Nasby, J. R. Schwank, M. S. Rodgers, and P. V.
a,= charge density in the silicon ( C/m2). Dressendorfer, J. Appl. Phys. 68, 6463 ( 1990).
4 = [Ei( CO) - Ei]/q = band bending (function of posi- 19S. L. Miller, J. R. Schwank, R. D. Nasby, and M. S. Rodgers, J. Appl.
Phys. 70, 2849 ( 1991).
tion) (V). ‘OS. M. Sze, Physics of Semiconductor Devices (Wiley, New York, 1981),
4S= [Ei( co ) - Ei]/q = band bending (at the silicon sur- p. 368.
face) (V). 2’C. B. Sawyer and C. H. Tower, Phys. Rev. 35, 269 (1930).

6010 J. Appl. Phys., Vol. 72, No. 12, 15 December 1992 S. L. Miller and P. J. McWhorter 6010