Abstract—We propose a model for the so-called constant- and bias temperature instability (BTI) [10], [11]. These in-
amplitude charge-pumping (CP) characteristics, giving the Elliot vestigations generally assume that the trap densities, which
Gaussian-like CP current curve (ICP –VL ) of lightly doped drain are given by CP and its variants from maximum CP current
(LDD) MOSFET with local oxidation of silicon (LOCOS). This
method is based on modulation of the contributing active-channel (ICPmax ), are uniform in the whole active channel area of the
area (AG ) to the ICP –VL curve, depending on the position of transistor. Unfortunately, this assumption is not experimentally
the high and low levels of the gate signal voltage. In addition, it valid, particularly for a small lightly doped drain (LDD) device
allows to separate and clarify the contribution of all MOSFET with field oxide isolation, where edge effects become signif-
regions (such as the effective channel, LDD, LOCOS, and LDD
subdiffusion under the LOCOS) to the amount of ICP –VL curves.
icant [12]–[14]. Recently, it has been shown that LDD and
We have simulated this model and compared with experimental local oxidation of silicon (LOCOS) regions pollute the ICPmax
CP data. The model shows a very good correlation with exper- of the main effective channel [15]–[17]. These papers clearly
imental ICP –VL curves, particularly for transistors with short demonstrate the contribution of different active channel regions
channel gate lengths (LG ≤ 1 µm). However, as the channel gate in the ICP –VL characteristics. They propose an experimental
length increases, the model matches only for rising and falling
ICP –VL curve edges, corresponding to the contribution of LDD methodology to separate the main effective channel CP current
and LOCOS regions, respectively. Moreover, we have demon- from the remainder CP current region for radiation-induced
strated that the deviation, which was observed between the CP traps in MOSFET. However, these methods need at least three
model and experimental data at the maximum plateau of ICP –VL transistors and do not give CP components of LDD and LOCOS
characteristics, depends on the gate pulse fall time and vanishes regions, separately. In addition, the unknown contribution of
for large fall time. This difference has been found to behave like a
geometric component, since it depends on gate length and fall time each region leads to unambiguous understanding of radiation
and disappears for both short gate lengths and long fall times. and BTI effects and subsequently conducts to unreliable mod-
els. Therefore, care has to be taken to distinguish between
Index Terms—Charge pumping (CP), lightly doped drain
(LDD), local oxidation of silicon (LOCOS). the LDD-CP, LOCOS-CP, and geometric-CP components to
develop a reliable model for radiation and BTI effects. There-
fore, it is important to determine the amount of CP curve by
I. I NTRODUCTION
modeling the individual contribution of each MOSFET part.
In this paper, we deeply investigate the different regions of
D UE TO its accuracy and sensitivity, the charge-pumping
(CP) technique [1] is extensively used to characterize
the MOSFET Si/SiO2 interface. Different CP variants have
LDD-MOSFET with LOCOS isolation. An empirical model
for LOCOS-CP, LDD-CP, and effective channel CP is devel-
been proposed in the literature to extract interface-, oxide-, oped and simulated to obtain the whole experimental ICP –VL
and border-trap densities and their energy distributions [2]–[5]. characteristic. Basically, the model employs the modulation
In fact, the CP technique has been used to study many MOSFET of the contributing active-channel area, which depends on the
reliability issues, such as radiation [6], [7], hot carrier [8], [9], position of the high and low level of the gate signal. The
approach used for this model is explicit and accurate, giving
a good correlation between simulated model and experimental
Manuscript received April 18, 2010; revised June 28, 2010; accepted results mostly for short channel length. However, as the channel
August 2, 2010. Date of publication September 20, 2010; date of current version
November 5, 2010. This work was supported by the High School Educational
increases, the geometric component starts to play an important
and Scientific Research Ministry of Algeria under the National Funding of role. Because of limitation on the paper’s length, this point will
Research. The review of this paper was arranged by Editor H. S. Momose. be investigated in another paper.
H. Tahi and B. Djezzar are with the Microelectronics and Nanotech-
nology Division, Centre de Développement des Technologies Avancées This paper is outlined as follows. Section II describes sim-
(CDTA), Algiers 16303, Algeria (e-mail: htahi@cdta.dz; hakimtahi@yahoo.fr; ulation and experiment tools. Model derivation is the subject
bdjezzar@cdta.dz; b_djezzar@yahoo.fr). of Section III. In Section IV, we analyze the simulation results
B. Nadji is with the Microelectroncs et microsystems Laboratory, Dé-
partement d’Automatisation et Electrification Laboratoire d’Electrification of the Elliot Gaussian-like ICP –VL curve of LDD MOSFET
des Entreprises Industrielles, Université M’hamed Bougara de Boumerdès, with LOCOS isolation and the individual CP curve associated
Boumerdès 35000, Algeria (e-mail: b_nadji@yahoo.com). with different regions. The simulated results are compared
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. and discussed regarding the experimental data in Section V.
Digital Object Identifier 10.1109/TED.2010.2068300 Conclusions are drawn in Section VI.
where q(C) is the electron charge, and Dit (in per square
centimeters per electronvolt) and Nit (in per square centime-
ters) are the trap densities. The Ep and Ee are the lower Fig. 2. Illustration of the Vth and Vfb CP distribution in the X- and
and upper boundaries of the active-CP energy interval. They Y -directions in the channel.
are equal to the nonsteady-state emission levels for holes and
According to the simulated LDD-NMOSFET structure given
electrons, respectively [2]. ΔF gives the probability that the
in [16] and illustrated in Fig. 1, (5) can be expressed as
trap at (x, y, E) contributes to CP current [21], [22]. From the
Shockley–Read–Hall statistics and by neglecting the emission ICP = qf ΔF C (x, y)NitC dxdy
of carriers, the evolution with time and space of the filling factor
WG LEFF (VL ,VH )
ΔF of an interface trap situated at an energy E in the silicon
bandgap during one cycle can be expressed as a function of + 2qf ΔF LDD (x, y)NitLDD dxdy
channel position by WG ΔL(VL ,VH )
ΔF (x, y) + 2qf ΔF LOCOS (x, y)
(1−exp(−cp (x, y)Tacc (x, y))(1−exp(−cn (x, y)Tinv (x, y)) ΔW (VL ,VH ) LEFF (VL ,VH )
=
1−exp−cn (x, y)Tinv (x, y)−(cp (x, y)Tacc (x, y) × LOCOS
Nit dxdy
(6) + 4qf ΔF LOCOS−LDD (x, y)
ΔW (VL ,VH ) ΔL(VL ,VH )
where cp (x, y) and cn (x, y) are the capture rate of hole and
electron as a function of channel position (x, y). Tinv (x, y) and × NitLOCOS−LDD dxdy (9)
Tacc (x, y) are the inversion and accumulation times, respec- where LEFF (VL , VH ) is the effective channel length, WG is
tively. Contrary to ΔF given by Bouza et al. [21], in our model, the gate width, ΔL(VL , VH ) is the LDD subdiffusion length
both cp(n) (x, y) and Tacc(inv) (x, y) are varied as a function of under the poly-silicon gate, and ΔW (VL , VH ) is the width of
position (x, y) in the channel due to the nonuniform doping the LOCOS edge. NitC , NitLDD , NitLOCOS , and NitLDD−LOCOS
in the channel. Consequently, we can express Tinv (x, y) and are the interface trap densities in the Channel-Eff, LDD-sub,
Tacc (x, y) as a function of the local Vth and Vfb , respectively, LOCOS-Eff, and LOCOS-LDD regions, respectively.
as (see Fig. 2) In the constant-amplitude CP technique, the CP current is
[(VL + ΔVG − Vth (x, y)] probed from different regions at the same time, depending
Tinv (VL , x, y) = TH + (tr + tf ) on Vth (x, y) and Vfb (x, y) distributions. Therefore, (9) can be
ΔVG
(7) expressed as
[(Vfb (x, y) − VL ] ICP = qf NitC WG LEFF (VL , VH )ΔF C (VL , VH )
Tacc (VL , x, y) = TL + (tr + tf ) (8)
ΔVG + 2qf NitLDD WG ΔL(VL , VH )ΔF LDD (VL , VH )
where TH , TL , tr , and tf are the time high level, time low level, + 2qf NitLOCOS ΔW (VL , VH )LEFF (VL , VH )
rise time, and fall time gate pulse, respectively. Vth (x, y) and × ΔF LOCOS (VL , VH )
Vfb (x, y) are the threshold and flatband voltage distributions in + 4qf NitLOCOS−LDD ΔW (VL , VH )ΔL(VL , VH )
the channel. × ΔF LOCOS−LDD (VL , VH ). (10)
⎡ ⎤
En
⎢ ⎥
ICP = q.f ⎣ ΔF (x, y, E)Dit (x, y, E)dE ⎦ dxdy (3)
active width active length Ep
TAHI et al.: MODELING AND SIMULATION OF CP CHARACTERISTICS FOR LDD-MOSFET 2895
⎧
⎪
⎪ Region(A) → ICP
LDD
= 2qf NitLDD WG ΔL(VL , VH )ΔF LDD (VL , VH )
⎨
Region(C) → ICP = qf NitC WG LEFF (VL , VH )ΔF c (VL , VH ) + 2qf NitLDD WG ΔL(VL , VH )ΔF LDD (VL , VH )
C
(11)
⎪
⎪ +2qf NitLOCOS ΔW (VL , VH )LEFF (VL , VH )ΔF LOCOS (VL , VH )
⎩
Region(E) → ICP
LOCOS
= 2qf NitLOCOS LEFF (VL , VH )ΔW (VL , VH )ΔF LOCOS (VL , VH )
2896 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 57, NO. 11, NOVEMBER 2010
Fig. 4. Determination of Vth (x, y) and Vfb (x, y) spatial distribution. (a) Cut plan in the LOCOS region. (b) and (c) Electron spatial distribution contour at cut
plan position. (d) Vth (x, y) and Vth (x, y) spatial distribution for two transistors. In (a), (b), and (c), the polarization conditions are gate voltage Vg = 0.4 V,
drain bulk, and source bulk voltage Vr = 0.1 V.
Fig. 9. Comparison between experimental and calculated ICP –VL characteristics of LDD-NMOSFET with WG /LG = 10/1. (a) Current in logarithmic scale.
(b) Current in linear scale. The CP conditions are: Vr = 0.1 V, ΔVG = 4 V, f = 1 MHz, tr = tf = 10 ns.
The rising edge of both experimental and total simulated higher the difference. This divergence for gate length greater
ICP –VL characteristics presents two tails. The first tail is just than 1 μm is mainly caused by the CP of the Channel-Eff
under Vth of the effective channel and located in regions (B) region, since only this region varies with channel length. The
and (C) [see Fig. 9(a)]. The second tail is located in region other regions are still unchanged since all transistors are fab-
(A) and extends from the end of the first tail deeply toward the ricated by the same process on the same chip in the same
negative bias. The same behavior has been reported by Habas production line.
[24]. They are mainly caused by nonuniformity of doping To deeply investigate the origin of this discrepancy, we have
distribution across Channel-Eff/LDD-Sub and LDD regions, calculated and measured ICP –VL curves for transistors with
respectively. The first tail only occurs in ICP –VL of the LDD different gate sizes at different tf and tr . All ICP –VL curves
devices and has no equivalent in ICP –VL of the conventional exhibit the same behavior with WG /LG and tf (tr ). In addition,
devices, and the second tail looks like the tail of the con- the ICP –VL curve edges are also the same for transistors with
ventional devices [24]. Indeed, Heremans et al. [25] reported 10/0.75, 10/0.65, 10/0.55, 1/10, 5/10, and 2/10 (ICP –VL curves
that the conventional device ICP –VL presents one current tail not shown here). We plot in Fig. 11 the differences between
for VL smaller than Vth − ΔVG . This tail was theoretically experimental and simulated ICPmax normalized with respect
calculated and compared with experiments [24], [25]. They to the gate area as a function of gate length and width. This
obviously found a very good correlation between calculated difference is equal to the leakage current for LG ≤ 1 μm but
and measured data. However, they did not include the LOCOS exhibits a linear behavior in logarithmic scale for LG > 1 μm.
edge effect in their calculations to evaluate the slope of the Moreover, Fig. 11 plainly shows that this difference does not
falling side of the ICP –VL curves. Some authors [21], [22] have depend on the gate width. To better understand the origin of
found a similar ICP –VL calculation and experiment agreement this deviation, a thorough analysis on its time dependence has
regarding the ICPmax at the interface and inside the oxide but been carried out. Fig. 12 shows the current differences as a
different ICP –VL rising and falling edges. function of pulse fall time for different gate lengths and widths.
In this paper, not only the rising edge of the ICP –VL curve is All time-dependent deviations reveal the same feature and are
calculated, but also the falling edge by including the LOCOS identical with gate width but different with gate length. In
effect. This last part of ICP –VL presents a good correlation addition, this difference is independent of the pulse fall time for
with the experimental ICP –VL characteristics for all transistors. 1-μm gate-length transistor and equal to the leakage current.
It is influenced by doping nonuniformity and the variation In other terms, the calculated and simulated CP curves are
of oxide thickness in the LOCOS region (bird-beak oxide). the same. All transistor curves tend to converge to the same
Nevertheless, our calculation model does not take into account value as tf increases. They coincide at 500 ns, where the
the CP at low frequencies to probe traps in the oxide. It would deviation is about leakage current (∼10−12 A). This means
be valuable to consider this point in a future work. that the difference disappears and the calculated model matches
In contrast, Fig. 10 shows that ICPmax of the calculated well the experiment for all transistors. According to the above
ICP –VL curves diverges from experiment data when the gate analysis, the deviation seems to have the same characteristics
length is increased. Fig. 10(a)–(d) shows plots for transistors as the geometric component regarding gate length and fall time
with WG /LG = 10/5 and WG /LG = 10/10 in log scale to dependences [1], [2]. This is why we believe that the geometric
show that the tails (LOCOS side and LDD side) remain un- current is the origin of the model and measurement mismatch.
changed and in linear scale to illustrate the vertical deviation Care has to be taken for this component not to pollute the CP
dependence on the gate length. The longer the gate length, the current and cause overestimation when it is used to evaluate
TAHI et al.: MODELING AND SIMULATION OF CP CHARACTERISTICS FOR LDD-MOSFET 2899
Fig. 10. Comparison between experimental and calculated ICP –VL characteristics of LDD-NMOSFET. (a) and (b) LDD-NMOSFET with WG /LG = 10/5.
(c) and (d) LDD-NMOSFET with WG /LG = 10/10. The CP conditions are the same as in Fig. 9.
device reliability such as BTI and radiation effect. Because provide more correct evaluation of BTI and radiation stresses,
of paper length, this issue will be separately analyzed and subconsequently enabling to develop accurate models for de-
addressed in next paper. vice life time prediction.
By computing each region CP current separately, the method In the future work, it would be beneficial to carry out analysis
presented in this paper can help to deeply analyze the transistor on geometric current. We will complete this model by adding
reliability, particularly the radiation- or BTI-induced traps in this component and also show how it should be integrated with
different regions of the channel. This approach could be able to different CP current regions.
2900 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 57, NO. 11, NOVEMBER 2010
Boualem Djezzar was born in Algiers, Algeria, on Bacharia Nadji was born in Béchar, Algeria, in
December 3, 1965. He received the Diplôme des 1961. She received the “Ingénieur” degree in elec-
Etudes Supérieures (DES) degree in solid-state phys- tronics from the University of Sidi Bel Abbes, Sidi
ics from the University of Constantine, Constantine, Bel Abbes, Algeria, in 1985 and the Ph.D. degree for
Algeria, the Diplome des Etudes Approfondies her work on anodic silicon oxide thin films from the
(DEA) degree in microelectronics from the Univer- University Joseph Fourier de Grenoble, Grenoble,
sity of Grenoble, Grenoble, France, in 1991, and the France, in 1990.
Master’s degree in thin film from the University of From 1990 to 1992, she was a physics Teacher
Algiers, Algiers, in 2000. with the University of Savoie Chambéry, Chambéry,
From 1990 to 1992, he worked on thin-film depo- France. Since 1992, she has been teaching with
sition under ultrahigh vacuum of rare earth (Erbium) the University of M’Hamad Bougara, Boumerdes,
Silicide on Si (111) with the LEPES Laboratory, CNRS, Grenoble. Since Algeria. In 2007, she became “maitre de conférences A” and has been the Head
1992, he has been with the Microelectronics and Nanotechnology Division, of the Microelectronic and Microsystems Laboratory. She is also responsible
Centre de Développement des Technologies Avancées, Algiers. He worked on for the infotronic postgraduate. Her current research focuses on material and
the technological process simulation and modeling of CMOS, BiCMOS, and device reliability.
CBiCMOS. In 1997, he started working with the Caracterisation des Dispositifs
à Semiconducteurs (CDS) Group on MOS device reliability and electrical
characterization of radiation-induced traps in silicon dioxide SiO2 and interface
Si/SiO2 in MOS devices. He is the author or coauthor of more than 40
papers published in refereed journal and refereed conference proceedings on
modeling/simulation devices and radiation effect on MOS devices. His current
research interests include CMOS technology reliability, particularly electrical
characterization and modeling of radiation and NBTI effects.