Received 22 September 2014; revised 12 November 2014 and 7 December 2014; accepted 11 December 2014.

Date of current version 22 April 2015.

The review of this paper was arranged by Editor S. S. Chung.
Digital Object Identifier 10.1109/JEDS.2014.2382759

1T Capacitor-Less DRAM Cell Based on

Asymmetric Tunnel FET Design
ARNAB BISWAS, (Student Member, IEEE) AND ADRIAN M. IONESCU, (Senior Member, IEEE)
Nanolab, Ecole Polytechnique Fdrale de Lausanne, Lausanne 1015, Switzerland
CORRESPONDING AUTHOR: A. Biswas (e-mail: arnab.biswas@epfl.ch)

This work was partly supported by the European STEEPER project under Grant Agreement No. 257267.

ABSTRACT In this work we propose and demonstrate the use of a Tunnel FET (TFET) as capacitorless
DRAM cell based on TCAD simulations and experiments. We report more experimental results on
Tunnel FETs implemented as a double-gate (DG) fully-depleted Silicon-On-Insulator (FD-SOI) devices.
The Tunnel FET based DRAM cell has an asymmetric body and a partial overlap of the top gate (LG1 )
with a total overlap of the back gate over the channel region (LG2 ). A potential well is created by biasing
the back gate (VG2 ) in accumulation while the front gate (VG1 ) is in inversion. Holes from the p+
source are injected by the forward-biased source/channel junction and stored in the electrically induced
potential well. Programming conditions and related transients are reported and the role of temperature is
investigated.

INDEX TERMS Tunnel FET, DRAM, Capacitorless memory, 1T/0C

I. INTRODUCTION in SOI transistors to store the information i.e., to hold

device geometries shrink, the scaling of the con-
A S
ventional 1Transistor/1Capacitor (1T/1C) Dynamic
RAM (DRAM) has become increasingly difficult, in par-
the charge. In a nMOSFET based 1T-DRAM, the 1-state
current level is triggered by the excess hole charge gen-
erated into the body by one of aforementioned techniques.
ticular due to the capacitor which has become harder to However the read out operation is based on a MOSFET oper-
scale. The memory industry has achieved a lot of success ation (drift/diffusion of carriers over a potential barrier) in
in packing more and more bits per unit area in a sili- all cases.
con die. We are fast approaching the scaling limits for the Tunnel FETs as steep slope switches are very promising
physical capacitor element in a 1T/1C DRAM, and new devices to respond to the demanding requirements of future
materials and new device and integration approaches are technology nodes [10], [11]. The benefits of Tunnel FETs are
required to meet the market demands. The concept of capac- specially linked to their potential for its sub-60mV/decade
itorless single-transistor (1T) DRAMs was first reported in subthreshold swing, a prerequisite for scaling the supply
the early 90s [1], [2], however they have attracted atten- voltage well below 1V. While high-performance Tunnel
tion [3][6] in more recent times. This is due to the ability FETs exploit InAs-Si hetero-structures on nanowires [12],
of (1T) DRAM to achieve higher memory cell density and the all-Si Tunnel FET family appears primarily to have a
to solve the problems associated with the scaling of a very high ION /IOFF and the lowest IOFF [13], but they fail to
physical capacitor. These memory cells can be classified achieve the requirements of high ION for a particular tech-
in different groups with according to the excess major- nology node [14]. An interesting property of a Tunnel FET
ity carrier generation method i.e., depending on how the is the fact that its sub-threshold slope is weakly dependent
1-state is programmed. This extra charge can be achieved on temperature [15]. Tunnel FETs can thus be considered
mainly by four ways: impact ionization [7], bipolar junction as very good candidates for operating at higher temperature,
transistor (BJT) effect [8], band-to-band (B2B) tunneling without compromising reliability. In Section IV-B, we will
generation [9] and gate tunneling current [5]. The (1T) see how the read-out operation of the memory cell is not
DRAMs take advantage of the parasitic floating body effects degraded by temperature.
2168-6734 2015 IEEE. Translations and content mining are permitted for academic research only.
Personal use is also permitted, but republication/redistribution requires IEEE permission.
VOLUME 3, NO. 3, MAY 2015 See http://www.ieee.org/publications_standards/publications/rights/index.html for more information. 217
 BISWAS AND IONESCU: 1T CAPACITOR-LESS DRAM CELL BASED ON ASYMMETRIC TUNNEL FET DESIGN

FIGURE 1. Simulated potential profile at 1 nm below the

gate-oxide/channel interface for both DG-MOSFET and DG-TFET.

II. BACKGROUND
In this paper, we further build upon [16] that showed
that all-Si double-gate (DG) Tunnel FET can serve
for building a new class of devices: the capacitor-
less Tunnel FET DRAM, where the very low IOFF is
offering low refresh rate and the zero-capacitor struc-
ture very high potential for scalability. We report a
detailed simulation study and some experimental results
on capacitorless Tunnel FET DRAM cell implemented as FIGURE 2. (a) SEM image of a fabricated FD-SOI Tunnel FET showing the
a double-gate fully-depleted Silicon-On-Insulator (FD-SOI) top gate partially covering the channel. (b) Independent gate FinFET device
structure for TFET based capacitorless DRAM.
device. The device has an asymmetric design, with
a partial overlap of the top gate (LG1 ) and with a
total overlap of the back gate over the channel region
(LG2 = LG1 +LIN ), which creates the necessary condition to Fig. 2(a) [13]. It is known that for efficient Tunnel FET
store holes injected from the source-to-body junction in operation a fully depleted body is required. However for
an electrically induced potential well near the drain. The charge storage in the body partial depletion is preferable. In
potential well is created by biasing the back gate (VG2 ) order to meet both conditions a step like fin is used, which
in accumulation while the front gate (VG1 ) is in depletion manages the trade-off between the electrostatic control of the
and/or inversion. Gate1 and the ability to store charges by Gate2. Simulations
In Fig. 1 we see a comparison between the potential pro- showed that 50-60nm fin widths for the partially depleted
files at the surface of the channel for a DG-MOSFET and part were optimum. Gate1 partially covers the channel region
DG-Tunnel FET. The green line shows that, for the DG- and will be used as the main control gate for Tunnel FET
MOSFET a potential well already exists which can hold the operation. Gate2 on the other hand covers the entire chan-
charges. The challenge in this case as discussed earlier is to nel region and as explained below will be used to induce a
generate the excess carriers. Whereas in case of a DG-Tunnel potential well in the channel region. This potential well will
FET as shown by the red curve, there is no potential well then be used to store charges for memory operation.
by default due to the two different doping in the source and
drain region. However, we do have a reservoir of both types A. PRINCIPLE OF OPERATION
of carriers in source/drain region which can easily brought to To understand the principle of operation let us take a 2D cut
the body area by forward biasing the source/channel junc- of the device shown in Fig. 2 along XY plane. Simulations
tions. The challenge in case of a Tunnel FET is thus to were done in Synopsys Sentaurus TCAD [17]. The devices
create and maintain a potential well which could hold these simulated were with fin width of 25nm in the source side
extra charges. In the following section we will discuss about and 50nm in the drain side, fin height = 65nm; Gate1
how we can design a structure which would facilitate the length (LG1 ) = 80nm; Gate2 length (LG2 ) = 200nm and
formation of a potential well and still work as conventional a pocket length LIN = 120nm, unless otherwise specified.
Tunnel FET. A source/drain doping of 1x1020 cm3 of Boron/Arsenic was
used. The channel was intrinsically doped. 3nm of SiO2 gate
III. FIN-TUNNEL FET AS 1T/0C DRAM oxide was used for both Gate1 and Gate2. The same metal
For this study the Tunnel FET architecture used is shown in work-function was used for both the gates. Fig. 3(a) shows
Fig. 2(b). It is basically an independent gate Fin-Tunnel the 2D hole-density plot with VG1 = 2V, VG2 = 1V,
FET implementation of the FD-SOI devices shown in VD = 0V, VS = 0V. The memory operation is based on the

218 VOLUME 3, NO. 3, MAY 2015

BISWAS AND IONESCU: 1T CAPACITOR-LESS DRAM CELL BASED ON ASYMMETRIC TUNNEL FET DESIGN

-0.08
-0.08 hDensity [cm^-3]
3.75e-14 1.83e-08 8.89e-03 4.32e+03 2.10e+09 1.02e+15 4.98e+20
-0.06
-0.06

-0.04
-0.04
Y

Y
-0.02
-0.02
GATE1
00
SOURCE DRAIN
0.02
0.02

0.04
0.04
GATE2
00 0.1
0.1 0.2
0.2 X 0.3
0.3 0.4
0.4 0.5
0.5

X [m]
(a)

(a)

(b)

FIGURE 3. 2-D cross-section of the Fin-TFET showing the hole density at

the end of a write 1 operation (top). Hole density and electrostatic
potential at a cross section 1nm above the Gate2 showing the formation of
potential well/hole pocket (bottom).

(b)

creation of an induced potential well achieved by biasing FIGURE 4. (a) Simulated transfer curves at various Gate2 bias.
(b) Simulated hysteresis curves observed with varying LIN with same
Gate2 with a negative potential. A small positive bias on biasing conditions. Devices with longer LIN shows stronger hysteresis.
the source can help to flood the body with excess holes. VG1 = 2 V, VD = 1 V.
These excess holes are then trapped in the potential well
close to Gate2 as shown in Fig. 3 (top). Fig. 3(b) shows length scaling. There will be a minimum LIN required for
the cross-section at 1nm above the Gate2. We can clearly the memory cell to work satisfactorily.
see the build-up of holes in the hole-density plot. The figure The dependence of charge storage on LIN is also observed
also shows an induced potential well where the holes can be in measurements from the devices described in [16]. The
stored. Hence, as the device turns on, the presence or absence relaxation time is observed to be more prominent for devices
of the excess carriers will affect the threshold voltage of the with longer LIN. Fig. 5(a) shows that depending on the length
device indicating two different memory states. of the intrinsic region (LIN ) the discharge time varies. The
The simulated transfer curves of the device in Fig. 1 is device with longer LIN = 500nm has a slower discharge
shown in Fig. 4(a). This is similar to what has been measured than the device with LIN = 200nm. This means that longer
in [16] for FD-SOI devices. The principle of charge storage the LIN , the more charge it stores and hence it takes longer
can be further verified by the hysteresis curves showed in to discharge. This experiment also demonstrates the impor-
Fig. 4(b). With VG1 = 2V and VD = 1V, as the back gate tance of charge storage in the un-gated region. The discharge
bias is swept back and forth the charge storage in the par- time can be related to the retention time of a conventional
tially depleted region results in two different current paths. capacitorless DRAM, as it indicates how long the device
When the VG2 goes from 1V to 2V in the first sweep, can hold the charge once the write cycle is complete. In
holes start to accumulate in the induced potential well in this context we observe a discharge time in the order of few
the body affecting the body potential, resulting in the lower milliseconds for devices with longer LIN . The same prin-
branch on current in the plot. As VG2 increases to +2V, the ciple can also be observed with steady state measurements
accumulated holes are evacuated to the drain at this state. in Fig. 5(b) and (c). The device with a longer LIN clearly
Now, as VG2 goes back to 1.5V in the following sweep, shows a stronger history effect. This observation validates
the drain current follows the top branch due to the lack of the simulation results shown in Fig. 4.
the excess charges in the body. Based on this observation,
we can set the design a programming scheme of a mem- B. PROPOSED PROGRAMMING SCHEME
ory cell with read potential at Gate2 at 1.5V. A fast sweep The proposed programming scheme for the device in
rate of 1mV/sec was used in this transient simulation for Fig. 3(a) is shown in Table 1.
convergence issues. (i) WRITE 1: The write 1 step involves biasing the
It is also interesting to note that the hysteresis is stronger source with a small positive voltage, together with a negative
for devices with longer LIN . This is expected as length of the bias on Gate2. This would push holes into the body which
partially depleted region greatly influences the charge storage will then be trapped in the induced potential well caused by
as also explained in [16]. So in this particular structure, the Gate2 as shown in Fig. 3(b). Bias values used in simulation,
memory cell will loose its retention characteristics with gate VG1 = 0V, VG2 = 1V, VS = 0.25V, VD = 0V.

VOLUME 3, NO. 3, MAY 2015 219

 BISWAS AND IONESCU: 1T CAPACITOR-LESS DRAM CELL BASED ON ASYMMETRIC TUNNEL FET DESIGN

TABLE 1. Proposed Programming Scheme of Fin-TFET Based Capacitorless DRAM.

(a)

(b) (c)
FIGURE 6. Simulated 2-D hole densities after a (a) WRITE 1 and
FIGURE 5. (a) Different relaxation times in the millisecond range are (b) WRITE 0 operation. (c) Simulated potential profile at 1 nm below Gate1
recorded when the back gate is biased at VG2 = 10 V (after writing 1), showing the hole pocket.
depending on the length of the LIN region. The TFET with LIN = 500 nm
shows the longer discharge time while it is negligible in devices with
LIN = 200 nm. (b) & (c) Measured drain current with respect to back gate
voltage at fixed front gate and drain voltages, VG1 = 4.5 V, VD = 4 V.
and WRITE 0 operations. The results are given in Fig. 6.
A clear difference is present showing that more holes are
(ii) WRITE 0/ERASE: The write 0 step would simply stored under the front gate after WRITE 1 compared to
mean putting a positive bias to the Gate2 (~2V). This will after WRITE 0 state. The same principle is also observed
remove any induced potential well from the previous state in the potential profiles at 1nm below the Gate1. After a
and the holes in the body diffuse back to the source or recom- WRITE 1 operation, there is a hole pocket present in the
bine to drain. Bias values used in simulation, VG1 = 0V, channel region. Similarly after a WRITE 0 operation, the
VG2 = 2V, VS = 0.25V, VD = 0V. hole pocket is much smaller. This hole pocket will primar-
(iii) HOLD: After the write 1 or 0 step the source goes ily act as a resistance to current flow. Now as the device
back to zero and a small positive bias on Gate2 holds on is turned on, the presence or absence of this excess carriers
to excess charges (if any) in the body. Bias values used in will affect drive current of the device indicating two different
simulation, VG1 = 0V, VG2 = 1.0V, VS = 0V, VD = 0V. memory states.
(iv) READ: The readout operation is carried out via Tunnel
FET operation with BTBT from p+ source to the intrinsic IV. EXPERIMENTAL RESULTS
channel. For this, the drain is biased at VDD and VG1 > A. TRANSIENT MEASUREMENTS
threshold voltage (VTH ) of the device. Source remains at Transient measurements were carried out FD-SOI Tunnel
zero bias for the read operation. Gate1 controls the read out FET devices [13] with 20 nm Silicon layer thickness and
current. The threshold voltage of the device is affected by 145 nm of BOX, according to the programming scheme is
the presence or absence of excess carriers in the body. Bias depicted in Table 2. This scheme is different from what was
values used in simulation, VG1 = 2V, VG2 = 1.5V, VS = 0V, reported in [16] in order to better understand the principle
VD = 1V. of operation. The timing diagrams of READ and WRITE
Transient simulations were done to observe the hole den- operations are shown in Fig. 7. The Gate1, Gate2 and drain
sity in the body in HOLD mode preceded by WRITE 1 were pulsed according to Fig. 7(a) below. The Agilent 4156C

220 VOLUME 3, NO. 3, MAY 2015

BISWAS AND IONESCU: 1T CAPACITOR-LESS DRAM CELL BASED ON ASYMMETRIC TUNNEL FET DESIGN

TABLE 2. Programming Conditions for Indicated Operations of

Capacitorless 1T TFET DRAM with LG1 = 400 nm, LIN = 200 nm.

State V D [V] V G1 [V] V G2 [V] V S [V]

WRITE 1 0.5 0 -10 0
WRITE 0 0 0 +10 0
READ 4 4 -10 0

semiconductor parameter analyzer was used in sampling

mode for the tranisent measurements. The source current was
recorded at every 60s (limited by the measurement setup).
During the WRITE cycle the Gate1/drain cycle is 50%
phase shifted to the Gate2 pulse. This was to ensure that
induced potential well stays as the device is turning ON after
a WRITE 1. As highlighted in Fig. 7(b), after a WRITE 1
operation the Gate2 switches from 10V to +10V, with the
device still ON the potential well ceases to exist and the
stored charges can now escape. In other words the capac-
itor discharges as the Gate2 switches back to +10V. This
capacitor discharge can be observed in the slow transient in
Fig. 7(b). This experiment demonstrates the importance of
the charge storage in the hole pocket outside the front gate
overlap, controlled by VG2 . As shown in Fig. 7(d) a memory
effect is thus observed.
During the ERASE cycle the Gate1/drain pulse is in oppo-
site phase to the Gate2 pulse. Finally the READ cycle has the
same biasing conditions (most negative Gate2 bias) for both
WRITE and ERASE. The device turns ON at the most nega-
tive Gate2 bias i.e., 10V. Depending on the previous state,
a clear difference in READ (source) current levels (IS ) of
10nA is observed (Fig. 7(d)).
In the experimental setup, the biasing conditions are
slightly different than what has been suggested in
Section III-B. Most importantly a HOLD state could not
be experimentally demonstrated, because of the very low
ON current drive (~10nA) of the devices measured. The
reported high values of VG2 in this section as compared to
Section III-B is due to the fact that the measured device
was a planar device on a SOI substrate with a 145nm BOX. FIGURE 7. Timing diagram for different operations described in Table 2.
Hence, a high VG2 was required to have the same effect as Consecutive READ and WRITE operations for both states. The difference in
READ current for states 1 and 0 is 10 nA, as highlighted, a memory effect
2V (VG2 ) in the simulated FinFET structure. Also it was is observed. Programming voltages are: VG2 = 10 V for TFET with
not possible to measure the source current and at the same LG1 = 400 nm, LIN = 200 nm.
time pulse the source potential (VS kept at a constant poten-
tial) due to the equipment limitations in the measurement
setup. In erase operation, the influence of positive source is B. TEMPERATURE DEPENDENCE
partially compensated by using highly positive gate bias at The transfer characteristics of the reported Tunnel FET
Gate2 (VG2 ). In read operation drain and Gate1 potentials devices have been measured from 25C to 85C and they
are chosen such that a desired ON current drive is achieved. show almost unchanged sub-threshold swing over the entire
Another difference in the biasing scheme can be observed range as highlighted in the inset of Fig. 8(a) and low IOFF .
in VD for write operation. This has been done to stop the This is expected as band-to-band tunneling current is only
holes to recombine at the drain side. Also the Gate2 bias weakly dependent on temperature [15]. Most importantly in
for read is different based on the observed hystory effect. Fig. 8(b), the discharge cycle (as highlighted in Fig. 7(b))
At last Gate1 (VG1 ) is biased positively in all the opera- of the source current after a READ 1 operation, remains
tion modes again due to the setup limitations. The reason unaffected (at T = 25C, 55C, 85C). This reflects the
of the low frequency choice is based on the ON current retention ability of the memory cell at high temperatures,
levels of the fabricated devices and the high parasitic capac- in contrast with results reported on MOSFET-based solu-
itance contributed by non optimized test pads and the test tions [8], [18], wherein the retention ability of the devices
instruments. degrade with increasing temperature.

VOLUME 3, NO. 3, MAY 2015 221

 BISWAS AND IONESCU: 1T CAPACITOR-LESS DRAM CELL BASED ON ASYMMETRIC TUNNEL FET DESIGN

[5] T. Tanaka, E. Yoshida, and T. Miyashita, Scalability study on a

capacitorless 1TDRAM: From single-gate PD-SOI to double-gate
FinDRAM, in Proc. IEEE Int. Electron Devices Meeting Tech. Dig.
(IEDM), San Francisco, CA, USA, 2004, pp. 919922.
[6] C. Hu and T. King, A capacitorless double-gate DRAM cell, IEEE
Electron Device Lett., vol. 23, no. 6, pp. 345347, Jun. 2002.
[7] S. Okhonin, M. Nagoga, J. Sallese, and P. Fazan, A SOI capacitor-
less 1T-DRAM concept, in Proc. IEEE Int. SOI Conf., Durango, CO,
USA, 2001, pp. 153154.
[8] S. Okhonin, M. Nagoga, E. Carman, R. Beffa, and E. Faraoni, New
generation of Z-RAM, in Proc. IEEE Int. Electron Devices Meeting,
Washington, DC, USA, 2007, pp. 925928.
[9] E. Yoshida and T. Tanaka, A capacitorless 1T-DRAM technology
using gate-induced drain-leakage (GIDL) current for low-power and
highspeed embedded memory, IEEE Trans. Electron Devices, vol. 53,
no. 4, pp. 692697, Apr. 2006.
[10] A. M. Ionescu and H. Riel, Tunnel field-effect transistors as energy-
efficient electronic switches, Nature, vol. 479, no. 7373, pp. 329337,
Nov. 2011.
[11] L. Knoll et al., Inverters with strained Si nanowire complementary
tunnel field-effect transistors, IEEE Electron Device Lett., vol. 34,
no. 6, pp. 813815, Jun. 2013.
[12] K. E. Moselund et al., InAs-Si nanowire heterojunction Tunnel
FETs, IEEE Electron Device Lett., vol. 33, no. 10, pp. 14531455,
FIGURE 8. (a) Measured transfer characteristics with different Gate2 Oct. 2012.
voltage at 25 C and 85 C temperature. As expected for a TFET the [13] F. Mayer et al., Impact of SOI, Si(1-x)Ge(x)OI and GeOI substrates
sub-threshold swing has a negligible temperature dependence. on CMOS compatible Tunnel FET performance, in Proc. IEEE Int.
(b) Measured relaxation times as a function of temperature after a READ Electron Devices Meeting, vol. 4. San Francisco, CA, USA, Dec. 2008,
1 operation. A minor dependence of the relaxation behavior on the pp. 15.
temperature is experimentally demonstrated. [14] ITRS. (2011). ITRS International Technology Working Groups.
International Technology Roadmap for Semiconductors. [Online].
Available: http://www.itrs.net
[15] M. Born and K. Bhuwalka, Tunnel FET: A CMOS device for high
V. CONCLUSION temperature applications, in Proc. Int. Conf. Microelectron. (MIEL),
In summary we have presented simulation results on a new Belgrade, Serbia, 2006, pp. 1417.
asymmetric Tunnel FET structure that can be used as a [16] A. Biswas et al., Investigation of tunnel field-effect transis-
tors as a capacitor-less memory cell, Appl. Phys. Lett., vol.
capacitorless DRAM. We have also experimentally demon- 104, no. 9, Mar. 2014, Art. ID 092108. [Online]. Available:
strated a scalable implementations of such 1T capacitorless http://scitation.aip.org/content/aip/journal/apl/104/9/10.1063/1.4867527
DRAM cell based on DG Tunnel FET with a IS of 10nA. [17] Synopsys Sentaurus TCAD User Guide Ver: H2012.03, Synopsys, Inc.,
Mountain View, CA, USA, 2012.
Retention times in the order of 100s of microseconds to [18] C. Kuo, T.-J. King, and C. Hu, A capacitorless double-gate DRAM
few milliseconds, at room temperature in devices with chan- cell design for high density applications, in Proc. IEEE Int. Electron
nel length varying from 400nm to 1m were observed. Device Meeting, San Francisco, CA, USA, 2002, pp. 843846.
The discharge cycle was shown to be unaffected when the
temperature varies between 25C to 85C. ARNAB BISWAS (S08) was born in Darjeeling,
India. He received the bachelors degree in elec-
tronics and communication from the Sikkim
ACKNOWLEDGEMENT Manipal Institute of Technology, Gangtok, India
and the masters degree in nanotechnology from
The authors would like to thank Cyrille Le Royer of VIT University, Vellore, India. He is currently pur-
CEA-LETI for providing the Tunnel FET devices mea- suing the Ph.D. degree from NANOLAB-cole
sured in this work. The authors also thank Nilay Dagtekin, Polytechnique Fdrale de Lausanne, Lausanne,
Switzerland, under the guidance of Prof. A. M.
Wladyslaw Grabinski and Antonios Bazigos for their help Ionescu. He later worked for Tata Consultancy
during the meaurements. Services Ltd., Mumbai, India, as a Software
Engineer, for two years. His current research interest focuses on modeling
and simulation of semiconductor devices.
REFERENCES
[1] M. Tack and M. Gao, The multistable charge-controlled memory ADRIAN M. IONESCU (SM06) received the B.S.
effect in SOI MOS transistors at low temperatures, IEEE Trans. and M.S. degrees from the Polytechnic Institute
Electron Devices, vol. 31, no. 5, pp. 13731382, May 1990. of Bucharest, Bucharest, Romania, and the Ph.D.
[2] H.-J. Wann and C. Hu, A capacitorless DRAM cell on SOI substrate, degree from the National Polytechnic Institute
in Proc. IEEE Int. Electron Devices Meeting, Washington, DC, USA, of Grenoble, Grenoble, France, in 1989 and
1993, pp. 635638. 1997, respectively. He is a Professor with the
[3] P. Agopian et al., Experimental analog performance of pTFETs as Swiss Federal Institute of Technology, Lausanne,
a function of temperature, in Proc. IEEE Int. SOI Conf., Napa, CA, Switzerland. He has held a staff and/or visit-
USA, 2012, pp. 12. ing positions at Laboratoire dlectronique des
[4] M. Bawedin, S. Cristoloveanu, A. Hubert, K.-H. Park, and technologies de linformation-Commissariat
F. Martinez, Floating body SOI memory: The scaling tourna- lEnergie Atomique et aux Energies Alternatives,
ment, in Semiconductoron-Insulator Materials for Nanoelectronics Grenoble, cole nationale suprieure dlectronique et de radiolectricit
Applications, A. Nazarov et al., Eds. Berlin, Germany: Springer, 2011, de Grenoble, Grenoble, and Stanford University, Stanford, CA, USA, in
pp. 393421. 1998 and 1999.

222 VOLUME 3, NO. 3, MAY 2015