Memory and Learning behaviour of ZnO based transparent synaptic thin film
transistors
Premlal B. Pillai and M. M. De Souza
Department of EEE, University of Sheffield, S1 3JD, Sheffield, UK
Introduction
:
The development in mimicking memory or learning behaviours of biological systems by
nanoscale ionic/electronic devices has spurred a great deal of interest in the scientific community in realising
neuromorphic systems
1-2
. Synaptic devices utilise a voltage pulse on the gate electrode as a presynaptic spike to
trigger an excitatory post synaptic current/conductance (EPSC) on the channel, similar to the dendritic synapses
in biological systems. The EPSC is measured as the time dependent channel conductance after the application of
a voltage pulse on the gate electrode. If the EPSC signal lasts from few seconds to tens of minutes, it is
considered as the analogue of a Short Term Memory (STM) in psychology whereas an EPSC signal lasting from
few hours to lifetime is considered as a Long-Term Memory (LTM) transition. Synaptic plasticity is the ability
of synapses to strengthen or weaken over time. Spike-Timing-Dependent Plasticity (STDP), measured as the
growth/decay of the EPSC and short-term memory to long-term memory transition are considered as the two
main synaptic learning rules. In the quest for realising physical devices with synaptic functions, three terminal
transistors with tuneable channel conductivities were initially proposed
3
. More recently, oxide based synaptic
transistors were demonstrated using nanogranular Silicon dioxide based proton conductor films
4
. The main
drawback of such synaptic devices are the requirements of a certain level of humidity to function as a synaptic
FET and considerably higher power consumption of CMOS based synaptic devices (900 pJ).
4,5
The performance
of the synaptic devices presented in this study are not dependent on the environmental factors and developing
such synaptic transistors with biocompatible materials such as ZnO offers new possibilities for realising
synaptic memory devices that feature lower processing time and cost.
Experiment and results
: The devices composed of radiofrequency sputtered semiconducting layers of ZnO and
insulating Tantalum Oxide (Ta
2
O
5
)
on Indium Tin Oxide coated glass substrates. The capacitance and electrical
characteristics were measured by using Agilent E4980A LCR meter and Keithley (4200 SCS) respectively. The
devices exhibit superior memory windows greater than 2V with in an operation voltage of ± 4V, utilizing the
mobile oxygen vacancies present in the insulator. A strong lateral modulation of the channel conductance using
a side gate is observed due to the oxygen vacancy related Electric double layer (EDL) capacitance effect. EPSC
signals for a range of pulse widths (6-250 ms), magnitude (0.2-3 V) and frequency (1-83 Hz) are analyzed and
compared with other device technologies (table 1).
Device
W/L (µm) Synaptic weight (EPSC@
N=10)
Energy
dissipation
(Vg=0.2V, 10 mS)
Memory
on-off ratio
Max.EPSC @
0.3V
IZO
4
150/300 45µS, Vs = 1V, 100mS
160 pJ
-
-
IZO
6
1000/180 1.2 µS, Vs = 0.5V, 10mS 45 pJ
-
30 nA
SnO
8
1000/10 -
-
<10
3
-
IGZO
9
500 (cir) 5-30 nS, Vs=6V, 100 mS -
-
<10 nA
This
work
300/10
50 µS, Vs=3V, 86 mS
38 pJ
10
6
300 nA
Table 1: comparison of the performance of the reported synaptic devices
4,6,8-9
Conclusion
:
Synaptic behavior of low temperature processed ZnO/Ta
2
O
5
thin film transistors analyzed for the
first time on the basis of spike timing dependent plasticity and EPSC
as the synaptic weight revealed enhanced
saturated EPSC signal > 300 nA at V
pulse
= 0.3V, significantly better than value of 10-30 nA reported for IZO
and IGZO synaptic devices
6,9
. The ZnO based synaptic devices proposed here are a viable, low cost alternative
to current CMOS based three terminal synaptic devices.
References:
1
C. Sanchez et.al., Nat. Mater. 4, 277 (2005),
2
D. Kuzum, et. al., Nanotechnology, 24, 382001 (2013),
3
S.
Brink, et. al., Proceedings of the IEEE International Symposium on Circuits and Systems (ISCAS)432 (2008),
4
C. J. Wan et. al., Nanoscale 5, 10194 (2013),
5
G Indiveri, et.al., IEEE Trans. Neural Networks 17, 211 (2006),
6
L. Q. Zhu et. al., Nat. Commun. 5,3158 (2014),
7
S-M. Yoon et.al., Adv. Funct. Mater. 20, 921 (2010),
8
J. A.
Caraveo-Frescas, et. al., Scientific Reports 4, 5243 (2014),
9
Z. Q. Wang et.al., Adv. Funct. Mater. 22, 2759
(2012).
O 10
-41-
