Functional Devices for Bio‐inspired Computing

by Adrian Petraru (Kiel University) , Hermann Kohlstedt (Kiel University)

109 (25b)




Information processing in biological nerve system is characterized by highly parallel,
energy efficient and adaptive architectures in contrast to clock driven digital Turing
machines. Even simple creatures outperform supercomputers when it comes to pattern
recognition, failure tolerant systems and cognitive tasks. Fundamental building blocks
leading to such remarkable properties are neurons as central processing units, which are
(with variable strengths) interconnected by synapses to from a complex dynamical three
dimensional network. The field of neuromorphic engineering aims to mimic such biological
inspired information pathways by electronic circuitries. The advent of novel functional
electronic components, such as memristive devices and piezoelectric field effect transistors,
may path the way to energy efficient and compact neuromorphic computing architectures.
However, any functional devices have to fulfill several requirements being a
candidate for neuromorphic circuits. Therefore an in‐depth study of the process procedure,
the electrical characteristics (I‐V curves, retention, fatigue etc.) and the electronic interfacial
structure, is of likewise importance. In particular compability with silicon technology is
mandatory. In this framework results on double barrier memristive devices (DBMD), e.g.
with the layer sequence Nb/Al/Al2O3/NbOx/Au, epitaxial SrTiO3/LaxSr1‐
xMnO3/doped:HfO2/SrRuO3 ferroelectric capacitors and piezo electric field effect transistors
comprizing AlN and AlScN will be presented. Such devices, combined with pulse‐coupled
non‐linear relaxation type oscillators may represent basic building blocks for neuromorphic
circuit working at biologically relevant frequencies, of about a 100 Hz.