Noise Exploitation and Adaptation in Neuromorphic Sensors - Engineered Biomimicry - page 37

CHAPTER

2

Noise Exploitation and
Adaptation in Neuromorphic
Sensors

Thamira Hindo and Shantanu Chakrabartty
Department of Electrical and Computer Engineering, Michigan State University,
East Lansing, MI 48824, USA

Prospectus
Even though current micro-nano fabrication technology
has reached integration levels at which ultra-sensitive
sensors can be fabricated, the sensing performance
(bits per Joule) of synthetic systems are still orders
of magnitude inferior to those observed in neuro-
biology. For example, the filiform hair in crickets
operates at fundamental limits of noise and energy
efficiency. Another example is the auditory sensor in
the parasitoid fly Ormia ochracea that can precisely
localize ultra-faint acoustic signatures in spite of the
underlying physical limitations. Even though many
of these biological marvels have served as inspirations
for different types of neuromorphic sensors, the main
focus of these designs has been to faithfully replicate
the biological functions, without considering the
constructive role of noise . In manmade sensors, device
and sensor noise are typically considered nuisances,
whereas in neurobiology noise has been shown to
be a computational aid that enables sensing and
operation at fundamental limits of energy efficiency
and performance. In this chapter, we describe some
of the important noise exploitation and adaptation
principles observed in neurobiology and how they can
be systematically used for designing neuromorphic
sensors. Our focus is on two types of noise exploitation
principles, namely, (a) stochastic resonance and (b)
noise shaping, which are unified within a framework

called ΣΔ learning. As a case study, we describe the
application of ΣΔ learning for the design of a miniature
acoustic source localizer, the performance of which
matches that of its biological counterpart ( O. ochracea ) .

Keywords
Acoustic sensors, Adaptation, Localization, Neuro-
morphic sensors, Neurons, Noise exploitation, Noise
shaping, Signal-to-noise ratio, Spike-time-dependent
plasticity (STDP), Stochastic resonance, Synapse

2.1 INTRODUCTION

Over the last decade significant research effort
has been expended in designing systems
inspired by biology. Neuromorphic engineer-
ing constitutes one such discipline in which the
objective has been to design sensors and systems
that mimic or model the physical principles
observed in neurobiology [1-3] . The key moti-
vation behind this effort has been to reduce the
performance gap that exists between neurobio-
logical sensors and their synthetic counterparts.
For instance, it has been known that biology, in
spite of its physical and fundamental limitations,