Date of Award


Degree Type


Degree Name

Master of Engineering Science


School of Engineering

First Advisor

Associate Professor Adam Osseiran

Second Advisor

Dr Alexander Rassau


The conventional Von Neumann architecture imposes strict constraints on the development of intelligent adaptive systems. The requirements of substantial computing power to process and analyse complex data make such an approach impractical to be used in implementing smart systems.

Neuromorphic engineering has produced promising results in applications such as electronic sensing, networking architectures and complex data processing. This interdisciplinary field takes inspiration from neurobiological architecture and emulates these characteristics using analogue Very Large Scale Integration (VLSI). The unconventional approach of exploiting the non-linear current characteristics of transistors has aided in the development of low-power adaptive systems that can be implemented in intelligent systems. The neuromorphic approach is widely applied in electronic sensing, particularly in vision, auditory, tactile and olfactory sensors. While conventional sensors generate a huge amount of redundant output data, neuromorphic sensors implement the biological concept of spike-based output to generate sparse output data that corresponds to a certain sensing event. The operation principle applied in these sensors supports reduced power consumption with operating efficiency comparable to conventional sensors. Although neuromorphic sensors such as Dynamic Vision Sensor (DVS), Dynamic and Active pixel Vision Sensor (DAVIS) and AEREAR2 are steadily expanding their scope of application in real-world systems, the lack of spike-based data processing algorithms and complex interfacing methods restricts its applications in low-cost standalone autonomous systems.

This research addresses the issue of interfacing between neuromorphic sensors and digital neuromorphic circuits. Current interfacing methods of these sensors are dependent on computers for output data processing. This approach restricts the portability of these sensors, limits their application in a standalone system and increases the overall cost of such systems. The proposed methodology simplifies the interfacing of these sensors with digital neuromorphic processors by utilizing AER communication protocols and neuromorphic hardware developed under the Convolution AER Vision Architecture for Real-time (CAVIAR) project. The proposed interface is simulated using a JAVA model that emulates a typical spikebased output of a neuromorphic sensor, in this case an olfactory sensor, and functions that process this data based on supervised learning. The successful implementation of this simulation suggests that the methodology is a practical solution and can be implemented in hardware. The JAVA simulation is compared to a similar model developed in Nengo, a standard large-scale neural simulation tool.

The successful completion of this research contributes towards expanding the scope of application of neuromorphic sensors in standalone intelligent systems. The easy interfacing method proposed in this thesis promotes the portability of these sensors by eliminating the dependency on computers for output data processing. The inclusion of neuromorphic Field Programmable Gate Array (FPGA) board allows reconfiguration and deployment of learning algorithms to implement adaptable systems. These low-power systems can be widely applied in biosecurity and environmental monitoring. With this thesis, we suggest directions for future research in neuromorphic standalone systems based on neuromorphic olfaction.