Date of Award

Summer 8-15-2021

Author's School

McKelvey School of Engineering

Author's Department

Biomedical Engineering

Degree Name

Doctor of Philosophy (PhD)

Degree Type



Insects are ideal candidates for developing bio-robotic systems owing to their ability to thrive in almost any environment. For example, neurons in their exquisite olfactory sensory systems can be tapped to create a sensing platform for standoff chemical monitoring. However, for enabling such cyborg systems, it is vital that the neural activity of a freely behaving organism can be measured for long periods of time. The current state-of-the-art neural recording techniques are power-intensive and they either need batteries, which make them too bulky for insects, or they have to maintain a continuous telemetry link to an external power source which restricts the mobility of the organism. In this dissertation, I explore algorithmic and device based approaches for overcoming the limitation of current neural activity recording-technologies in terms of energy efficiency and longevity.

In the first part of my thesis I show that changes in energy content of the extracellular neural recordings contain sufficient information for specific stimulus identification tasks. Measuring the energy content of the signal can be achieved using sampling rates that are significantly lower than what is required for spike-sorting and hence could lead to significant improvements in energy-efficiencies. This hypothesis was investigated and verified using the locust olfactory system for odor identification. We could reliably classify odors based on energy information collected from the extracellular medium in the locust antennal lobe.

To eliminate the high power consumption of long range wireless communication, I investigated energy-efficient approaches for storing data in non-volatile memory. I implemented ultra-low-power compressive sensing-storage circuits that can efficiently store a compressed signal in floating-gate memory cells. The compressed information could be retrieved later and the events of interest reconstructed offline. I further reduced the power consumption of the memory cells themselves. By modulating the trajectory of synchronized dynamical systems, I show that information can be stored by expending less than 1pj of energy per bit, a four orders of magnitude improvement. These dynamical systems are based on the physics of quantum-mechanical tunneling of electrons through thin-oxide layers. Because these devices are thermodynamically driven, the recorder based on this principle can operate at femto-watt power levels.

Finally, I investigate the feasibility of a completely self-powered neural recording paradigm in which the neuronal action-potentials modulate the dynamical memory cells directly. Through modeling studies, I show that modulation of the quantum-tunneling barrier by a neural activity can be registered as a change in the response of the self-powered dynamical system. However, the signals are smaller than the detection limits of current readout circuits. I propose techniques for recovering the information and reconstructing the time history of the input signal.


English (en)


Shantanu Chakrabartty

Committee Members

Baranidharan Raman, Dan Moran, Keith Hengen, Erik Henriksen,