Events

[Thesis Colloquium] CeNSE : Biologically Inspired Electronic Nose: Innovations and Developments

Name of the Student:  Mr. Deepak Sharma

Degree Registered: Ph.D. Engineering

Advisor/s: Prof. Navakanta Bhat and Prof. Sreetosh Goswami

Date:  3rd October 2024 (Thursday), 3:30PM
Venue: : Hybrid : Seminar Hall / https://shorturl.at/IhCH9

Abstract
The olfactory system is a highly efficient and complex network for detecting and interpreting a wide range of chemical signals. It consists of specialized sensory cells located in the nasal mucosa that detect odor molecules and transmit signals to the brain. Drawing inspiration from this naturally efficient system, this thesis emulates similar behavior in an artificial context by connecting dots from multiple disciplines, mainly including:

Semiconductor technology
Material engineering
Mixed signal electronics
Brain-inspired computing

The first part of the presentation will focus on the development of various room-temperature operable gas sensors, each targeting a specific gas. In this thesis, we fabricated interdigitated electrode (IDE) structures on silicon and flexible substrates, serving as common platforms for electrical stimulation and for sensing changes in material conductivity upon exposure to gaseous molecules. We used wet chemistry methods to synthesize a variety of different material compositions to enhance orthogonal selectivity, ensuring that each material primarily responds to a specific gas analyte while minimizing interference from others. Advanced semiconductor characterization techniques were exploited to understand and enhance the performance of adsorption-desorption kinetics of gas molecules, with concentrations down to parts per billion. We also developed an ultrafast humidity sensor that responds and recovers in milliseconds, using 0D/2D heterostructures to compensate for the effects of environmental humidity fluctuations on the sensor array.
The second part of this presentation will demonstrate the scalability aspect of solid-state gas sensors where we fabricated unfolded crossbar array structure consisting of various (16 and 64 here) IDE based sensors. For diverse sensing capabilities, we engineered a variety of materials including 0D, 1D, 2D nanostructures, metal oxides, nanoparticles, quantum dots, and their heterostructures. The next phase of this study aimed for higher-dimensional data handling and algorithmic development with brain-inspired approaches. An Artificial Neural Network (ANN) was implemented on offline data collected from a 10-sensor array in a laboratory from various gaseous exposures for classification (testing accuracy > 92%) and prediction (mean square error < 4e-3) of respective gas concentration levels. To further improve testing accuracy to 98%, LSTM-based sequential models were implemented on offline data, The final part of this presentation will focus on neuromorphic computing systems. Here, we developed a 64x64 memristor crossbar array using molecular complexes. We designed a customized mixed-signal printed circuit board capable of producing nanosecond-level electrical pulses and acquiring data with high precision (>15 effective number of bits), simultaneously characterizing up to 64 independent parallel channels. Each memristor in this crossbar array acts as a synapse and can be linearly programmed without any additional selector, storing up to 16,520 non-volatile analog levels (equivalent to 14 bits). We demonstrated high-resolution computing with approximately 460 times less energy consumption than traditional digital computers.