Silicon Olfactory System Implementation
Schematic layout of silicon implementation of the olfacotory system, showing the odour delivery, chemFET array, integrate and fire elements, olfactory bulb model and output decoder. GL - glomerulus cell model, PG - periglomerulus cell model, GR- granule cell model, M/T - mitral/tufted cell model. Excitatory connections shown as filled circles. Inhibitory connections shown as empty circles

Project Overview

This research project aims to combine recent advances in our understanding of biological olfactory systems with CMOS-compatible chemical microsensors, in order to produce the first neuromorphic analogue VLSI chemical-sensing chip and hence a micropower, fully integrated, electronic nose, with the capability of adapting to changes in its operating conditions.

The research is a collaborative venture between three Universities: novel integrated silicon microsensors will be developed by the University of Warwick, the neuromorphic olfactory model by the University of Leicester, and the analogue VLSI circuitry by the University of Edinburgh. In the last 15 years, machine olfaction, or electronic nose development has seen rapid progress both in terms of improvements and diversification of chemical sensing technology as well as novel methods for sensor-array signal processing - much of which has been pioneered by Warwick University and the start-up companies of Neotronics Scientific Ltd (now part of Marconi Applied Technologies) and Alpha MOS (recently floated on the French stock market).

Besides providing the basis for an expanding worldwide odour-sensing instrumentation industry, some of the application-specific challenges have been overcome in order to improve, amongst other factors, system sensitivity, resolving power, and the variety of chemical stimuli that may be appreciated. However, odour-sensing applications demand high levels of system sensitivity and stability that have not yet been obtained using traditional engineering approaches. Commercial electronic nose systems tend to suffer from sensor drift and poisoning and therefore require frequent supervised calibration (sometimes on a daily basis). They also have limited sensitivity and an inability both to deal with non-stationary background odours and to discriminate between odour quality and odour intensity. While for some applications these technological limitations may be overcome through the use of large headspace conditioners and autosamplers or may not be critical, a new generation of electronic nose systems are required in order to meet the future demands for low-cost, reliable, real-time, portable odour measurement that is invariant to interfering background odours and can independently resolve intensity and quality.

Recently, biomedical research has led to a new understanding of chemosensory information processing in olfaction systems and a number of computational models now exist for odour delivery, chemical transduction, and sensory signal processing. What is clear from these investigations is that the biological olfactory system already has all of the properties that we seek to impart to the next generation of handheld electronic noses. The focus of the neuromorphic modelling research in this proposal is therefore to extract the engineering design principles that underpin the biological olfactory system and apply them to artificial nose technology. New neuromorphic models will be generated that are optimised for temperature and humidity invariance, sensor drift and poisoning, adaptation, and embodied in an analogue VLSI artificial nose.

The focus of the sensor research in this proposal will be in the novel implementation of 2-d arrays of gas-sensitive FET devices combined with a diffusion microchannel to provide odour sensing. Individual sensors will be made sensitive to different odours by using a selection of conducting polymers to create an array of different polymer gate FETs across a square array. The focus of the analogue VLSI research will be in the implementation of sensor interface and adaptive olfactory neuromorphic circuits using highly parallel, physically small, relatively low precision and micropower adaptive analogue electronics operating in real time.

In combination, this research proposal promises to deliver leading edge research results in the areas of CMOS-based micro-arrays for odour sensing, adaptive spatio-temporal neuromorphic models of the olfactory system, and novel analogue circuits and techniques for sensor interfacing and implementation of the smart neuromorphic interface. It aims to demonstrate a commercially viable role for analogue sensor systems and neuromorphic-engineered analogue VLSI through the direct involvement with our commercial partners, Osmetech, PLC. The project will, we believe, prod uce the first ever adaptive silicon-based olfactory system.

This project has been funded by EPSRC

Project Partners:

Tim C. Pearce, Lecturer in Bioengineering, University of Leicester, UK (Neuronal Modelling and Co-ordinator)

Julian W. Gardner, Professor of Electronics, University of Warwick, UK (Sensors and Microchannel)

Alister Hamilton, Lecturer in Electronics, University of Edinburgh, UK (aVLSI)

Research Team:

James Covington, Postdoctoral Research Assistant, University of Warwick, UK

Carlo Fulvi-Mari, Postdoctoral Research Assistant, University of Leicester, UK.

Thomas Jacob Koickal, Postdoctoral Research Assistant, University of Edinburgh, UK

Forest Su Tan, PhD Student, University of Warwick, UK.

Collaborators:

Osmetech PLC, UK.

Paul F.M.J. Verschure, Group Leader, ETH, Switzerland (Neuronal Simulation)

Thom Cleland, Postdoctoral Researcher, Cornell, USA (Olfactory Modelling)

Log into project pages

Results will be posted here as they become available ...

Author: Tim Pearce , last updated 5th September 2001.
Any opinions and views expressed are the author's and not those of the University