A bio-inspired electronic nose micro-system based on integrated gas sensor array and log-spike processing

Electronic Nose (EN) systems based on integrated gas sensor arrays have attracted increased interest during the past two decades. SnO2-based gas sensors with Micro-HotPlate (MHP) structures as the heating elements are commonly used for such applications. However, due to fabrication difficulties, large dimension gas sensor arrays are rarely seen in the literature. Surface micro-machining is preferred due to its simplified process and CMOS compatibility. However, the limited sacrificial layer thickness usually leads to higher power consumption. Besides power limitation, micro electronic gas sensors also suffer from lack of selectivity and post-processing is typically required for reliable detection. This thesis first describes the design, fabrication and characterization of a monolithic 4x4 tin oxide gas sensor array. A novel convex MHP structure is developed to increase the thermal efficiency. Different post-treatments are used to modify the characteristics of the tin oxide gas sensing films in order to improve the overall selectivity. The monolithic sensor array with its pre-processing circuitry has been implemented using a 5 μm process in NFF-HKUST (Nanoelectronics Fabrication Facility, Hong Kong University of Science and Technology). Comparison of the power consumption with other designs and experimental results to different analyte gases illustrate the effectiveness of the proposed process. Besides the gas sensor array, an EN system needs a decision-making component to analyze the data and perform gas recognition. Normally such a system is implemented using a PC or dedicated complex system level implementation. Human's olfactory system shows remarkably high computational efficiency and accuracy. However, such bio-inspired algorithm and its implementation are really new and little work has been seen. In this thesis, a logarithmic spike-timing encoding scheme is developed to translate the output of the integrated gas sensor array into spike sequences (rank ordering). Simulation results illustrate that a particular analyte gas generates a unique spike pattern with an invariant rank order to gas concentrations. Gas recognition can thus be achieved by encoding the rank order of this particular spike train. Implementing such a biologically inspired processing into a custom VLSI chip is also explored