(156b) A Computational Roadmap for Better Electronic Noses

Most gas sensors can only detect a single gas species (e.g., a household carbon monoxide detector), with a small subset able to distinguish among a few gas species (e.g., commercial “electronic nose” sensors). However, few attempts have been made – and none successfully – at creating a device that could detect a very broad spectrum of gases under a wide range of operating conditions. This seemingly challenging task is carried out every day by our own biological noses, however. Creating an artificial nose as capable as a biological nose would have enormous socioeconomic impact; analogous to the invention of the camera (i.e., the “artificial eye”). It would be transformative for a wide range of industrial and military applications, but, in particular, it would revolutionize health diagnostics.

So what would it take to build a better electronic nose? Electronic noses rely on arrays of chemically distinct materials that each interact uniquely with different gas species, which is closely analogous to how biological noses work. Current electronic noses, however, use small arrays with typically less than 20-30 materials. Dogs, on the other hand, have thousands of distinct olfactory receptors, each of which are replicated tens of thousands of times throughout the interior of their noses. Recent evidence suggests that building larger arrays can significantly improve electronic nose performance, but only if the sensing materials are carefully chosen. For large arrays this becomes a classic “big data” problem, requiring significant computational power and clever algorithms to address the combinatorial explosion of possible arrays one could construct.

In this talk, we discuss recent work on the computational design of gas sensing arrays, based on modeling gas adsorption in metal-organic frameworks (MOFs), and lay out a roadmap for how we might eventually build an electronic nose that could rival a dog’s one.