The Latest Electric Nose
Remember losing your sense of smell with COVID? I lost my sense of taste, too, to the point that I couldn’t tell whether the food I cooked was well seasoned.
Researchers at UC Berkeley just built something that doesn’t have that problem. It is an “electronic nose” with a chip carrying 16 sensors, each coated with a different sensing material.
Foods release mixtures of volatile compounds into the air, and each sensor responds to those mixtures slightly differently. Machine learning then uses the combined response as a chemical fingerprint to classify what the device is “smelling.”
Across the foods tested, the system achieved an overall classification accuracy of 92.6%. It could also distinguish fresh raw chicken, milk, and boiled eggs from samples left at room temperature for 24 or 48 hours.
The trick was using carbon nanotubes as sensors. Most gas sensors rely on metal oxides that must be heated to function, ruling out many useful sensing materials. Nanotubes work at room temperature, so the team could pack a much wider range of them onto a single chip.
It can also distinguish four nut allergens: walnut, hazelnut, cashew, and peanut. This is the first time a gas sensor array has been shown to do that.
Still to be optimized: the chip distinguishes categories well (nuts vs. fruit vs. spoiled meat) but struggles within them. For example, it confuses peanuts with hazelnuts, likely because foods in the same category emit similar chemical profiles, leading to overlapping sensor signatures. Improving that distinction is important for allergy applications.
I have been writing about health technology and wearable sensors a lot lately, and papers like this help explain what underlies the final number or classification. Sensor performance is not only about the machine-learning model. It begins with the chemistry of the sensing material: which molecules interact with it, how that interaction translates into an electrical signal, and whether similar samples produce signals distinct enough to interpret.
Understanding health technology often means understanding materials science too.
source: Carla Bassil et al., Scalable multiplexed machine-learning gas sensor chips for food classification. Sci. Adv. 12, eaec7965 (2026).
