Electronic Nose

A new chip has been developed that can identify a chemical by smell after a single exposure to it. Known as Loihi, the chip goes beyond typical deep learning architectures to more closely mimic the way the human brain operates.

Loihi falls into a category of what are known as neuromorphic chips. Instead of requiring a large number of previous data sets and time-consuming training, neuromorphics employs “one-shot” learning. With regard to the sense of smell, the human brain can smell something once and immediately recognize it again. In fact, mammals can learn hundreds and even thousands of smells without forgetting those they previously learned.

When this principle is applied to electronics, it eliminates the power-hungry training process that artificial intelligence (AI) systems require. It also means that the system does not require retraining every time a new sample is encountered, enabling detection up to a thousand times faster than existing methods.

The latest version of Loihi is a 768-chip, multi-board system representing the equivalent of 100 million neurons. Systems like this could one day be used to sniff out hazardous chemicals and explosives, or even aid in diagnosing medical conditions.