From chemical traces to neural signals
Loihi 2 does not contain a nose or directly sense molecules. A separate chemical sensor produces electrical readings as substances in the air interact with it. Those readings form patterns that a trained network on the chip can classify—for example, as matching a chemical it has learned to recognize.
The neuromorphic part is how the network processes those patterns. Rather than repeatedly moving data between a processor and separate memory, Loihi 2 uses networks of small computing units that send brief electrical-like events, called spikes, when their activity crosses a threshold. This is loosely inspired by how neurons communicate; it is not a literal copy of a biological brain.
Why process the signal on the chip?
Chemical sensors can produce a stream of changing readings, and waiting to send every reading to a remote server adds communication time and requires a network connection. Running the classifier locally lets a device respond at the point of sensing, which can matter when an operator needs an immediate warning or when connectivity is unreliable. Event-based processing can also avoid spending energy on calculations when little is changing, though the actual power and speed depend on the task and system.
The chip’s role is narrower than a general-purpose chemical detector: it recognizes patterns represented in its training and sensor inputs. A new substance, a changed sensor, or a different environment may produce unfamiliar readings and require additional training or testing. The practical promise is therefore not that silicon can smell anything, but that a compact sensor-and-chip system can make a useful first-pass decision without routing every measurement through the cloud.
