While machines have mastered both sight and sound, the sense of taste has proved harder to digitize. We’ve seen the creation of highly specialized artificial tongues targeting sweetness, chocolate, beer, wine or whisky, but now researchers in Beijing have developed a more generalist graphene oxide “tongue” that doesn’t just detect chemicals, it learns them. During laboratory tests, the system identified sour, salty, bitter, and sweet with nearly 99% accuracy, demonstrating that taste can be captured in digital form.
Researchers at the National Center for Nanoscience and Technology in Beijing, together with colleagues across China, have built a neuromorphic device that mimics one of our most personal senses: taste. Their “artificial gustatory system” uses layered graphene oxide membranes that not only sense chemicals in solution, but process the signals directly, echoing how biological taste buds and neurons work together.
Unlike most artificial senses built from solid-state electronics, taste must operate in liquid, where ions – not electrons – can carry the signal. The team tackled that challenge with a graphene oxide ionic sensory memristive device (GO-ISMD).
Inside the device’s nanoconfined channels, ions undergo interfacial adsorption and desorption that slow their motion and create a memory-like, hysteretic electrical response. This volatile short-term memory allows the same component to both detect chemicals and perform in-sensor computation in a wet, physiological environment, the first of its kind to achieve this.
When tested with voltage pulses, the device behaves much like a synapse: it can strengthen or weaken its response, show memory effects, and even remember two signals that arrive close together. The thicker the membrane, the longer this memory lasts; in some cases up to about 140 seconds, far beyond what simple ion movement would predict. To turn those dynamics into perception, the group used reservoir computing.
“Inspired by the biological taste system, we developed a smart system using our devices to ‘recognize’ chemicals based on their flavors,” explains Yon Yang, in an email to New Atlas. “The system includes three key components: a sensing input, a reservoir layer, and a single-layer fully connected neural network. The sensing input and reservoir layer are both realized through our hardware (devices). These signals are then processed by the reservoir layer, which converts them into unique digital patterns. These patterns are fed into the single-layer fully connected neural network.”
In practice, the sensing module detects flavors and converts them into electrical signals before they reach the reservoir layer. The neural network is then trained on a computer to recognize these digital patterns and save the key parameters, effectively giving the system a “memory” of different flavors it can later recall.
In their proof-of-concept, the researchers tested four representative tastants: sour (acetic acid), salty (NaCl), bitter (MgSO₄), and sweet (lead acetate). Signals from the device fed into the trained neural network achieved about 98.5% accuracy in distinguishing the tastants, with binary test accuracies ranging from 75% to 90% depending on the sample. Even beverages such as coffee, Coke, and their mixtures could be classified with strong performance.
Despite these successes, the authors emphasize that this is still a proof-of-concept demonstration. The current setup is noted as bulky, requiring large amounts of energy to function, and further miniaturization and circuit integration will be required before such systems are practical outside the lab.
“This technology perfectly bridges brain-inspired computing, chemical detection, and biologically-inspired systems,” explains Yan. “With further advances in scaling up production, enhancing power efficiency, integrating multi-sensor arrays, and developing compatible neuromorphic hardware, we anticipate transformative applications in healthcare technology, robotics, and environmental monitoring within the next decade.”
By combining sensing and computing in one aqueous device, the graphene oxide system marks a notable step for biomimetic gustation and neuromorphic engineering, as well as hints at future tools that may extend, or even reconstruct, the sense of taste.
The new study was published in the journal Proceedings of the National Academy of Sciences.
