We are entering an era where artificial intelligence (AI) mirrors human perception. While brain-inspired chips already excel at processing vision, sound, and touch, they have struggled to "taste" chemicals in wet conditions, the characteristic of the human mouth, which affects their potential life-saving uses in medical and environmental applications.

In a study published in PNAS, a research team led by Prof. YAN Yong from the National Center for Nanoscience and Technology (NCNST) of the Chinese Academy of Sciences developed an artificial gustatory system which is capable of precisely distinguishing different flavors in aqueous environments.

Researchers constructed a memristive device using layered graphene oxide (GO) membranes. The device has two unique functions: adapting brain-like synapses for learning and detecting chemicals as a sensor. Its functionalities arise from dynamic ion migration within the nanofluidic channels of the GO membrane. The ion adsorption-desorption process within GO membranes was found to produce a memory effect.

"When ions 'hesitate' within the graphene interlayer, they get 'stuck' temporarily through surface interactions. This prolonged interfacial interaction simultaneously enables both memory retention and sensing capabilities," said ZHANG Yuchun, one of the first authors of the study. 

Capitalizing on this dual functionality, the researchers developed an AI-based artificial taste perception system which, after training, could distinguish basic tastes including sour, bitter, salty and sweet, and complex flavors such as coffee and cola with accuracies above 90%.

Conventional architectures suffer from significant energy and temporal overhead due to the von Neumann bottleneck between discrete sensing and processing units. The device could compute where it senses, suggesting that neuromorphic in-sensor computing in liquid environments is possible, which would enable the development of smarter, faster, and more energy-efficient systems.

This work opens up new possibilities for the application of brain-inspired chips in medical therapy to reconstruct taste for taste-loss patients, and in autonomous machines to "taste" their environment, which represents a significant leap toward truly intelligent and sense-enabled artificial systems. 

How to maintain computational performance of artificial gustatory systems while miniaturizing these systems for compatible complementary metal-oxide-semiconductors (CMOS) integration is the challenge that lies ahead.