A Graphene “Tongue” That Tastes: How Scientists Built an Artificial Sense of Flavor

When Machines Learn Flavor: Building a Graphene-Based Artificial Tongue

Imagine losing the ability to taste your favourite foods — the sweetness of ripe mangoes, the bitterness of dark chocolate, or the sour punch of lemonade. Now imagine a device that could restore that, or better yet, that could learn tastes the way our tongues and brains do.

That’s exactly what a team of researchers in China has built: a first-of-its-kind artificial tongue that doesn’t just detect taste chemicals, but does so in liquid, learns from what it experiences, and recognizes flavors with accuracy close to our own. This isn’t sci-fi. It’s neuroscience, materials science, and machine learning coming together.

The Challenge: Sensing Flavor in Wet Conditions

Taste is essentially chemistry in liquid form. In our mouths, taste buds detect dissolved substances (ions, molecules), and our brains interpret them. But building a sensor that can reliably detect “taste” in a moist, ion-rich environment is difficult. Many electronic sensors fail when wet; others require dry conditions or rely on external circuits.

The Innovation: Graphene Oxide Membranes + Reservoir Computing

Here’s how the researchers overcame those hurdles:

Graphene Oxide (GO) Membranes: They built ultra-thin layered graphene oxide membranes with nanofluidic channels (tiny pores and paths for ions) that slow down ion movement. That “slowing down” matters. It gives the system time to generate electrical signatures (changes) in response to different dissolved chemicals.
GO-ISMD Device: The sensing device is called a “graphene oxide ionic sensory memristive device” (GO-ISMD). It isn’t just a sensor — it has memory-like behavior. The GO membranes allow ions to adsorb/desorb at interfaces, meaning the device’s response depends not only on the current taste chemical but also recent exposures. In other words, it can “remember” short-term flavors.
Neuromorphic / Reservoir Computing: The system includes both sensing and processing. Signals from the graphene sensor feed into a reservoir computing framework (a kind of brain‐inspired model). Patterns generated by chemical stimuli (ion flows, electrical changes) are processed to characterize taste categories. This enables the device to generalize to tastes it hasn’t encountered before.

The Performance: Near-Human Taste Recognition

The results are quite impressive:

When trained on four basic tastes (sweet, salty, bitter, sour), the system reached ~ 98.5 % accuracy on tastes it had seen before.
For flavours the sensor hadn’t encountered before — more complex mixtures or drinks — it still achieved between 75 % and 90 % accuracy.
It could also distinguish more complex beverages like coffee and cola based on their unique flavor signatures.

These are very strong results for something operating in conditions similar to those inside the human mouth.

What’s New & Why It Matters

This work marks several breakthroughs:

Sensing + Processing in one Wet System: Previous “electronic tongue” systems needed the sensor and the computing (to interpret signals) separate, often with dry, non-physiological environments. This work merges them in one device operating in liquid.
Memory Effects from Ions: By slowing ion movement, the device exhibits hysteresis and short-term memory, which helps in distinguishing flavors not just based on instantaneous input but recent input history. That mimics sensory biology more closely.
Generalization: Recognizing new flavors (that weren’t part of training) shows potential for real-world applications — food quality, beverage testing, or even restoring taste sensations.

Limitations & Next Steps

No device is perfect yet. Here are some of the current limitations and what the future holds:

Size and energy use: The current prototype is not yet miniaturized or optimized for low power, which would be needed for practical (e.g. wearable or implantable) uses.
Complexity of flavors: There’s infinite variety in food and drink; many flavors are mixtures of many chemicals. More work is needed to scale up to more diverse and subtle tastes.
Durability and stability: Operating in wet, ionic conditions can degrade materials, cause drift, etc. Long-term stability will be a challenge.
Integration with biological systems: To help restore taste (e.g. after neural injury), such devices would need safe, biocompatible integration, signal interfaces to nerves, or prosthetic systems.

Looking Ahead: Applications & Impacts

This research opens doors in multiple areas:

Medical restoration of taste: For people who’ve lost taste due to illness, stroke, or neurodegenerative disease.
Food and beverage industry: Quality control, flavor calibration, and detecting adulteration or spoilage—all possible with precise “taste sensors.”
Robotics & AI sensory devices: Robots or devices that can monitor liquid chemicals (for safety, environmental sensors, etc.) could benefit.
Neuromorphic engineering: Integrating sensing with computation in realistic, harsh conditions (liquid, ionic) is a big step toward artificial biological sensory systems.

Conclusion

The creation of a graphene-oxide artificial tongue that senses, processes, and learns flavor in liquid is more than a technical trick. It’s a glimpse into a future where machines feel — at least taste — like we do. By combining material innovation with neuromorphic computing, researchers have built a proof-of-concept that pushes the boundaries of sensor technology and neurobiology. The challenges ahead are nontrivial, but so are the possibilities: restoring lost senses, improving food safety, and giving devices taste.

Reference

Zhang, Y., Liu, L., Qiao, Y., Yao, T., Zhao, X., & Yan, Y. (2025). Confinement of ions within graphene oxide membranes enables neuromorphic artificial gustation. Proceedings of the National Academy of Sciences, 122(28), e2413060122. https://doi.org/10.1073/pnas.241306012