A leaf can now run inference. Not just report its own state — compute on it, in situ, the way a synapse does. Researchers at the University of Texas at Austin have built a graphene-based device that mounts directly on leaf surfaces and performs artificial synaptic processing on collected data, without transmitting raw signals back to any server. The energy cost: 23 attojoules per conductance update. A single solar panel could power millions of them indefinitely.
The device is a graphene field-effect transistor array applied to leaf surfaces via transfer printing. A small voltage pulse drives ions in the leaf toward or away from the device, shifting its conductance — a proxy for hydration level. The synaptic part is not metaphorical: under repeated light exposure, the device's conductance response strengthens in a manner that mirrors long-term potentiation in biological neural circuits, the same mechanism the brain uses to form memories. Each light pulse doesn't just measure the leaf — it updates the device's internal state, the way a spiking neural network adjusts its weights. The raw data never leaves the leaf.
"We are able to gather so much more information than what our current technology can, and in a much easier way," Ashley Matheny, an associate professor in the Jackson School of Geosciences' Department of Earth and Planetary Sciences at UT Austin, said in UT Austin's news release.
What's harder to dismiss is the power number. The sensors require 23 attojoules per conductance update and 0.23 microwatt of power for reading data, according to UT Austin. A modest solar panel could power millions of them simultaneously — making large-scale deployment in remote forests not just possible but economically realistic, a threshold that matters more than any benchmark number.
The graphene tattoo originated in Jean Anne Incorvia's lab at UT Austin, which was working on sensing proton movement using the material. The leaf application came from Maya Borowicz, a visiting undergraduate student who is one of the paper's authors. "It was a fun summer project that got a second life," Incorvia said. The project found its application partner when Incorvia met Matheny through UT's Associate Professor Experimental program, which connects newly tenured faculty across departments. Matheny had spent years tracking hydration in wood and soil. What she needed was leaves.
"Leaf water levels represent the best indicator of live fuel moisture content," Matheny said in the UT release. That content is one of the leading predictors of wildfires — and one of the hardest to measure continuously, because the current methods are destructive. Most researchers gather the data by snipping branches off trees, or in some cases shooting them down.
The graphene sensors change what continuous means. They can monitor the same living leaf across days, seasons, and stress events — the early-morning recovery phase after a hot windy day, the slow drawdown over a drought. Matheny plans to combine the leaf hydration record with her existing soil and wood hydration data to model what is happening in a forest from canopy to root system. "If I know something about the leaves, I can better predict what is going on with the wood," she said. "We are looking at everything from stress responses to what is happening in the forest right now to understand the risk to the public."
The paper, "Graphene In-Sensor Compute Device for Plant Hydration Monitoring," was published in Nano Letters on February 16, 2026 (Volume 26, Issue 7, Pages 2432-2440). Utkarsh Misra was lead author. Jean Anne C. Incorvia and Deji Akinwande, both in the Chandra Family Department of Electrical and Computer Engineering at UT Austin, were corresponding authors. Other authors include Maya Borowicz, Philip Varkey, Ning Liu, Samuel Liu, and Benjamin K. Keitz from UT Austin's engineering departments, Ashley M. Matheny from Earth and Planetary Sciences, and Dmitry Kireev, who completed the work as a postdoctoral researcher in Akinwande's lab and is now an assistant professor of biomedical engineering at the University of Massachusetts Amherst.
The work remains in the lab-validation stage — outdoor deployment in active wildfire-prone forests is next, and the team is honest about what that transition requires. But the physics is solid. 23 attojoules per update is not a projection. It's a measured number on a working device. Whether it survives a full wildfire season attached to a pine needle is a different question — one the researchers are now equipped to answer.
