A University of Chicago team built a soft, brain inspired computer that runs AI directly on the body, spotting a fatal heart rhythm in milliseconds without sending data to a remote server.
Today's best smartwatches can sense a dangerous heart rhythm, but they cannot read it on the wrist. The signal is shipped to a phone, then to a cloud server, where an algorithm decides whether the wearer is in trouble. The round trip can take seconds. For ventricular fibrillation, the chaotic electrical failure that causes sudden cardiac arrest, that delay is the difference between a shock that restarts a heart and a body that does not make it to the hospital.
A University of Chicago team has now moved that decision onto the skin. In a paper published in Nature Electronics, the group led by Sihong Wang, associate professor at the university's Pritzker School of Molecular Engineering, demonstrates a stretchable, skin-worn computer that runs AI inference on the body itself, spotting ventricular fibrillation in real time without sending a single sample off the patch.
The mechanism is the news. The device is built from printed organic electrochemical transistors, a class of flexible electronics where each transistor moves ions through a gel-like layer and computes while it stores, much like a synapse. Because the array is printed on a soft, conformable sheet rather than a rigid chip, it bends with the skin. Because each transistor doubles as a memory element, the array processes the electrocardiogram signal where it is collected, in milliseconds, and only then decides what to do.
That is the shift. Today's consumer wearables are sensors, not computers. They capture the heart's electrical trace with high fidelity, but they outsource the thinking. A UChicago summary of the work frames the gap as a latency problem: cloud-dependent inference is too slow for a life-threatening rhythm where every millisecond counts. The new patch is designed to close that gap by running a brain-inspired, or neuromorphic, classifier on the patch itself. The result is an answer on the skin, with no server in the loop.
The manufacturing route matters as much as the chip. The transistors were printed using techniques developed with Argonne National Laboratory, a U.S. Department of Energy national lab that co-developed the device. Printing, rather than the photolithography used on a silicon wafer, is what lets the array be stretchable and large-scale at the same time. The reference design and inference code are available openly on GitHub, which is unusual for a high-profile neuromorphic device and means other groups can replicate the demonstration on their own hardware.
What the patch is not, yet, is a clinical product. The work is a research-stage demonstration, validated in bench tests against known arrhythmia waveforms rather than on patients in a hospital. The paper does not address power: a stretchable, always-on computer that runs inference on the skin will need a battery or a harvested-energy source, and the release does not say what that looks like. It does not address biocompatibility over days or weeks of wear, durability under sweat and motion, manufacturing yield at scale, or the regulatory path that any real cardiac monitor would have to clear before a doctor could prescribe it. The phrase "instant personal doctor" appears in the SciTechDaily writeup as a paraphrase of the researchers' vision, not as a description of what the device is today.
The honest framing is that the latency problem, the one that made cloud-dependent wearables too slow for ventricular fibrillation, is now technically solvable on the body. The remaining work, powering a soft computer around the clock, proving it safe on real skin for real durations, and running it through clinical trials, is the same work any new medical monitor would face. Watch for the next paper: whether the same array can run on a harvested or printed power source, and whether the on-body inference holds up on a live animal or human heart, not just on stored waveforms.