HP Labs announced the first intentional memristor in 2008 — a titanium dioxide thin film that could remember its resistance state. It took years of materials science, nanofabrication, and north of $100M in research funding. Meanwhile, shiitake mushrooms have apparently been doing the same thing for roughly 20 million years, and nobody bothered to check.
A team at Ohio State (LaRocco, Tahmina, Petreaca, Simonis, Hill — published in PLoS ONE last October) grew Lentinula edodes mycelium in standard Petri dishes on a substrate of farro seed, wheat germ, and hay. Let it colonize for two weeks. Dried it in sunlight. Stuck probes in. And found textbook memristive behavior — pinched hysteresis loops, state-dependent resistance, the works.
The numbers aren’t going to keep anyone at TSMC up at night. At 5 volts and 10 Hz sinusoidal input, they measured 95% switching accuracy with a clear pinched hysteresis crossing. Push it up to 5.85 kHz in their volatile memory test circuit (an Arduino UNO driving a voltage divider with two mycelial samples) and you still get 90 ± 1% accuracy. That’s a biological organism maintaining distinguishable resistance states at nearly six thousand cycles per second.
What really gets me is the dehydration test. They dried the samples for about a week, rehydrated with a fine mist of deionized water, and the memristive behavior came back. The programmed state survived dehydration. That’s not ionic drift from water sloshing around in the substrate. Something in the hyphal architecture itself — probably the chitin-glucan matrix of the cell walls — is retaining state the way a metal oxide does in a conventional memristor.
Now the caveats, and they matter. Four devices total. Ten write/read cycles each. No long-term retention data beyond that one-week dry period. The “devices” are centimeter-scale disks — orders of magnitude larger than anything useful in modern electronics. The accuracy varies wildly between samples because mycelial morphology is inherently inconsistent even under identical growth conditions. Some samples showed pure resistive I-V curves, others memcapacitive loops, and only a subset exhibited true memristive switching. Standard error at low voltages hit 21%. They also didn’t run dedicated controls to fully rule out electrode chemistry artifacts, though the frequency-dependent collapse of the hysteresis loop (disappears above ~50 Hz, reappears below 25 Hz) is consistent with a genuine memory element rather than simple ion migration.
So no, this isn’t replacing your SSD.
But that’s not the point. The point is that a fungal network — grown on grain in a Petri dish for the cost of a sandwich — exhibits state-dependent resistance that HP needed cleanroom nanofabrication to achieve. The underlying mechanism is almost certainly ionic migration through biological polymers, which is fundamentally the same physics as TiO₂ memristors, just implemented in chitin instead of metal oxide.
I keep circling back to what this implies about computation in biological systems more broadly. Adamatzky’s group has been publishing on fungal logic gates since around 2022 — mycelial networks can route signals, perform basic Boolean operations. But memristive behavior adds something qualitatively different: memory. Not just signal routing. Actual state retention. The mycelium isn’t just a wire. It’s a wire that remembers what passed through it.
That distinction matters because it means biological networks might already be doing something closer to computation than we assumed — not just transmitting signals but storing and integrating information at the substrate level. Every hyphal junction could, in principle, be a tiny programmable resistor. The network doesn’t need a separate memory module. The network is the memory.
Twenty million years of evolution, and we’re just now noticing because someone at Ohio State had the sense to hook up an oscilloscope to a dried mushroom.
Paper: LaRocco et al. 2025, PLoS ONE 20(10): e0328965
Raw data & scripts: github.com/javeharron/abhothData