Your brain runs on roughly 20 watts - about as much as a dim light bulb. Meanwhile, training GPT-4 consumed enough electricity to power a small town for a month. Somewhere between those two numbers lies the holy grail of computing, and a team in China just got us closer using molecules that change color when you zap them.

The Problem With Silicon Brains

Traditional computers have a fundamental design flaw that your brain solved four billion years ago: they keep memory and processing in separate rooms. Every time your laptop needs to think about something, it has to run down the hall, grab the data, run back, do some math, then put the data away again. Repeat a trillion times per second. It's exhausting.

Your neurons? They store and process information in the same place. No hallway. No running. Just electrochemical signals doing both jobs simultaneously. This is why a bee can navigate a garden on a brain the size of a sesame seed while Boston Dynamics robots still fall over sometimes.

Neuromorphic computing tries to steal this trick - building chips that work more like neurons than spreadsheets. The challenge has been finding materials that can actually pull this off.

Enter the Color-Changing Molecule

Researchers at Xi'an Jiaotong University just published work in Angewandte Chemie using something called thienoviologen - a molecule that's basically a viologen (already known for switching between electrical states) with thiophene groups bolted on. If that sounds like chemistry word salad, here's what matters: when you apply voltage, these molecules shift between stable states and their electrical conductivity changes in ways you can precisely control.

That "precisely control" part is doing heavy lifting. Regular viologens flip between states, sure, but they're temperamental. The radical cation state (the molecule with an unpaired electron, for the chemistry-curious) tends to form dimers and generally misbehave. Adding thiophene fixes this - it stabilizes the radical, reduces the energy gap between states, and makes the whole system more predictable.

The team built what they call a BEND - bioinspired electrochemical neuromorphic device. It's a two-terminal device where the thienoviologen electrolyte acts as a tunable conductor. Hit it with voltage pulses and its conductivity changes. Stop the pulses and it remembers where it was. Sound familiar? That's basically what synapses do.

Teaching Molecules Pavlov's Trick

Here's where it gets wild. The BEND doesn't just store information - it learns. The device demonstrates spike-timing-dependent plasticity (STDP), which is how biological synapses strengthen or weaken connections based on the precise timing of electrical signals. If neuron A fires just before neuron B, their connection strengthens. Reverse the order, and it weakens. This is thought to be fundamental to how brains learn.

The thienoviologen device does this naturally, through the electrochemistry of ion movement. No software simulation required.

Even better: the researchers demonstrated Pavlovian conditioning. They trained the device to associate two unrelated stimuli, like Pavlov trained dogs to drool at bells. The molecule literally learned an association.

When hooked into a simulated convolutional neural network and tested on Fashion-MNIST (a benchmark dataset of clothing images that's basically the "Hello World" of image recognition), the system hit nearly 80% accuracy. That's not going to replace your phone's camera app, but for a handful of molecules doing the heavy lifting, it's impressive.

Why This Matters (Beyond Cool Chemistry)

The energy consumption of AI is becoming genuinely alarming. MIT researchers have called it "unsustainably increasing." Electrochemical neuromorphic devices like BENDs could eventually process information at a fraction of the power cost - the USC team working on similar approaches talks about reducing chip size and energy consumption "by orders of magnitude."

These devices also work at room temperature, in ambient conditions, without needing exotic cooling. The thienoviologen BEND specifically showed excellent stability over time, which has been a weakness of many organic electronics.

The Catch

We're nowhere near replacing GPUs with jars of colorful liquid. The Fashion-MNIST accuracy is a proof of concept, not a product roadmap. Scaling from a single device to millions working together coherently is a problem nobody has fully solved. And while "nearly 80% accuracy" sounds good, modern neural networks on that same dataset hit 99%+.

But the trajectory matters. Organic electrochemical transistors have achieved energy consumption as low as 2 femtojoules per synaptic event - that's 0.000000000000002 joules. For comparison, one flick of a light switch uses about one joule. We're talking about efficiency that approaches biological levels.

Looking Forward

The thienoviologen work represents a specific approach in a rapidly expanding field. Other teams are exploring memristors, conducting polymers, and various electrochemical systems. What they share is the insight that maybe we don't need to simulate neurons in software running on conventional hardware - maybe we can build hardware that's inherently neuron-like.

Your brain figured this out before your great-great-great-times-a-billion-grandmother was a single cell. We're just now catching up.

References

1. Sun, S., Zhang, Y., Han, Z., Liu, C., Sun, B., Zhang, W., & He, G. (2026). Bioinspired High-Performance Neuromorphic Devices Enabled by Thienoviologen-Based Electrochemical Ion Gating. Angewandte Chemie International Edition. DOI: 10.1002/anie.202523345. PMID: 41432180

2. Fan, Z., et al. (2023). Short-term synaptic plasticity in emerging devices for neuromorphic computing. iScience. PMCID: PMC10025973

3. Flexible Organic Electrochemical Transistors for Energy-Efficient Neuromorphic Computing. (2024). Nanomaterials, 14(14), 1195. https://www.mdpi.com/2079-4991/14/14/1195

4. Xiao, H., Rasul, K., & Vollgraf, R. (2017). Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms. arXiv. https://github.com/zalandoresearch/fashion-mnist

5. MIT News. (2025). The brain power behind sustainable AI. https://news.mit.edu/2025/brain-power-behind-sustainable-ai-miranda-schwacke-1024

6. USC Viterbi School of Engineering. (2025). Artificial neurons developed by USC team replicate biological function for improved computer chips. https://viterbischool.usc.edu/news/2025/10/artificial-neurons-developed-by-usc-team-replicate-biological-function-for-improved-computer-chips/

Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.