Just out todayAugmented reality: The AR trainees were faster on the model they trained on, and no better anywhere elseVirtual reality: They matched the exercise dose exactly, and the VR advantage fell below the line that mattersVirtual reality: Both groups used the same headset. One watched a video first and picked up 10.8 more blocks

Semiconductors & chips/China

One molecule, ten conductance states, and 6.34 attojoules per switch

A single-molecule device switches for 6.34 attojoules and holds more than ten conductance states at room temperature.

What the paper found

A bio-inspired single-molecule neuromorphic device consuming approximately 6.34 attojoules per operation was built by electrochemically gating molecule-ion electrostatic interactions. It holds over 10 distinct conductance states at room temperature and emulates neural plasticity from short-term to long-term memory, demonstrated on Pavlovian conditioning and Morse code pattern recognition.

Grade 13 reading level6 min read

What happened

Training a neural network costs a great deal of energy, and a large share of it goes into switching transistors. The long-term answer most of the field is chasing is neuromorphic hardware: devices that behave like synapses, storing and computing in the same place.

The limit of that idea is a single molecule. You cannot build a smaller switch, and a molecular switch should need extraordinarily little energy to flip. The obstacle has always been that a single molecule at room temperature is being battered by thermal noise, and getting it to hold several distinguishable states reliably has not worked.

A team at the State Key Laboratory of Physical Chemistry of Solid Surfaces at Xiamen University, the Beijing National Laboratory for Molecular Sciences at the Institute of Chemistry of the Chinese Academy of Sciences, and the School of Chemical Sciences at the University of Chinese Academy of Sciences built one that holds more than ten.

It switches for about 6.34 attojoules.

The test

The mechanism is electrochemical gating of molecule-ion electrostatic interactions. Rather than trying to change the molecule itself, the device modulates the electrostatic environment around it by moving ions, which shifts the molecule’s conductance.

That is what gives programmable multi-state behaviour at room temperature, where previous approaches have been defeated by thermal fluctuations.

Characterisation covered the energy per operation and the number of distinguishable conductance states. Function was then demonstrated through two tasks: Pavlovian conditioning, which tests associative learning, and Morse code processing, which tests pattern recognition.

The device was also shown to emulate the biological transition from short-term to long-term memory.

The numbers

Measure Result
Energy per switching operation approximately 6.34 attojoules
Distinct conductance states over 10
Operating temperature room temperature
Mechanism electrochemical gating of molecule-ion electrostatic interactions
Plasticity emulated short-term to long-term memory
Demonstrations Pavlovian conditioning; Morse code pattern recognition
Data figshare, accession code

For scale: a silicon logic gate in a modern process switches for somewhere around a femtojoule. 6.34 attojoules is roughly a thousandth of that, in a device whose active element is a single molecule.

The result

The device consumes approximately 6.34 attojoules per operation and realises biomimetic emulation of neural plasticity from short-term to long-term memory, featuring over 10 distinct conductance states.

It was demonstrated in Pavlovian conditioning for associative learning and in pattern recognition for Morse code processing.

The authors frame the contribution as enabling multi-state synaptic emulation using an individual molecule, towards energy-sustainable artificial intelligence.

How this compares to what came before

Single-molecule electronics has been an active field for over two decades, and it has produced molecular diodes, switches and transistors. Nearly all of those operate as two-state devices, and many require cryogenic temperatures to suppress thermal noise.

Neuromorphic devices need more than two states. A synapse that can only be on or off cannot represent a weight, and a network of binary synapses learns poorly. Getting to ten or more programmable, distinguishable levels is what separates a switch from a synapse.

Doing that in a single molecule at room temperature is the specific advance here, and the electrochemical gating approach – changing the environment rather than the molecule – is the reason it works.

What it means

Take this as physics rather than engineering.

What it establishes is a floor. If a single molecule can hold ten states and switch for attojoules at room temperature, then the energy cost of computation is not bounded anywhere near where silicon sits. That matters for the long argument about whether AI’s energy consumption is a fixed property of computation or a property of our current hardware.

What it does not establish is any route to a device. One molecule is not a circuit. There is no array, no yield, no fabrication process, and no indication of how you would place and contact a billion of these reproducibly. Every single-molecule electronics result for twenty years has stalled at exactly that step.

The demonstrations – Pavlovian conditioning, Morse code – are the standard toy tasks for a new synaptic device. They show the device behaves like a synapse. They do not show it can do anything useful.

Business ideas from this paper

Molecular placement and contacting. The bottleneck for every single-molecule device is reproducibly putting molecules where you want them and connecting to them. Whoever solves that unlocks a field, and it is a materials and process business rather than a device one.

Electrochemical gating in larger devices. The mechanism may transfer to devices larger than one molecule, where fabrication is tractable. Applying it to thin-film or nanowire synapses is a more immediate research-to-product path.

Energy floor benchmarking. As AI energy consumption becomes a policy question, credible figures for the theoretical minimum energy per operation become valuable to anyone arguing about data centre power. This paper is a data point in that argument.

Multi-state device characterisation. Measuring and verifying how many conductance states a device really holds, reliably, over time, is a methodological specialism that neuromorphic hardware development needs and lacks.

What could be wrong with it

This is one device. There is no array, no yield figure and no fabrication route in what we could verify.

Retention is not reported in the abstract. A device holding ten states is only useful if it holds them for a useful length of time, and single-molecule systems are not known for stability.

The over-10-states claim at room temperature is the central technical result and rests on measurements of this single device.

Pavlovian conditioning and Morse code are demonstrations chosen to show synaptic behaviour. Neither is a computational benchmark.

One research consortium, no independent replication.

How sure can you be?

We score this 8 out of 10.

It is a peer-reviewed paper in Nature Communications, with three named institutions – Xiamen University, the Institute of Chemistry of the Chinese Academy of Sciences, and the University of Chinese Academy of Sciences – reporting measurements from a fabricated device, with relevant data deposited in figshare under an accession code.

It loses points because there is no independent replication and the evidence is a single-device laboratory demonstration.

Our score reflects the quality of the reporting and the provenance, not the readiness of the technology. On readiness this is as early as research gets, and the article says so plainly.

Do this today

If you argue about AI energy consumption, this paper is a useful reference point for what the physical floor looks like. Three orders of magnitude below silicon, in principle, at room temperature.

If you work in neuromorphic hardware, the electrochemical gating mechanism is the idea to examine. Changing the environment rather than the device is a different lever from the one most synaptic devices pull.

If you read single-molecule electronics results, check for an array. Twenty years of this field has produced excellent single devices and almost no circuits, and that gap is the whole story.

Source: Single-molecule neuromorphic device with aJ-level power consumption per switching, Nature Communications, 2026. PMID 41917022 · doi.org ().

Just Out Tech explains new research in plain language. This article was drafted with AI assistance and checked by a human against the original source.

What to remember
  • Approximately 6.34 attojoules per switching operation - three orders of magnitude below silicon logic.
  • Over 10 distinct conductance states at room temperature, which is the hard part.
  • The mechanism is electrochemical gating of molecule-ion electrostatic interactions.
  • Data are deposited in figshare under an accession code.
Where this really is

The work reports a fabricated single-molecule device with measured per-operation energy and conductance states at room temperature, with associative learning and pattern recognition demonstrations.

Questions people ask

What is an attojoule?

A billionth of a billionth of a joule. Silicon logic gates switch for femtojoules, which are a thousand attojoules, so this device is about a thousand times more efficient per switching event.

Why is room temperature hard?

Thermal energy at room temperature is enough to randomly perturb a single molecule. Holding ten distinguishable conductance states against that noise is the central technical achievement.

What is neural plasticity emulation?

Mimicking how biological synapses strengthen and weaken with use, including the transition from short-term to long-term memory. Demonstrated here through Pavlovian conditioning and Morse code pattern recognition.

Can this be manufactured?

Not remotely. This is one molecule in a laboratory measurement setup. There is no array, no yield figure and no fabrication route in what we could verify.

About the author

Mark Alex

Mark Alex is the founder and Managing Director of Real Biz Digital, a technology company operating out of Nairobi since 2018. He works in agentic AI and the Model Context Protocol, AI governance, enterprise software architecture and cybersecurity. He holds an MSc in Mechatronical Engineering from Obuda University in Budapest and a BSc in IT, Forensic Technology and Cybercrime, from USIU-Africa in Nairobi, and has published IEEE conference research on an AI-powered digital twin for greenhouse systems. He is the author of seven books. Between 2020 and 2024 he mentored more than 200 university students and interns in Nairobi. He writes every Just Out Tech article from the original research paper.