Artificial Neurons That Work Like Real Ones: From Mimicking Biology to Communicating With It

A new study in Nature Communications demonstrates that artificial neurons now operate with the same physical and functional parameters as real nerve cells, marking a pivotal step toward biohybrid intelligence

The algorithm of life has been solved; it’s now in mass production.

Introduction: Where the Synthetic Meets the Biological

Artificial intelligence has always relied on the abstraction of “artificial neurons,” simplified computational units that mimic how biological neurons process information. Yet these artificial constructs have remained physically distant from real biology. They operate at voltages and energy levels far higher than the brain, lack chemical sensitivity, and cannot interact directly with living tissue.

Researchers at the University of Massachusetts Amherst have now bridged that divide. Their new artificial neurons not only emulate the behavior of biological ones but also share their electrical, energetic, and chemical characteristics. Operating at biological voltage levels of around 60 millivolts, consuming energy in the picojoule range, and responding to neurotransmitter-like molecules such as dopamine and sodium ions, these devices represent a new generation of neuromorphic components that physically behave like living cells rather than just simulating them.

Designing a Physically Realistic Neuron

At the core of this breakthrough lies a memristor-based circuit that reproduces the fundamental electrophysiological behavior of neurons. A memristor or memory resistor changes its resistance depending on the electrical history of the current that passes through it, mimicking the way biological synapses adjust their conductivity through ionic movement.

In the UMass design, the memristor is coupled with a capacitor and a resistor to form a true leaky integrate-and-fire circuit, where electrical charge builds up on the capacitor until a threshold is reached, producing a voltage spike that discharges through the memristor. This interaction between charge integration and leakage emulates the timing and refractory dynamics of a real neuron. The system operates at approximately 60 mV, closely matching the biological range of neuronal action potentials, and exhibits an “off” resistance of about 200 megaohms, equivalent to the membrane resistance of real cells. Each firing event requires only a few picojoules (pJ) of energy, placing it squarely within the same energy efficiency scale as natural neural processes.

Integrating Chemistry Into Electronics

A defining aspect of this work is that the artificial neurons are not purely electronic they are chemically active as well. The researchers integrated chemiresistive elements that can detect and respond to the concentration of specific ions and neurotransmitter analogs in the surrounding medium.

When sodium ions (Na⁺) were introduced, the neurons’ firing frequency increased, reproducing the excitatory effect of sodium influx in biological neurons. In contrast, exposure to dopamine reduced the firing frequency, mirroring dopamine’s inhibitory function in living systems. These modulations occurred without increasing energy consumption, indicating that chemical feedback can dynamically tune the electrical behavior of the artificial neurons. This fusion of chemical and electrical responsiveness brings neuromorphic systems closer than ever to genuine biological intelligence.

Communication With Living Cells

To validate the biocompatibility and functionality of their design, the researchers connected their artificial neurons to cultures of cardiomyocytes heart muscle cells that generate rhythmic electrical pulses. The neurons successfully detected and responded to these pulses, firing in synchrony with the living tissue. When the cardiomyocytes were exposed to noradrenaline, a neurotransmitter that naturally increases heart rate, the artificial neurons exhibited a proportional increase in their firing frequency.

This demonstrated true bidirectional communication between artificial and biological systems. For the first time, an electronic neuron was able not only to read a biological signal but also to respond to it appropriately in real time, proving that direct neural interfacing is possible at the single-cell level.

Quantitative Performance and Biological Parity

Parameter	Artificial Neuron	Biological Neuron Range	Match
Operating Voltage	~60 mV	50–80 mV	Fully matched
Membrane Resistance	~200 MΩ	100–300 MΩ	Comparable
Energy per Spike	1–3 pJ	~1 pJ	Same order
Frequency Range	1–50 Hz	1–60 Hz	Consistent
Chemical Modulation	Dopamine / Na⁺	Present	Matched
Live-Cell Interaction	Real-time	—	Achieved

The close correspondence between these parameters confirms that the devices achieve parametric parity with biological neurons. Instead of abstractly imitating neural activity, these circuits recreate it under the same physical and chemical conditions, bridging the gap between silicon and biology at the most fundamental level.

Implications: Toward Biohybrid Intelligence

The implications of this work extend across neuroscience, computing, and bioengineering. In neuromorphic computing, low-voltage, energy-efficient neurons could lead to processors that match the brain’s computational density while consuming only a fraction of current power levels. In brain–machine interfaces (BMIs), these neurons could communicate directly with living tissue without causing electrical or thermal damage, allowing seamless neural prosthetics and real-time neural augmentation.

Beyond that, the creation of biohybrid neural networks, where living and artificial neurons coexist in a shared information system, becomes a tangible goal. Such systems could revolutionize biomedical sensing, adaptive robotics, and even experimental cognitive science by creating circuits that are both synthetic and alive. The convergence of organic and inorganic intelligence is no longer a metaphor it is a laboratory reality.

Challenges and Future Directions

Despite the success, significant challenges remain. The long-term stability of memristive nanostructures must be tested under biological conditions where ions, moisture, and temperature fluctuations can affect performance. Scalability is another key issue; fabricating uniform arrays of thousands of identical neurons with consistent thresholds will require precise nanoscale manufacturing.

Furthermore, these artificial neurons currently lack plasticity the ability to strengthen or weaken connections through learning. Achieving synaptic adaptability is essential for building self-organizing, intelligent networks. Finally, researchers must ensure biocompatibility over extended periods to avoid tissue rejection or chemical degradation when interfaced with living systems. Addressing these challenges will determine how quickly this technology can move from the lab to practical applications.

Conclusion: From Imitation to Integration

This study marks a turning point in artificial intelligence and neuroscience. For the first time, artificial neurons have achieved not only functional but also physical equivalence with their biological counterparts. They operate at the same voltages, consume the same energy, and respond to the same chemical cues.

By combining electronic precision with biological sensitivity, the UMass Amherst team has built the foundation for living electronics systems that no longer just simulate the brain but participate in its language. The boundary between organic and artificial intelligence is fading, giving rise to a new paradigm: integration rather than imitation.

References

1. Fu, S., Gao, H., Wang, S., et al. (2025) - Constructing artificial neurons with functional parameters comprehensively matching biological values.

   Nature Communications, 16, 8599. DOI: 10.1038/s41467-025-63640-7

2. ScienceDaily (October 13, 2025) - Scientists build artificial neurons that work like real ones.

   https://www.sciencedaily.com/releases/2025/10/251013040335.htm