Definition
Action potential: An action potential is a brief, all-or-none electrical pulse that travels along a neuron's membrane when its voltage crosses a threshold. Driven by the opening and closing of sodium and potassium channels, it is the basic unit of signaling in the nervous system and the event biocomputing systems record and stimulate.
Also known as: spike, nerve impulse, neuronal firing
An action potential is a brief, all-or-none electrical pulse that a neuron fires when the voltage across its membrane crosses a threshold. Lasting roughly a millisecond or two, it is produced by voltage-gated sodium and potassium channels and travels along the axon to trigger communication at synapses. It is the basic unit of neural signaling and of biocomputing.
Key takeaways
- An action potential (often just called a “spike”) is the discrete event neurons use to send information over distance.
- It is all-or-none: information is carried by when and how often neurons fire, not by spike size.
- Hodgkin and Huxley’s 1952 model of the squid giant axon explained the ionic mechanism and still underpins computational neuroscience.
- Microelectrode arrays detect action potentials from outside the cell as small voltage deflections, which are then sorted and analyzed.
- In organoid intelligence, action potentials are both the output we read and the language we use to stimulate living networks.
How does an action potential work?
At rest, a neuron keeps its inside negative relative to the outside, typically somewhere around minus 60 to minus 70 millivolts in mammalian central neurons (Bean, 2007). This resting potential is maintained by ion pumps and by the membrane’s selective permeability, mainly to potassium.
When incoming signals at the dendrites push the membrane voltage up far enough, a cascade begins. The sequence below is the textbook picture that Hodgkin and Huxley first described quantitatively in The Journal of Physiology (1952).
- Depolarization to threshold. Excitatory inputs raise the membrane voltage. If it reaches threshold, voltage-gated sodium channels begin to open.
- Rising phase. Sodium ions rush in, which depolarizes the membrane further and opens more sodium channels. This positive feedback drives the voltage sharply upward, past zero.
- Peak and repolarization. Sodium channels inactivate while slower voltage-gated potassium channels open. Potassium flows out and the voltage falls back down.
- Afterhyperpolarization. Potassium channels close slowly, so the voltage briefly undershoots the resting level.
- Refractory period. For a short time the neuron cannot fire again, or needs a stronger push to do so.
Why is it called “all-or-none”?
Because of the sodium channel feedback loop, once threshold is crossed the spike runs to completion. A bigger input does not produce a bigger spike. Neurons encode stimulus strength through firing rate and precise timing instead. This is central to neural encoding and decoding: to understand what a network is “saying,” you look at patterns of spikes across time and across cells.
How does the spike travel?
The depolarization at one patch of membrane spreads to the next, opening its sodium channels in turn. The refractory period behind the wave stops it from reversing. In myelinated axons, the spike jumps between gaps in the myelin sheath, which speeds conduction. When the spike reaches the axon terminal, it triggers neurotransmitter release onto the next neuron.
What do the phases of an action potential look like?
| Phase | Main ion movement | What the voltage does |
|---|---|---|
| Resting state | Potassium leak, pumps active | Stable, negative inside |
| Depolarization | Sodium flows in | Rises rapidly past zero |
| Repolarization | Potassium flows out, sodium channels inactivate | Falls back toward rest |
| Afterhyperpolarization | Potassium channels still open | Dips below resting level |
| Refractory period | Channels resetting | Firing blocked or harder |
Exact values for threshold, peak and duration vary between neuron types. Bean’s 2007 review in Nature Reviews Neuroscience describes how different central neurons use different mixtures of ion channels to shape spikes, which is one reason some neurons fire in fast trains while others fire slowly.
How are action potentials recorded in a dish?
There are two broad ways to observe spikes, both part of electrophysiology.
- Intracellular (patch clamp) recording puts an electrode in contact with the inside of a single cell. It sees the full voltage swing in detail but usually only for one or a few neurons at a time.
- Extracellular recording places electrodes outside cells. Each electrode picks up much smaller voltage deflections from nearby firing neurons. This is how a microelectrode array works, and it scales to many neurons at once.
On a high-density microelectrode array, thousands of electrodes sit beneath the tissue. Software first detects candidate spikes by threshold crossing, then uses spike sorting to estimate which signals came from which neuron, based on waveform shape and location. Sharf et al. (2022, Nature Communications) used this kind of high-density recording to characterize single-neuron activity and network oscillations in human brain organoids.
For a deeper tour of the hardware, see our guide to how microelectrode arrays listen to and talk with neurons.
Why do action potentials matter for biocomputing?
In biocomputing with living neurons, the action potential is the shared currency between biology and software.
- Output: The computer reads the network’s state as spike times and rates across electrodes.
- Input: Electrical stimulation through the same electrodes evokes action potentials, which is how information is written into the culture.
- Learning signal: Changes in connection strength depend on the relative timing of spikes, a process described by spike-timing-dependent plasticity.
- Collective dynamics: Synchronized firing across many cells produces network bursts, which can dominate activity in cultures and organoids.
Kagan et al. (2022, Neuron) showed this loop in practice: cultured neurons on an MEA received stimulation encoding the position of a ball in a simplified Pong game, and their evoked spiking was decoded into paddle movements. Over a short period of play, performance improved. Our article on how neurons learn in a dish walks through that experiment.
Action potential vs artificial neuron activation: what is the difference?
Artificial neurons in deep learning output a continuous number at every step. Biological neurons output discrete spikes at specific moments. Spiking neural networks sit in between: they simulate spike events with simplified models. The leaky integrate-and-fire model drops most ion channel detail, while Izhikevich’s 2003 model in IEEE Transactions on Neural Networks reproduces many real firing patterns with only two equations.
| Feature | Biological action potential | Spiking model neuron | Deep learning unit |
|---|---|---|---|
| Signal type | Discrete voltage spike | Discrete simulated spike | Continuous value |
| Timing matters | Yes, precisely | Yes | Usually no |
| Mechanism | Ion channels | Equations | Weighted sum plus activation function |
| Energy cost | Metabolic, very low per event | Depends on hardware | Depends on hardware |
What are the limitations of working with action potentials?
- Extracellular signals are indirect. Electrodes see a blurred, attenuated version of the spike, mixed with noise and signals from neighboring cells.
- Spike sorting is imperfect. Assigning spikes to individual neurons involves statistical guesswork, especially in dense 3D tissue.
- Sampling is partial. Even high-density arrays only see neurons close to the electrode surface, which matters for thick organoids.
- Stimulation is blunt. Electrical pulses activate many cells at once and create artifacts that can mask nearby spikes for a short time.
These are engineering problems as much as biological ones. Making spike data reliable, time-aligned and reproducible is part of what we are building at Organoid Programming, alongside open data formats such as Neurodata Without Borders.
Conclusion
The action potential is the simplest and most important event in neural computing: a millisecond-scale spike that carries information through timing and rate. Understanding how it is generated, recorded and evoked is the starting point for every other concept in organoid biocomputing, from plasticity to closed-loop training.
Frequently asked questions
What is an action potential in simple terms?
An action potential is a short electrical pulse a neuron produces when it is excited enough. The inside of the cell briefly flips from negative to positive voltage and then returns, all within a few milliseconds. The pulse travels down the axon and triggers the release of chemical messengers at synapses, which is how one neuron passes a signal to the next.
Why is the action potential called all-or-none?
Once the membrane voltage crosses threshold, voltage-gated sodium channels open in a self-reinforcing way and the spike runs its full course. A stronger stimulus does not make a bigger spike. Instead, neurons convey intensity through how often and when they fire. This is why spike timing and spike rate, rather than spike size, carry most of the information in neural codes.
How are action potentials recorded from brain organoids?
Brain organoids are usually placed on a microelectrode array, a chip with many small electrodes. When nearby neurons fire, the electrodes pick up small extracellular voltage deflections. Software then detects these events and, through spike sorting, attempts to assign them to individual neurons. The result is a time-stamped list of spikes that can be analyzed or used to drive closed-loop experiments.
What is the refractory period?
The refractory period is the short window after an action potential during which a neuron cannot fire again (absolute refractory period) or needs a stronger push to fire (relative refractory period). It happens because sodium channels are inactivated and potassium channels are still open. The refractory period limits maximum firing rates and helps spikes travel in one direction along the axon.
Do artificial neural networks use action potentials?
Most deep learning networks do not. They pass continuous numbers between units. Spiking neural networks and neuromorphic chips do model discrete spike events, often with simplified neuron models such as leaky integrate-and-fire. Biocomputing goes one step further by using real action potentials produced by living neurons as the computational signal.
Sources
- Hodgkin & Huxley (1952). A quantitative description of membrane current and its application to conduction and excitation in nerve. The Journal of Physiology.
- Bean (2007). The action potential in mammalian central neurons. Nature Reviews Neuroscience.
- Kagan et al. (2022). In vitro neurons learn and exhibit sentience when embodied in a simulated game-world. Neuron.
- Sharf et al. (2022). Functional neuronal circuitry and oscillatory dynamics in human brain organoids. Nature Communications.
- Izhikevich (2003). Simple model of spiking neurons. IEEE Transactions on Neural Networks.
