The action potential as the neuron's digital-ish signal

In BCI-1.1 you left the neuron sitting at rest, its inside about 70 millivolts negative relative to its outside, written -70 millivolts. That resting voltage is a charged capacitor, the membrane battery from S9.1 topped up and idling: sodium ions (Na+) piled up outside, potassium ions (K+) piled up inside, and the pumps quietly paying to keep it that way. A battery on a shelf does no work. This lesson is about the one event where the neuron closes the circuit and spends that charge in a controlled, explosive spike, then recharges. That spike is the action potential, and it is the fundamental unit of information the nervous system signals with, and the richest thing a brain-computer interface can hope to read. We are going to build it from the resting state up, one channel at a time.

The threshold: a spark that catches

Start with the trigger. Something (a signal from another neuron, or an experimenter's electrode) nudges a patch of membrane in the depolarizing direction, meaning it makes the inside less negative, say from -70 up toward -60 millivolts. Depolarize is just the word for the voltage moving toward zero from its resting negative. So far this is a passive nudge and it fades on its own.

Now add one new kind of protein that BCI-1.1's resting neuron did not lean on: the voltage-gated sodium channel. It is an ion channel exactly as S9.1 defined one, a gated water-lined pore, except its gate is wired to the membrane voltage itself. Below a certain voltage it stays shut. Push the membrane past a threshold, around -55 millivolts, and the gate springs open.

Watch what that does, because it is the whole trick. When the channel opens, Na+ pours in. It is driven two ways at once, both downhill: sodium is far more concentrated outside than in, and the inside is negative, so charge pulls the positive ion inward too. That inrush of positive charge depolarizes the patch further. But the gates are voltage-controlled, so more depolarization opens more sodium channels, which lets in more Na+, which depolarizes even further. That is positive feedback, a spark catching fire. Once threshold is crossed, the process runs away on its own and slams the membrane voltage upward, past zero and on toward +30 to +40 millivolts, in well under a millisecond. The inside of the cell, negative a moment ago, is briefly positive. That near-vertical upstroke is the action potential's rising edge.

The energy for that spike is not made here. It is the sodium gradient discharging, the store the pumps spent minutes building, dumped through opened channels in a flash. S9.1 called this out: charging costs ATP and is slow, discharging is free and fast.

The reset: potassium out, then a dip, then recharge

A spark that only catches never stops. Two things bring the voltage back down, and their timing is the point.

First, the sodium channels do not just close, they inactivate. Each one has a second gate, a kind of plug, that swings shut a fraction of a millisecond after the channel opens, and it jams the pore even while the voltage is still high. So the sodium inrush shuts itself off automatically. This inactivation matters again in a moment, so hold onto it.

Second, a slower channel joins in: the voltage-gated potassium channel. The same depolarization that opened the sodium gates also opens these, but they are sluggish and lag behind. By the time they are fully open, sodium is already inactivating. Now K+ flows, and it flows outward, because potassium is concentrated inside and the inside has just turned positive, so both the concentration and the charge push it out. Positive charge leaving the cell drags the voltage back down. This is repolarization, the falling edge.

The potassium channels are slow to close as well as slow to open, so they overshoot. For a brief window potassium keeps leaving after the voltage has already returned to rest, and the membrane dips more negative than -70, a small undershoot (also called hyperpolarization or the afterhyperpolarization). Then the voltage-gated potassium channels finish closing and the leak conductances settle the patch back to its resting -70 millivolts, recharged and ready, while the sodium-potassium pump quietly restores the concentrations over the longer run. The whole spike, up and down and dip, takes roughly one to two milliseconds.

All-or-none: the spike has exactly one size

That warning is worth stating as a law, because it shapes everything downstream. The action potential is all-or-none. Below threshold, nothing fires and the nudge fades. At or above threshold, the positive feedback guarantees a full-size spike. There is no half spike and no giant spike. The size is set by the ion gradients and the channel physics, not by the input.

For a programmer this is the first genuinely digital-flavored property. It is a thresholded, discretized event: analog voltage goes in, and out comes a clean go or no-go. But notice what it costs. If every spike is identical, then a single spike cannot tell you how bright the light was or how hot the stove. That information has to live somewhere else, and it does: in how many spikes per second the neuron fires (the rate) and exactly when they arrive (the timing). A brighter light makes the same-size spikes come faster. We will pull this apart at the end, because it is the crux of what a BCI is actually reading.

Travel without fading: a spike is not a voltage in a wire

A signal that fires but cannot travel is useless. An axon (the neuron's output cable) can be long, up to about a meter from your spinal cord to your foot. Here is the problem you would expect from S9.1's capacitor picture: if you just injected a voltage at one end of that cable, it would leak and decay as it spread, the way a passive electrical signal dies out over distance in a lossy wire. A membrane leaks charge, so a passive bump would be a whisper by the time it traveled even a millimeter. That cannot carry a foot-long signal.

The action potential dodges this by refusing to be passive. When a patch of membrane spikes, the Na+ flooding in there spills sideways just far enough to depolarize the immediately adjacent patch past its threshold. That neighbor then fires its own full-size spike from its own gradients. Which depolarizes the next patch. Which fires. The signal is not passed along, it is regenerated, freshly rebuilt at full amplitude at every step. A programmer can borrow the idea of a repeater on a long line, a device that receives a weakening pulse and retransmits a clean one, so no matter how long the run, the far end sees full strength. The spike is a chain of repeaters, each patch of axon relighting the next.

Because every step is a fresh full spike, the action potential arrives at the end of a meter-long axon exactly as tall as it started. Propagation without amplitude loss is not a feature bolted on. It falls straight out of "regenerate rather than conduct."

The refractory period: one-way travel and a speed limit

Now cash in the sodium inactivation you were told to hold onto. Right after a patch fires, its sodium channels are not merely closed, they are inactivated, plugged, and they stay that way for a couple of milliseconds until the voltage has returned to rest and resets them. During that window the patch physically cannot fire again no matter how hard you push it. That is the absolute refractory period.

This dead window does two jobs, and both are load-bearing. First, it enforces one-way travel. When a spike moves forward into fresh membrane, the patch it just left is refractory, so the wave cannot double back. The signal can only propagate away from where it started, which is why axons carry information in one direction. Second, it caps the firing rate. A patch has to recover before it can spike again, so a neuron cannot fire faster than its channels can reset, which sets a ceiling of very roughly a few hundred spikes per second. There is no way to squeeze two spikes closer than that floor allows.

After the absolute window there is a softer relative refractory period: the potassium channels are still finishing up and the membrane is still hyperpolarized (that undershoot), so the patch can fire, but only if a stronger-than-usual stimulus drags it back to threshold. The neuron is reluctant, not frozen.

Myelin: insulation with gaps, and jumping between them

Regeneration is reliable but slow, because relighting the channels at every single patch takes time. Nature's speed hack is myelin, a fatty sheath that other cells (glia) wrap in many layers around the axon like insulation around a wire. Fat is a good electrical insulator, so a myelinated stretch leaks far less charge.

The sheath is not continuous. It comes in segments with small bare gaps between them called nodes of Ranvier, and the voltage-gated channels cluster densely at those nodes. Here is the payoff. Along an insulated segment the signal travels the fast, passive way, as a voltage that barely decays because the myelin has plugged the leaks. It only stops to regenerate at the next node, where the channels fire a fresh full spike and re-launch it. So the action potential effectively leaps from node to node instead of crawling continuously, which is called saltatory conduction (from the Latin for "jumping").

This is both faster, up to something like ten to fifty times faster than an equivalent bare axon, and cheaper, because far fewer channels open and so far less Na+ and K+ have to be pumped back afterward. It buys speed and saves the metabolic bill at once. It also names a real, fragile dependency: when myelin is damaged, as in multiple sclerosis, the insulated segments leak, the passive signal decays before it reaches the next node, and conduction slows or fails outright. The nervous system's speed is only as good as its insulation.

Digital-ish, and exactly where the word lies

Now pay off the title honestly, because this track cares about not overselling. It is fair to call the action potential digital-ish, and it is a real mistake to call it digital.

The digital-ish part is genuine. The spike is all-or-none, a thresholded go or no-go, so a lot of the analog mush at the input is cleaned into a discrete event. And it regenerates without degrading, exactly like a repeated digital pulse and unlike an analog voltage that decays. Those two properties, discretization and lossless relay, are why the metaphor is worth reaching for at all.

Here is where it breaks, and the failure edge matters as much as the analogy. There is no clock. A digital computer marches to a global timing signal that says "now sample every wire," and the brain has nothing of the kind. Spikes land at whatever continuous, real-valued instant they land, and that timing is itself the message, a spike two milliseconds earlier means something different from one two milliseconds later. The firing rate varies smoothly, 43 spikes per second shading into 44, not snapping between fixed levels. And a spike is not a bit in a word: there is no fixed word length, no framing, no byte boundary. The information is smeared across the precise timing of many spikes and across the pattern over a whole population of neurons at once. Underneath, the neuron is doing analog computation, continuously summing its inputs, and it merely uses the all-or-none spike as a clean way to shout the result down a long wire. Digital-ish output, analog everything else.

So the honest one-liner: a neuron thresholds an analog computation into discrete, non-degrading spikes, but the meaning lives in their continuous timing and smoothly varying rate, not in their amplitude and not in any clocked pattern of ones and zeros.

Key terms

action potential

The neuron's signal: a fast, all-or-none swing of membrane voltage from about -70 millivolts up toward +30 to +40 and back, produced by voltage-gated channels discharging the ion gradients.

voltage-gated sodium channel

An ion channel whose gate opens when the membrane depolarizes past threshold, letting Na+ rush in and drive the spike's rising edge through positive feedback.

threshold

The membrane voltage, around -55 millivolts, past which the positive feedback becomes self-sustaining and a full spike is guaranteed to fire.

all-or-none

The rule that a stimulus below threshold fires nothing while any stimulus at or above threshold fires one identical full-size spike, so spike amplitude carries no information about stimulus strength.

refractory period

The window right after a spike when sodium channels are inactivated and the patch cannot fire again (absolute), then can fire only with extra push (relative), which forces one-way travel and caps the firing rate.

saltatory conduction

Fast propagation in which the spike jumps between nodes of Ranvier, the bare gaps in the myelin insulation, regenerating only at the nodes instead of at every patch.

myelin

The fatty insulating sheath glial cells wrap around an axon, leaving nodes of Ranvier as gaps, which speeds conduction and lowers its energy cost.

rate and timing code

The idea that neural information is carried by how many spikes fire per second and precisely when, rather than by the amplitude of any single spike.

How little of the battery a single spike actually spends

It is easy to assume a spike drains the sodium gradient the way a discharge empties a battery. It barely dents it. The membrane is a thin capacitor, so flipping its voltage from -70 up to +40 millivolts takes only a whisper of ions, a tiny fraction of a percent of all the sodium waiting outside. That is why the sodium-potassium pump can comfortably keep up: it is not refilling an emptied tank after each spike, just topping off the trickle that crossed. This is also why myelin's savings compound. Fewer channels opening per meter of axon means fewer ions to shuttle back, so the pump's ATP bill drops. The neuron can fire hundreds of times a second for a lifetime precisely because each spike is a cheap sip from a large, constantly refilled store, not a full drain and recharge. The all-or-none spike looks dramatic, but energetically it is thrifty, and the real ongoing cost is the pump quietly holding the gradient steady between spikes.