Lab-in-a-Tab

How Memory Works

How does your brain store 2.5 petabytes of memories - and why do some last forever while others vanish overnight?

β–Ά Run the interactive simulation
MemoryLTPPlasticity

Your brain rewires itself when you learn!

Junior level β€” plain language, no maths

Ever wondered how you learned to ride a bike? At first it was hard - your brain had never done it. But with practice something remarkable happened: your brain literally rewired itself. The neurons handling balance and pedalling began talking to each other more efficiently, forging stronger connections, until one day the skill ran itself and you could do it without a thought. That physical change in the brain is what a memory really is.

Scientists sum it up in a lovely phrase: "neurons that fire together, wire together." Every time two neurons light up at the same instant, the synapse between them grows a shade stronger. Repeat the experience enough - practise the piano, revise for a test, replay a happy afternoon in your head - and the connection turns rock-solid. That's synaptic plasticity.

Not all memory is the same. Short-term memory clutches a scrap of information for seconds or minutes, like a phone number you just glanced at. Long-term memory can last a lifetime. To move a memory from one to the other, your brain replays it - above all during sleep, when the hippocampus re-runs the day's events and slowly hands them off to the cortex for permanent keeping.

Things worth knowing

  • The human brain can store an estimated 2.5 petabytes - equivalent to about 3 million hours of TV. You'd never run out of space.
  • Sleep is essential for memory. Students who sleep after studying retain up to 40% more than those who stay up all night revising.
  • Memory champions who memorise hundreds of random digits use "method of loci" - imagining walking through a familiar place. It exploits spatial memory circuits in the hippocampus.

Long-Term Potentiation and the Hebbian Synapse

Student level β€” the core equations

Long-term potentiation - LTP - is the cellular act of remembering. Bliss and LΓΈmo found in 1973 that a burst of high-frequency stimulation leaves a synapse lastingly stronger, and the molecule at the heart of it is the NMDA receptor. It works as a coincidence detector: it only opens when two things happen together - the incoming neuron releases glutamate and the receiving neuron is already depolarised enough to knock a blocking magnesium ion out of the channel's mouth.

When both conditions coincide, calcium pours into the receiving cell and sets a cascade in motion. Kinases - chiefly CaMKII - crank up the existing AMPA receptors and ferry new ones to the synapse, boosting its strength within minutes. Push the stimulation harder and the change becomes permanent: genes switch on, fresh proteins are built, and the dendritic spine physically swells. It all maps neatly onto Hebb's rule, \(\Delta w_{ij} = \eta\, x_i x_j\) - joint activity strengthens the link.

Where these changes matter most is the hippocampus, the brain's gateway for facts and events. The starkest proof came from patient H.M., whose hippocampus was surgically removed in 1953: he kept his old memories and his skills, but could never again lay down a new fact or experience. During sleep the hippocampus replays the day in compressed bursts called sharp-wave ripples, gradually shipping each memory out to the cortex for long-term keeping - a slow migration known as systems consolidation.

Key formulas

Hebbian rule\(\Delta w_{ij} = \eta\, x_i x_j\)co-activation strengthens
NMDA coincidence\(\text{glutamate} \;\wedge\; \Delta V > 0 \;\Rightarrow\; \text{Ca}^{2+}\ \text{influx}\)
LTP induction\([\text{Ca}^{2+}]_i \uparrow \;\to\; \text{CaMKII} \;\to\; \text{AMPA insertion}\)
STDP\(\Delta t < 0 \Rightarrow \text{LTP};\quad \Delta t > 0 \Rightarrow \text{LTD}\)Ξ”t = t_pre βˆ’ t_post

Things worth knowing

  • Patient H.M. (Henry Molaison) lost his hippocampus in 1953. He couldn't form any new declarative memories until his death 55 years later, advancing our understanding of memory more than any other case.
  • During sleep, hippocampal sharp-wave ripples replay the day's events up to 20Γ— faster than real time, consolidating memories into the cortex.
  • Playing a musical instrument restructures the brain's motor and auditory cortices - a 10-year-old pianist has measurably thicker grey matter than a non-musician.

Synaptic Plasticity Rules, Engrams, and Systems Consolidation

Scholar level β€” full mathematical depth

01Timing is everything: STDP

Hebb's rule says co-active neurons strengthen their link, but experiments revealed something sharper: the exact order of firing matters, down to milliseconds. In spike-timing-dependent plasticity, a synapse is strengthened when the presynaptic spike arrives just before the postsynaptic one - plausibly causal - and weakened when it arrives just after, within a narrow \(\pm 40\) ms window, \(\Delta W = A^+ e^{\Delta t/\tau^+}\) for potentiation and \(-A^- e^{-\Delta t/\tau^-}\) for depression. The brain, in effect, rewards synapses that helped predict the next spike.

02Keeping plasticity stable: the sliding threshold

Pure Hebbian strengthening is unstable - strong synapses would only grow stronger without bound. The BCM theory fixes this with a sliding threshold \(\theta_M\) that itself tracks recent activity: a neuron that has been firing a lot raises its bar for potentiation, so plasticity self-regulates. Calcium is the physical readout - a large influx drives LTP, a moderate one LTD, and this graded, self-adjusting response keeps a lifetime of learning from saturating the network.

03The engram: catching a memory in the act

The physical trace of a specific memory - the engram - turns out to be a sparse ensemble, only ~5–10% of the neurons in a region, bound together by potentiated synapses. This was speculation for a century until optogenetics let researchers tag the exact cells active during an experience and then switch them back on with light. Reactivating an engram makes an animal behave as if recalling the event; silencing it blocks the recall. The memory really does live in that identifiable set of cells.

04Writing and erasing memories with light

Once you can address an engram, you can edit it. In a landmark 2013 experiment, researchers implanted a false memory in a mouse by artificially co-activating an engram for a place with an aversive shock, so the animal later feared a location where nothing bad had happened. And in mice made amnesic by blocking protein synthesis, directly reactivating the engram restored the "lost" memory - showing the trace was stored but merely inaccessible. Memory became something you could not just observe but manipulate.

05Storage finer than the cell

Remarkably, the unit of storage is smaller than a neuron. A single dendritic spine can be potentiated or depressed independently of its neighbours a few microns away on the same branch, so one cell can hold thousands of near-independent memory elements. This vastly multiplies the brain's capacity and explains how ~86 billion neurons store an estimated petabytes' worth of experience - the real bit is the synapse, not the cell.

06The sleeping dialogue that makes memories last

Systems consolidation is the slow handover of a memory from the fast-learning hippocampus to durable neocortical storage, and it happens largely in sleep through an exquisitely timed conversation between brain rhythms. Hippocampal sharp-wave ripples, cortical sleep spindles and slow oscillations nest inside one another in a precise hierarchy, and this coordination is thought to replay the day's traces to the cortex just when it is primed to receive them - gradually embedding a fleeting experience into the distributed fabric of long-term memory.

Key formulas

STDP (potentiation)\(\Delta W = A^+ e^{\Delta t/\tau^+}\)Ξ”t < 0
STDP (depression)\(\Delta W = -A^- e^{-\Delta t/\tau^-}\)Ξ”t > 0
BCM rule\(\dfrac{dw}{dt} = \phi(v_{\text{post}}, \theta_M)\,v_{\text{pre}}\)
Sliding threshold\(\tau\,\dfrac{d\theta_M}{dt} = v_{\text{post}}^2 - \theta_M\)
Calcium control\([\text{Ca}^{2+}] > \theta_{\text{high}} \Rightarrow \text{LTP}\)
Engram\(S = \{\, n_i : w_{ij} \gg \text{baseline} \,\}\)sparse ensemble

Things worth knowing

  • Optogenetics - controlling neurons with light - allowed researchers to implant a false memory in a mouse in 2013 (Ramirez et al., Science), directly testing engram theory.
  • A single dendritic spine can potentiate or depress independently of its neighbours on the same dendrite - the basic unit of storage is finer than the neuron itself.
  • BDNF Val66Met, a common genetic variant (~30% of the population), impairs activity-dependent BDNF secretion and is associated with reduced hippocampal memory consolidation.

Sources

Full article on Wikipedia β†—