Most sleep advice treats sleep as a single thing you either got enough of or you didn't. The bedtime is 10:30, wake at 6:30, that's eight hours, done. The wearable on your wrist gives you a number and a colored ring and you feel good or bad about it.
Sleep is not a single thing. It is four distinct stages your brain cycles through in roughly 90-minute blocks, and each stage does specific biological work. Two of those stages consolidate memories. One clears metabolic waste from the brain. Another handles emotional processing and creative recombination. Skip enough of any one of them and something specific breaks, whether or not the total number of hours looks fine.
This is the basics, honestly told. What each stage does, how the cycles fit together, why architecture matters more than the hour count, and where consumer wearables actually help versus mislead.
The four stages
Modern sleep research divides sleep into two broad categories: non-REM (NREM), split into three substages, and REM. The current staging system (AASM 2007) replaced the older four-NREM-stage system by folding the deepest two stages together into what is now called N3.
N1 — light sleep, the transitional bridge
Duration: about 1 to 7 minutes per cycle, roughly 5 percent of a night.
N1 is the entry stage. Alpha brain waves give way to lower-amplitude theta, muscles relax, eyes start slow rolling movements. Wake someone from N1 and they will often insist they weren't asleep. Functionally, N1 is a bridge — the shift from wake to sleep, not a stage that does much work on its own. If you spike out of it (a jerk, a car horn, an intrusive thought), you go back to wake and try again.
Two things happen in N1 that are worth knowing. First, hypnic jerks — those falling sensations that snap you awake — happen here as motor systems relax faster than the arousal system. Second, if you are severely sleep-deprived, N1 collapses; you skip it and drop into deeper stages within a minute or two. That is a signal your body is short on sleep, not a success.
N2 — the workhorse most people ignore
Duration: about 45 to 55 percent of the total night. It is the stage you spend the most time in.
N2 is defined electrographically by two features that only appear here: sleep spindles (short 11 to 16 Hz bursts, half a second to two seconds long) and K-complexes (brief high-amplitude waves that appear spontaneously and in response to external stimuli). These are not decoration. Sleep spindles have been linked directly to memory consolidation — specifically procedural and motor learning. Practice a piano scale, sleep, and the improvement the next morning tracks spindle density from that night. Rasch and Born's 2013 review in Physiological Reviews walks through the evidence in detail.
N2 is also where body temperature drops further and heart rate slows. It is not "junk" sleep between the interesting stages. Cut N2 short and you cut motor learning and consolidation with it.
N3 — slow-wave sleep, the "deep" one your tracker fixates on
Duration: about 15 to 25 percent of the night, concentrated in the first third to first half.
This is what your wearable calls "deep sleep." It is dominated by delta waves — slow (0.5 to 4 Hz), high-amplitude oscillations that mark the deepest homeostatic sleep state. Blood pressure, heart rate, and breathing all drop. Growth hormone secretion peaks. Wake someone from N3 and they will be groggy, confused, disoriented for minutes — sleep inertia is worst here.
Two functions matter most.
First, memory consolidation, specifically declarative memory — facts, events, places. Slow oscillations coordinate hippocampal replay of the day's experiences and transfer them to neocortical storage. This is Rasch and Born's core thesis, backed by decades of consolidation experiments and formalized by Diekelmann and Born's earlier 2010 Nature Reviews Neuroscience paper.
Second, glymphatic clearance. Xie's 2013 Science paper — "Sleep drives metabolite clearance from the adult brain" — showed that the space between brain cells expands during sleep (particularly slow-wave sleep) and cerebrospinal fluid flushes through, clearing metabolic waste including beta-amyloid. The clearance rate during sleep was roughly double the rate during wake. This is the biological basis for the connection between chronic poor sleep and neurodegenerative risk that has been discussed for the last decade. The full story is still being written, and the mouse-to-human translation is not perfect, but the flushing mechanism itself is real.
N3 is front-loaded. Most of your deep sleep happens in the first three to four hours after sleep onset. Go to bed at 11 and sleep until 7, and most of your N3 was between 11 and 2. This has consequences we will get to.
REM — the paradoxical stage
Duration: about 20 to 25 percent of the night, concentrated in the second half.
REM (rapid eye movement) sleep is a strange state. The EEG looks nearly identical to wake — fast, low-amplitude activity. Your eyes dart under closed lids. Your brain is metabolically as active as when you are awake. Your voluntary muscles are essentially paralyzed (atonia), which is why you don't act out dreams. When that atonia system breaks — REM behavior disorder — people punch, kick, and jump out of bed while dreaming.
Functionally, REM handles emotional processing, complex procedural consolidation, and something like a dedicated creativity function. Cai's 2009 PNAS study on REM and creative problem-solving showed participants who reached REM during a nap improved measurably on associative tasks that non-REM nappers and quiet-rest controls did not. Wagner's 2004 Nature paper on sleep-dependent insight showed the same class of effect on hidden-rule discovery.
REM is where most vivid dreaming happens (dreams occur in other stages too, less vividly). It is when the amygdala is most active. And it is the stage that gets cut first when you shorten sleep. That is a whole separate article — the case that REM is the more undervalued stage.
The 90-minute cycle
Sleep does not proceed linearly through the four stages once. It cycles. A typical cycle runs about 90 minutes and looks roughly like this:
- N1 → N2 → N3 → back up through N2 → REM
- Repeat, roughly every 90 minutes
- Four to six full cycles across a normal eight-hour night
What changes across the night is the composition. Early cycles are heavy on N3 and light on REM. Later cycles flip — REM stretches get longer, N3 gets shorter or disappears entirely. The last REM period of the night can be 30 to 45 minutes long.
This is why waking naturally in the last third of your sleep window often catches you in or near REM, and why alarms that pull you out of that stage feel worse than alarms that catch you in N2. Some wearable "smart alarm" features try to time waking to lighter stages within a set window; the evidence they materially improve grogginess is mixed, but the underlying principle — sleep inertia depends on which stage you were pulled from — is real.
Why architecture matters more than hours
An eight-hour night with heavily fragmented sleep — one where you cycle in and out of N1 repeatedly, get little N3, and get shortened REM periods — is not equivalent to a seven-hour night with clean architecture and consolidated stages.
The clinical example is obstructive sleep apnea. A person with untreated moderate apnea can spend eight hours in bed, sleep the whole time on paper, and wake exhausted because breathing disruptions cause dozens of micro-arousals per hour that fragment N3 and REM. Their total sleep time on a tracker looks fine. Their sleep is not.
The subtler example is alcohol. A couple of drinks before bed shorten sleep latency, suppress REM in the first half of the night, and cause rebound REM plus fragmentation later. Eight hours in bed after drinking is not eight functional hours of sleep. The number your wearable gives you is misleading in a specific direction.
The prescriptive version: total sleep time is a necessary condition, not a sufficient one. If you are consistently getting seven to eight hours in bed but waking unrestored, the answer is more often about what is happening inside those hours than about adding more hours.
What consumer wearables can and cannot do
The gold standard for measuring sleep architecture is polysomnography (PSG) — EEG, EOG, EMG, ECG, respiratory bands, oxygen saturation, all run in a sleep lab or in a full home study. That is what every published study on the four stages uses. It is the ground truth.
Consumer wearables — wrist trackers, rings, mattress mats — do not measure EEG. They infer stages from proxies: heart rate variability, movement, respiratory rate, sometimes skin temperature. The algorithms are trained against PSG in validation studies. How well they do depends on which stage you ask about.
The current honest picture from validation literature (de Zambotti and colleagues have published the cleanest reviews):
- Wake vs sleep detection: generally good — agreement with PSG in the 85 to 95 percent range for most modern devices.
- REM detection: moderate — devices tend to over- or under-estimate depending on the individual, with agreement often in the 60 to 80 percent range.
- Deep sleep (N3) detection: weakest — many consumer devices over-report deep sleep against PSG. If your ring says you got 90 minutes of deep sleep, the real number is often meaningfully less.
- N1 vs N2 discrimination: essentially not done meaningfully at the consumer level. Most apps lump these together as "light sleep."
None of this makes wearables useless. They are useful for trend, not truth. If your deep sleep number drops by 30 percent over a week starting the week you began drinking wine every night, that trend probably reflects a real drop — the absolute number was never the point. If your device says you got zero REM, something is happening, whether the specific minute count is accurate or not.
The mattress-based sensors have a different profile. They do not touch your body, they capture ballistocardiography and respiration, and for people who will not tolerate a wristband or ring at night they are the practical option. The Withings Sleep Tracking Mat is the current best-in-class option in the Vyvata catalog — validated against PSG for sleep vs wake, screens for sleep apnea, and its stage estimates are as good as most wrist devices.
Honest caveat: for anyone with actual sleep concerns — chronic insomnia, suspected apnea, unexplained daytime sleepiness — a home tracker is not a diagnostic tool. It can flag a pattern worth investigating. Take that flag to a sleep physician, get a proper study, get a real answer.
The takeaway
Sleep is not one thing you either got or didn't. It is four distinct biological processes stacked in cycles across the night, each responsible for specific work — motor learning in N2, declarative memory and metabolic clearance in N3, emotional and associative processing in REM. Understand which stage does what, and the vague concept of "good sleep" becomes something you can reason about.
The next step is not to obsess over the numbers on your wrist tomorrow morning. It is to protect the conditions that let architecture assemble itself — consistent bedtime, dark and cool room, no alcohol at night, enough hours in bed for four to six full cycles. The architecture takes care of itself if the container is right. Every one of the interventions with the strongest evidence protects that container.
The companion piece to this one makes the case that the stage most people undervalue is not the one their tracker tells them to obsess over.
