REM vs deep sleep: what each does and how much you need

Deep sleep (N3, slow-wave) is the quiet, high-amplitude, hard-to-wake stage that front-loads the first third of the night and drives physical restoration, declarative-memory consolidation, and glymphatic clearance. REM (rapid eye movement) is the active, awake-like-EEG, dreaming stage that back-loads the last third of the night and supports emotional and procedural processing. They are not points on a single 'deeper' scale: they are two distinct brain states with opposite electrical signatures, and a night can be normal in one and short in the other. For the full staging map see the sleep stages pillar.

Deep sleep is defined by delta waves (0.5-2 Hz) occupying more than 20% of an epoch, the slowest, highest-amplitude EEG of the night, with the highest arousal threshold; heart rate and breathing are slow and regular. REM looks almost the reverse: a fast, low-amplitude, desynchronized EEG that resembles wakefulness, paired with vivid dreaming, irregular heart rate and breathing, and skeletal-muscle atonia (the body is paralyzed except the eyes and diaphragm). This is why a person woken from N3 is groggy and disoriented (sleep inertia) while a person woken from REM often recalls a dream and feels alert.

In healthy adults the most-cited normative dataset, Ohayon et al. 2004 (a meta-analysis of 65 polysomnography studies), puts deep sleep (N3) at roughly 13-23% of total sleep time and REM at roughly 20-25%. On a 7-8 hour night that is about 60-110 minutes of N3 and 90-120 minutes of REM. Both are age-dependent in opposite ways: N3 declines steeply across adulthood while REM stays relatively stable. There is no single target number to hit for either; consistent total sleep time is what protects both fractions. See how much deep sleep you need for the N3 detail.

The two stages occupy opposite ends of the night, which is the single most useful fact for reading a hypnogram. Slow-wave pressure is highest at sleep onset and dissipates with each cycle (Dijk 2009), so N3 is concentrated in the first one or two 90-minute cycles. REM is reciprocally back-loaded: the first REM period may last only ~10 minutes, the last (near morning) 30-60 minutes, because REM propensity rides the circadian temperature rhythm and peaks near the core-body-temperature minimum toward dawn. The practical consequence: going to bed earlier adds mostly deep sleep, while sleeping in or waking late adds mostly REM.

Deep sleep is the consolidation-and-clearance stage: slow oscillations and spindles coordinate hippocampal-to-cortical transfer of declarative memory (Rasch & Born 2013), interstitial space expands during NREM to accelerate glymphatic clearance of solutes including amyloid-beta (Xie et al. 2013), and most growth-hormone secretion occurs here. REM is the emotional-and-procedural stage: it supports emotional-memory processing, overnight recalibration of emotional reactivity, and motor/procedural learning (Walker; Rasch & Born 2013). Both are homeostatically defended: selectively depriving either produces a rebound of that specific stage on recovery nights, evidence the brain protects a quota of each.

The two stages are vulnerable to different things. Aging and sleep deprivation hit deep sleep hardest, with N3 minutes dropping steeply across adulthood (Ohayon et al. 2004). REM is preferentially lost by cutting the night short, since it is back-loaded toward morning, and is blunted by alcohol (which suppresses first-half REM then causes second-half rebound and fragmentation; Ebrahim et al. 2013) and by THC and many antidepressants. So 'low deep sleep' and 'low REM' point to different causes: low N3 suggests short or late sleep and age, while low REM suggests a truncated morning or evening alcohol. And both numbers on a wearable are estimates inferred from heart rate and movement, not EEG, best read as weekly trends. The lever for both is the same: more total sleep on a consistent schedule.