The last two hours of sleep are where most REM lives

The Short Version

Sleep is not spread evenly across the night: Matt Walker says most deep non-REM sleep comes in the first half and most REM sleep in the second half, "particularly in the last quarter of the night" [2]. Gina Poe describes the same slope, with more and more REM the later the night gets and longer REM periods in the second half [4]. That means an early alarm and a late bedtime are different losses: the alarm mostly removes REM, while a much later bedtime mostly removes deep slow-wave sleep and, on Poe's account, the first cycle's growth hormone surge and brain cleanup [1][4]. Andrew Huberman's advice on getting more REM starts from the same fact, which is why both he and Walker tell people to add minutes at the morning end of sleep [2][3]. None of the sources puts an exact figure on the last two hours, so treat that phrase as shorthand for Walker's last quarter of a normal night [2].

Table of contents

What They Said

Matt Walker: the night has two different halves

Walker, a professor of neuroscience and psychology at UC Berkeley, describes his own night to show the pattern [1]. He goes to bed around 10:30 p.m. and wakes naturally just before 7, with an alarm set for 7:04 as a backup to protect regularity [1]. He passes through light stages 1 and 2, drops into deep stages 3 and 4 after about 20 minutes, rises again, and has a first short REM period after about 80 minutes [1]. The cycle then repeats roughly every 90 minutes, but the ratio inside each cycle changes: the first half of the night is dominated by deep non-REM sleep, the second half by stage 2 and more and more REM [1].

In the later series episode on mood he says it most directly: "you get most of your deep sleep in the first half and you get most of your REM sleep in the second half, and particularly in the last quarter of the night" [2]. He calls the end of the night "the REM sleep rich phase" and adds that "the later that you sleep into the morning, the more of that REM sleep that you will have" [2].

His proof from everyday life is the weekend. People who sleep in on Saturday and Sunday notice more vivid, more intense dreams that they remember better [2]. Walker says there is nothing magical about dream recall on weekends; they have simply slept into the REM-rich morning phase and increased their REM sleep [2].

Walker also answers the reverse case. Huberman asked whether someone forced to stay up until 3 a.m. would begin with deep sleep or jump to REM [1]. Walker's answer is that it is mostly the latter: you get some deep sleep because of built-up pressure, but you end up significantly short on it, because the brain does not simply start its program again from the top [1]. He adds that sleep laboratories use exactly this timing trick when they want to deprive people of one stage and not the other [1].

He is specific about what each loss costs. Losing deep sleep brings higher blood pressure, autonomic dysfunction and poorer blood-sugar regulation, since deep non-REM acts as a form of natural blood pressure medication [1]. Losing REM hits emotional stability, some aspects of learning, and hormones: he says growth hormone appears more REM-linked than many assume, and testosterone peaks just before and during REM in the second half of the night [1]. Asked which loss he would choose, he prefers neither, on the argument that every stage has survived evolution [1]. To people chasing one favoured stage, his line is "It's all the good sleep" [1].

Why the REM loss matters emotionally is the subject of the mood episode. Walker describes REM as overnight therapy that strips "the bitter emotional rind" from "the informational orange", so you keep the memory and lose the pain [2]. In his study, people who formed emotional memories and then slept showed a clear drop in amygdala reactivity when retested, people who stayed awake the same length of time did not, and the more REM sleep people had, the bigger the benefit [2]. He also recounts William Dement's classic Stanford experiments, in which people were woken every time they entered REM and over successive nights showed signs bordering on severe psychosis [2].

Gina Poe: go to bed late and you miss the early work

Poe, a professor of integrative biology and physiology at UCLA, lays out the same architecture and puts most of her weight on the front of the night [4]. A perfect night, she says, is four or five cycles of about 90 minutes, seven and a half to eight hours in all [4]. The first cycle runs longer, about 105 to 110 minutes, and later cycles get shorter and consist mostly of stage two and REM [4].

On the back half she is brief but clear: "There's more and more REM sleep the later the night we get" [4]. Growth hormone and melatonin are declining by the second and third cycles while other hormones pick up, and she calls this "a really different stage that you also don't want to shortchange yourself on" [4]. Studies, she says, point to these later cycles as the time when the most creativity can happen, when dreams put old and new material together and schemas are built [4]. Her image is of the brain opening folders and comparing documents, finding things that are only slightly linked and strengthening the link [4]. She adds that "in the second half of the night you have longer REM sleep periods", and she argues those can be considered the deepest sleep [4].

Her warning about late bedtimes is the mirror image of the early-alarm problem. The first deep slow-wave period brings a large bolus of growth hormone, in men and women equally, along with heavy protein synthesis that helps build memories [4]. Huberman asked why going to bed later would not simply shift that window, and Poe answered that every cell has a circadian clock, the clocks are synchronized, including with melatonin timing, and if you go to sleep late your cells have already moved on to the next phase [4]. The same holds for the slow-wave cleanup she compares to a bilge pump: asked whether a late bedtime delays it, she says no, "you miss the washout" [4]. Sleep that begins at 1 or 2 a.m. will still be dominated by stage two and REM, not by slow-wave sleep [4].

From this she and Huberman conclude that bedtimes should be consistent, not just wake times; in her words, "one of the best markers of good neurological health when we get older is consistent bed times" [4]. Her one caveat is that someone who habitually sleeps from 1 or 2 a.m. to 10 a.m. may be fine if everything, including melatonin release and the morning hormone rise, has shifted with them [4].

Poe also has something to say about how the morning ends. "It's called sleep inertia when we wake up out of the wrong state", she says, comparing a cycle to a washing machine you should not open while the clothes are soaking and soapy [4]. But the last cycles are mostly stage two and REM, and "if you wake up out of REM sleep there's usually no problem cognitively" [4].

Andrew Huberman: more REM means more morning

Huberman's solo material rests on the same structure. In his episode on dreams he explains that whatever the total length, sleep divides into roughly 90-minute cycles; early cycles are dominated by shallow and slow-wave sleep with little REM, and with each cycle the share of REM grows, so the hours toward morning are rich in REM [5]. He stresses that this holds even when sleep is broken by a trip to the restroom, and that the more total sleep you get, the more REM you get [5].

He uses himself as the example. He describes falling asleep around 10:30 or 11, waking around 3 or 4 a.m., then using a non-sleep deep rest (NSDR) protocol to fall back asleep until about 7 a.m., a stretch in which he gets a lot of REM [5]. On the nights he does not get back to sleep, he notices what he says lab studies confirm: REM deprivation makes people emotionally irritable, makes little things feel like big things, and leads to catastrophizing [5].

When a listener asked how to get more REM, his first answer was the morning. REM is more concentrated toward the end of the night, so he suggests adding 10 to 30 minutes of sleep in the morning, which he jokes is a case for hitting snooze and going back to sleep [3]. He adds three more ideas: a bigger surge of adrenaline early in the day through morning exercise or deliberate cold exposure, which he says increases REM that same night; a 10- to 30-minute NSDR protocol, which he hypothesizes mimics REM sleep and which he and Walker's lab plan to test; and trusting REM rebound, the brain's habit of taking extra REM the night after a shortfall if you sleep enough and avoid late caffeine and alcohol [3]. He says there is currently no clear pharmacology that specifically increases REM [3].

On what REM is for, Huberman agrees with Walker and Poe on emotion. He calls REM "its own form of trauma therapy" and says "REM sleep is powerful and has this amazing capacity to eliminate the meanings that don't matter" [3][5]. He gives the early night its due as well: studies that selectively deprive people of slow-wave sleep show that motor learning and the learning of specific details happen mostly in that stage, early in the night [5].

Where They Agree, and Where They Don't

All three describe the same slope: deep slow-wave sleep is front-loaded and REM is back-loaded [1][2][4][5]. All three treat the two stages as doing different jobs, so that losing one is not made good by keeping the other [1][4][5]. All three name alcohol as a suppressor of REM [1][2][3][4]. And both Walker and Poe say a late bedtime does not replay the night from the beginning [1][4].

Growth hormone. This is the sharpest of the differences. Poe ties the big pulse of growth hormone to the deep slow-wave sleep of the first cycle, and that is why she warns about late bedtimes [4]. Huberman likewise links growth hormone release to early-night slow-wave sleep [5]. Walker says growth hormone appears more REM-linked than many assume, and cites a study in which alcohol near the legal limit cut growth hormone release by well over 50% [1]. The summaries do not show the two views being put side by side, so a reader cannot tell from these sources how much of the hormone belongs to each end of the night [1][4].

Which sleep is "deep". Walker uses deep sleep for non-REM stages 3 and 4 [1]. Poe argues that late-night REM can be considered the deepest sleep; slow-wave sleep earned the label because people are hard to rouse from it and wake confused, but slow waves shrink in adults and older people, and as they fade REM becomes the deepest stage [4]. This is mostly a difference of terms, though it changes how one thinks about the morning hours [1][4].

How much sleep to aim for. Walker speaks of a night as seven to nine hours, notes that mortality curves rise below seven, and stresses how quickly short sleep does emotional harm; a Japanese group found the same amygdala over-reaction in people limited to less than six hours for five nights [1][2]. Huberman cites a study from Robert Stickgold's lab at Harvard in which regularity of sleep duration mattered more for exam performance than total amount, and says that for him a consistent six to six and a half hours beats swinging between five and nine [5]. Huberman's point is about consistency, and he still says more sleep is always better than less, but his personal number is below Walker's range and would leave less room for the REM-rich end of the night [1][5].

Sleeping in. Walker's best tool for more REM is to sleep 15 to 25 minutes later [2]. Yet after a bad night he follows colleague Michael Perlis's advice to "do nothing", including not sleeping in [1]. The two can be reconciled by scale and regularity, a small standing shift versus an irregular lie-in, but that reconciliation is ours; the summaries do not record Walker making it [1][2]. Huberman's snooze-button joke sits a little awkwardly beside Poe's advice not to wake in mid-cycle, though her own remark that waking from REM is usually fine suggests the conflict is small late in the night [3][4].

Can lost REM be recovered? Huberman says REM rebound will give you extra REM the next night if you sleep enough [3]. Walker, speaking about people who stop alcohol or THC, says the brain tries to reclaim lost REM but never gets it all back [1].

Where nightmares happen. Walker reads the repetitive nightmares of PTSD as REM's overnight therapy failing, linked to elevated noradrenaline [2]. Poe's hypothesis is similar: when norepinephrine is not turned down during REM, sleep re-sews the emotion into the memory each night [4]. Huberman says the research suggests intense fearful dreams probably happen in slow-wave sleep, not REM, because norepinephrine is absent in REM [5]. The sources do not resolve this.

Why It Works

The mix inside each cycle changes. The night is built of cycles of about 90 minutes, and what changes is the proportion of each stage within them: early cycles are mostly deep non-REM, later cycles mostly stage two and REM [1][4][5]. Cutting the end of the night therefore removes the cycles that hold the most REM, and Walker's point about the last quarter means the loss is out of proportion to the minutes given up [2].

The clock sets the menu. Poe's explanation for why a late bedtime does not simply shift everything is circadian: cells throughout the body keep synchronized clocks and expect certain events at certain times relative to melatonin [4]. Walker's version is that the brain does not start the program from the top, though built-up sleep pressure still buys some deep sleep [1]. He describes the morning hours as the time when the brain most prefers REM [2].

REM has a chemistry found nowhere else in the day. Walker says REM is the only time in the 24 hours when noradrenaline is completely shut off in the brain; serotonin drops too, acetylcholine rises to about 30% above waking levels in some regions, and memory and emotion centres such as the hippocampus and amygdala are highly active [2]. Poe locates the switch in the locus coeruleus, which fires at about 2 Hz awake, about 1 Hz in sleep, and stops completely only in REM [4]. She believes that silence is what lets the brain erase synapses that no longer serve it, like wiping a thumb drive after copying it to the cortex [4]. Huberman makes the same point about the body: "It's probably one of the few times in our life that epinephrine is essentially at zero activity within our system" [5]. The result, on all three accounts, is that experiences can be replayed without the chemistry of fear [2][4][5].

The front of the night has its own work. Poe describes the first 20 minutes or so of sleep as rebuilding ATP, followed by slow waves in which neurons firing and falling silent together pump debris out of the brain, mainly in the first third of the night [4]. She adds that the first four hours appear especially important for processing new memories [4]. Walker's team found that deep non-REM sleep, not REM, predicted overnight reductions in anxiety [2].

The evidence on stakes, with its caveats. Walker points to a Harvard analysis, replicated in two large populations by his account, in which REM was the sleep stage most predictive of lifespan, with roughly 13% higher mortality risk for every 5% reduction in REM; he says himself that he may have the numbers backwards, and the finding is an association [1]. Huberman says even 20 more minutes of REM sleep would do more for testosterone than the supplement shilajit, which is his judgment, not a cited trial [3]. Sleep deprivation, he says, "is deprivation of self-induced therapy every time we go to sleep" [5].

Where It Applies, and Where It Doesn't

The pattern describes a person on a regular schedule. Poe allows that a consistently late sleeper may have a fully shifted clock [4]. Walker recommends learning your chronotype with the free Morningness-Eveningness Questionnaire and matching your schedule to it as far as life allows [2]. Huberman reports that the same amount of sleep feels very different to him depending on timing: 8:30 or 9 p.m. to about 4:30 or 5 a.m. works like an antidepressant, while midnight to 8 or 9 a.m. leaves him with a low-level depression [2].

The advice to add morning minutes has a ceiling. Walker warns against "social jet lag", going to bed and waking two or three hours later on weekends, which he compares to flying between Los Angeles and New York every weekend [2]. Huberman's consistency study points the same way [5].

Not every short night is a REM problem. Huberman tells athletes and performers that if skills are already trained and they wake after only three or four hours before an event, they have already received much of the motor learning and recovery benefit, which comes early [5]. That is reassurance about one night, and he does not offer it as a way to live [5].

Substances can remove REM whatever the clock says. Walker's top "don't" is alcohol, which he says abolishes REM sleep, with THC close behind; people who stop cannabis often report an explosion of dreams as REM rebounds [2]. Alcohol also fragments sleep in the first four hours, when deep sleep dominates, so it damages both ends [2]. Poe says alcohol suppresses REM and the stage-two spindles until it is metabolized and cleared [4]. Huberman adds late caffeine as something that undermines REM rebound [3].

REM is not simply a case of more being better. Walker notes that people with depression tend to enter their first REM period much earlier in the night and that many antidepressants delay or reduce REM, and he says the picture is not settled [2]. Anyone taking such medication, or dealing with depression, trauma or persistent nightmares, should talk to their doctor before trying to adjust sleep around it [2].

Some people cannot choose their wake time. Walker says red-eye flights "should be abolished" for how they wreck him for days, and both he and Huberman note the harm of long shifts for physicians [1]. Huberman argues for later school start times on the grounds that the rewiring behind learning happens during deep and REM sleep [3].

Finally, the limits of the sources. No one here states how much REM falls in the last two hours, or how much an alarm set one hour early removes; the strongest wording is Walker's "most" in the second half and "particularly" the last quarter [2]. Poe speaks at more length about what a late bedtime costs than about what an early alarm costs [4]. Huberman's NSDR idea is, in his own framing, a hypothesis awaiting a test [3]. And home trackers will not settle the question for an individual: Poe says the best are about 70% effective at staging sleep [4].

What To Do With It

1. If you want more REM, add the time at the end of the night: wake 15 to 25 minutes later, as Walker suggests, or 10 to 30 minutes, as Huberman does, instead of going to bed that much earlier [2][3].
2. Keep the change small and steady. Do not swing two or three hours later on weekends, which Walker calls social jet lag [2].
3. Protect the other end too by holding a consistent bedtime, because on Poe's account a much later bedtime misses the first cycle's growth hormone pulse and brain cleanup and does not postpone them [4].
4. Before you set an earlier alarm for a workout or work, know what you are trading: the end of the night is where REM is richest, and REM is tied to emotional steadiness, learning and testosterone [1][2].
5. Skip evening alcohol and THC if REM matters to you; both suppress it, and alcohol also breaks up early deep sleep [1][2][4].
6. Avoid late caffeine, especially the day after a short night, so REM rebound can do its work [3].
7. Try a stronger morning signal, such as exercise or deliberate cold exposure early in the day, which Huberman says increases REM that night [3].
8. If you wake at 3 or 4 a.m., try to get back to sleep, for example with an NSDR script as Huberman does, since the hours that follow hold a lot of REM [5].
9. Do not read one bad night as a crisis: Huberman's advice is to return to a consistent schedule, and Walker's reminder is that "Sleep is not something that we do", so straining for it pushes it away [2][5].
10. If you regularly cannot sleep through to your natural wake time, or you feel irritable and unrested despite enough time in bed, raise it with a doctor [2][5].

Sources

1. Dr. Matt Walker: The Science & Practice of Perfecting Your Sleep · Huberman Lab
2. Dr. Matt Walker: Improve Sleep to Boost Mood & Emotional Regulation · Huberman Lab
3. AMA #18: Cold Therapy Advice, Skin Health Tips, Motivation, Learning Strategies & More · Huberman Lab
4. Dr. Gina Poe on How Each Stage of Sleep Shapes Learning, Memory, Creativity, and Emotional Recovery (Huberman Lab Essentials) · Huberman Lab
5. Understand and Use Dreams to Learn and Forget · Huberman Lab