Sleep notes

Is a Sleep Cycle Always 90 Minutes? What Research Actually Shows

By ToSleep AI · 2026-09-19

Research reviewed September 14, 2026; product implementation rechecked September 19, 2026. This article concerns the biological sleep cycle, not a review of the similarly named app. Published by the product’s developer; not independently clinically reviewed. Cited researchers do not endorse ToSleep.

The short answer: A sleep cycle is not a fixed timer. Its duration and composition can vary between people and within a person’s sleep. A reference duration can help with planning, but cannot identify an exact best moment to wake tonight. Cycle-duration research

What is a sleep cycle?

A sleep cycle describes the progression through non-rapid eye movement (NREM) and rapid eye movement (REM) sleep. NREM includes lighter stages and deep sleep. These stages are identified from physiological recordings, rather than by counting time from when someone gets into bed. Deep sleep is generally more prominent earlier in the night, with REM becoming more prominent later. NHLBI explanation

A row of identical blocks is a convenient way to draw a night. It should not be mistaken for a precise picture of how an individual body will behave.

What does research say about the 90-minute rule?

A typical duration is not a personal deadline

Christian Cajochen and colleagues, including researchers at the Centre for Chronobiology at the University Psychiatric Clinics in Basel, examined laboratory sleep recordings. Their retrospective study reported a median cycle duration of 96 minutes, alongside substantial differences within and between participants. The analysis covered 369 participants and 6,064 scored cycles. These are study observations, not a prediction for every sleeper. Abstract and results, Sleep Health

The useful conclusion is not to replace every 90-minute calculator with a 96-minute calculator. A population median still cannot tell you the length of your next cycle. The study was published online in 2023 and in a 2024 journal volume; the statistics here are from that historical study, checked September 14, 2026.

Personal patterns can coexist with nightly variation

Akifumi Kishi and Hans P. A. Van Dongen, affiliated with the University of Tokyo and Washington State University, studied the dynamic structure of sleep under controlled conditions. They found evidence of individual differences in aspects of sleep-stage transitions and cycle structure, including observations across baseline and recovery sleep. Original paper, 2023

Personal history is therefore worth examining, but a recurring pattern is not an exact schedule. This research did not validate ToSleep’s algorithm, nor show that an app can learn a perfect wake-up time after a few nights.

Where does a fixed-cycle calculator become misleading?

A promise to wake someone at precisely the end of a cycle requires more than arithmetic. The tool would need a reliable sleep-onset time, an accurate estimate of each upcoming cycle, and evidence that its chosen boundary improves the waking experience.

The conceptual problem is treating an approximate pattern as an exact forecast. A planning aid can remain useful while acknowledging uncertainty. Our editorial recommendation is not to cut a night’s available sleep simply to reach a neat-looking whole number of cycles.

Does waking from light sleep guarantee feeling refreshed?

Sleep inertia is the temporary impairment that can follow awakening. Frank Scheer and colleagues at Brigham and Women’s Hospital and Harvard Medical School investigated its relationship to the body’s internal circadian timing in a controlled experiment. They found that biological time influenced post-awakening cognitive performance. This was not a trial of a consumer smart alarm. Original research, 2008

That distinction matters for product claims. Choosing a reminder time and improving how someone feels afterwards are separate questions. A quieter reminder is a design feature; reliably reducing morning grogginess is an outcome that needs its own evidence. We should not promise the latter simply because we built the former.

What can a wrist-worn device estimate?

Laboratory sleep staging uses signals such as brain activity, eye movements and muscle activity. Wrist motion and heart-rate measurements are different inputs. Research by Michele Guagnano and colleagues at Politecnico di Torino and Sleep Advice Technologies explored real-time scoring from heart rate, heartbeat intervals and accelerometer information, with comparisons against polysomnography. Original paper, 2025

That study used a limited sample, and its inputs and algorithm are not ToSleep’s implementation. Sleep-versus-wake classification is also not the same task as distinguishing individual sleep stages. The paper’s performance numbers cannot be presented as ToSleep’s accuracy, and a sleep-staging test is not a test of whether people wake feeling better.

How does ToSleep approach variable sleep cycles?

The following is a description of the development implementation rechecked on September 19, 2026, not a clinical validation or a promise about every installed build. Check the current access information and your build instructions.

Planning and waking are separate decisions

ToSleep’s planning implementation retains 90 minutes per cycle as a fallback reference when there is insufficient usable personal history. With suitable records, it can use estimated REM intervals and historical patterns at different cycle positions. Some cycle-count displays also fall back to duration-based arithmetic when the required REM markers are unavailable.

Our product principle is to distinguish a measured signal, an algorithmic estimate and a default assumption. A fallback helps a planner get started; it should not be described as a measured personal cycle.

The wake window uses the current night’s signals

In the reviewed implementation, wrist motion and available heart-rate data feed a rule-based sleep-stage estimate. Once the allowed wake window begins, local decision logic considers recent stage labels and their freshness. It does not wait for a multiple of the planning cycle length, and a cloud conversation model does not make the real-time wake decision.

Qualifying consecutive light-sleep or awake estimates can trigger an earlier reminder. There is also a REM branch late in the window. Deep-sleep estimates, insufficient recent labels or stale labels cause the policy to wait; the latest wake time takes priority when the deadline is reached.

These are estimates, not direct brain measurements. The implementation does not use a separate stage-confidence threshold as a hard wake-decision gate. Reminder delivery depends on the device, permissions, power and monitoring session remaining operational. Keep an independent backup alarm when waking on time is important.

Gentler describes the reminder, not a guaranteed outcome

The reviewed wrist-reminder implementation begins with haptic cues only. If the user does not stop it, soft app audio begins after two minutes and ramps up gradually; the reminder has a maximum duration. These are product-timing choices, not proof that every person will wake or that a bed partner will remain undisturbed.

Our aim is to use the available signals within a time range you permit, then remind you more gently. “The perfect moment every morning” would go beyond the evidence. The product walkthrough explains the broader workflow.

What should you look for in a sleep report?

We suggest asking whether the recording is complete, which outputs are estimates, and whether any values came from defaults. It also helps to distinguish a stage-based reminder from a deadline reminder, and compare the report with your own waking experience.

A precise-looking chart is not necessarily a precise measurement. Transparency about how a number was produced is more useful than displaying extra detail without explaining it.

Frequently asked questions

Is the 90-minute rule wrong?

It can be a rough planning reference. Treating it as a constant that every person follows throughout every night is the misleading step. Replacing it with another fixed number does not resolve that uncertainty.

Will ToSleep always wake me in light sleep?

No such guarantee is made. The reviewed logic uses estimated stages and includes other triggers, including a deadline. Device operation and data quality also affect what is possible.

Has the cited research validated ToSleep?

No. The papers explain sleep biology or evaluate other methods. We have not established ToSleep-specific clinical validation in this article. Agreement with another wearable’s labels would not, by itself, substitute for validation against laboratory recordings.

Where can I learn about availability and my data?

See current access information, the FAQ and privacy information. ToSleep is intended for everyday wellbeing, not diagnosis or treatment.

Research claims are linked to their sources. Interpretations and product-design judgments are identified by context rather than presented as clinical findings. Product implementation statements are the developer’s account, not findings of the cited institutions. For questions or corrections, contact us.

Sources

Wellness information, not medical advice.