Transcranial electrical stimulation has an old, stubborn problem. Current applied at the scalp spreads. It follows the path of least resistance, which in practice means it does most of its work in skin, muscle, and superficial cortex and arrives at deep structures attenuated and smeared. You can raise the current, but then the scalp tingles, the overlying cortex gets stimulated whether you want that or not, and the targeting you hoped for dissolves. For decades the standard answer was that if you want to reach the subthalamic nucleus or the hippocampus with electricity, you implant an electrode.
Temporal interference stimulation (TIS, sometimes tTIS) proposes a way around that limit. Two or more pairs of scalp electrodes deliver sinusoidal currents at slightly different high frequencies—say 2,000 Hz on one pair and 2,005 Hz on another. Each frequency is far too fast for neurons to follow. Where the two fields overlap, however, they beat against each other, and the amplitude of the combined field rises and falls at the difference frequency, 5 Hz in this example. If that envelope matters to neurons, and if it is largest at the place where the two fields cross, then in principle you can drive a deep target at a physiological rhythm while the individual fields stay individually inert everywhere else.
The idea is elegant, and elegance is a reason for caution. Whether the brain actually reads the envelope—rather than simply responding to the summed fields, or blocking them—is contested, and the answer determines whether TIS is a new class of deep brain stimulation or a cleverly marketed variant of ordinary transcranial current.
The mechanism, stated carefully
The foundation is a 2017 paper in Cell by Nir Grossman and colleagues, working with the Boyden lab and collaborators. They reasoned that if two high-frequency fields are applied through separate electrode pairs, the amplitude of their superposition is modulated at the difference frequency, and that modulation is greatest where the two fields have comparable magnitude and align. Crucially, because each field is individually too fast to entrain spiking, the only place the envelope is physiologically meaningful is the overlap region. They demonstrated the effect in the mouse hippocampus, showing they could drive activity there without recruiting the overlying cortex, and they showed the envelope could be steered by changing the ratio of currents delivered to each electrode pair.
The mouse result does not establish that the same effect can be obtained safely in a human brain. Head size, tissue conductivity, electrode placement, and the strength of the intracranial field all affect that translation. Later human studies must establish their own physiological effect and safety at the currents actually used. The gap between driving mouse neurons and treating a human patient cannot be closed by citing the animal demonstration alone.
The mechanism rests on a particular assumption: that a neuron responds to the low-frequency envelope rather than to the high-frequency carriers. For that to be true, the membrane must, in effect, demodulate—either by passive filtering that strips the carrier and leaves the envelope, or by some active nonlinearity in ion channels that rectifies the fast signal. Which of these is happening, if either, is where the field’s disagreement lives.
The human evidence so far
The most important human result came in 2023. Ines Violante and colleagues, publishing in Nature Neuroscience, combined cadaver measurements, computational modeling, and a study of twenty healthy participants. Using carrier frequencies of 2,000 and 2,005 Hz, they measured the envelope field directly in a human cadaver head and found that modulation amplitude was about 75 percent larger in the hippocampus than in the overlying cortex—the first direct physical demonstration of the depth selectivity the technique predicts. They estimated roughly 0.1 V/m at about 44 mm depth per milliamp of applied current, showed that changing the current ratio from 1:1 to 1:3 shifted the envelope, and reported that hippocampal stimulation modulated activity and improved accuracy on an episodic-memory task.
Two caveats belong next to that result. The first is infrastructure: the same group and several of its members are inventors or co-founders of TI Solutions AG, the company commercializing the method, and the paper discloses those competing interests. That does not falsify anything, but it means the results should be read with the same skepticism applied to any lab with a commercial stake. The second is that a cadaver head is not a living brain. It provides the geometry and conductivity for a field model but not the physiology that determines whether the envelope does anything.
Beyond that, human work has largely been subthreshold modulation rather than driving. Andreas Wessel and colleagues used theta-burst TIS to modulate striatal activity in a Nature Neuroscience paper in 2023, and Vassiliadis and colleagues demonstrated effects on reinforcement learning in a 2024 Nature Human Behaviour study. These are real results, and they establish that TIS can be applied safely in awake humans and can produce measurable behavioral change. But they are studies of how stimulation biases an existing circuit, not demonstrations that TIS can command a deep structure the way an implanted electrode can.
A 2026 perspective in Nature Biomedical Engineering by Vassiliadis, Beanato, Wessel, and Hummel makes the distinction explicitly. It frames temporal interference as an emerging method for focal non-invasive deep brain stimulation, and it characterizes the human regime as subthreshold and tACS-like—closer to nudging oscillations than to firing neurons. It also states that the mechanisms of action remain uncertain and that validation work is still ahead. That is the field’s own current self-assessment, and it is more sober than much of the coverage.
The demodulation problem
If the envelope is what matters, neurons should respond differently to two fields whose difference frequency falls in a physiological range than to two fields that do not. Several independent lines of work suggest that the simplest version of this picture is wrong.
In 2023, Budde, Williams, and Irazoqui reported that temporal interference in peripheral nerves is not driven by envelope extraction. Studying the response directly, they argued that the effect could be explained by neurons acting as linear membrane integrators, without any ability to extract the low-frequency envelope. If that is right, then TIS’s selectivity advantage is weaker than advertised in the peripheral case, and the burden shifts to showing that central neurons behave differently.
The same year, Iszak and colleagues published a pilot study in Biomedicines titled, bluntly, “Why Temporal Interference Stimulation May Fail in the Human Brain.” They found that peripheral nerve stimulation responded at the modulated frequency, but they could find no corresponding central effect—no phosphenes, no change in alpha power—and concluded that central activation thresholds may be higher, or that the carrier components may exert their own inhibitory influence that masks the envelope.
The most pointed challenge to the deep-control narrative comes from direct single-neuron recording. In 2024, Vieira, Krause, and Pack recorded 234 neurons in the primate brain during temporal interference and found that it altered spike timing rather than firing rate. That is a meaningful distinction: controlling when neurons fire is not the same as making them fire, and it is a weaker form of control than DBS provides. They also reported that the method was roughly 80 percent weaker than other non-invasive stimulation techniques and argued that the combination of high carrier frequencies, multiple electrodes, and amplitude-modulated waveforms limits its effectiveness.
A more fundamental objection came earlier, from Mirzakhalili, Barra, Capogrosso, and Lempka in Cell Systems in 2020. Their biophysical modeling argued that passive membrane filtering cannot produce the TIS effect at all, because the membrane time constant is far too slow to pass the carrier while retaining the envelope. Any real effect, they argued, would require ion-channel-mediated rectification—and the same rectification that enables stimulation at the target would produce high-frequency conduction block in off-target axons. In other words, the mechanism that makes TIS work might also make it less selective than its field geometry suggests, because conduction block depends on which fibers the carrier fields cross, not on where the envelope peaks.
On the other side of the argument, Esmaeilpour and colleagues, publishing in Brain Stimulation in 2021 with an experimentally constrained model, found that temporal interference could target deep regions by modulating neural oscillations rather than by directly driving spikes. Their results also show the price. To modulate gamma rhythms, a 100 Hz carrier regime would require roughly 5 V/m, about 13 mA of scalp current; a 1 kHz carrier would require about 60 V/m, around 160 mA; a 2 kHz carrier would require about 80 V/m, around 220 mA. Those are large currents for a scalp electrode montage, and they explain why human TIS studies emphasize modulation of ongoing activity rather than supra-threshold driving.
Taken together, the mechanistic literature does not support the clean story that neurons demodulate the envelope at the target. What it supports is more modest: the field superposition is real, the envelope amplitude does peak in the overlap region, some neurons in some preparations respond to modulated waveforms, and the response may involve nonlinear membrane properties that carry their own side effects. That is a genuine phenomenon and an unresolved mechanism, which is a different thing from an established principle.
Selectivity, off-target effects, and what “deep” buys you
Even granting the envelope, the targeting is less clean than the diagram suggests. Head anatomy is heterogeneous, with layers of scalp, skull, cerebrospinal fluid, gray matter, and white matter of very different conductivities. Boundaries between tissues distort fields, and the strongest membrane polarization often happens where a fiber crosses a conductivity boundary or changes direction—not where the envelope is largest. This means the site of neural activation can drift away from the site of maximal modulation, particularly for axons running perpendicular to a tissue interface.
The Mirzakhalili analysis sharpens this: if stimulation depends on ion-channel rectification, then any axon crossed by the carriers is a candidate for conduction block, whether or not it sits near the envelope peak. The scalp currents that make deep modulation possible are large enough that peripheral nerve stimulation, scalp discomfort, and off-target effects in overlying cortex become real design constraints. Human studies report that TIS produces less scalp sensation than conventional transcranial alternating current stimulation at comparable currents, which is genuinely useful—fewer painful twitches means higher tolerable currents and less chance subjects can tell real stimulation from sham. But reduced sensation is a property of the stimulus waveform, not evidence of deep selectivity.
A 2025 paper by Botzanowski and colleagues on multipolar temporal interference points at where the engineering is heading: more electrode channels, more complex modulation patterns, better control of the envelope shape. This is promising as engineering, and it also illustrates how much of TIS’s targeting currently depends on models rather than measurement. The envelope is computed from a head model; the resulting behavior is measured. If the model is wrong about conductivities or about the demyelination of specific tracts, the predicted focus and the actual one diverge, and the experiment cannot tell.
Safety and dosing
Because TIS uses currents in the milliamp range delivered through scalp electrodes, its safety profile is related to that of transcranial direct and alternating current stimulation, with the added question of what sustained kilohertz carriers do to tissue. A two-part effort by Cassarà, Newton, Zhuang, Regel, Achermann, Pascual-Leone, Kuster, and Neufeld, published in Bioelectromagnetics in 2025, addressed this directly. Part I reviewed adverse events across tDCS, tACS, deep brain stimulation, and TIS. Part II used multiphysics simulations in an anatomically detailed head model to compare TIS exposures against levels known to be safe for tACS and DBS, and proposed frequency-dependent thresholds below which applied voltages and currents are unlikely to pose risk from heating or unwanted stimulation.
The authors are explicit that their thresholds are a first approximation that needs verification in human studies, and several of them are affiliated with TI Solutions, so the guidance shares an interest with the technology. The important structural point is that TIS safety cannot simply borrow tDCS limits. Because kilohertz currents are less effective at directly driving neurons—that is the premise of the whole method—higher amplitudes can be tolerated before discomfort or unwanted low-frequency effects appear, but the deep structures still absorb energy and still heat slightly. Where exactly the safe upper bound sits is a question that requires human safety data, not just simulation.
What would make this real
The technique’s promise and its uncertainty are tightly coupled, because the same property that makes TIS promising—selective action at the envelope peak—is the property whose mechanism is least settled. Three kinds of evidence would resolve it.
First, independent measurement of the envelope’s neural effect at depth in humans, using something other than the model that designed the stimulation. Invasive recordings in patients with existing electrodes, simultaneous fMRI in the target, or reliable physiological readouts tied specifically to the deep structure would let the field separate “the envelope is large there” from “the brain responds there.”
Second, falsifiable tests of demodulation. If neurons are envelope extractors, then changing the carrier frequencies while holding the difference frequency constant should preserve the effect, and carrier pairs that produce no low-frequency difference should produce none of it. Some of these experiments have been run, and the mixed results are precisely why the mechanism is contested. A clean, preregistered version across preparations would settle more than another modeling paper.
Third, comparison against the alternatives. TIS should be judged not against no stimulation but against the best available non-invasive method for the same target, and where possible against implanted DBS. If the behavioral effect is a modest bias in a laboratory task, that is worth knowing, but it is not the same as the clinical control an electrode provides.
The most defensible current summary is this: temporal interference is a real physical method for shaping the amplitude envelope of electrical fields deep in a human head, it has been applied safely in dozens of participants, and it produces measurable effects on brain activity and behavior in some tasks. Its central mechanistic claim—that neurons respond to the envelope because they demodulate it—is plausible, unproven, and disputed by serious work. And its human effects are currently in the subthreshold range, biasing circuits rather than commanding them.
That leaves TIS in an interesting position. It is more than speculation, because the physics and the human recordings are real. It is less than a therapy, because the mechanism and dose have not been nailed down. The temptation will be to describe it as “non-invasive deep brain stimulation” and let the phrase do the work of the evidence. The more accurate description is a promising, mechanism-contested method for reaching deep circuits with electricity, whose ultimate value depends on experiments the field has not yet run.
Sources and further reading
- Grossman N, Bono D, Dedic N, et al. Noninvasive deep brain stimulation via temporally interfering electric fields. Cell 169(6):1029–1041 (2017).
- Violante IR, Alania K, Cassarà AM, et al. Non-invasive temporal interference electrical stimulation of the human hippocampus. Nature Neuroscience 26(11):1994–2004 (2023).
- Vieira PG, Krause MR, Pack CC. Temporal interference stimulation disrupts spike timing in the primate brain. Nature Communications 15:4558 (2024).
- Mirzakhalili E, Barra B, Capogrosso M, Lempka SF. Biophysics of temporal interference stimulation. Cell Systems 11(6):557–572.e5 (2020).
- Esmaeilpour Z, Kronberg G, Reato D, Parra LC, Bikson M. Temporal interference stimulation targets deep brain regions by modulating neural oscillations. Brain Stimulation 14(1):55–65 (2021).
- Budde RB, Williams MT, Irazoqui PP. Temporal interference current stimulation in peripheral nerves is not driven by envelope extraction. Journal of Neural Engineering 20(2) (2023).
- Iszak K, Gronemann J, Meyer S, et al. Why temporal interference stimulation may fail in the human brain: a pilot research study. Biomedicines 11(7):1813 (2023).
- Cassarà AM, Newton TH, Zhuang K, et al. Recommendations for the safe application of temporal interference stimulation in the human brain, Part II: biophysics, dosimetry, and safety recommendations. Bioelectromagnetics 46(1):e22536 (2025).
- Vassiliadis P, Beanato E, Wessel MJ, Hummel FC. Temporal interference stimulation for deep brain neuromodulation in humans. Nature Biomedical Engineering 10:1279–1294 (2026).
Loading comments…