Neural dust and ultrasonic implants: revision 1 to 2

Revision 1

Revision 2

categories: ["cybernetics", "nanomedicine"]

tags: ["implants", "ultrasound", "wireless", "miniaturization", "bci", "neural recording"]

summary: "A proposed class of millimetre-scale wireless implants powered and read out by ultrasound, intended to replace tethered electrode arrays with distributed untethered sensors."

summary: "A proposed class of millimetre-scale wireless implants powered and read out by ultrasound, intended to replace tethered electrode arrays with distributed untethered sensors."

updated: "2026-07-27"

updated: "2026-10-29"

humanEvidence: "No human study of ultrasonic neural dust has been published; the in vivo work is in anaesthetized rats and covers peripheral nerve and muscle, not cortex, while related magnetoelectric stimulators have reached first-in-human study."

humanEvidence: "No human study of ultrasonic neural dust has been published; the in vivo work is in anaesthetized rats and covers peripheral nerve and muscle, not cortex, while related magnetoelectric stimulators have reached first-in-human study."

access: "Nothing to obtain: these are laboratory devices with no approved product, no commercial supplier, and no trial of an ultrasonic mote in a person."

access: "Nothing to obtain: these are laboratory devices with no approved product, no commercial supplier, and no trial of an ultrasonic mote in a person."

reversibility: "difficult"

**Neural dust and ultrasonic implants** are miniature, battery-free devices that draw their power

from an external ultrasound beam and report their measurements by modulating the sound they reflect

back. The concept was proposed in 2013 as a way past the central failure mode of implanted

back. The concept was proposed in 2013 to get around the central failure mode of implanted

recording: not the electrode itself but the wires, connectors, and skull penetrations that tether it

recording, which lies in the wires, connectors and skull penetrations that tether the electrode to

to the outside world. Despite the name, the demonstrated devices are millimetre-scale, not nanoscale,

the outside world rather than in the electrode itself. Despite the name, the demonstrated devices

and the published in vivo work is confined to peripheral nerve and muscle in rodents.

are millimetre-scale, not nanoscale, and the published in vivo work is confined to peripheral nerve

and muscle in rodents.

## The problem it addresses

Chronic neural recording fails at its interfaces. Penetrating arrays such as the

Chronic neural recording fails at its interfaces. Penetrating arrays such as the [[utah-array]] lose

[[utah-array]] lose channels over months to years as tissue encapsulates the shanks, but the more

channels over months to years as tissue encapsulates the shanks, but the more mundane failures are

mundane failures are mechanical: a rigid array tethered to a skull-mounted connector transmits every

mechanical. A rigid array tethered to a skull-mounted connector transmits every micromotion of the

micromotion of the brain into the tissue, insulation on the wire bundle delaminates, and a

brain into the tissue, insulation on the wire bundle delaminates, and a percutaneous pedestal is a

percutaneous pedestal is a permanent infection route. Scaling up channel count makes all of this

permanent infection route. Raising the channel count makes all of this worse, because more channels

worse, because more channels mean more wires through the same skull opening.

mean more wires through the same skull opening.

The neural dust proposal inverts the architecture. Instead of one large array with many wires,

The neural dust proposal inverts the architecture. In place of one large array with many wires, it

distribute many independent, sealed, wireless motes, each recording locally and each small enough

distributes many independent, sealed, wireless motes, each recording locally and each small enough

that the tissue response around it is minimal. There is nothing to tether and nothing to route. The idea belongs to a broader

that the tissue response around it is minimal, with no tether and no wiring to route. The idea

push in [[neuroprosthetics]] toward fully implanted wireless systems, and it is the most aggressive

belongs to a broader push in [[neuroprosthetics]] toward fully implanted wireless systems, and it is

version of it.

the most aggressive version of that push.

## How ultrasonic backscatter works

Each mote contains a piezoelectric crystal, a pair of recording electrodes, and — in the simplest

Each mote contains a piezoelectric crystal, a pair of recording electrodes and, in the simplest

designs — a single transistor. An external transducer sends ultrasound pulses; the crystal converts

designs, a single transistor. An external transducer sends ultrasound pulses, and the crystal

the incident acoustic energy into a voltage, which powers the device. The local extracellular

converts the incident acoustic energy into a voltage that powers the device. The local extracellular

potential at the electrodes modulates the transistor's conductance, which changes the electrical

potential at the electrodes modulates the transistor's conductance. That changes the electrical load

load on the crystal, which changes how much of the incident ultrasound is reflected. The external

on the crystal, which in turn changes how much of the incident ultrasound is reflected. The external

transceiver listens for the echo and reads the neural signal off its amplitude. Time-of-flight

transceiver listens for the echo and reads the neural signal off its amplitude. Time of flight

distinguishes echoes from motes at different depths, so many devices can be interrogated with one

separates echoes from motes at different depths, so one beam can interrogate many devices.

beam.

The scheme's elegance is that the mote needs no oscillator, no radio, no battery, and almost no

The scheme is attractive because the mote needs no oscillator, radio or battery, and almost no

digital logic. Its cost is bandwidth: backscatter is a low-rate uplink, and the signal must compete

digital logic. The cost is bandwidth: backscatter is a low-rate uplink, and the signal must compete

with reflections from tissue interfaces.

## Why not radio

The choice of ultrasound over electromagnetic power transfer follows from physics rather than

Ultrasound is chosen over electromagnetic power transfer for physical reasons. At the frequencies

preference. At the frequencies where an antenna a millimetre across would radiate efficiently, the

where an antenna a millimetre across would radiate efficiently, the wavelength in tissue is short

wavelength in tissue is short and absorption is high, so the power that reaches an implanted device

and absorption is high, so little power reaches an implanted device and most of the transmitted

is small and most of the transmitted energy heats the intervening tissue. Ultrasound at a few

energy heats the intervening tissue. Ultrasound at a few megahertz has a wavelength on the order of

megahertz has a wavelength on the order of a millimetre in soft tissue and attenuates far less per

a millimetre in soft tissue and attenuates far less per centimetre. A millimetre-scale piezoelectric

centimetre, so a millimetre-scale piezoelectric receiver is well matched to the wavelength and the

receiver is therefore well matched to the wavelength, and safe acoustic intensity limits still

safe acoustic intensity limits permit useful power delivery at depth.

permit useful power delivery at depth.

Bone breaks this argument. The skull attenuates, reflects, and refracts ultrasound severely and

Bone breaks this argument. The skull severely attenuates, reflects and refracts ultrasound, and it

heats under it, which is why proposals for cortical neural dust place a sub-cranial transceiver — a

heats under it. Proposals for cortical neural dust therefore place a sub-cranial transceiver, a thin

thin device implanted beneath the skull — that relays between the motes and an external unit. That

device implanted beneath the skull, to relay between the motes and an external unit. That relay is

relay is itself a conventional implant requiring surgery, which removes much of the claimed

itself a conventional implant requiring surgery, which removes much of the claimed simplicity for

simplicity for brain applications and explains why every published in vivo result to date sits

brain applications and explains why every published in vivo result to date sits outside the skull.

outside the skull.

## Development history

[

  { "year": "2013", "title": "Concept published", "text": "A Berkeley group proposes ultrasonically powered sub-millimetre motes as a route to chronic brain-machine interfaces, arguing that electromagnetic power transfer does not scale to that size in tissue." },

  { "year": "2013", "title": "Concept published", "text": "A Berkeley group proposes ultrasonically powered sub-millimetre motes as a route to chronic brain-machine interfaces, arguing that electromagnetic power transfer does not scale to that size in tissue." },

  { "year": "2016", "title": "First in vivo demonstration", "text": "Roughly millimetre-scale motes implanted on the sciatic nerve and gastrocnemius muscle of anaesthetized rats record nerve and muscle activity and report it by ultrasonic backscatter." },

  { "year": "2016", "title": "First in vivo demonstration", "text": "Roughly millimetre-scale motes implanted on the sciatic nerve and gastrocnemius muscle of anaesthetized rats record nerve and muscle activity and report it by ultrasonic backscatter." },

  { "year": "2018–2020", "title": "Stimulation added", "text": "A millimetre-scale ultrasonically powered stimulator with bidirectional communication delivers current to peripheral nerve, showing the link can carry commands as well as data." },

  { "year": "2018-2020", "title": "Stimulation added", "text": "A millimetre-scale ultrasonically powered stimulator with bidirectional communication delivers current to peripheral nerve, showing the link can carry commands as well as data." },

  { "year": "2021", "title": "Distributed microimplant networks", "text": "A separate approach demonstrates dozens of sub-millimetre radio-frequency-powered chiplets recording from rodent cortex as a coordinated network, without ultrasound." },

  { "year": "2021", "title": "Distributed microimplant networks", "text": "A separate approach demonstrates dozens of sub-millimetre radio-frequency-powered chiplets recording from rodent cortex as a coordinated network, without ultrasound." },

  { "year": "2020s", "title": "Magnetoelectric and clinical spin-offs", "text": "Magnetoelectric films are used to power millimetre-scale stimulators, and companies begin first-in-human studies of pea-sized wireless stimulators for psychiatric indications." }

  { "year": "2020s", "title": "Magnetoelectric and clinical spin-offs", "text": "Magnetoelectric films are used to power millimetre-scale stimulators, and companies begin first-in-human studies of pea-sized wireless stimulators for psychiatric indications." }

]

Two published results anchor the field. The 2016 demonstration implanted millimetre-scale motes on

rat peripheral nerve and muscle and recovered electroneurogram and electromyogram signals

wirelessly.[^seo2016] A later device shrank an ultrasonically powered

wirelessly.[^seo2016] A later device shrank an ultrasonically powered *stimulator* to a few cubic

*stimulator* to a few cubic millimetres with a bidirectional link, showing that the same channel can

millimetres with a bidirectional link, showing that the same channel can deliver instructions as

deliver instructions as well as carry data.[^piech2020]

well as carry data.[^piech2020]

Neither has recorded single-neuron action potentials in cortex, which is the application the concept

Neither has recorded single-neuron action potentials in cortex, the application the concept was

was proposed for. The gap is substantial: extracellular spikes are tens of microvolts against a

proposed for, and the gap is substantial. Extracellular spikes are tens of microvolts against a

noise floor that a passive backscatter link struggles to resolve, and they require kilohertz sampling

noise floor that a passive backscatter link struggles to resolve, and they need kilohertz sampling,

rather than the slow envelope signals that peripheral nerve recording can tolerate.

whereas peripheral nerve recording can tolerate slow envelope signals.

Parallel work reaches for the same goal by other routes. Networks of tens of sub-millimetre chiplets

Parallel work pursues the same goal by other routes. Networks of tens of sub-millimetre chiplets

powered by radio-frequency fields have recorded from rodent cortex as a coordinated

ensemble.[^lee2021] Magnetoelectric films, which convert an external magnetic field into a local

ensemble.[^lee2021] Magnetoelectric films convert an external magnetic field into a local electric

electric field, power millimetre-scale stimulators without the skull's acoustic penalty, and this

field and power millimetre-scale stimulators without the skull's acoustic penalty. This approach has

approach has moved fastest toward humans: pea-sized wireless cortical stimulators for psychiatric

moved fastest toward humans: pea-sized wireless cortical stimulators for psychiatric indications

indications have entered first-in-human studies in the mid-2020s, aiming at indications currently

entered first-in-human studies in the mid-2020s. They target indications currently served by

served by [[deep-brain-stimulation]] or by [[non-invasive-neuromodulation]] with far less hardware

[[deep-brain-stimulation]] or by [[non-invasive-neuromodulation]], with far less hardware than the

than the former and more precision than the latter.

former and more precision than the latter.

## Limits

Four constraints govern how far the approach can scale.

**Power.** A mote's harvested power scales with its cross-sectional area. Shrinking a device by a

factor of ten in linear dimension cuts available power by a hundred, while the amplifier and

factor of ten in linear dimension cuts the available power by a factor of a hundred, while the

digitizer needed for spike-band recording have a floor set by thermal noise. This is the reason

amplifier and digitizer needed for spike-band recording have a floor set by thermal noise. This is

motes have not shrunk to the tens of microns the original name implies.

why motes have not shrunk to the tens of microns the original name implies.

**Uplink bandwidth.** Backscatter carries little information per interrogation pulse. Recording many

channels at spike-band rates requires either many interrogations per second, which raises the

acoustic dose, or on-mote compression, which requires power the mote does not have.

acoustic dose, or compression on the mote, which needs power the mote does not have.

**Addressing and registration.** With hundreds of motes, the interrogator must know where each one

is, and motes move as tissue moves. Localization by time-of-flight works for a handful and becomes

is, and motes move as tissue moves. Localization by time of flight works for a handful and becomes

an inverse problem for many.

**Surgery does not disappear.** Distributing hundreds of motes through cortex requires inserting

**Surgery does not disappear.** Distributing hundreds of motes through cortex means inserting them,

them, which means either an injection needle or an insertion tool per mote, plus a sub-cranial relay.

with either an injection needle or an insertion tool per mote, plus a sub-cranial relay. The tissue

The tissue trauma of many small insertions is not obviously less than that of one array, and it is

trauma of many small insertions is not obviously less than that of one array, and it is not well

not well characterized.

characterized.

A fifth consideration is not technical. A recording device with no wires, no external marker, and no

A fifth consideration is not technical. A recording device with no wires, no external marker and no

battery to replace is also a device that is difficult to audit, which sharpens the questions raised

battery to replace is also hard to audit. That sharpens the questions raised under

under [[mental-privacy]] about who holds neural data and under what terms it is read.

[[mental-privacy]] about who holds neural data and on what terms it is read.

> [!caution] Naming versus scale

> "Neural dust" invites comparison with the devices imagined under [[medical-nanorobots]], which are

> proposed to be a thousand times smaller and to move under their own power. The demonstrated motes

> are passive millimetre-scale electronics, closer in kind to a very small pacemaker than to anything

> are passive millimetre-scale electronics, closer in kind to a very small pacemaker than to

> nanoscale, and the mismatch between the name and the hardware has repeatedly caused the work to be

> anything nanoscale. The mismatch between the name and the hardware has repeatedly led to the work

> reported as more advanced than it is.

> being reported as more advanced than it is.

## Outlook

The most defensible near-term application is peripheral: chronic, untethered monitoring or

stimulation of nerves in the abdomen, chest, or limbs, where there is no skull, the signals are

stimulation of nerves in the abdomen, chest or limbs. There is no skull in the way, the signals are

larger and slower, and the clinical need — closed-loop control of bladder function, inflammation, or

larger and slower, and the clinical need is real, for example closed-loop control of bladder

cardiac rhythm — is real. This is bioelectronic medicine rather than a

function, inflammation or cardiac rhythm. This is bioelectronic medicine rather than a

[[brain-computer-interface]], and it is where ultrasonic implants are most likely to reach patients.

It also overlaps with the untethered devices covered under

It also overlaps with the untethered devices covered under [[microrobots-in-medicine]], which are

[[microrobots-in-medicine]], which are steered through the body rather than fixed in place, and with

steered through the body rather than fixed in place, and with the delivery-focused work in

the delivery-focused work in [[targeted-drug-delivery]].

[[targeted-drug-delivery]].

The cortical version competes with approaches that are further along. Penetrating arrays behind a

fully implanted wireless can, as pursued by [[neuralink]] and others; surface arrays as used in

fully implanted wireless can, as pursued by [[neuralink]] and others, surface arrays of the kind

[[ecog-interfaces]]; and the endovascular route of the [[stentrode]] all address the tethering

used in [[ecog-interfaces]], and the endovascular route of the [[stentrode]] all address the

problem without requiring a new power-transfer physics. Whether distributed motes offer enough

tethering problem without new power-transfer physics. Whether distributed motes add enough coverage

additional coverage to justify their difficulty depends on a question that

to justify their difficulty depends on a question that [[neural-decoding]] research has partly

[[neural-decoding]] research has partly answered in the negative: information from motor cortex

answered in the negative. Information from motor cortex saturates well before channel count does, so

saturates well before channel count does, so a thousand scattered sensors may not decode much better

a thousand scattered sensors may not decode much better than a hundred well-placed ones. The case

than a hundred well-placed ones. The case for neural dust is stronger where broad, sparse sampling

for neural dust is stronger where broad, sparse sampling across many regions matters more than dense

across many regions matters more than dense sampling of one — the regime relevant to

sampling of one. That regime is relevant to [[whole-brain-emulation]] arguments and, more

[[whole-brain-emulation]] arguments and, more immediately, to any attempt at a

immediately, to any attempt at a [[memory-prosthesis]] spanning distributed circuits.

[[memory-prosthesis]] spanning distributed circuits.

## See also