Nanoparticle BCIs: Fact or Fiction
Subsense's $17M Funding and Iota Biosciences' 2020 Acquisition Raise New Questions Around BCI Nanoparticles
Brain-computer interfaces (BCIs) enable direct communication between neural circuits and external devices, traditionally requiring either invasive implanted electrodes or non-invasive surface sensors. Nanoparticle-based BCI technology has emerged as a novel approach promising high-fidelity neural interfacing without major surgery. Rather than placing macro-scale electrodes via open-brain procedures, this approach uses tiny injectable micro/nano-scale particles (“neural dust,” “neurograins,” etc.) that can transmit and/or receive neural signals from within the brain. Pioneering efforts in this domain include academic projects like neural dust from UC Berkeley and magnetoelectric nanoparticles from MIT/Rice, as well as new startups such as Subsense Inc., which in 2025 announced development of a “non-surgical, nanoparticle-based bidirectional BCI.” These technologies aim to achieve a seamless brain interface by distributing countless microscopic sensors or stimulators inside neural tissue via injection or bloodstream delivery, thereby reading or modulating neural activity wirelessly.
These technologies hold the promise of making the interface to the brain as invisible and ubiquitous as the microchips that revolutionized computing, essentially turning neural interfacing into a distributed “nano-electronics” problem rather than a surgical hardware problem.
I’ve compiled my notes over the last few years detailing nanoparticle-based BCIs, covering their operating mechanisms, material science foundations, neural interfacing at the cellular level, current limitations, experimental progress, commercial outlook, and future potential applications.
As always, I don’t have insider knowledge into these companies, and can only derive my notes from former employees, interviews, papers, and patents/patent applications.
Technical Mechanisms of Nanoparticle BCIs
Nanoparticle BCI systems function by sensing neural electrical signals and/or stimulating neurons using devices on the scale of a cell or smaller. Because of their minuscule size, these devices must operate wirelessly and be powered remotely. Key aspects of their mechanism include:
Neural Signal Detection: Nanoparticle sensors detect the electrical activity of neurons (e.g. action potentials or local field potentials) typically from the extracellular space near the neuron. For example, “neural dust” sensors contain nanoelectrodes attached to a tiny transistor or piezoelectric element that picks up voltage changes in nearby neurons. When a neuron fires, the local electric field alters the electrical properties of the nanoparticle device. In one design, a change in neural voltage modulates an electrical circuit’s impedance, which can be detected via changes in an emitted signal (more on transmission below). Another approach uses nano-optoelectronic sensors: for instance, plasmonic nanoparticles can act as “nano-antennae” that transduce electrical fields into optical signals by shifting their light-scattering properties in response to voltage (described in detail in Section 2). Each nanoparticle effectively serves as a microscale EEG electrode, but one that can be present right at the neuron’s membrane, vastly improving spatial resolution compared to scalp sensors.
Neural Stimulation: For bidirectional BCI capability, nanoparticles can also actively influence neurons. Stimulation typically involves generating a local electric or thermal field to depolarize neurons. One prominent method is via magnetoelectric nanoparticles (MENPs): these are particles engineered to produce an electric potential when exposed to an external magnetic field. By applying a magnetic field pulse from outside the head, each MENP can emit a tiny electric current in its immediate vicinity, stimulating any neuron to which it is adjacent. Other concepts include ultrasonic stimulation (using acoustic energy to make a piezoelectric particle generate an electric field) and optogenetic-like stimulation (e.g. using light on light-sensitive nanoparticles), though magnetoelectric and electrical stimulation are most common in current designs. Importantly, because each particle is untethered, addressing specific brain regions for stimulation may rely on targeting (guiding particles to a region or cell type) or using field focusing externally.
Wireless energy delivery is central to nanoparticle BCI function, as on-board batteries are unfeasible at micro/nanoscales. Different energy transfer modalities are used:
Ultrasound: Neural dust devices pioneered the use of ultrasonic acoustic waves to power implants. An external ultrasonic transducer sends pulses into the body; nanoparticles with piezoelectric crystals convert ultrasound vibrations into electrical energy. This powers the sensor electronics momentarily (in bursts synchronized with the pulses). Ultrasound is advantageous because it penetrates deep tissue and can be focused, while piezoelectric materials efficiently convert sound to electricity at small scales.
Radiofrequency (RF) Electromagnetic Fields: Some microscale devices use inductive or RF coupling. For example, “neurograins” (grain-of-salt-sized wireless chips from Brown University) harvest power from an RF transmitter placed on the scalp. Similarly, any nanoparticle with a small antenna or coil could rectify RF energy. However, electromagnetic waves at high frequency attenuate in tissue, limiting depth; thus RF is more often used for slightly larger microscale implants rather than sub-millimeter nanoparticles.
Magnetic Fields: Magnetically coupled power is used with magnetoelectric particles. A time-varying magnetic field can induce voltages in a magnetostrictive/piezoelectric core-shell (as described later in Section 2), effectively delivering energy to generate electric pulses for stimulation. Magnetic fields (especially in the low radiofrequency range) have good tissue penetration, enabling deep-brain energy delivery.
Light (Photonic Energy): In optical nanoparticle systems like UC Santa Cruz’s approach, near-infrared (NIR) light serves as both power and readout medium. Nanosensors absorb NIR photons and use that energy to probe local voltage (via electro-optical effects), then emit a modified optical signal. NIR light in the 1000–1700 nm range can pass through skull and tissue with relatively low attenuation, enabling a form of in-vivo optical powering and data retrieval
Data Communication (Wireless Readout): Once powered and sensing, the nanoparticle must transmit the neural data out to an external receiver (on the scalp or nearby). The strategies mirror the powering methods:
Backscatter Modulation: This is used by passive devices like neural dust. The external transmitter (ultrasound emitter) not only powers the particle but also listens for changes in the return signal. When the neural dust’s piezoelectric crystal is driven by neural voltage, it slightly alters how it vibrates. This causes a measurable change in the echo or backscatter of the ultrasound signal. In effect, the neural spike modulates the ultrasonic reflections. By analyzing the returning ultrasound, the system decodes the neural signals – all without the particle actively generating its own acoustic wave (saving energy).
Active Radio Transmission: In more complex microscale chips (like neurograins), onboard electronics can actively transmit a radio signal with the recorded data. In the Brown University neurograin demo, each 0.25 mm chip included an amplifier and transmitter that sent out bursts of data at a rate of about 1 Mb/s to a receiver hub on the surface of the brain. This requires more complex circuitry and higher power draw, so it’s feasible for slightly larger “micro” particles but challenging for true nanoscale particles.
Optical Scattering/Emission: With optonanoparticle BCIs, the communication is optical. As illustrated in Figure 1, a plasmonic nanoparticle can scatter an externally supplied NIR light beam. The scattering intensity or wavelength shift encodes the local voltage (for instance, an increase in membrane potential might cause a small resonant wavelength shift in the nanoparticle’s optical response). A detector capturing the far-field scattered light can thus pick up fast neural dynamics. This method leverages the high bandwidth of photonics – multiple sensors could potentially be distinguished by different optical spectra or by imaging their locations if resolution permits.
Magnetic Field Modulation: Though still theoretical, a reciprocal use of magnetoelectric particles could allow neural activity to modulate a particle’s magnetic properties, which could then be read by a sensitive external magnetometer or an MRI-like device. Simulations suggest that when a neuron fires and creates an electric field, a magnetoelectric nanoparticle nearby might slightly change its magnetization (via the inverse magnetostrictive effect), which in principle could be detected by advanced magnetic sensing tech. This concept remains to be proven experimentally but is envisioned as a future pathway to magnetic readout of neural signals.
Deployment Mechanisms: A crucial aspect of nanoparticle BCIs is how the particles are delivered and positioned in the brain:
Injection: Because of their size, these devices can be delivered via minimally invasive injection rather than open surgery. One approach is intravenous injection – nanoparticles injected into the bloodstream that then cross the blood-brain barrier (BBB) to enter neural tissue. Some studies have shown that suitably coated nanoparticles (e.g. ~100 nm iron oxide based) can cross the BBB or even be taken up intranasally and transported into the brain. Applying a static magnetic field can help concentrate magnetic nanoparticles in a target brain region after systemic injection. Another approach is direct intracerebral injection via a fine needle, implanting a bolus of nanoparticles in a specific area (or infusing via microcatheter). Because the particles are microscopic, the injection can be done through a small bore needle with far less tissue damage than placing a large device.
Surface Sprinkling: In some experimental setups (e.g. early neurograin tests), the particles were placed by opening the skull and distributing them on the cortex surface. This is less ideal (since it reverts to surgery), but was used to demonstrate the concept. In the future, even distribution throughout the brain might be achieved by cerebrospinal fluid (CSF) infusion – injecting nanoparticles into the CSF (e.g. via lumbar puncture or cistern injection) to let them circulate and settle in brain spaces.
Targeting Specific Cells: Advanced designs include functionalizing nanoparticles with biological molecules (antibodies, peptides) that bind to particular neural cell types or receptors. For example, Subsense Inc reports that their nanoparticles “bind with receptors in the brain” to achieve stable attachment at neural synapses or cell membranes. By selecting a receptor unique to a neuron subtype, one could target the particles to, say, motor neurons or dopamine neurons specifically. This targeting increases both the specificity of interfacing and the likelihood that each particle is optimally located to sense or stimulate its intended neuron.
Best way to think about it is as this: nanoparticle BCIs replace traditional wired electrodes with distributed, free-floating “sensor/effector dust” that is powered and read by remote fields. The technical principle is to exploit physics at small scales, converting external physical energy (ultrasound, EM waves, light, magnetism) into local electric interactions with neurons, and vice versa.
Material Science and Nanotechnology Considerations
Building functional electronics on the scale of a cell (tens of microns or smaller) requires advanced materials and nanofabrication techniques. These nanoparticle BCI systems draw from innovations in nanoelectronics, biomaterials, and MEMS/NEMS (micro- and nano-electromechanical systems). Key material components include:
Piezoelectric Crystals: Many wireless neural interfaces rely on piezoelectric materials, which generate electric charge when mechanically deformed (and conversely, deform when an electric field is applied). Lead zirconate titanate (PZT) or barium titanate (BaTiO₃) crystals are common choices. In neural dust prototypes, a tiny PZT crystal serves as both the sensor and actuator: it converts incoming ultrasound into electricity to power the device, and it converts neural voltage changes into mechanical vibrations that modulate the reflected ultrasound. Piezoelectric nanocrystals can be fabricated by thin-film deposition or nanoprinting techniques and then etched or released into freestanding particles. A challenge is making these crystals biocompatible – PZT contains lead, so often it is fully encapsulated in a non-toxic coating (like an epoxy or polymer) to avoid exposure to tissue
Semiconductor Nanoelectronics: To process and transmit signals, nanoparticles often incorporate miniature semiconductor components. For instance, neural dust motes include a microscopic CMOS transistor and electrodes bonded to the piezoelectric element. This custom transistor can amplify and filter neural signals and control the backscatter modulation. These electronics are typically built on a silicon chip that is only a few hundred micrometers in size. Advanced lithography is used to pattern circuits at this scale, essentially creating a system-on-chip that is then diced into tiny free chips. Similarly, the Brown “neurograin” chips used standard CMOS processes shrunk into 0.25 mm specks containing an amplifier, stimulator, and wireless transceiver. As technology progresses, even smaller integrated circuits (on the order of tens of microns) could be made using semiconductor nanofabrication, though handling and packaging such chips is extremely challenging.
Magnetostrictive and Magnetoelectric Materials: Magnetoelectric nanoparticles (MENPs) are a special class of engineered material combining a magnetostrictive phase and a piezoelectric phase intimately. Magnetostrictive materials (like cobalt ferrite, CoFe₂O₄, or Terfenol-D) change shape when exposed to a magnetic field. Coupling this with a piezoelectric (like BaTiO₃) yields a core–shell nanoparticle that converts magnetic fields to electric fields (and vice versa). Researchers fabricate MENPs by coating magnetostrictive nanoparticle cores with a piezoelectric layer, for example, CoFe₂O₄ nanoparticles synthesized via sol-gel methods, then growing a BaTiO₃ shell around them.
When an external magnetic field oscillates, the CoFe₂O₄ core strains and deforms the BaTiO₃ shell, generating a localized electric voltage. This magnetoelectric effect can stimulate nearby neurons by depolarizing them. Tuning the size (typically 20–100 nm) and crystallographic alignment of these particles optimizes the effect. Magnetoelectric nanomaterials are a major area of development because they offer wireless deep brain stimulation capability with no wires or implanted coils. They must be made of chemically stable compounds – fortunately, ferrites and titanates are generally insoluble and can be made biocompatible (coating with silica or polymers if needed to prevent any ion leaching).
Plasmonic Nanoparticles and Electrochromic Polymers: For optical readout BCIs, plasmonic nanostructures are used. Plasmonic materials (typically noble metals like gold or silver) have conduction electrons that oscillate collectively at optical frequencies, producing strong scattering or absorption at a resonant wavelength. Nanoscale gold, for example, can be engineered into rods or shells that resonate in the near-infrared. Crucially, if the local electric field (like a neuron’s membrane potential) changes the dielectric environment near the particle, it can shift the plasmon resonance. To enhance this effect, researchers incorporate electrochromic coatings such as the conducting polymer PEDOT:PSS on the nanoparticle surface. PEDOT:PSS can change its optical absorption properties in response to slight voltage changes, effectively amplifying the influence of a neuron’s voltage on the particle’s optical behavior. One implementation called NeuroSWARM³ uses a multi-layer design: a silica and magnetite core (for stability and optional magnetic guidance), a gold plasmonic shell, and a PEDOT:PSS electrochromic outer layer. The entire particle is <200 nm in diameter. When a neuron fires near this particle, the voltage drop across the PEDOT layer alters the gold shell’s optical scattering spectrum, which can be picked up outside the brain with NIR optical sensors. The use of gold and silica provides a highly biocompatible and chemically inert structure (gold and silica are commonly used in biomedical nanoparticles for drug delivery and imaging). These “nanoantenna” sensors are fabricated via colloidal nanoparticle synthesis (for the core and shell) and polymer chemistry for the coating. They represent a marriage of nanophotonics and neurotechnology.
Carbon Nanomaterials (Graphene, Nanotubes): While not yet a core part of free-floating nanoparticle BCIs, carbon nanomaterials are worth noting. Graphene, a one-atom-thick sheet of carbon, has extraordinary electrical conductivity, flexibility, and biocompatibility. Graphene has been explored for ultra-small neural electrodes and field-effect transistors that can sense neural activity at very high resolution. For example, graphene microelectrode arrays on flexible substrates can record neural signals with minimal tissue response due to their thin, transparent nature. In principle, a graphene-based nanosensor could be created – e.g., a tiny graphene transistor that modulates its conductance with local electric fields. Carbon nanotubes (CNTs) are another candidate for nanoelectrodes; they can act as needle-like electrodes that penetrate cell membranes or as antennas for radio frequency coupling. Some research has used CNTs to form nanoelectrode coatings that improve interface impedance and stability of larger electrodes. While graphene and CNTs have outstanding electrical properties, controlling and deploying them as free particles in the brain is non-trivial (they tend to form aggregates and are hard to biochemically target). However, they might be integrated into hybrid nanoparticles or used to connect networks of nano-sensors in future designs.
Biocompatible Coatings and Encapsulation: A critical materials issue is ensuring that nanoparticles can reside in the brain without causing damage or being corroded. The brain’s immune cells (microglia) react strongly to foreign objects, especially those above a certain size. Studies have found that implant feature sizes below ~15 microns elicit significantly less chronic immune response. Thus, the nanoscale dimensions of these particles inherently help – a 100 nm particle is often too small to activate macrophages/microglia aggressively. Still, surface chemistry matters: nanoparticles are typically coated with biocompatible layers such as PEG (polyethylene glycol) or lipid shells to make them “stealthy” and water-dispersible. For example, magnetic or gold particles might be coated with a phospholipid layer or a synthetic polymer to avoid protein adsorption and immune recognition. Some neural particles use PEDOT/PSS coatings, which not only serve an electrical function but also are known to improve biocompatibility and reduce scar formation on electrodes. Encapsulation in an inert matrix (silica or a hydrogel) can further prevent any toxic elemental leaching. Long-term durability is also needed: materials like silicon, gold, and metal oxides can remain stable for years. In one neural dust prototype, the entire 1 mm sensor was coated in medical-grade epoxy to survive in vivo for extended periods. Ongoing research is evaluating how these nanoparticles fare over months to years in brain tissue – whether they migrate, degrade, or trigger any delayed immune reactions.
Fabrication Techniques: The interdisciplinary nature of these systems means fabrication can involve a combination of microelectronics fab (for any silicon chips), wet chemistry (for nanoparticle synthesis), self-assembly, and nano-imprinting. For instance, the Brown neurograins were made by first manufacturing wafers with hundreds of tiny microchips, then dicing and packaging them into freestanding grains. In contrast, magnetoelectric nanoparticles are produced in solution via chemical methods (precipitation, sol-gel, etc.) similar to how one makes magnetic nanoparticles for MRI contrast agents. The optical nanoantennae can be made via colloidal synthesis for the core-shell, then functionalized in chemical reactors with polymers and targeting ligands. These “bottom-up” approaches allow mass production of billions of particles, but achieving uniformity (in size, shape, response) is an ongoing challenge – variations in particle size or coating thickness could lead to differences in resonant frequency or magnetoelectric coupling, which complicates calibration of the BCI system. Ensuring tight manufacturing tolerances and quality control at the nanoscale will be vital for any commercial deployment.
Neural Interface at the Cellular Level
For any BCI, the crux is how the device interfaces with neurons. Nanoparticle-based BCIs operate at or near the cellular scale, which offers both advantages and unique considerations in terms of resolution, invasiveness, and biointegration.
Because nanoparticles are comparable in size to neurons or even synapses, they can achieve extremely high spatial resolution. Traditional microelectrode implants (like Utah arrays) have prongs spaced hundreds of microns apart, each averaging signals from perhaps tens of neurons. In contrast, a 50 µm neural dust mote or a 200 nm nanoantenna can reside adjacent to a single neuron (or even attached to a specific membrane receptor on that neuron). In principle, if one nanoparticle could be allocated per neuron of interest, one could record from or stimulate thousands of individual neurons in parallel, mapping activity with single-cell precision. In practice, achieving a one-to-one mapping is challenging – the particles might distribute stochastically. However, their sheer numbers (an injection could deliver millions of nanoparticles) mean coverage could be dense. For example, researchers envision sprinkling neural dust throughout the cortex such that every ~50–100 µm there’s a sensor present, enough to capture mesoscale population activity in fine detail. Some targeting strategies (using cell-specific binding molecules) could further improve which cells get a nanoparticle attached, enhancing selectivity.
Being so close to the neuron’s signal source gives a strong signal, but it is still an extracellular recording in most designs. A neuron’s action potential might be 100 mV across its membrane, but the extracellular field a few microns away could be only microvolts. The nanoparticle sensors must be extremely sensitive to pick this up above thermal noise. Fortunately, many designs show excellent sensitivity: the optical nanoantennas have high signal-to-noise ratios due to bright plasmonic signals and inherent filtering of background light. In the first neural dust experiments, the system reliably detected neural firing from a peripheral nerve, indicating the backscatter modulation was discernible above noise. Still, comparing to gold-standard microwire recordings, one challenge is separating multiple neuron signals – if two neurons fire near one nanoparticle, the signals may mix. Researchers are investigating algorithms to decode mixed signals or calibrate each particle’s “listening radius.” On the stimulation side, fidelity means being able to reliably excite the intended neuron without affecting too many neighbors. The electric field from a single nanoparticle stimulator likely spreads over tens of microns; this could stimulate a small cluster of neurons rather than one. However, by adjusting the strength of the external stimulus (e.g. magnetic field amplitude for MENPs), one could titrate how strongly and how many particles activate, to tune the effect. In summary, nanoparticles promise improved coupling efficiency (by literally sitting at the neuron), but maintaining a high signal-to-noise ratio and single-neuron specificity requires careful engineering of sensitivity and deployment density.
Although these approaches are described as “non-surgical” or “minimally invasive,” they do introduce foreign objects into the brain. Each nanoparticle is a foreign body that the nervous system may react to. Fortunately, their small size dramatically reduces invasiveness. A particle on the order of the size of a virus (100 nm) might be practically invisible to the immune system or cause only a transient response. Empirical studies have shown that implants below a certain size induce negligible glial scarring. For instance, in a study of nanoelectrode coatings, probes with nanoscale dimensions had much reduced microglial activation long-term compared to larger electrodes. Additionally, by avoiding any tethering wires or bulk, nanoparticles move with the brain’s micromotion, which might reduce chronic tissue irritation. In contrast, rigid electrode implants often cause inflammation as the brain shifts slightly (e.g., with heartbeat), rubbing against the fixed device. Nanoparticles, being freely dispersed or loosely bound, can accommodate such movement without shearing tissue. Thus, at the cellular interface, these tiny devices might be considered quasi-invisible implants. Subsense’s CEO has emphasized that their nanoparticle BCI takes a “far gentler approach” to interfacing by eliminating the trauma of surgery and large implants.
Some nanoparticle systems are designed to bind to neuron membranes (for example, via ligand-receptor binding). This ensures the particle remains at a fixed site on the cell, much like a protein, rather than drifting away. When bound, the nanoparticle effectively becomes part of the cell’s microenvironment. One concern could be whether the binding or presence interferes with normal cell function. For instance, if a particle binds to an ion channel or receptor, does it block that receptor? Ideally, targeting is done to a non-critical site or uses a reversible attachment. Early studies of magnetoelectric nanoparticles in brain tissue showed that neurons remained viable and functional with MENPs present, and that the particles could stimulate neurons without causing cytotoxic damage. This suggests that, at least in short term, neurons tolerate these nano-objects. Over longer term, cells might internalize some particles (through endocytosis) or transport them along axons; whether that poses issues (or could even be beneficial for targeting specific circuits) is an area of ongoing research.
The brain’s immune response (neuroinflammation) can damage both the interface device and neural tissue. As noted, small size mitigates this. Moreover, materials like gold, silica, and certain polymers are largely inert in the brain – for example, gold nanoparticles have been used in biological settings without immune rejection. In animal tests, magnetoelectric nanoparticles were found to be biocompatible and caused no significant inflammation or heat damage during stimulation experiments. Coating electrodes with PEDOT to make “nanoelectrodes” has also been shown to reduce astrogliosis and scar formation on implanted probes. Nonetheless, if millions of particles are introduced, the body may attempt to clear them over time via phagocytosis or draining them through lymphatic pathways. Some fraction might be carried out of the brain and end up in the liver or spleen (as happens with other nanoparticles). Ensuring that any cleared particles or degradation products are non-toxic is critical. For example, if a magnetic nanoparticle dissolves, it releases iron ions which in excess can be harmful; hence stable coatings are used to prevent that. Immune safety profiles from mouse studies have so far been encouraging: one study injecting MENPs in mice observed normal behavior and tissue health weeks afterward, and histology confirmed minimal microglial activation beyond controls.
Important to consider that because these interfaces ultimately sense electrical events, they offer millisecond or better temporal resolution – comparable to invasive electrode recordings and far better than techniques like fMRI or even calcium imaging. In tests, neural dust devices provided continuous real-time readouts of peripheral nerve activity, with the ability to sample on the order of 10 kHz (every 100 µs). There is essentially no inherent speed limitation imposed by the particles themselves; the bottleneck is the external telemetry (e.g., how fast one can pulse ultrasound or scan an optical signal). Optical and electromagnetic methods can easily achieve MHz-scale bandwidths, so capturing fast oscillations or high-frequency bursts is feasible. This makes nanoparticle BCIs suitable for applications like detecting high-frequency neural oscillations or precise spike timing patterns that are important in neural coding.
An interesting aspect at the cellular interface is the potential to combine recording and stimulation in the same locale. A single nanoparticle could, in theory, both sense a neuron’s activity and then deliver stimulation to it if commanded. For instance, a magnetoelectric particle inherently is bidirectional: it could act as a receiver (neuron’s field tweaks its magnetization) and as a stimulator (external magnetic field drives it to stimulate the neuron). Achieving true closed-loop control at the nanoscale (where a particle detects a neuron firing and then provides feedback stimulation within milliseconds) is a futuristic idea, but one uniquely enabled by this technology. It would mimic the natural synaptic feedback loops but with an external controller in the mix.
Limitations and Challenges
Despite the excitement around nanoparticle BCIs, several key hurdles must be addressed before these technologies can reach widespread, practical use. These challenges span technical, biological, and ethical domains.
Delivering adequate power to potentially thousands or millions of free-floating devices in the brain is non-trivial. For ultrasonic power, one must ensure the ultrasound penetrates the skull and reaches all regions of the brain evenly. Strong ultrasound intensities risk tissue heating or cavitation; hence there is a limit to how much power can be sent in. If too many devices draw energy simultaneously, the external source may not be able to sustain them. Similar issues arise with magnetic powering: strong time-varying magnetic fields (> Tesla range) might induce currents in tissue or heat the tissue if the particles have any hysteresis losses. Techniques like using resonant circuits or optimizing field frequencies can help, but careful safety evaluations are needed. Also, coordinating power delivery – for example, scanning an ultrasound focus to different brain regions sequentially – might be required to support large numbers of devices without interference. This complicates the system design and could limit the true concurrency of recordings.
Reading out data from perhaps thousands of neural sensors is a major challenge. Each neural dust mote or nano-sensor generates a stream of bits representing neural spikes or field potentials. If one had 1,000 sensors each sampling at 1 kHz with even just 8-bit resolution, that’s 8,000 bits per millisecond, or 8 Mbps – already comparable to USB data rates – and high-end visions involve far more than 1,000 sensors. The Brown neurograin study with 48 sensors transmitted on the order of 1 Mb/s and anticipated scaling to ~770 sensors with ~16 Mb/s requirement. Optical and RF links can handle tens of Mb/s in theory, but as we scale toward several thousand channels, bandwidth could become a bottleneck. The external receiver must demodulate many signals potentially occurring at once. One mitigation is using multiplexing: e.g., each nanoparticle could be assigned a unique frequency channel (frequency-division multiplexing) or unique time slots (TDMA). Magnetoelectric or ultrasonic backscatter systems might allow frequency multiplexing by tuning each particle’s resonance slightly differently. However, manufacturing such variability deliberately is complex. Alternatively, sophisticated coding schemes and on-particle preprocessing (e.g., detecting spikes on the particle and only sending a binary event rather than continuous data) can reduce bandwidth. Nonetheless, achieving real-time decoding of large-scale brain activity from distributed particles will demand advanced signal processing hardware and algorithms, possibly employing machine learning to sift signal from noise and from the cacophony of multiple simultaneous channels.
In current BCI arrays, each electrode has a known fixed position in the brain. With free nanoparticles, knowing which sensor is where is difficult. If all the signals come back mixed (e.g., as one optical blob or one magnetic field), distinguishing spatially where an event came from is challenging. Some approaches inherently provide localization – for instance, optical imaging of particles could theoretically pinpoint active spots in a field of view. But through the skull, resolving individual 200 nm particles is not currently feasible (it’s beyond diffraction limits and scattering blurs the image). Magnetic or ultrasonic signals also don’t inherently give location. This is a major issue: if you record a neural spike, you’d like to know from what brain region or neuron it originated. One idea is to use triangulation: sending energy from different angles or detecting from multiple sensor arrays to infer the position of a responding particle. Another approach is to deliberately restrict dispersion: for example, inject particles only in a specific nucleus or cortical area, so that any signal is known to come from within that volume. Still, for high-density mapping, better localization methods are needed – perhaps using field gradients (so that response frequency encodes position) or integrating location sensors (like tiny gyromagnetic locators) on the particles. At the moment, this remains an open problem and one reason neural dust-type systems have so far been limited to stimulating/recording general regions rather than pinpoint single-neuron addresses.
Getting nanoparticles to the brain parenchyma non-invasively faces the obstacle of the blood-brain barrier. The BBB prevents most foreign particles larger than ~nm-scale and non-lipophilic molecules from entering the brain from blood. While some experiments achieved BBB crossing with certain coatings or via intranasal routes, doing so reliably in humans is uncertain. Temporary BBB disruption methods (like focused ultrasound opening or osmotic opening) could be used in a clinical scenario to let the particles in, but these techniques carry their own risks (edema, unwanted substance entry). Direct injection via a very fine needle is another option – for example, if a patient is undergoing a procedure, a doctor could inject a nanoparticle suspension at a target site through a small burr hole. This is far less invasive than placing a device but is still a procedure. Navigating particles to desired regions is also complex: systemic injection might lead to many particles getting trapped in peripheral organs or random brain regions. Magnetic targeting can herd magnetic particles, but focusing a magnetic field on deep brain structures from outside is imprecise (one can concentrate them broadly, but not select, say, the hippocampus versus cortex easily without affecting intervening areas). Thus, delivery and targeting remain significant challenges for translating this to human use. A partial workaround is focusing on applications that don’t require whole-brain distribution – for instance, treating Parkinson’s by targeting the basal ganglia: one could inject magnetoelectric particles and use magnets to gather them in the subthalamic nucleus, then stimulate. This has been done in rodents with some success. But broad “brain-wide” neural interfaces (like for a full brain-computer interface) will require creative solutions to overcome biological barriers.
Any new technology implanted in the brain must undergo rigorous safety testing. Nanoparticle toxicology is an active field: certain nanomaterials can cause oxidative stress or DNA damage if they interact improperly with cells. While gold and silica are generally benign, long-term effects of leaving a large quantity of foreign particles in the brain are unknown. Could they trigger late-stage inflammation? What if they slowly dissolve or wear down? The clearance pathways in the brain (the glymphatic system) might remove some fraction over time – where do those end up and are they safely excreted? Additionally, we must ensure that activation methods are safe: e.g. high-intensity RF fields can cause tissue heating (specific absorption rate concerns), strong magnetic fields could induce peripheral nerve stimulation or heat nanoparticles if they aren’t truly lossless, and powerful ultrasound could cause microcavitation if misused. So far, studies show that with moderate field strengths, these methods are safe in small animals, but scaling to humans (with larger head sizes) might require more power, thus more risk. Regulatory bodies will likely treat nanoparticle BCIs as a combined drug-device product, meaning both the material and the emitting hardware will be scrutinized. It will take time and extensive trials to demonstrate that, for example, a nanoparticle BCI for a given therapy can stay in a patient’s brain for years without adverse effects.
We expect medical BCIs to function reliably for years. Traditional implants face issues like electrode degradation or encapsulation by scar tissue. Nanoparticles, being so small, won’t have the exact same failure modes, but new ones emerge: particles might get clustered together over time (reducing coverage), or get phagocytosed by immune cells and sequestered away from neurons, or gradually lose their functionalization (e.g., antibodies detaching, or polymer coatings swelling and changing properties). The long-term stability of each functional layer (magnetic, piezo, polymer) in the brain’s biochemical environment is uncertain. Replacing or removing nanoparticles is also challenging – once injected, one can’t easily retrieve them (short of perhaps using strong magnets to pull out magnetic ones, though complete removal is unlikely). Therefore, one-shot delivery must ideally last the lifetime of the patient or at least many years. This places high demands on material robustness.
Having a high-bandwidth stream from many neural channels necessitates heavy real-time data processing and decoding algorithms. To truly be useful (e.g. controlling a prosthetic limb or decoding speech from thoughts), the system must interpret the raw signals rapidly. This often requires machine learning models or complex neural decoding algorithms, which in turn might need substantial computing power and training data. Implementing this in a portable device (say a wearable that the patient carries) is another layer of complexity beyond just capturing the signals. Essentially, the software challenge grows alongside the hardware capabilities. It’s one thing to record 1,000 neurons; it’s another to make sense of what those 1,000 neurons’ firing patterns mean in real time. Progress is being made in neural decoding (e.g., using neural networks to translate brain signals to text or movement), but coupling that with a nanoparticle BCI system will require robust interfaces between the hardware output and the decoding software.
On the regulatory side, these systems blur the line between device and injectable material. Approval pathways are not well-established – it might require demonstrating both the safety of the particles (like a new type of neuroprosthetic material) and the safety of the external energy system. Public acceptance is also a factor: some people may feel uneasy about “nanobots in the brain,” so there is a challenge in communicating the safety and reversibility (or irreversibility) of these interventions.
Development and Testing: From Lab Experiments to Trials
Nanoparticle BCI technology is still largely in the research and preclinical testing phase. Here, I summarize key milestones and the current state of development, including animal studies and any steps toward human use.
The foundational work on neural dust was performed at UC Berkeley. In 2016, Maharbiz, Carmena, and colleagues published in Neuron the first demonstration of ultrasonic wireless sensors implanted in living tissue. They implanted a 1 mm neural dust mote next to the sciatic nerve of a rat and showed that when the nerve fired (stimulated by an electrode for testing), the wireless device detected it and transmitted the signal out via ultrasound. This validated the core concept of an ultrasound-powered, batteryless implant. Following that, they have worked on shrinking the devices further. The long-term vision mentioned was to reach ~50 micron sized dust that could be safe for brain deployment, although that size has not been reached in published experiments yet. Another milestone was coating these devices for biocompatibility: initially epoxy-encased, later versions aim to use thin-film encapsulations to last for years.
In 2021, researchers including Kozielski, Anikeeva, and Sitti published studies demonstrating wireless deep brain stimulation with magnetoelectric nanoparticles. In one notable study (Kozielski et al., Science Advances, 2021), MENPs were injected into the mouse brain (targeting the subthalamic nucleus, relevant for Parkinson’s disease). By applying a magnetic field from outside, they were able to induce controlled neural activation in that deep region, which led to observable changes in the mouse’s motor behavior. This was a breakthrough showing that “remote-controlled” particles could have a therapeutic effect comparable to an implanted electrode for deep brain stimulation (DBS). Importantly, they reported that freely moving mice tolerated the stimulation well and that the MENPs did not cause significant tissue damage or heating. Anikeeva’s group at MIT also demonstrated stimulating the brain and peripheral nerves with magnetoelectric or magnetic nanoparticles (her work includes magnetothermal stimulation – using magnetic nanoparticles to generate heat and activate temperature-sensitive ion channels in neurons – and magnetomechanical stimulation – using magnetic torque to open mechanosensitive channels). While those methods (magnetothermal, etc.) involve genetic modification (to express heat-sensitive channels), the magnetoelectric approach does not require any genetic alteration, making it more viable for eventual human use. By now, multiple groups have shown that magnetoelectric nanoparticles can be delivered to rodent brains non-surgically (via injection) and can stimulate neurons at targeted sites, with behavioral outcomes.
In 2021, a team from Brown University demonstrated the “neurograins” system in a rodent model (A. Nurmikko et al., Nature Electronics, 2021). They placed 48 microscale neurograin sensors on the cerebral cortex of a rat and recorded neural signals wirelessly from them. They also showed that these chips could provide micro-stimulation, proving bidirectional operation in principle. Although this experiment still required a craniotomy to place the chips on the brain surface, it was a key step toward fully distributed BCI nodes. The wireless networking of the grains, coordinated by an external head-mounted hub, was shown to scale up to potentially hundreds of grains. This work was part of a DARPA program and indicated that even conventional silicon technology can be pushed to very small implants that operate as a network.
Ali Yanik’s group at UC Santa Cruz, in collaboration with Notre Dame, demonstrated electro-plasmonic nanoantenna arrays for sensing electrical activity in vitro. In 2019, they reported recording cardiac cell action potentials optically using a glass substrate dotted with 2.25 million nanoantenna sensors. This was not yet free particles, but it proved the sensing mechanism: the nanoantennas detected the field from beating heart cells with high SNR and temporal precision. Building on this, they proposed the NeuroSWARM³ injectable format. Much of that work has been theoretical or in simulation for the fully wireless nanoparticle probes. Simulations published by Yanik’s team (Hai et al., 2021) suggest that gold-PEDOT nanoparticles could detect single-neuron action potentials via NIR light with sufficient fidelity. The next step is in vivo testing of these optical probes. Subsense Inc’s partnership with Yanik implies that animal studies with optical nanoparticle BCI probes may be underway or starting soon, using perhaps a combination of magnetic and optical readouts.
Some recent research addresses how to guide nanoparticles to the brain and specific regions. A study by Khizroev’s group demonstrated that applying a static magnetic field could pull magnetite-based nanoparticles across the BBB and concentrate them in targeted brain areas in mice. Another by Kaushik et al. showed intranasal administration (sniffing in nanoparticles) can deliver them to the brain along olfactory pathways. These methods were tested with imaging to confirm particle presence in the brain and with histology to check for any damage. Results indicated that significant quantities of nanoparticles can be delivered non-invasively, opening the door for human-compatible delivery routes (e.g., an IV infusion plus a magnetic helmet to guide particles to a brain region).
So far, short-term animal studies (over weeks to a few months) have not shown catastrophic issues. Mice and rats with implanted nanoparticles continued normal behavior, and tissue analysis often shows no major inflammation beyond control levels. Efficacy-wise, we’ve seen that particles can indeed substitute for electrodes in achieving functional outcomes: e.g., in a mouse Parkinsonian model, magnetoelectric stimulation via nanoparticles improved motor function similarly to a wired DBS implant. These are promising signs. However, long-term studies (6+ months, which might correspond to several years of human lifespan) are still needed to be sure particles don’t migrate or cause late-onset effects. Also, larger animal models (like non-human primates) will likely be necessary before human trials, to account for scaling issues (brain size, skull thickness for signal transmission, immune differences, etc.). The MedCity News review noted that “all the pieces are present” for magnetoelectric nanoparticle BCI and that now is the time to begin translational research in earnest, implying that moving to more clinically relevant models is the current focus.
As of 2025, there have not been publicly reported human trials of fully injectable nanoparticle BCI systems. However, the interest is clearly there: DARPA’s N³ (Next-Generation Nonsurgical Neurotechnology) program (2018–2022) funded several groups to develop high-bandwidth, non-surgical BCIs for military use. The outcomes of N³ have not all been published, but the program likely accelerated the technology readiness. Companies like Subsense Inc have formed to carry this forward. Subsense, having secured funding, will presumably aim for first-in-human studies once they demonstrate safety in animals. They mention therapeutic targets like Parkinson’s, epilepsy, depression, etc., and these are indications that typically require years of validation (e.g., for a new DBS approach). It’s reasonable to expect initial human feasibility studies (perhaps in patients already indicated for some neurosurgery) within the next 5 years if all goes well. Another company, Iota Biosciences (spun out of the Berkeley neural dust project), was acquired by a pharma company (Astellas) in 2020, indicating industry confidence in the concept. Iota is reportedly pursuing “electroceutical” applications – tiny implants to treat disease – which might first target peripheral nerves or organs (a simpler path to market than brain BCIs). Those developments can indirectly benefit brain nanoparticle BCI efforts by proving the core technologies (like ultrasound power or magnetoelectric stimulation) in medical applications.
In testing these systems, researchers have to develop new protocols beyond what’s used for typical electrodes. For example, verifying that nanoparticles have reached their target often involves imaging techniques: MRI can visualize iron-oxide based particles; optical imaging or two-photon microscopy can track fluorescently labeled nanoparticles in superficial brain areas. Electrophysiological validation is also important – e.g., using traditional electrodes to confirm that when a nanoparticle reports a spike, the electrode also sees it (ground-truthing the new sensor). Behavioral tests are the ultimate proof for stimulators: the rotorod test for motor function in mice (for Parkinson’s models), or conditioned behavior changes when certain brain regions are stimulated. Some groups also look at whether nanoparticles can record neural activity without any external electrode for ground truth – in these cases they might stimulate the brain (with a known stimulus) and see if the particles detect the expected evoked activity. Testing has to cover biocompatibility: typically by histological staining for microglia, astrocytes, and neurons after weeks of implantation to check for inflammation or neurodegeneration.
Adoption/Commercial Viability
Despite the rapid progress in research, nanoparticle-based BCI technologies have not yet seen broad adoption or commercialization. There are several reasons for this, tied to both the maturity of the technology and the broader BCI landscape. Let’s focus on my favorite word: adoption!
At present, the most advanced BCIs in humans are either invasive electrode arrays (e.g. Utah arrays used in BrainGate clinical trials, or emerging high-channel devices like Neuralink’s thread implants) or non-invasive devices (EEG headsets, fNIRS caps). Nanoparticle BCIs are in between – minimally invasive – but they are new and unproven in humans. Neuralink (though requiring surgery) is moving quickly toward human trials with thousands of electrodes and has demonstrated high-bandwidth recording in monkeys. Synchron’s Stentrode (a stent-based array delivered via blood vessels) has already been tested in a few human patients for basic communication. In contrast, nanoparticle BCIs might be 5-10 years behind in terms of demonstration in humans. Early adopters (patients, clinicians) will not have access to these until safety and efficacy are clearly established. So, current BCI needs (like restoring communication to paralyzed patients) are being tentatively met by other approaches, reducing immediate pressure for nanoparticle solutions. However, in the long run, if nanoparticles can achieve comparable performance with lower risk, they could outcompete these established players.
For a company to commercialize a nanoparticle BCI, they must navigate regulatory approval which can be stringent for medical devices, especially those that interface with the brain. A traditional implanted BCI (like a DBS device or electrode array) is reviewed as a medical device; a nanoparticle system might be considered a drug-device combination or an advanced biologic device, given it involves injecting a particulate suspension. This could complicate and lengthen the approval process. Companies will need to do extensive clinical trials, first proving safety (Phase I), then efficacy (Phase II/III) in specific indications. That means many years and substantial funding. Venture capital has started to flow (Subsense’s $17M seed is an example), but compared to the hundreds of millions invested in Neuralink, Paradromics, and other BCI firms, nanoparticle BCI startups have relatively modest funding so far. The uncertainty in regulatory classification and the novelty of the tech might make investors cautious until there are clear pilot results.
Market Need and Use Cases:
Neuromodulation Therapies: Treating neurological and psychiatric disorders by brain stimulation (depression, OCD, chronic pain, Parkinson’s, epilepsy). Currently, techniques like transcranial magnetic stimulation or implanted DBS are used. Nanoparticle BCIs could offer a “middle ground” – e.g., an outpatient procedure to inject nanoparticles and then an external wearable cap provides ongoing deep brain stimulation without an implanted pulse generator or battery changes. If demonstrated safe, this could be very attractive to patients who are unwilling or unable to undergo brain surgery. So the therapeutic BCI market might adopt nanoparticle tech sooner than the augmentative BCI (like for paralysis or communication), because the risk/benefit trade-off is favorable for treating serious illness.
BCI for Paralysis/Communication: Here, speed and reliability of control are paramount. A non-surgical high-channel alternative would be revolutionary. But to be adopted, it must prove it can achieve similar control fidelity as, say, a Utah array, which currently can record signals to allow a paralyzed person to type or move a robotic arm. If nanoparticles can’t quite match that performance yet, clinicians will stick with what works, even if it’s surgical. Early adoption might involve a scenario like: a patient cannot undergo open-brain surgery due to health conditions, so a nanoparticle BCI is tried as a last resort to give them some communication ability. Such compassionate-use cases could pave the way if successful.
Cognitive Enhancement and Consumer Use: This is a more speculative market – healthy individuals using BCIs for VR/AR control, gaming, or cognitive boosts. Right now, non-invasive devices (headbands, etc.) are the only ethical option here. If nanoparticle BCIs became incredibly safe and easy (e.g., a single injection gives you a high-bandwidth interface), it could open a massive consumer market. But that’s far off; it would require overwhelming proof of safety and likely a shift in public attitude toward elective brain interfaces. Until then, adoption is confined to medical necessities.
In the medical realm, nanoparticle BCIs stand to revolutionize neuromodulation therapies. Consider epilepsy: one could inject nanoparticles that detect the electrical signature of a seizure starting deep in the brain, then immediately provide on-the-spot stimulation to abort the seizure – all wireless and automated. Or Parkinson’s disease: instead of implanting a DBS electrode in the subthalamic nucleus (a risky surgery), a patient might receive magnetoelectric nanoparticles in that region via an arterial catheter, then wear a small coil around their head that continuously provides tailored stimulation, reducing tremors. Chronic pain or depression: treatments like transcranial magnetic stimulation (TMS) could be vastly improved if magnetic nanoparticles were first delivered to the specific networks involved, allowing more focal and intense modulation from the external magnet fields. Essentially, any condition currently managed by implantable neurostimulators (DBS for movement disorders, vagus nerve stimulators for epilepsy/depression, spinal cord stimulators for pain) could eventually be addressed by “wireless injectable stimulators.” This not only avoids surgeries but could allow more granular control (since you could distribute many micro-stimulators through a target region, not just a couple of electrode contacts). It’s conceivable that in 10–20 years, standard practice in neurosurgery/neurotherapy might include injecting therapeutic nanoparticles as commonly as implanting electrodes today.
Afterthoughts
In my opinion, the future outlook for nanoparticle-based BCIs is highly promising. These technologies hold the promise of making the interface to the brain as invisible and ubiquitous as the microchips that revolutionized computing, essentially turning neural interfacing into a distributed “nano-electronics” problem rather than a surgical hardware problem. If that happens, it could usher in a new era of neurotechnology with broad impacts on medicine, communication, and human experience. Achieving this will require interdisciplinary innovation and careful navigation of challenges, but the trajectory set by current research suggests that nanoparticle BCIs do have a meaningful place in the future of neurotechnology – potentially dominating applications where we need a large-scale, precise interface to neural circuits without the burden of surgery.
Select Academic Sources
Seo, D. et al. (2016). “Wireless Recording in the Peripheral Nervous System with Ultrasonic Neural Dust.” Neuron, 91(3): 529-539. – First demonstration of ultrasonic-powered neural dust in live rats, showing wireless recording of nerve signals.
Kozielski, K. et al. (2021). “Nonresonant powering of injectable nanoelectrodes enables wireless deep brain stimulation in freely moving mice.” Science Advances, 7(5): eabd4590. – Shows magnetoelectric nanoparticles used for deep brain stimulation in mice, with behavioral effects comparable to wired electrodes.
Nurmikko, A. et al. (2021). “Wireless network of synchronized microchips for interfacing with neural circuits.” Nature Electronics, 4: 102-111. – Brown University team’s “neurograins” system, demonstrating 48 wireless microscale implants recording and stimulating rat cortex.
Habib, A. et al. (2018). “Electro-Plasmonic Nanoantenna Arrays for High-Performance Neural Recording.” Science Advances, 4(8): eaat4225. – Development of nanoantenna optical probes for neural activity, showing high-sensitivity voltage detection in cardiomyocyte and neural cultures.
Khizroev, S. et al. (2020). “Microscopic brain network activation enabled by magnetoelectric nanoparticles.” Bioelectronic Medicine, 6: 3. – Theoretical and simulation work suggesting how magnetoelectric particles could both stimulate and record neural activity, and describing methods for delivering nanoparticles across the blood-brain barrier.
Yanik, A. A. et al. (2021). “NeuroSWARM³: A System-on-a-Nanoparticle for Remote Neural Recording.” Proceedings of IEEE Nano Bio Conference. – Introduces the concept of NeuroSWARM³ particles (silica-magnetic core, gold plasmonic shell, PEDOT coating) and presents simulations for NIR optical readout of neural signals.
Subsense Inc. (2025). “Subsense Emerges from Stealth with $17M in Seed Funding for a Non-Surgical Brain-Computer Interface.” BusinessWire Press Release, Feb 2025. – Announces Subsense’s development of the first non-surgical bidirectional BCI using nanoparticles that bind to brain receptors, highlighting collaborations with UCSC and ETH Zurich.
MedCity News (2025). “Biocompatible Nanoparticles: Tiny Antennae with Huge Potential for Brain-Computer Interfaces.” – Overview article discussing the state of nanoparticle BCI research, including magnetoelectric stimulation results, delivery methods, and theoretical frameworks for neural reading (magnetic and optical approaches).
This is such a helpful primer Aryan. Appreciate your taking the time to pull it together.