Why Brain Implant Electrodes Degrade Over Time, From Scar Tissue to Cracked Insulation

The microelectrodes that record from individual neurons are among the most delicate objects ever placed in the human body. A typical intracortical array is a few millimetres across, with shanks thinner than a hair. The tips are positioned within tens of micrometres of the neurons they are meant to hear. And they are expected to work for years inside tissue that is soft, wet, salty, mechanically restless and biologically active.

Over time, most of them hear less. The number of electrodes picking up clear neural spikes tends to fall, signal amplitudes shrink, and some arrays fail outright. This is often summarised as "scar tissue", as if the body simply walls the device off. The evidence shows something more complicated. Degradation comes from several distinct processes, biological, mechanical, electrical and system-level, which unfold on different timescales and have different engineering solutions.

This article works through those mechanisms using the longest-running evidence available. That includes nearly two decades of data from people with implanted arrays, explanted human devices examined under the microscope, and a large failure analysis in non-human primates.

The device in question

Most long-term human data come from the Utah array, a silicon microelectrode array that has been used in human brain-computer interface research since 2004. It is a grid of 100 silicon shanks spaced 400 micrometres apart. In most human recording versions, 96 of them are wired as electrodes. The shanks are 1.0 or 1.5 millimetres long, enough to reach the cortical layers containing the neurons of interest.

The distinction between parts matters for the rest of this article. Each shank is coated with an insulating polymer, Parylene C, except at its tip. The exposed tip, coated with platinum or with sputtered iridium oxide, is the electrode proper: the conductive surface that exchanges charge with the surrounding fluid. The array sits on the brain's surface and is connected by a bundle of fine gold wires to a connector. In current human research systems that connector is a pedestal fixed to the skull and passing through the skin, which links to external amplifiers and computers.

Each of these components can fail independently. The electrode tip can degrade, the insulation can crack, the wire bundle can be damaged, the connector can fail, and the tissue around the shanks can change in ways that affect what the electrode can detect.

What the long-term human data show

The largest analysis of long-term human performance was released in 2025 as a preprint by BrainGate investigators. It examined 20 Utah arrays in 14 participants with paralysis enrolled in BrainGate trials between 2004 and 2019. The arrays were implanted for between 296 and 2,780 days. The longest is about seven and a half years, and the mean is just under three years.

The main measure was yield: the proportion of electrodes recording spiking activity from nearby neurons. Average yield was 35.6%, with large variation between arrays. Average yield fell from 41% in the first three months to 34% in the last three months of recording, a decline of about seven percentage points. More than half the arrays kept at least 20% yield over three years. Not all did well. Three of the 20 arrays never provided consistently useful signals for decoding, and two showed steep declines in their first year.

Two findings complicate the usual narrative of steady decline. First, spike and noise amplitudes both tended to fall over time, so the ratio between them stayed roughly constant. Second, a measure of decoding signal-to-noise ratio, which captures how much movement-related information the array provided, was on average slightly higher at the end of recording than at the beginning. Spike-band power, which reflects activity of neurons too distant to be isolated as individual units, often supported useful decoding even when yield was low.

The authors note the limits of their data. The analysis is retrospective, and the tasks varied between participants. Manufacturing processes changed over 15 years, and earlier arrays used shorter electrodes. Disease, implant location, surgery and the individual tissue response could not be separated as causes of variation. Even so, the picture is clearer than the "scar tissue kills implants" story suggests. Utah arrays in people usually keep providing usable signals for years, with gradual rather than catastrophic decline, but with substantial differences between individuals and arrays.

Mechanism 1: the tissue response

Inserting a rigid shank into cortex is an injury. It ruptures blood vessels, damages the blood-brain barrier, displaces and kills some cells, and releases blood proteins into brain tissue. What follows is a sequence of biological processes, not a single "rejection".

The acute injury response. Within minutes to hours, microglia, the brain's resident immune cells, extend processes toward the implant. Over the following days, damaged tissue is cleared and inflammatory signals are released. Some of the early changes in recordings, such as the sharp rise in electrode impedance seen soon after implantation in both human and animal studies, coincide with this phase.

The foreign-body response. Because the device remains, the injury does not resolve as a normal wound would. Microglia and macrophages persist at the tissue-device interface, and inflammation can continue at a low level for as long as the implant is present. This chronic foreign-body response is the body's reaction to a persistent object it cannot break down. It is different from immune rejection of transplanted tissue.

Glial encapsulation. Over weeks, astrocytes become reactive and form a dense layer around the shanks, commonly called a glial scar. This layer changes the electrical environment around the electrode. It can also increase the distance between the electrode tip and the nearest active neurons.

Neuronal changes. Animal studies, largely in rats, have repeatedly found reduced neuronal density in a zone on the order of 100 micrometres around chronically implanted silicon probes. Neurons near the tip may also become less active or change their firing properties. Because a microelectrode can only isolate spikes from neurons very close to it, losing or silencing even a small number of nearby neurons can reduce yield substantially.

Meningeal encapsulation. A separate process occurs on the brain's surface. The meninges, the membranes covering the brain, can grow fibrous tissue around the base of an array. In a large primate study discussed below, this encapsulation was the most common biological cause of failure, in some cases pushing arrays partly out of the cortex. Explanted human arrays examined by University of Pittsburgh researchers showed similar collagenous tissue, including membrane pulled away with the device.

These processes are biologically distinct. A glial layer, a fibrous meningeal capsule, neuronal loss and an infection have different causes, timescales and remedies. Treating them all as "scar tissue" hides the engineering problem.

Mechanism 2: the mechanical mismatch

Brain tissue is extremely soft. Its stiffness is usually measured in kilopascals. Silicon is stiffer by roughly seven orders of magnitude, with a stiffness measured in gigapascals. Even flexible polymers used in newer probes, such as polyimide, are much stiffer than the brain, though far less stiff than silicon.

The brain also moves. It pulses with each heartbeat and breath, and shifts with changes in posture and head movement. A rigid device anchored to the skull, or tethered by a wire bundle, cannot move with it. The resulting small relative motion, called micromotion, is thought to repeatedly irritate the surrounding tissue and contribute to a sustained foreign-body response. Researchers have long argued that reducing this mismatch, with softer, thinner or more flexible devices, should reduce chronic tissue damage.

Animal studies support that idea in part. Flexible and ultrathin probes have produced smaller tissue responses in rodents than conventional silicon shanks, and polymer-based "mesh" electronics have recorded stably for months in mice. But flexibility brings its own costs. A very soft probe cannot be pushed through the brain's surface without a stiff insertion shuttle or a surgical robot. Thin polymer insulation can be more permeable to moisture than thicker coatings. And the long-term human evidence for flexible designs is still accumulating. Flexible should not be read as automatically more biocompatible. It changes the trade-off rather than removing it.

Mechanism 3: materials that break down

The third family of mechanisms concerns the device itself. The body is a warm salt solution containing reactive oxygen species and enzymes, and implant materials slowly degrade in it.

The most direct human evidence comes from a 2021 study in Frontiers in Bioengineering and Biotechnology by researchers at the University of Pittsburgh. They examined six Utah arrays explanted from two people who had used intracortical brain-computer interfaces. One participant's two platinum-tipped recording arrays had been implanted for about 980 days. The other participant had two platinum recording arrays and two iridium-oxide stimulating arrays, implanted for 182 days.

The longer implant showed more damage. On the platinum arrays implanted for about 980 days, 27.8% of electrode tips showed degradation, against 11.1% on the arrays implanted for 182 days. Damage to the shanks relative to the Parylene C insulation, including cracking and peeling, affected 15.2% of electrodes after 980 days and 1.8% after 182 days. Tissue encapsulation was also more extensive on the longer implant.

On one stimulating array, iridium-oxide coating was lost mainly from electrodes that had been used for stimulation. Damaged tips were found on 41.1% of stimulated sites, compared with 5.6% of unstimulated ones. The damage was associated with stimulation that drove the electrode to more negative voltages. Notably, electrodes with stimulation damage still recorded as well as undamaged ones over the period studied. That result is a reminder that visible material damage does not translate directly into immediate functional loss.

The primate evidence points the same way. Insulation failure matters because it can expose more conductive surface than intended, change electrode impedance, create leakage paths between neighbouring electrodes and let fluid reach underlying layers. Researchers analysing long-term primate data concluded that insulation degradation was the most important factor in long-term signal decline among the cases they could explain.

Mechanism 4: everything outside the brain

Some of the most common failures have nothing to do with the brain-electrode interface.

The clearest evidence comes from a 2013 study in the Journal of Neural Engineering by John Barrese, John Donoghue and colleagues at Brown University. It analysed 78 silicon microelectrode arrays implanted in 27 rhesus macaques between 1996 and 2012. Of these, 62 arrays failed completely during the study. Their mean recording life was 387 days and their median 182 days, and about 56% stopped recording within the first year.

Most failures were sudden, not gradual. Of the 62 failures, 45 (73%) were acute, a loss of signal within about a week. Acute mechanical failures accounted for 30 of these, and 83% of those involved the connector. Other acute failures were caused by infection, connector shorting or unknown causes. Seventeen failures were chronic and progressive. They were dominated by biological processes, mainly meningeal encapsulation, and by causes that could not be determined. Arrays that failed slowly for unknown reasons had lasted longest, an average of 1,311 days.

Those are animal data, and primate implants face different handling and risks from human research implants. But the pattern is instructive. In this dataset, the connector and surgical handling were a bigger practical problem than the brain's biological response.

Human research systems face related issues. The skull-mounted pedestal that brings signals out of the body passes through the skin, which carries a risk of infection and skin problems. A 2020 systematic review in Neuromodulation by Autumn Bullard and colleagues at the University of Michigan counted 48 people implanted with Utah arrays by September 2018: 30 temporarily and 18 chronically. One array was explanted after 987 days because of skin retraction around the pedestal, without infection. The authors noted that the number of chronic participants was far too small to estimate complication rates reliably. They argued that safety would need to be established in a properly powered prospective trial. In the 2025 BrainGate analysis, some sudden declines followed specific events. One was a forceful knock to the pedestal, and another followed a pedestal repair.

html

<table>
  <thead>
    <tr>
      <th>Failure class</th>
      <th>Mechanism</th>
      <th>Typical timescale</th>
      <th>Evidence</th>
      <th>Main engineering responses</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>Biological</td>
      <td>Acute injury, chronic foreign-body response, glial encapsulation, neuronal loss near the tip</td>
      <td>Days to months, then persistent</td>
      <td>Animal histology; human explant analysis</td>
      <td>Smaller and softer devices; coatings; anti-inflammatory strategies</td>
    </tr>
    <tr>
      <td>Biological</td>
      <td>Meningeal fibrous encapsulation lifting the array</td>
      <td>Months to years</td>
      <td>Primate failure analysis; human explants</td>
      <td>Surgical technique; managing the dura and meninges</td>
    </tr>
    <tr>
      <td>Biological</td>
      <td>Infection at the skin or implant site</td>
      <td>Any time, often linked to percutaneous connectors</td>
      <td>Primate and human reports</td>
      <td>Fully implanted wireless systems; skin care</td>
    </tr>
    <tr>
      <td>Mechanical</td>
      <td>Micromotion between rigid device and soft tissue</td>
      <td>Continuous</td>
      <td>Largely animal and modelling studies</td>
      <td>Flexible or thinner probes; floating designs</td>
    </tr>
    <tr>
      <td>Mechanical</td>
      <td>Connector or wire bundle damage</td>
      <td>Often sudden</td>
      <td>Primate failure analysis; human pedestal events</td>
      <td>Robust connectors; eliminating percutaneous connections</td>
    </tr>
    <tr>
      <td>Material and electrical</td>
      <td>Insulation cracking or delamination</td>
      <td>Years</td>
      <td>Human explants; primate analysis</td>
      <td>Better insulation materials and adhesion</td>
    </tr>
    <tr>
      <td>Material and electrical</td>
      <td>Degradation of electrode tip metal, including stimulation-induced damage</td>
      <td>Months to years</td>
      <td>Human explants</td>
      <td>Coating adhesion; safe stimulation limits; voltage monitoring</td>
    </tr>
  </tbody>
</table>

Recording and stimulating are not the same problem

Electrodes are used both to record and to stimulate, and the two uses degrade differently.

The most detailed human comparison comes from the University of Pittsburgh. A participant with a spinal cord injury had two arrays in motor cortex used only for recording, and two arrays in somatosensory cortex used to deliver small electrical pulses that evoked sensations in the hand. The team reported results over 1,500 days.

Recording quality declined on both types of array, but less on the stimulated electrodes. Between day 200 and day 1,500, the number of electrodes recording high-amplitude signals fell by 47% on stimulated electrodes and by 72% on unstimulated ones. Stimulation remained effective. The median current needed for the participant to detect a sensation fell from 31.5 microamperes at day 100 to 10.4 microamperes at day 1,500, and the stimulated electrodes continued to evoke percepts throughout. Impedance rose after implantation and then fell steadily over the following years on both array types.

That finding should not be over-generalised: it involves one participant. But it matches the evidence above that recording and stimulation depend on different properties. Recording single neurons requires the electrode to remain very close to healthy, active neurons, and is sensitive to small changes in the surrounding tissue. Stimulating a sensation requires delivering enough charge to activate neurons in a broader region, and can tolerate more change in the immediate environment. A device may become a poor recorder while remaining a useful stimulator.

Why impedance is not a health check

Electrode impedance, the opposition to current flow measured at a given frequency, is the most commonly reported measure of electrode condition, because it is easy to measure. The evidence suggests it should be read with care.

In both the human and primate data described here, impedance typically rises sharply in the weeks after implantation and then declines gradually over years. The rise coincides with the acute tissue response. The later fall can reflect several things, including changes in the tissue around the tip and degradation of insulation that exposes more conductive surface. In the human explant study, impedance did not correlate with recorded signal amplitude on platinum electrodes. In the primate analysis, impedance was a weak predictor of recording quality.

A falling impedance can therefore mean a better electrode-tissue contact, or a cracked insulating layer. On its own, it cannot tell the two apart.

What more electrodes do and do not solve

The newest commercial systems increase electrode counts from 96 per array to more than 1,000, using thin polymer threads or thin-film arrays. More channels can carry more information, and the 2025 BrainGate analysis suggested decoding performance rises roughly with the logarithm of electrode count. On that estimate, the best-performing arrays might approach the performance of able-bodied control with about 1,000 electrodes, while poorly performing arrays would need far more. The authors note that this extrapolation went beyond the range they measured.

Higher channel counts do not change the underlying mechanisms. Every additional electrode is still subject to the tissue response, the mechanical mismatch, insulation degradation and the risks of the connection to the outside world. More electrodes in a smaller volume also bring their own problems: more heat from on-board electronics, more interconnects, and more surfaces at which insulation can fail. Long-term human data for the high-channel-count designs now entering clinical studies are measured in months to a few years, not the decade or more available for the Utah array.

The trade-offs that remain

The evidence points to a few conclusions that hold across devices.

Intracortical electrodes can work in people for years. The common account of rapid failure because "the body rejects" the device is not supported by long-term human data. Degradation is real, but it is usually gradual and highly variable between people and arrays.

Degradation has several causes. Biological processes reduce the number of neurons an electrode can hear. Materials degrade, especially insulation and, under stimulation, electrode coatings. And many sudden failures in long-term animal studies arise from connectors and handling rather than from the brain itself.

Each engineering response involves a trade-off. Softer, smaller probes may reduce tissue response but are harder to insert and to insulate durably. Fully implanted wireless systems remove the skin connector and its infection risk, but add electronics, power supplies and sealed packages that must last for years in the body. Higher electrode counts add information but do not remove any of the underlying failure mechanisms.

The decisive evidence for the next generation of devices will come from long-term follow-up in people. That means not only recordings in the first year, but explant analysis, impedance and yield trends, and the pattern of failures over a decade. Until that evidence exists, claims about how long new neural interfaces will last remain projections.