Utah array: revision 1 to 2 — Future Human Wiki

Revision 1

Revision 2

trl: 7

categories: ["cybernetics"]

tags: ["bci", "microelectrodes", "neural recording", "braingate", "implants", "neurotechnology"]

summary: "A silicon microelectrode array of 100 needles on a 10-by-10 grid, the standard sensor for chronic single-neuron recording in human brain–computer interface research."

summary: "A silicon microelectrode array of 100 needles on a 10-by-10 grid, the standard sensor for chronic single-neuron recording in human brain-computer interface research."

updated: "2026-07-27"

updated: "2026-10-29"

humanEvidence: "People with paralysis have used implanted arrays under investigational device exemptions to control cursors, robotic arms and speech decoders; some implants recorded usable signals beyond a thousand days."

humanEvidence: "People with paralysis have used implanted arrays under investigational device exemptions to control cursors, robotic arms and speech decoders; some implants recorded usable signals beyond a thousand days."

access: "Not a therapy: the US clearance covers recording for up to thirty days, and every long-term implant is placed in a research study under an investigational device exemption."

access: "Not a therapy: the US clearance covers recording for up to thirty days, and every long-term implant is placed in a research study under an investigational device exemption."

reversibility: "difficult"

}

```

**Utah array** is a silicon microelectrode array consisting of a grid of sharpened needles etched

**Utah array** is a silicon microelectrode array: a grid of sharpened needles etched from a single

from a single block of silicon, inserted into the cortex so that each tip records the extracellular

block of silicon and inserted into the cortex, so that each tip records the extracellular voltage

voltage near a small number of neurons. Introduced at the University of Utah around 1990, it became

near a small number of neurons. Introduced at the University of Utah around 1990, it became and

and remains the standard sensor for chronic single-neuron recording in humans, and nearly every

remains the standard sensor for chronic single-neuron recording in humans. Nearly every major human

major human [[brain-computer-interface]] result involving individual action potentials has used

[[brain-computer-interface]] result involving individual action potentials has used one.

one.

Its persistence is notable given how little it has changed. A device designed before the field had

It has changed little since. A device designed before the field had a single human participant still

a single human participant still sets the performance benchmark that newer, higher-channel-count

sets the performance benchmark against which newer designs with higher channel counts are measured.

designs are measured against.

```keyfacts

[

  { "value": "1991", "label": "First manufacturing description", "note": "Campbell et al., IEEE Transactions on Biomedical Engineering" },

  { "value": "1991", "label": "First manufacturing description", "note": "Campbell et al., IEEE Transactions on Biomedical Engineering" },

{ "value": "96", "label": "Channels typically wired", "note": "of 100 electrode sites on the grid" },

{ "value": "~4 mm", "label": "Footprint edge length", "note": "shanks usually 1.0–1.5 mm long" }

{ "value": "~4 mm", "label": "Footprint edge length", "note": "shanks usually 1.0-1.5 mm long" }

]

```

The array is machined from a monolithic block of doped silicon. Dicing saws cut a grid of columns,

which are then acid-etched into tapered needles on a 400-micrometre pitch, giving a device roughly

four millimetres on a side with 100 electrode sites. The shanks, typically one to one and a half

millimetres long, reach the cortical layers where pyramidal-cell somata are dense. The whole

millimetres long, reach the cortical layers where pyramidal-cell somata are dense. Everything except

structure except the tips is insulated with parylene-C; the exposed tips are coated with platinum or

the tips is insulated with parylene-C, and the exposed tips are coated with platinum or iridium

iridium oxide to lower impedance.[^campbell1991]

oxide to lower impedance.[^campbell1991]

Insertion is the tricky part. Pressed slowly, the array dimples the cortical surface rather than

Insertion is the difficult step. Pressed in slowly, the array dimples the cortical surface instead

penetrating it, damaging tissue and pial vessels. A pneumatic inserter fires the array in at several

of penetrating it, damaging tissue and pial vessels. A pneumatic inserter fires the array in at

metres per second, which produces cleaner penetration.[^maynard1997] Wires from each shank run

several metres per second, which produces cleaner penetration.[^maynard1997] Wires from each shank

through a bundle to a connector; in the standard research configuration this is a percutaneous

run through a bundle to a connector. In the standard research configuration this is a percutaneous

titanium pedestal fixed to the skull, through which amplifiers are plugged in during sessions.

titanium pedestal fixed to the skull, into which amplifiers are plugged during sessions.

What the array measures is the extracellular field near each tip. Filtered above roughly 250 Hz and

The array measures the extracellular field near each tip. Filtered above roughly 250 Hz and

thresholded, this yields action potentials from one to a few nearby neurons per channel; filtered

below about 300 Hz, it yields local field potentials reflecting activity over a larger volume. Both

feed the decoders described under [[neural-decoding]].

## Development history

Richard Normann's group designed the array in the late 1980s with a cortical visual prosthesis in

mind — a grid of stimulating electrodes in visual cortex intended to evoke a pattern of phosphenes,

mind: a grid of stimulating electrodes in visual cortex intended to evoke a pattern of phosphenes.

the same goal that continues to motivate work described under [[retinal-implant]]. The

The same goal still motivates the work described under [[retinal-implant]]. The manufacturing

manufacturing process was published in 1991 and the recording characterization followed through the

process was published in 1991, and the recording characterization followed through the 1990s.

1990s.

Commercialization passed through several hands: Bionic Technologies, then Cyberkinetics

Neurotechnology Systems, then Blackrock Microsystems, now Blackrock Neurotech. Cyberkinetics ran

Neurotechnology Systems, then Blackrock Microsystems, now Blackrock Neurotech. Cyberkinetics ran the

the BrainGate pilot trial, in which Matthew Nagle, paralysed by a knife wound to the spinal cord,

BrainGate pilot trial, in which Matthew Nagle, paralysed by a knife wound to the spinal cord,

received an array in his motor cortex in 2004 and used it to move a cursor, open email, and operate

a prosthetic hand.[^hochberg2006] The BrainGate2 consortium — Brown, Massachusetts General Hospital,

a prosthetic hand.[^hochberg2006] The BrainGate2 consortium (Brown, Massachusetts General Hospital,

Stanford, and later UC Davis — has run the successor studies since.

Stanford, and later UC Davis) has run the successor studies since.

In the United States the array holds a clearance for temporary recording of up to thirty days,

In the United States the array holds a clearance for temporary recording of up to thirty days, which

which supports intraoperative and epilepsy-monitoring use. Every long-term human implant is

supports intraoperative and epilepsy-monitoring use. Every long-term human implant is conducted

conducted under an investigational device exemption, not as approved therapy.

under an investigational device exemption, not as approved therapy.

## Role in human BCI research

Because the array resolves individual neurons, it supports the highest-dimensional control yet

demonstrated. Robotic-arm control with multiple degrees of freedom, high-rate cursor typing, and

several of the leading [[speech-neuroprosthesis]] results rest on Utah arrays, usually two to

several of the leading [[speech-neuroprosthesis]] results rest on Utah arrays, usually two to four

four implanted across motor and premotor areas; the fastest reported speech rates have come from

implanted across motor and premotor areas. The fastest reported speech rates have come from surface

surface grids rather than penetrating arrays, and the two approaches sit close together. It is also the sensor behind most attempts at

grids rather than penetrating arrays, though the two approaches sit close together. The array is

writing information back in through intracortical microstimulation, the route by which touch

also the sensor behind most attempts to write information back in through intracortical

percepts have been delivered to users of [[neuroprosthetics]].

microstimulation, the route by which touch percepts have been delivered to users of

[[neuroprosthetics]].

A variant with shanks of graded length, the Utah slanted electrode array, is designed for

A variant with shanks of graded length, the Utah slanted electrode array, is designed for peripheral

peripheral nerve rather than cortex, and has been used to give sensation to users of

nerve rather than cortex. It has been used to give sensation to users of

[[myoelectric-prosthetics]] — an early instance of the closed-loop principle that

[[myoelectric-prosthetics]], an early instance of the closed-loop principle that

[[sensory-augmentation]] research generalizes to senses the body never had.

[[sensory-augmentation]] research extends to senses the body never had.

Depth electrodes of other designs, rather than the Utah array, carry most of the human hippocampal

Most of the human hippocampal work behind the [[memory-prosthesis]] literature uses depth electrodes

work behind the [[memory-prosthesis]] literature, because the relevant structures lie centimetres

of other designs rather than the Utah array, because the relevant structures lie centimetres below

below the cortical surface that the array is built to sample.

the cortical surface the array is built to sample.

> [!note] What "96 channels" does and does not mean

> Each channel may record several neurons whose spikes must be sorted, or none at all. The number

> Each channel may record several neurons whose spikes must be sorted, or none at all. The number of

> of channels yielding usable single units is generally well below the number wired, falls over

> channels yielding usable single units is generally well below the number wired, falls over time,

> time, and varies by implant site and participant. Reported channel counts are a property of the

> and varies by implant site and participant. A reported channel count describes the hardware, not

> hardware, not of the signal obtained.

> the signal obtained.

## Failure modes and longevity

years.[^simeral2011]

The causes are mixed. Insertion injury and chronic micromotion provoke a foreign-body response:

microglia activate, astrocytes form an encapsulating sheath, and neurons retreat from the tip.

microglia activate, astrocytes form an encapsulating sheath, and neurons retreat from the tip. The

Independently, the device itself deteriorates — parylene insulation cracks and delaminates, tip

device also deteriorates independently. Parylene insulation cracks and delaminates, tip

metallization corrodes, wire bonds and connectors fail. A systematic analysis in non-human primates

metallization corrodes, and wire bonds and connectors fail. A systematic analysis in non-human

found that material and mechanical failures accounted for a large share of losses, which matters

primates found that material and mechanical failures accounted for a large share of losses. That

because it implies that improving biocompatibility alone will not fix

finding implies that better biocompatibility alone will not fix longevity.[^barrese2013]

longevity.[^barrese2013]

Percutaneous pedestals add infection risk and prevent the user from living with the system

Percutaneous pedestals add infection risk and keep the user from living with the system untethered.

untethered. This is the specific limitation that motivates fully implanted wireless designs,

This limitation is what motivates fully implanted wireless designs, including [[neuralink]]'s sealed

including [[neuralink]]'s sealed package and the surface arrays discussed under

package and the surface arrays discussed under [[ecog-interfaces]].

[[ecog-interfaces]].

## Alternatives

Three lines of work aim to displace the array. Higher-density silicon probes, most prominently

Neuropixels, put hundreds to thousands of recording sites along a single thin shank and have been

used acutely in human cortex during neurosurgery.[^jun2017][^paulk2022] Flexible polymer threads,

used acutely in human cortex during neurosurgery.[^jun2017][^paulk2022] Flexible polymer threads, as

as used by Neuralink, trade rigidity for reduced chronic tissue reaction but require robotic

used by Neuralink, trade rigidity for a reduced chronic tissue reaction but require robotic

insertion. Non-penetrating approaches — subdural films and the endovascular [[stentrode]] — give up

insertion. Non-penetrating approaches, namely subdural films and the endovascular [[stentrode]],

single-neuron resolution entirely in exchange for stability and lower surgical risk.

give up single-neuron resolution entirely in exchange for stability and lower surgical risk.

A fourth approach avoids metal electrodes altogether: [[optogenetics]] can read and write neural

A fourth approach avoids metal electrodes altogether. [[optogenetics]] can read and write neural

activity optically, but requires genetically modifying the target neurons, which places it far from

activity optically, but it requires genetically modifying the target neurons, which places it far

routine human use.

from routine human use.

None has yet accumulated a comparable human track record. That is the array's main remaining

None of these has yet built up a comparable human track record, and that is the array's main

advantage: two decades of implants across multiple sites, with known failure statistics, against

remaining advantage. It has two decades of implants across multiple sites, with known failure

competitors whose chronic behaviour in humans is largely unmeasured.

statistics, while its competitors' chronic behaviour in humans is largely unmeasured.

## Outlook

The realistic future for penetrating arrays is not that they win or lose outright but that they

Penetrating arrays are more likely to specialize than to win or lose outright. Applications that

specialize. Applications demanding many independent control dimensions — dexterous limb control,

demand many independent control dimensions, such as dexterous limb control, fluent speech, and any

fluent speech, and any bidirectional system delivering fine sensory feedback — need single-neuron

bidirectional system delivering fine sensory feedback, need single-neuron resolution that surface

resolution that surface and vascular electrodes cannot supply. Applications needing a reliable

and vascular electrodes cannot supply. Applications that need a reliable switch do not.

switch do not.

Whether that specialization survives depends on an unresolved empirical question: whether a

Whether that specialization survives depends on an unresolved empirical question: can a penetrating

penetrating device can be built that still records well after ten years in a human brain. The

device be built that still records well after ten years in a human brain? The best-established

best-established implanted neurotechnologies, among them the [[cochlear-implant]] and

implanted neurotechnologies, among them the [[cochlear-implant]] and [[deep-brain-stimulation]],

[[deep-brain-stimulation]], stimulate rather than record and sit in tissue that tolerates

stimulate rather than record, and they sit in tissue that tolerates them for decades. No recording

them for decades. No recording array has demonstrated equivalent durability, and the failure

array has shown equivalent durability, and the failure analyses suggest the answer depends as much

analyses suggest the answer depends as much on materials engineering as on neuroscience.

on materials engineering as on neuroscience.

## See also