QE Prize 2026: Nine engineers who wired machines to the nervous system

The 2026 Queen Elizabeth Prize for Engineering honours nine researchers whose collective work created the four pillars of modern neural interface technology.

In recognition of the contributions made by nine individuals who developed the foundational pillars of contemporary neural interface technology - including the development of the four primary types of neural interfaces (cochlear implants, brain computer interfaces, deep brain stimulation and spinal cord stimulation) - which have collectively helped millions of individuals around the world regain their ability to hear, move and maintain their independence through these technologies. On 3rd February 2026, the £500,000 QE Prize was announced at the Science Museum in London with the award being presented to the above-mentioned nine laureates in recognition of "the design and development of modern neural interfaces that restore human function." As such this represents the first occasion that the QE Prize has been awarded to multiple recipients - in total there are now nine - and is an indicator that the field of neural engineering has progressed from a curiosity in the laboratory to a mainstream clinical discipline which is changing how medicine addresses neurological disease/disability.

The four types of neural interface technologies that were developed as part of the project had a common engineering challenge - i.e., how to create a bridge between the flow of electrons in metal and the flow of ions in biological tissue. However, each of the groups developed a different approach - for example for the ear, the motor cortex, deep brain nuclei, and the spinal cord - to overcome the engineering challenge; however, all of the solutions were based upon the same three fundamental engineering components - namely the use of biocompatible materials, the miniaturization of electronic systems and the application of real-time signal processing techniques.

From seashells to a million bionic ears

Graeme Clark, operating professor at the University of Melbourne, drew design inspiration from a seashell’s spiral geometry when he built the first multi-channel cochlear implant prototype in 1978. His device used 20 platinum ring electrodes threaded into the spiral of the cochlea, exploiting its tonotopic organisation - the natural frequency map that runs from high-pitched sounds at the base to low-pitched sounds at the apex. In parallel, Erwin and Ingeborg Hochmair at the Technical University of Vienna had already implanted the world’s first microelectronic multichannel cochlear implant in December 1977: an eight-channel device with wave-shaped platinum wires for flexibility. By 1979, their patient “Connie” became the first person to understand spoken sentences through a cochlear implant alone, without lip-reading.

The engineering of a modern cochlear implant is deceptively elegant. An external microphone captures sound and routes it to a behind-the-ear processor that splits the signal into 12 to 22 frequency bands using digital bandpass filters. Each band’s envelope is extracted, compressed from the roughly 120 dB acoustic dynamic range into the narrow 6-20 dB electrical dynamic range that neural tissue can resolve, and encoded as charge-balanced biphasic current pulses. These instructions cross the skin via inductively coupled coils - carrier frequencies range from 5 MHz (Cochlear Ltd) to 49 MHz (Advanced Bionics) - to a hermetically sealed titanium receiver-stimulator implanted in the mastoid bone. From there, platinum-iridium electrodes on a silicone carrier deliver current to auditory nerve fibres along the cochlea’s length.

The transformation in outcomes came from Blake Wilson’s Continuous Interleaved Sampling (CIS) strategy, developed at the Research Triangle Institute and published in 1993. Earlier compressed-analog processors stimulated multiple electrodes simultaneously, producing unpredictable electric-field interactions that scrambled channel independence. Wilson’s insight was to fire only one electrode at a time in rapid succession - interleaving pulses at rates of 800 to 2,000+ pulses per second per channel. This eliminated cross-channel interference while preserving the full spectral envelope. In Wilson’s initial study, every test subject scored higher with CIS than with the older approach, with some reaching 100% on open-set sentence tests. Wilson and his colleagues donated all intellectual property from their NIH-funded work to the public domain, accelerating adoption by every manufacturer.

Today, over one million cochlear implants have been placed worldwide, with roughly 60,000-70,000 new surgeries annually. Cochlear Ltd holds approximately 50% of the global market, MED-EL 25%, and Advanced Bionics (Sonova) 20%. Modern recipients average 60-80% on open-set sentence recognition in quiet - sufficient for telephone conversation. The latest devices, such as Cochlear’s Nucleus Nexa system (FDA-approved July 2025), feature upgradeable implant firmware and direct Bluetooth streaming. Implants weigh as little as 7.6 grams (MED-EL SYNCHRONY 2) and measure just 3.9 mm thick. Yet performance in background noise remains the field’s stubborn frontier, and only 5-10% of the estimated 430 million people with disabling hearing loss who could benefit have received an implant.

Decoding thought into movement with BrainGate

John Donoghue’s contribution sits at the opposite end of the neural interface spectrum - not stimulating nerves but reading from them. At Brown University, Donoghue built the BrainGate system around the Utah array, a silicon microelectrode device created by Richard Normann at the University of Utah and manufactured by Blackrock Neurotech. The array is a 10 × 10 grid of silicon shanks spanning just 4 mm × 4 mm - roughly the size of a baby aspirin - with 96 active electrodes tipped in platinum or iridium oxide, each penetrating about 1.5 mm into the motor cortex. A pneumatic inserter drives the array into cortical tissue at high velocity to minimise displacement.

The signal chain begins at those electrode tips, which transduce extracellular voltage changes of 50-500 microvolts caused by neuronal action potentials. Gold wires carry signals to a titanium pedestal protruding through the scalp, then through a cable to external amplifiers that bandpass filter between 0.3 Hz and 7.5 kHz, digitising at 30 kilosamples per second per channel. Feature extraction computes threshold-crossing spike rates every 20 milliseconds. A Kalman filter - a state-space model predicting intended velocity from neural firing rates - translates these features into cursor movements or device commands in real time.

In June 2004, Matthew Nagle became the first human to receive a BrainGate implant. Paralysed from the neck down by a spinal cord injury, Nagle controlled a computer cursor, drew shapes, changed television channels, and operated a prosthetic hand - all by imagining movement. “I just thought it,” he said. “I said, ‘Cursor go up to the top right.’ And it did.” The landmark results were published as a Nature cover story in 2006. Then in 2012, Cathy Hutchinson, paralysed for 15 years by a brainstem stroke, used BrainGate to control a robotic arm, reach for a bottle of coffee, and drink unaided - the first time she had done so since her stroke.

Subsequent milestones have been remarkable. In 2021, a participant achieved 90 characters per minute by imagining handwriting - decoded from motor cortex signals by a recurrent neural network. By August 2024, a New England Journal of Medicine paper reported a speech neuroprosthesis achieving 97.5% accuracy on a 125,000-word vocabulary, with a patient using the system for over 248 hours of self-paced conversation and text-to-speech output matching his pre-disease voice.

The central engineering challenge remains the foreign body response. Silicon’s Young’s modulus of 130-185 GPa is roughly six orders of magnitude stiffer than brain tissue at ~0.5-10 kPa. This mismatch produces chronic micromotion-induced inflammation: microglia encapsulate electrode shanks within weeks, astrocytes form a dense glial scar, and neuronal density drops within the recordable radius. Some patients maintain usable signals for over five years - the longest-serving BrainGate recipient has had a functional implant for nine years - but others see degradation within months. A wireless transmitter demonstrated in 2021 eliminated the tethered cable, transmitting full-bandwidth single-neuron signals at 3.3-3.8 GHz, though it still connects to the percutaneous pedestal rather than being fully implanted.

Benabid’s frequency threshold that tamed Parkinson’s

Deep brain stimulation emerged from a single observation during surgery. In 1987, Alim Louis Benabid, a neurosurgeon at the University Hospital of Grenoble, was performing a thalamotomy - the standard destructive procedure for severe tremor - when he noticed that high-frequency electrical stimulation of the ventral intermediate nucleus (VIM) caused the patient’s tremor to stop instantly. Below approximately 100 Hz, stimulation excited tissue and worsened symptoms. Above that threshold, it functionally inhibited neural activity, mimicking a lesion without destroying tissue. This was the birth of modern DBS.

Working with neurologist Pierre Pollak, Benabid refined the technique across hundreds of patients. Their 1991 Lancet paper demonstrated long-term tremor suppression in 26 patients, with 85% achieving substantial tremor reduction from VIM stimulation. By the mid-1990s, the pair had extended targeting to the subthalamic nucleus (STN), showing in a landmark 1998 NEJM paper that DBS could address all three cardinal Parkinson’s symptoms - bradykinesia, rigidity, and tremor - while enabling radical medication reduction. The neurologist David Marsden called STN-DBS “the most important discovery since levodopa.”

A DBS system comprises three implanted components: a quadripolar electrode lead with platinum-iridium contacts (1.27 mm diameter, 1.5 mm contact length), an extension cable, and an implantable pulse generator (IPG) placed subcutaneously in the chest. Standard stimulation parameters are 130 Hz frequency, 60 microsecond pulse width, and 1-5 volts amplitude, delivering rectangular biphasic charge-balanced pulses. The mechanism remains debated - theories include depolarisation blockade, disruption of pathological beta-band (13-35 Hz) oscillations, and an “informational lesion” that overwrites pathologically encoded signals - but clinical effectiveness is unequivocal. Meta-analyses across 39 studies show STN-DBS improves motor scores by 50.5% on the UPDRS-III scale.

More than 244,000 patients worldwide have received DBS implants across indications including Parkinson’s disease, essential tremor, dystonia, epilepsy, and obsessive-compulsive disorder. The technology continues to evolve rapidly. Boston Scientific’s Cartesia directional leads use segmented contacts in a 1-3-3-1 configuration, enabling current steering that widens the therapeutic window by over 41% compared with omnidirectional stimulation. Most significantly, in February 2025 the FDA approved the first adaptive closed-loop DBS system - Medtronic’s BrainSense Adaptive - which senses local field potential beta-band power in real time and automatically adjusts stimulation amplitude, reducing energy delivery by up to 76% while maintaining symptom control.

A digital bridge across the broken spinal cord

The youngest technology among the four laureate groups - and arguably the most audacious - comes from Grégoire Courtine, a neuroscientist at EPFL, and Jocelyne Bloch, a neurosurgeon at Lausanne University Hospital. Their approach exploits a century-old finding in spinal physiology: the lumbar spinal cord contains locomotor circuits capable of producing coordinated walking patterns without input from the brain. The challenge was how to reactivate them.

Their epidural electrical stimulation (EES) platform places a 16-electrode paddle array (in a 5-6-5 configuration) over the dorsal surface of the lumbosacral spinal cord, targeting the dorsal root entry zones - the lateral points where sensory nerve roots enter the cord. Unlike conventional spinal cord stimulators designed for pain (which target midline dorsal columns), this approach recruits the proprioceptive feedback circuits that coordinate muscle activation during movement. Computational models showed that each electrode’s position relative to specific root entry points determines which muscle groups are activated, enabling spatially selective, temporally precise stimulation that mirrors natural motor neuron firing patterns during gait.

Results have been dramatic. In the 2018 STIMO trial, published in Nature, three participants with chronic incomplete spinal cord injuries walked with body-weight support within one week of starting stimulation. After five months of training, all three regained voluntary leg movement even with stimulation switched off - evidence of activity-dependent neuroplasticity. A 2022 follow-up in Nature Medicine pushed further: three patients with complete sensorimotor paralysis - no voluntary movement or sensation - could stand, walk, cycle, and swim within a single day of stimulation personalisation. One patient subsequently walked 500 metres independently.

The team’s most striking achievement appeared in Nature in May 2023: a brain-spine interface (BSI) that creates what they call a “digital bridge” across the injury. Patient Gert-Jan Oskam, paralysed from a cycling accident a decade earlier, received two WIMAGINE cortical implants - each a 64-electrode ECoG grid in a 50 mm titanium disc matching skull thickness - placed epidurally over the sensorimotor cortex. External antennas power the implants inductively at 13.56 MHz and receive decoded brain signals at 402-405 MHz. AI-based algorithms running on a portable processor decode movement intentions with approximately 100 milliseconds end-to-end latency and wirelessly command the spinal stimulator to activate the appropriate leg muscles. “The stimulation before was controlling me,” Oskam said. “Now I am controlling the stimulation by my thoughts.” After 40 rehabilitation sessions, his voluntary leg control improved even with the BSI switched off.

Courtine and Bloch co-founded ONWARD Medical (Euronext: ONWD) to commercialise their technology. The company’s ARC-EX, a non-invasive transcutaneous spinal stimulator for upper-limb recovery, received FDA de novo authorisation in December 2024 - the first FDA-approved non-invasive spinal cord stimulation device for spinal cord injury - and began US commercial sales in January 2025. Its pivotal Up-LIFT trial showed 72% of participants improved in both hand strength and function. The implantable ARC-IM system is in clinical trials, with the first lumbar lead implanted in March 2025, and a next-generation 60-electrode paddle array with embedded active multiplexing is under development.

What this prize means for neural engineering’s trajectory

The QEPrize, launched in 2012 and worth £500,000, has previously recognised the internet (2013), GPS (2019), LED lighting (2021), PERC solar cells (2023), modern wind power (2024), and machine learning (2025). Its 2026 selection reflects a deliberate choice to honour not a single technology but an interconnected engineering discipline. Cochlear implants proved that the nervous system could reliably interface with electronics over decades. DBS demonstrated that stimulation could modulate deep brain circuits reversibly. BrainGate showed that cortical signals could be decoded in real time. Spinal cord stimulation united all three principles - recording, decoding, and stimulating - into a closed-loop system that bridges brain to body.

The field is converging on several frontiers. Adaptive closed-loop systems are now clinically approved for DBS and under development for cochlear implants. Fully implantable BCIs from companies like Neuralink and Paradromics promise to eliminate percutaneous connections. ONWARD Medical’s brain-spine interface is pursuing commercial development with European Innovation Council support. And AI-driven decoding - already achieving near-perfect speech recognition from cortical signals - is collapsing the gap between neural intention and digital action. The nine laureates did not merely invent devices. They established that the nervous system is an engineering substrate, amenable to the same disciplined approach of measurement, modelling, and design iteration that built every other engineered system. That principle, more than any single electrode or algorithm, is what the 2026 QEPrize recognises.

Engineering explainer: how neural interfaces bridge biology and silicon

The fundamental challenge of neural interface engineering is transduction between two incompatible electrical systems. Neurons communicate using ions - sodium, potassium, and chloride flowing through membrane channels - generating voltage changes of 50-500 microvolts for action potentials (spikes lasting ~1 millisecond) and 10-100 microvolts for population-level signals. Electronic circuits use electrons in metal and semiconductor. The electrode-tissue interface is where these two worlds meet, and its engineering determines everything about a device’s performance and longevity.

Signal types form a hierarchy of invasiveness and resolution. Intracortical microelectrodes (like the Utah array) detect individual neuron spikes with sub-millimetre precision but require brain-penetrating surgery. Electrocorticography (ECoG) grids placed on the brain surface capture aggregate signals from neural populations at millimetre resolution without penetrating tissue - the approach used in the brain-spine interface’s WIMAGINE implants. Scalp EEG avoids surgery entirely but offers only centimetre-scale resolution and signals heavily attenuated by the skull. The trade-off is absolute: better signals demand closer proximity to neurons.

Electrode materials must simultaneously conduct charge, resist corrosion in warm saline, and avoid damaging tissue. Platinum-iridium alloy is the clinical workhorse, used in cochlear implants and DBS leads, with a charge injection capacity of 50-150 µC/cm². Iridium oxide coatings push this to 750-5,000 µC/cm², critical for microelectrodes that must deliver more charge per unit area. The Shannon safety limit - expressed as k = log(D) + log(Q), where D is charge density and Q is charge per phase - sets the boundary beyond which stimulation risks tissue damage, typically at k = 1.75 for macroelectrodes. Impedance matters too: a DBS electrode at roughly 6 mm² presents 1-10 kΩ at 1 kHz, while a 300 µm² microelectrode reaches ~1 MΩ, demanding lower-noise amplifiers for recording.

Biocompatibility is a time-dependent problem. Implanting any foreign object triggers an immune cascade: microglia activate within hours, astrocytes proliferate over weeks, and a dense glial scar encapsulates the device within six weeks. For recording electrodes, this scar increases impedance and pushes neurons away from electrode tips, degrading signal quality. The mechanical mismatch between rigid silicon (Young’s modulus ~130 GPa) and brain tissue (~0.5-10 kPa) - a factor of roughly one million - drives chronic inflammation through micromotion. Cochlear implants and DBS leads sidestep this partly through larger electrodes and stimulation rather than recording, but intracortical arrays remain vulnerable. Hermetic sealing protects implanted electronics from the body’s corrosive environment: titanium enclosures with alumina ceramic feedthroughs achieve helium leak rates below 10⁻¹² mbar·L/s, qualifying packages for projected lifetimes exceeding 100 years.

Power and data cross tissue without wires through inductive coupling - a primary coil outside the body drives an alternating magnetic field that induces current in a secondary coil beneath the skin. Cochlear implants operate their RF links at 5-49 MHz with practical efficiencies around 49%. The brain-spine interface powers its cortical implants at 13.56 MHz and receives data at 402-405 MHz. For fully implanted systems like DBS pulse generators, rechargeable lithium-ion batteries with wireless recharging via electromagnetic induction provide operational lifetimes of nine or more years. Tissue heating constrains all wireless systems: FDA guidelines require implants to raise local tissue temperature by no more than 1°C, limiting intracranial power dissipation to approximately 7.7 mW.

Decoding neural signals into useful commands has progressed from simple linear filters to sophisticated real-time machine learning. The BrainGate system digitises 96 channels at 30 kHz (roughly 23 Mbps of raw data), extracts spike-rate features every 20 milliseconds, and applies a Kalman filter to predict intended movement velocity. Modern speech neuroprostheses use recurrent neural networks to decode phonemes every 80 ms, achieving near-human accuracy. Closed-loop systems - record, decode, stimulate - now operate clinically: adaptive DBS senses pathological beta oscillations and adjusts stimulation automatically, while the brain-spine interface decodes cortical movement intentions and delivers precisely timed spinal stimulation with 100 ms latency. The boundary between neural and digital is, for the first time, becoming bidirectional, adaptive, and fast enough to feel natural.