Seven readings of the question, constructed to span the space of answers, each with its strongest case and its real cost, and eleven cases that force them apart. Built for argument, not for settlement.
Neuroengineering has no agreed definition. That is not a failure of tidiness. It is what you should expect from a field assembled in the last few decades out of electrophysiology, materials science, control theory, clinical medicine, and computer engineering, each of which arrived with its own account of what the work is for.
The usual shorthand is applied neuroscience, or applying engineering principles to solve neuroscience problems. Both are defensible. Both are also contested, and the arguments against them are not pedantic: they change what belongs in a curriculum, what counts as a contribution to the field, which proposals a study section considers in scope, and whether a given laboratory is doing neuroengineering or something adjacent to it.
This page sets out seven positions, gives each its strongest form, names what each one costs, and then tests all seven against cases chosen because they pull the definitions apart. It does not tell you which to hold.
Where the seven come from matters, so it is stated plainly. They were not collected from a survey of what practitioners believe. They were built here, as an attempt to span the space of defensible readings, and only then checked against what the field has actually published. Three of the seven map onto published definitions, one onto journal practice that is enacted rather than stated, and three have no published source at all. Each card below says which it is, and the sourced definitions further down show the checking.
There is an 19 slide opener deck built to introduce this discussion, with facilitator notes and timings on every slide (press S in the deck for speaker view). It runs about 50 to 60 minutes with breakout discussion and deliberately ends without an answer.
For a seminar: read the seven positions, pick the one closest to your own view, then work the matrix and find the case that embarrasses it most. Everyone will have one. The productive argument is usually not "which definition is right" but "which exclusions are you willing to live with."
Neuroengineering is the deployment of principles drawn from nervous systems toward practical ends, whatever those ends happen to be.
It captures the dependency practitioners actually feel. It is also the only reading that comfortably contains neuromorphic engineering, where organizing principles are taken from nervous systems and applied to computation. Any definition that has to throw out neuromorphic work is in trouble, and this one does not.
It admits neuromarketing, neurolaw, neuroeconomics, brain-inspired deep learning, and neuroarchitecture, all of which deploy neural principles toward practical ends in exactly this sense. If those are outside the field, this reading cannot say why.
Support: none located. No definition in the sample below states this reading outright; the published ones are consistently narrower. It survives as a paraphrase rather than as anyone’s stated position.
Neuroengineering is the deployment of neuroscience findings toward biological, clinical, or neural ends.
Adding the biological end condition fixes the over-inclusion problem cleanly. Neuromarketing and neuroarchitecture fall out, because a sales figure and a corridor are not biological outcomes, while cochlear implants and scaffolds stay in.
It struggles wherever the artifact preceded the finding. The voltage clamp made the Hodgkin-Huxley result possible rather than applying it. DBS was delivering therapy long before anyone could say why it worked. On a strict reading these are excluded, and they are not marginal cases.
Support: none located as a standalone definition. Closest to the "study nervous system function" clause in the dictionary and editorial definitions below, which pair it with a clinical clause rather than letting it stand alone.
Neuroengineering is the application of engineering principles and methods to problems in the study of the nervous system.
It handles the instrument cases that Position 2 cannot. The voltage clamp, high-density probes, and pulse-sequence design are all engineering aimed squarely at a scientific question, and it credits the field for building the tools that made modern neuroscience possible.
It puts the problem in the hands of a scientist. Most of the field's visible output serves a patient instead: prosthetics, implants, stimulation therapies, regeneration. It also cannot easily exclude computational neuroscience, which applies mathematical method to neuroscience questions and is usually considered a different discipline.
Support: the "analyze neurological function" half of the Journal of Neural Engineering editorial board’s definition, which also names "solving neuroscience-related problems" as one of the field's two main goals, and the "study" clause in the Merriam-Webster Medical entry. Neither lets it stand alone.
Neuroengineering is the design of systems that exchange information or energy across the boundary between engineered devices and nervous systems, whether to read, to write, or to replace.
It matches the field's textbook structure almost exactly: acquisition, modulation, prosthesis. It is concrete, it is teachable, and practitioners recognize themselves in it without argument.
It has no room for work where nothing crosses a boundary. Neuromorphic computing is out. So, arguably, is most computational modeling, and so is tissue engineering where the coupling is mechanical and chemical rather than informational.
Support: stated in the review literature as an empirical generalization rather than a definition, that "in most cases neural engineering involves the development of an interface between electronic devices and living neural tissue". That describes what the work usually is, which is weaker than a criterion for membership.
Neuroengineering is engineering directed at restoring, replacing, or augmenting nervous system function in living organisms.
It names the field's actual purpose and its actual accountability. It explains why regulatory pathway, chronic reliability, and patient outcome are first-class engineering concerns here in a way they are not in adjacent disciplines, and it draws a bright line against work that only produces papers.
It excludes diagnostics and measurement, which is to say most of neuroimaging. An EPI sequence restores no function. It also excludes basic instrument development, and it makes the field's own history unintelligible, since the tools came first.
Support: the "design solutions to problems associated with neurological limitations and dysfunction" half of the editorial board’s definition, and close to the NINDS funding criterion, though that one also covers normal function.
Neuroengineering is engineering whose binding design constraint is a living nervous system, meaning that removing the nervous system from the problem would materially change the design or remove its reason to exist.
Engineering disciplines are usually individuated by their governing constraints rather than their subject matter, as chemical engineering is by scale-up and transport rather than by chemistry. This account handles instruments, therapies, and materials in one move, and it explains why the field developed metrics no neuroscientist needs.
"Binding" does a great deal of work and resists sharp definition. The account also splits neuromorphic engineering, keeping interfacing work and excluding pure accelerators, which many practitioners will reject as carving up a coherent research community to save a definition.
Support: none located. This is a reconstruction. No source in the sample below defines the field by its design constraints, so it is offered to complete the space of options rather than reported from anyone.
Neuroengineering is what the field's institutions treat as neuroengineering: the work published in its journals, funded under its programs, and taught in its departments.
It is the only account that never miscounts, because it takes the practice as given rather than legislating it. Disciplines are social facts, and family resemblance may be all that holds any of them together. It also predicts membership better than any rival here.
It is circular, and it cannot criticize. If a department teaches something incoherent, this account has no standing to say so, and it gives no guidance at the boundary, which is the only place a definition is ever needed.
Support: the strongest of the seven, and from an unexpected direction. The editorial that states the field's official definition ends by saying the definition and scope "are best determined by the scientists and engineers that practice it", and its twenty-one entry scope list claims neuromorphic engineering, computational and systems neuroscience, neuroinformatics and neuro-diagnostics as territory.
Since the seven above were constructed rather than collected, the check that matters is what the field has actually put in writing. A short search turns up the following. This is a sample, not a survey, and the pattern in it is more interesting than any single entry.
| Source | What it says | Maps to |
|---|---|---|
| Durand,
"What is Neural Engineering?" J. Neural Eng. 4(4), 2007 |
The editorial board's definition, written because "the scope of the field has evolved and a clear definition of neural engineering is needed": an emerging interdisciplinary research area that brings to bear neuroscience and engineering methods "to analyze neurological function as well as to design solutions to problems associated with neurological limitations and dysfunction." It names the goal as two things at once, to "solve neuroscience-related problems and to provide rehabilitative solutions", and it does its own boundary work in both directions, separating the field from neurophysiology by its "emphasis on engineering and quantitative methodology" and from other engineering disciplines "such as artificial neural networks" by its integration with neuroscience. | 3 and 5, closing on 7 |
| The standard encyclopedic and review definition | A discipline within biomedical engineering using engineering techniques to "understand, repair, replace, enhance, or otherwise exploit" the properties and functions of neural systems. | 3, 5, and a catch-all wide enough to reach 1 |
| Merriam-Webster Medical | The application of engineering principles and techniques to neuroscience, especially to study, restore, or enhance nervous system function. | 3 and 5 |
| Journal of Neural Engineering scope | The scope list printed in the same editorial runs to twenty-one entries. Alongside the expected neural interfacing, neuromodulation and neural prostheses it claims neuromorphic engineering, theoretical and computational neuroscience, systems neuroscience, neuroinformatics, neuroimaging and neuro-diagnostics. | 7, and it claims territory 3, 4, 5 and 6 each exclude some of |
| NINDS funding scope | Technology development relevant to normal or disordered neural function and to the prevention, repair or replacement of compromised neural function. | 5, widened to include normal function |
| The first IEEE EMBS Neural Engineering
conference Capri, March 2003 |
Convened as the first gathering for a field understood as distinct from both neuroscience and biomedical engineering. | Contradicts the encyclopedic definition above |
| Review literature | Generalizes rather than defines: in most cases the work involves developing an interface between electronic devices and living neural tissue. | 4, as description rather than criterion |
No published definition here is a pure instance of any single position. The most considered attempt, from the flagship journal's editorial board, is a conjunction of Positions 3 and 5, joining a scientific purpose to a clinical one without saying how they relate. That same journal's scope is then considerably broader than the definition, including work the definition does not obviously cover.
There is also a flat contradiction in the record. The encyclopedic definition calls neuroengineering a subdiscipline of biomedical engineering; the field's founding conference was convened on the premise that it was distinct from biomedical engineering. Both statements remain in circulation, and neither has displaced the other.
Most striking is how the editorial ends. Having just given the field's definition, the editor-in-chief declines to legislate it: "the definition and scope of neural engineering are best determined by the scientists and engineers that practice it and this is only an overview of the field as it is understood today." The future of the field, it closes, "will be determined not by what we believe neural engineering should be but by its success in improving human health and quality of life." That is Position 7 offered by the very document that states the official definition, with a final lean toward Position 5 as the test that will actually decide the matter.
The same editorial also does boundary work that cuts finer than any position here. It separates the field from neurophysiology by its "emphasis on engineering and quantitative methodology", and from other engineering disciplines "such as artificial neural networks" by its integration with neuroscience. Note what that pairing implies: artificial neural networks are outside, while neuromorphic engineering is listed inside the scope. The line is drawn not at whether the work touches tissue but at whether the engineering and the neuroscience are still coupled, which is a criterion none of the seven readings above states in those terms.
If you came here looking for the field to have settled this, it has not, and the document that tried says so itself.
Eleven cases, chosen because they separate the accounts. No two positions agree across the whole table. The shaded rows are the hardest four, where no reading commands more than three of the seven votes, and those are the ones worth an argument. The rows that are nearly unanimous are worth noticing too, since agreement on a case that no single definition explains well is itself a finding.
Select any cell to read why that position classifies that case as it does, or a case name to compare all seven at once.
Select a definition to read its reasoning case by case. Nothing here is scored against a correct answer, because that is the thing in dispute.
Four bodies of evidence come up repeatedly. Each supports some positions and embarrasses others, and each has a standing reply.
Hodgkin and Huxley needed a voltage clamp before they could write their 1952 equations. Single-channel recording did not exist until the patch clamp was built in 1976. Human functional imaging went from a handful of well funded PET centers to a mass activity once BOLD contrast, described in 1990, removed the radiotracer, the cyclotron and the radiation dose from the requirements. If many of the field's largest contributions to neuroscience have been instruments, then engineering is generative here rather than derivative, which pressures Positions 1 and 2.
The replyInstrument builders were themselves steeped in the science of their day, and "application" need not mean applying a finished result. Cole and Marmont were applying an understanding of membrane behavior that already existed, just not yet in quantitative form. The imaging case is messier still: BOLD contrast was itself a scientific finding, later engineered into a method, so the arrow runs science to engineering to science rather than in one direction. Note also what the claim does not say. Cognitive neuroscience was not created by fMRI; the term predates it by roughly two decades, and PET activation studies, event-related potentials and lesion work had established the questions long before. The honest version of the claim is about scale and access, not about founding a field. On this reading the pipeline story survives; it is simply less tidy than the caricature.
The cochlear implant was pursued over the objections of auditory scientists who had principled reasons to think crude electrical stimulation could not carry speech. DBS at roughly 130 Hz has treated Parkinson's in well over 200,000 patients while its mechanism remains disputed. A field whose successes outrun its explanations is not straightforwardly applying anything.
The replyBoth cases did apply real neuroscience, just selectively. Cochlear implants exploit tonotopic organization, which was well established. DBS targeting rests on basal ganglia circuit anatomy. What was missing was a mechanism for the therapeutic effect, and no engineering discipline waits for complete mechanistic accounts before building.
Chronic yield, safe charge density limits, charge-balanced biphasic pulsing, encapsulation thickness as a design target, decoder recalibration burden, information transfer rate, mean time to explant. No neuroscientist needs any of these; every implant engineer does. A discipline that has invented its own metrics is not merely deploying someone else's results.
The replySpecialized vocabulary emerges inside subfields all the time without establishing independence. Analytical chemistry has terminology general chemistry does not need, and remains chemistry. The existence of native metrics shows maturity, not autonomy.
Neuromorphic work takes principles from nervous systems and applies them to computation. It is housed in neuroengineering departments and published in the field's venues, yet on several accounts here it is out, sometimes because it never touches tissue and sometimes because it serves a computing problem. Every position on this page has to say something uncomfortable about it.
The replyOne response is that neuromorphic work simply is applied neuroscience and the accounts that exclude it are wrong. Another is that the category should be split, with interfacing work inside the field and pure accelerators counted as computer engineering with a biological inspiration. A third is that this is a live disciplinary boundary rather than a puzzle any definition should be expected to settle.
The definitions differ; these constraints are largely common ground. Whatever account you hold, they are rarely the thing anyone argues about, and they are a large part of why the field is worth distinguishing from its neighbors at all.
Glial encapsulation begins within days and matures over weeks, impedance climbs, and the population recorded on day 1 is not the population on day 400. The system reacts to having been designed against.
In a closed-loop interface the decoder adapts to the user while the user adapts to the decoder. Two adaptive systems share one loop, which makes "it worked in testing" a weaker claim here than in most domains.
There is no non-invasive way to watch a single human neuron. Every measurement trades resolution against invasiveness against coverage against longevity, and no budget buys a way out of that trade.
There is one skull, one cochlea, one spinal cord. The failure mode is not a product return, which is why regulatory pathway thinking belongs in the first design meeting rather than the last.
For a bridge the load is given. For a speech neuroprosthesis nobody can hand over the correct neural code, so the specification is co-discovered while the device is built.
Silicon sits near 150 GPa and brain tissue near 1 kPa, roughly eight orders of magnitude apart, in warm salt water that attacks electronics, inside an organ that moves with every heartbeat.
Closed-loop stimulation parameters can shift mood, impulsivity, and the sense of agency. Chile amended its constitution in 2021 to protect neural rights. See Neuroethics.
Cochlear implants ran from early human trials in the 1960s to approval in the mid-1980s. DBS ran from Benabid's 1987 observation to approvals in 1997 and 2002. The feedback loop is measured in decades.
On method and sourcing. The seven positions were constructed for this page rather than gathered from practitioners, then checked against the published record. They are stated in the strongest form they can be given rather than quoted from single sources; each carries a note saying what published support was located for it and, in three cases, that none was. Nobody should cite them as a finding about what the field believes. The definitions in the sample were gathered in a short targeted search rather than a systematic review. The Journal of Neural Engineering editorial is quoted from the primary text and can be relied on; the remaining entries are as reported by the sources linked, and their wording should be checked against the originals before being quoted in print. Dates and figures are given at the resolution the historical record supports, and mechanisms described as contested reflect genuine ongoing disagreement. Corrections, better sources, and additional positions are all welcome.