Seeing the human body as it truly is
There is one band of the electromagnetic spectrum we have never learned to use. It passes through clothing and packaging, reads water and biology with molecular precision, and carries far too little energy to damage a single chemical bond. Elpis is opening it — and using it to give people their bodies back.
From the moment of injury to a life returned
Elpis builds terahertz sensing systems: instruments that look inside living tissue without radiation, without contrast agents, and without cutting. We are pursuing one capability along three deployments, each funding the next, and all of them converging on the same end.
Every one of those problems is the same problem wearing different clothes: we cannot see inside a human being quickly, safely, cheaply, or in enough detail. X-rays ionise. MRI costs millions and cannot move. Ultrasound is blind to what it cannot couple into. Terahertz is the band that could answer all of it — and it is the one band we never built the hardware for.
The last unclaimed band of the spectrum
Between the radio waves we generate with electronics and the light we generate with lasers lies a stretch of spectrum where neither approach works well. Engineers have called it the terahertz gap for forty years. It is not a gap in nature. It is a gap in our toolmaking.
Coming from below: electronics run out of speed
Every transistor has a ceiling. Charge takes time to cross the device, and parasitic capacitance fights you harder the faster you switch. Push past roughly 100 GHz and output power falls away steeply — by the time you reach 1 THz you are working in microwatts, and every stage of frequency multiplication throws most of your power away as heat.
Coming from above: photons get too faint to make
A laser needs an energy step for electrons to fall down. At 1 THz that step is about 4 meV — while room-temperature thermal energy is about 26 meV. The heat in the material is six times larger than the signal you are trying to create. The transition drowns in its own noise.
Source side. Solid-state electronic sources (Gunn, IMPATT, resonant tunnelling diodes, Schottky multiplier chains) are transit-time and parasitic limited. Output power above ~100 GHz falls approximately as 1/f², and multiplier chains compound conversion loss at every stage. On the photonic side, interband semiconductor lasers cannot reach the band because no practical material has a bandgap of a few meV. Quantum cascade lasers sidestep this with intersubband transitions in engineered superlattices, but thermal backfilling and non-radiative LO-phonon scattering force cryogenic or thermoelectrically-cooled operation for most of the band; sustained room-temperature operation below ~2 THz remains a live research frontier.
Detector side. The same energy scale defeats direct detection. With hν ≈ 4.1 meV at 1 THz against kBT ≈ 25.7 meV at 300 K, photon-counting is not available: thermal fluctuation dominates. Practical detection means bolometers and pyroelectrics (sensitive but slow, often cooled), Schottky or field-effect rectifiers (fast, less sensitive), or heterodyne receivers — which need a local oscillator, and so inherit the source problem they were meant to escape.
Why it matters commercially. The gap is not a physics prohibition, it is a technology maturity trough. Every adjacent band has had decades of industrial investment; this one has not. That is precisely the argument for building here: the science is settled, the engineering is not, and the first organisation to close it owns the instrument class.
Why this band reads biology better than any other
Terahertz is not a slightly different X-ray. It interacts with the body through an entirely different mechanism, and that difference is the whole opportunity.
Safe by physics, not by protocol
Radiation safety in medicine is usually a matter of budgeting harm: a CT scan delivers a real dose, and clinicians weigh that against the value of the picture. Terahertz removes the trade-off at its root. The energy per photon is thousands of times below the threshold to ionise anything, so there is no dose to accumulate and no lifetime limit to track. A terahertz scan can be repeated every hour, on a child, on a pregnant patient, on the same wound daily for a month.
Dielectric contrast dominated by water. Liquid water has an absorption coefficient on the order of 200–250 cm⁻¹ at 1 THz, driven by Debye relaxation of the hydrogen-bond network plus intermolecular librational modes. Soft-tissue penetration is therefore sub-millimetre in transmission. Consequently Elpis's imaging architecture is reflection-mode and computational, not transmission — we reconstruct depth from time-of-flight and dispersion in the returned pulse rather than shining through the patient.
THz time-domain spectroscopy. THz-TDS gates a photoconductive or electro-optic detector against a femtosecond pulse and samples the electric field as a function of time, not merely intensity. Recovering amplitude and phase together yields the complex refractive index directly, with no Kramers–Kronig inversion and no assumption about the sample. That is the single most under-exploited measurement advantage in the band.
Collective and conformational modes. Below ~6 THz, biomolecular spectra are dominated by intermolecular and low-frequency intramolecular collective modes — phonon-like vibrations of the whole structure and of its hydration shell. In vitro, THz-TDS has distinguished single- from double-stranded DNA, resolved hybridisation state without fluorescent labelling, and shown sensitivity to methylation and conformational change. These are structural and state-level readouts rather than sequence readout, and the distinction matters.
We state this plainly because the field is full of overclaiming, and because anyone technical will ask. Terahertz does not today image DNA inside a living body. Water absorption limits in-tissue penetration to fractions of a millimetre, and molecular-state discrimination has been demonstrated on thin films and prepared samples, not through skin. What is real today is superb near-surface structural and hydration imaging, with molecular specificity, at zero radiation risk. Extending that inward — through better sources, coherent reconstruction, and endoluminal delivery — is the engineering programme, and we would rather be measured against an honest roadmap than a fashionable one.
Three deployments, one capability
Each phase is a real market that pays for itself, and each one hardens the same core technology the next phase depends on. This is deliberate: the sources and detectors that survive a battlefield are the ones cheap and rugged enough to end up in a clinic, and precise enough to eventually map a nerve.
- Standoff threat detection. Terahertz passes through fabric, paper, plastic and ceramic, and stops at skin and metal. Concealed-object screening without an X-ray backscatter dose and without a physical search.
- Field triage. Burn depth, wound geometry, foreign-body location and tissue viability assessed at the point of injury by a handheld instrument, with no shielding, no consumables and no radiological officer.
- Search and rescue. Paired terahertz and millimetre-wave sensing to find breathing bodies through rubble, smoke and darkness — the millimetre band for penetration, the terahertz band for identification once close.
- Materials integrity. Non-destructive inspection of composite armour, airframes and coatings, where terahertz already outperforms ultrasound on layered dielectrics.
Defence funding is the only capital patient enough to close a hardware gap forty years wide. The specific deliverables — compact room-temperature sources with useful output power, sensitive uncooled detector arrays, and real-time coherent reconstruction — are exactly the components that gate every civilian application downstream. We are not detouring through defence; we are building the instrument there because that is where the instrument can be afforded.
- Burn depth, objectively. Depth of a burn is still judged largely by a clinician's eye, and that judgement decides whether tissue is excised or left. Terahertz reads the hydration gradient directly — the physical quantity that actually defines the injury.
- Surgical margins in real time. Delineating tumour from healthy tissue while the patient is still open, rather than waiting days for pathology and returning for a second operation.
- Dentistry without radiation. Caries and enamel demineralisation imaged with no dose at all, which changes what is reasonable to do routinely and to children.
- Wound and ulcer monitoring. Daily, quantitative healing assessment for diabetic ulcers and grafts — impossible when each look costs a radiation dose or a laboratory.
- Cost and access. The unit economics of a solid-state instrument are nothing like a cryogenic magnet in a shielded room. The goal is diagnostic imaging that a rural clinic can own outright.
- Map the wiring. Peripheral nerve pathways imaged non-invasively and patient-specifically, so an interface is placed against known anatomy rather than surgical estimate.
- Fine motor control. Prosthetics driven from many independent nerve channels — individual finger control, graded force, coordinated grasp.
- Sensation returned. Stimulating sensory fibres to close the loop, so a limb reports pressure, texture and position back to its owner. A hand you cannot feel is a tool. A hand you can feel is yours.
- Walking again. Bridging damaged spinal pathways to restore volitional control of the legs.
- Strength beyond baseline. Powered exoskeletal augmentation driven by the same intent signals — for rehabilitation first, and for capability after.
The nerve is the interface. Not the brain.
The most consequential engineering decision Elpis has made is a decision about where not to operate. We do not interface with the brain, and we do not intend to.
Why peripheral, not cortical
A cortical implant requires opening the skull, accepts a permanent infection and haemorrhage risk, provokes glial scarring that degrades the signal over years, and asks the brain to learn an artificial code. A peripheral nerve already carries the exact signals a limb used to obey — the motor commands are still being sent. The wiring is intact right up to the point of loss.
Working peripherally means no neurosurgery, a reversible procedure, no cortical remapping, and a signal that is already in the native language of the limb. It is the lower-risk path and the higher-fidelity one.
Why terahertz is the enabling instrument
Peripheral interfaces fail for a mundane reason: surgeons cannot see the fascicles. Nerve topography varies between individuals, and a fascicle carrying thumb flexion in one person sits elsewhere in the next. Electrodes are placed, then mapped by trial and error afterward.
Terahertz is exquisitely sensitive to myelin and to bound versus free water — precisely the contrast that separates fascicle from connective tissue. Imaging that structure before and during placement turns nerve interfacing from a craft into a procedure.
Closing the loop: sensation
Restoring movement is the visible half of the problem and the easier one. The half that determines whether someone actually uses a prosthesis is feedback. Without it, every action demands constant visual attention, grip force is guesswork, and the limb never stops feeling like equipment. Stimulating the sensory fibres that once served the missing hand produces referred sensation the user localises to the hand itself — and with it, embodiment: the limb stops being worn and starts being inhabited.
Electrode strategy. The design space runs from extraneural cuffs (safest, least selective) through interfascicular designs such as flat-interface and transverse intrafascicular electrodes, to penetrating arrays offering the highest channel counts at the greatest chronic risk. Surgical approaches including targeted muscle reinnervation and regenerative peripheral nerve interfaces amplify efferent signals into recordable muscle territory rather than fighting for microvolts inside the nerve. Elpis's contribution is upstream of this choice: patient-specific fascicular maps that make selective placement deterministic rather than exploratory.
Bidirectional encoding. Afferent stimulation must respect biology to be interpreted as sensation rather than paraesthesia. Biomimetic charge-modulated encoding — matching the transient-dominated firing profile of real mechanoreceptors rather than delivering tonic pulse trains — yields more natural, better-localised percepts and measurably improved force control.
Power. Percutaneous leads are the dominant chronic failure and infection mode, so the implant must be sealed and wireless. The architecture combines mid-field resonant inductive coupling for scheduled high-rate transfer with continuous ambient harvesting — thermoelectric gradient across tissue planes, piezoelectric and triboelectric capture of gait and joint motion — sized so that routine use trickle-charges the device and the wearer never thinks about it. No connector through the skin. No nightly ritual.
| Capability | What it restores | Gate |
|---|---|---|
| Fascicular nerve mapping | Deterministic electrode placement, patient-specific | THz depth & resolution |
| Multi-channel efferent decode | Individual digit control, graded force | Channel count & stability |
| Biomimetic afferent encode | Pressure, texture, proprioception — embodiment | Encoding fidelity |
| Spinal pathway bridging | Volitional standing and walking | Clinical trial pathway |
| Sealed wireless power | Chronic implant with no percutaneous lead | Harvest energy budget |
What stands between here and there
We publish our obstacles because the people we most want to hear from — physicists, RF engineers, neural interface researchers, clinicians — will identify them within minutes anyway. If you can solve one of these, we want to talk to you.
01 — Penetration depth
Water absorption confines terahertz to fractions of a millimetre in soft tissue. Every deep-imaging ambition runs into this first. Our approaches: reflection-mode coherent reconstruction, frequency-agile operation trading resolution for depth, endoluminal and catheter-borne delivery, and hybrid operation with millimetre-wave bands that penetrate further at coarser resolution.
02 — Source power at room temperature
The gap itself. Useful output power from a compact, uncooled, manufacturable source remains the single largest constraint on every application on this page. This is where defence funding is directed first, because nothing downstream improves until it does.
03 — Detector arrays and frame rate
Sensitive terahertz detection has historically meant cooled, single-pixel, raster-scanned instruments. Clinical and field use demands uncooled focal-plane arrays at video rates. Progress here converts a laboratory technique into an instrument someone can hold.
04 — Chronic interface stability
Neural interfaces degrade. Foreign-body response, micromotion and encapsulation erode signal quality over months to years. A restoration that works for one year and fails in the fifth is not restoration. Materials, mechanical compliance and surgical technique all have to hold.
Elpis will not claim a capability it has not measured, and will publish the limits of every instrument it ships alongside its performance. Medicine and defence both have long histories of technologies oversold into the gap between a demonstration and a deployment. We would rather be trusted than admired.
A person should not lose their life to an injury we could have seen
Elpis exists because the distance between an injury and a life returned is measured in things we could not see in time. A body under rubble. A bleed inside a chest. A nerve that still carries every signal a hand would need, waiting for something to listen. The band that could answer all of it has been sitting unused between radio and light for a century. We are going to use it.