Horizons

What This Becomes, and What Would Prove Us Wrong

Most deep-tech forecasts are wishful thinking with a timeline attached. This page keeps apart three things that usually get blurred: what is already moving, what we are betting on, and what physics will not allow no matter who funds it.

Dependency, Not Prediction

Each horizon depends on the one before it. Nothing in the second is reachable without the hardware from the first, and the third is as much a clinical and regulatory problem as a technical one.

Three horizonsEach gated by the last

Each stage is a precondition, not a phase of a marketing plan. If uncooled detector arrays do not arrive, the clinical horizon does not slip. It does not happen.

Hardware: Closing the Gap

0-5 years, already in motion

Portable sources. Quantum cascade lasers now reach temperatures a compact thermoelectric cooler can hold. The remaining work is packaging, yield and lifetime rather than new physics, which is a very different and far more tractable kind of problem.
Uncooled focal-plane arrays. The single highest-leverage development in the field. Terahertz imaging has meant a cooled single pixel raster-scanned over minutes. Video-rate uncooled arrays turn that into an instrument a person can hold and point.
Terahertz in silicon. CMOS and SiGe transceivers working in the hundreds of gigahertz mean front ends that cost tens of dollars rather than tens of thousands. Cheap changes what applications are conceivable.
The 6G tailwind. Telecommunications is investing heavily in sub-terahertz components for the next wireless generation. We do not need that standard to succeed. We need the supply chain it creates, and somebody else is paying to industrialise it.
Computational imaging. Compressive and single-pixel methods trade computation for detector count, and learned reconstruction extracts more from a starved signal than classical inversion does.

Clinic: Imaging Without a Dose

5-15 years Gated by hardware

Burn depth as a measurement. The first application where terahertz should become standard of care, because it answers a question clinicians currently answer by eye, and that answer decides whether tissue is cut away.
Intraoperative margins. Tumour boundaries found while the patient is still open, rather than days later in pathology followed by a second operation. The clinical and economic case is unusually clean.
Endoluminal probes. The honest answer to penetration depth is to stop shining through a person and deliver the sensor to the tissue: catheter and endoscope-borne terahertz. This is where terahertz reaches organs.
Wound and ulcer monitoring. Zero dose makes daily quantitative imaging reasonable, so diabetic foot ulcers and graft viability can be followed on a schedule no radiation-based modality could support.
Cost collapse changes geography. A solid-state instrument has nothing like the economics of a cryogenic magnet in a shielded room. The target is diagnostic imaging a rural clinic owns outright, which is where “For All Mankind” stops being a slogan.

Restoration: The Wiring Problem

15-30 years Gated by hardware and clinic

Fascicular mapping as routine imaging. Patient-specific nerve topography before the incision, turning electrode placement from exploratory craft into a deterministic procedure.
Bidirectional prosthetics as standard of care. Not a research demonstration for a handful of participants but the default fitting: limbs with graded force control that report pressure and position back to their owner.
Sensation as the metric that matters. The field’s own data is clear that feedback, not degrees of freedom, decides whether a prosthesis is used or abandoned. Embodiment is the outcome measure.
Bridged spinal pathways. Walking restored by decoding intent and stimulating below the lesion. It has been shown in individuals; the work is making it robust, implantable and affordable.
Sealed, ambient-powered implants. No wire through the skin, which is the dominant chronic failure route, and no charging ritual. Routine living tops the device up.

Five Assumptions We Are Exposed To

These are positions, not predictions. Being wrong about any one of them costs years, and writing them down is how you find out early.

Supply chain

Someone Else Industrialises Our Components

6G research and automotive sensing pull sub-terahertz devices into volume manufacturing, and we inherit the supply chain. A terahertz company that must fabricate everything itself is a far worse business.

Imaging

Computation Beats Penetration

Water absorption will not improve. Reflection-mode coherent reconstruction, frequency agility and learned priors can extract clinically useful depth from a signal classical inversion treats as hopeless.

Interfaces

Peripheral Interfaces Win Over Cortical Ones

No craniotomy, no glial scarring against a cortical array, no asking the brain to learn an artificial code. The lower-risk path is also the higher-fidelity one.

Capital

Defence Funding Is the Patient Capital

Closing a forty-year hardware gap is not a venture-timescale problem. Government programmes fund the components, and the same components then serve medicine, the path GPS, the internet and modern imaging all took.

Culture

Honesty Is a Competitive Advantage

Terahertz has a history of overclaiming, and the field carries scar tissue from it. Publishing our limits attracts the physicists, programme managers and clinicians who can tell the difference.

What Would Show This Is Wrong

A roadmap that cannot fail is not a roadmap. Here is what should make us stop, and what should make you stop believing us.

IfUncooled Arrays Do Not Arrive

If sensitive terahertz detection still needs cooling and single-pixel scanning in ten years, the clinical horizon does not slip. It closes. Every portable application assumes video-rate uncooled imaging.

IfReconstruction Plateaus at the Surface

If computational methods cannot push useful imaging meaningfully past a millimetre in living tissue, terahertz remains a superb dermatology and intraoperative tool and never becomes a general diagnostic one. That is a real business. It is not this one.

IfChronic Interfaces Keep Degrading

If foreign-body response and micromotion still erode neural signal quality over a handful of years, restoration is a temporary intervention rather than a life returned. Materials, not algorithms, decide this.

IfNerve Contrast Is Insufficient

The fascicular mapping claim rests on terahertz sensitivity to myelin and bound water being strong enough to resolve individual fascicles in vivo. If that contrast does not hold outside excised tissue, the deterministic-placement argument collapses.

What No Amount of Money Changes

These constraints belong to the universe, not to our engineering. Any roadmap that quietly assumes they will soften is not worth your time.

4.1meV

Terahertz will always be non-ionising. Photon energy at 1 THz is roughly two thousand times below the threshold to strip an electron. That is permanent.

~200cm⁻¹

Water will always absorb it. The absorption of liquid water at terahertz frequencies is a material property. Engineering routes around it; nothing removes it.

λ/2

Diffraction sets the resolution. At 1 THz the wavelength is 300 µm, so far-field imaging resolves around 150 µm. Finer detail needs near-field probing, which only works at surfaces.

1×1

No aperture means no direction. A single antenna cannot infer bearing. Position needs many spatially separated radios or genuine array bandwidth; better processing of one link will never produce it.

Where that hardware gets pointed is a choice.

Two hundred years of electromagnetism left one band we never learned to use, sitting exactly where biology is most legible and radiation damage is impossible. The hardware to open it is arriving now, mostly paid for by people who want faster wireless. We would rather it were pointed at finding people under rubble, at telling a surgeon where a burn actually ends, and at giving someone back a hand they can feel.

How We Got Here Current Hard Problems