THz Communication: Why 6G Is Looking Past mmWave

The appeal of terahertz spectrum is arithmetic. A single channel at 300 GHz could be tens of gigahertz wide — more bandwidth in one carrier than most operators hold across their entire licensed portfolio.

The problem is equally arithmetic, and it's the reason THz gets discussed for 6G rather than deployed. Path loss scales with frequency squared. Water vapour absorbs specific frequencies with enthusiasm. And the semiconductors that would generate useful power at these frequencies mostly don't exist yet.

Worth separating two things that get conflated: sub-THz (roughly 100–300 GHz) is what 6G research realistically targets. True THz (0.3–10 THz) is a longer-term research field with a different set of problems.

---

Why the bandwidth is tempting

Shannon's capacity relationship is unforgiving in one direction and generous in the other. You can chase spectral efficiency through better modulation and MIMO, but those gains are logarithmic. Bandwidth scales capacity linearly.

Mid-band 5G gives you 100 MHz. mmWave gives you 400 MHz per carrier. Sub-THz could plausibly offer 5–20 GHz in a single channel.

That's not an incremental improvement — it's a change of category, and it's why the band keeps attracting attention despite everything below.

There's a second motivation that's easy to miss: sensing resolution. Range resolution improves with bandwidth, and at 10 GHz of bandwidth you're resolving objects at roughly 1.5 cm. That makes sub-THz interesting for integrated sensing somewhat independently of its communication value.

---

The propagation problem

Free-space path loss

Free-space loss scales with the square of frequency. Moving from 3.5 GHz to 300 GHz costs roughly 39 dB of additional loss over the same distance.

Array gain buys some back — and at these wavelengths you can pack enormous numbers of elements into a small aperture, which helps considerably. But you don't get 39 dB for free, and you spend it on beams so narrow that pointing them becomes its own engineering problem.

Molecular absorption

This is the part that has no equivalent at lower frequencies, and it's what makes THz genuinely different rather than just "mmWave but worse."

Atmospheric molecules — water vapour primarily, oxygen secondarily — have rotational resonances in this range. At those frequencies the atmosphere doesn't just attenuate the signal, it absorbs it.

The result is a spectrum that isn't uniformly usable. Strong water absorption sits around 183 GHz, 325 GHz, and 380 GHz, with additional peaks above. Between them lie transmission windows where attenuation is far more tolerable.

Two consequences follow:

Usable spectrum is fragmented. You don't get a continuous swathe from 100 to 300 GHz — you get windows, and channel planning has to work around the absorption lines.

Absorption depends on humidity, which means link budget varies with weather in a way that's much more pronounced than rain fade at lower bands. A link engineered for a dry day may not close on a humid one.

Rain adds further attenuation, and at these wavelengths raindrops are comparable in size to the wavelength itself, producing scattering as well as absorption.

Effectively line-of-sight only

At mmWave, diffraction and reflection still provide some non-line-of-sight coverage — degraded, but present. At sub-THz, surfaces that look smooth at 3.5 GHz are rough relative to a sub-millimetre wavelength, so reflections scatter rather than reflect coherently. Diffraction around obstacles becomes negligible.

The practical reading: if you can't see it, you can't reach it. Foliage, walls, vehicles, and human bodies are effectively opaque. A hand over the wrong part of a device blocks the link entirely.

Beam alignment gets hard

The high array gain needed to close the link produces beams a fraction of a degree wide. That creates a problem the specifications haven't fully solved: misalignment loss.

A beam that narrow needs continuous, precise tracking. Small movements — a device rotating in a hand, a vehicle changing lane, thermal expansion in a fixed installation — produce pointing errors that cost significant gain. Beam management overhead scales badly, because you're searching a much finer angular grid.

---

The hardware problem

Propagation is difficult. Hardware is arguably the harder constraint, and it's the reason timelines keep slipping.

Power amplifiers

Output power falls sharply with frequency, and power-added efficiency falls with it. A device delivering watts at 3.5 GHz might deliver milliwatts at 300 GHz — while the link budget is asking for more power, not less.

This is the single biggest blocker. Everything else has a plausible path; transmit power at sub-THz does not currently have a good answer.

Semiconductors

Silicon CMOS runs out of usable gain before 300 GHz. The alternatives each have drawbacks:

SiGe BiCMOS integrates reasonably and reaches usable frequencies, but with limited output power.

InP HBT and HEMT deliver the best power and noise performance at these frequencies, and are expensive with poor integration density.

GaN offers high power and is still maturing above 100 GHz.

There's also the THz gap — the region where electronic approaches run out of speed and photonic approaches (uni-travelling-carrier photodiodes, photonic mixing) haven't yet become efficient or compact. Bridging it is an active research area with no settled winner.

Antennas and packaging

Sub-millimetre wavelengths mean tiny elements, which is convenient for packing large arrays into small apertures. It's inconvenient for everything else.

Feed network losses become severe because conductor loss rises with frequency. Manufacturing tolerances that are irrelevant at 3.5 GHz become significant fractions of a wavelength. Packaging parasitics that could be ignored now dominate. Getting the signal from the die to the radiating element without losing most of it is a genuine engineering challenge.

Data converters

A 10 GHz channel needs sampling rates in the tens of gigasamples per second. ADC power consumption scales roughly with sampling rate and exponentially with resolution.

The consequence is that the converters alone can dominate the power budget, which pushes designs toward low-resolution converters, analog processing, and hybrid architectures — the same pressures that made hybrid beamforming the standard choice at mmWave, only sharper.

---

What it's actually good for

Given line-of-sight operation, short range, and high power cost, the credible applications are the ones that don't need mobility or coverage:

Fixed wireless backhaul and fronthaul. Short, engineered, line-of-sight links between fixed points. Antennas can be large, aligned once, and mains-powered. This is the most realistic near-term deployment by a wide margin.

Wireless data centres. Rack-to-rack links replacing cabling, over metres, in a controlled environment with no weather.

In-room and short-range high capacity. Kiosk-style bulk transfer, XR headsets tethered wirelessly to a nearby compute unit, board-to-board links inside equipment.

Sensing and imaging. High-resolution radar, material characterisation, and security imaging — where the bandwidth matters more than the range.

Notice what's absent: wide-area mobile coverage. Nothing in the propagation or hardware picture suggests sub-THz will provide it, and treating THz as a replacement for mid-band 5G misreads the physics badly. It's a capacity layer for specific, constrained situations.

---

Regulation and timeline

Spectrum work is further along than the hardware. WRC-19 identified bands above 275 GHz for land mobile and fixed services — including portions of 275–296, 306–313, 318–333, and 356–450 GHz — subject to protecting passive Earth-exploration satellite services, which use several of these frequencies for atmospheric science and are extremely sensitive to interference.

That protection constraint matters more than it sounds. Some of the most useful transmission windows sit close to bands radio astronomy and Earth observation depend on, and coexistence rules will shape what's actually deployable.

A realistic sequence:

Now to 2028 — component research, channel measurement campaigns, 3GPP study work. Sub-THz appears in 6G studies, not normative specifications.

WRC-27 — the spectrum decisions that determine what's commercially usable.

2030 onward — first 6G deployments, almost certainly in the upper mid-band (7–24 GHz) rather than sub-THz. That's where the coverage-versus-bandwidth trade-off actually favours deployment.

Beyond 2030 — sub-THz in fixed links and specialised short-range applications, if the power amplifier problem yields.

The honest summary: sub-THz is unlikely to be a mainstream 6G access technology. The interesting 6G spectrum story is the upper mid-band. THz is a research frontier with real applications at the edges — and it makes for better slides than it does for coverage.

---

The takeaway

Bandwidth scales capacity linearly — that's the whole attraction, and it's a real one.

Path loss plus molecular absorption fragments the spectrum into transmission windows and makes link budget humidity-dependent.

Effectively line-of-sight only. Diffraction and coherent reflection largely disappear.

Power amplifier output is the hardest blocker, with no clear solution yet.

Realistic use is fixed links, short range, and sensing — not wide-area mobile coverage.

The upper mid-band, not sub-THz, is where 6G will actually deploy.

---

Further reading

• ITU-R Radio Regulations — WRC-19 identifications above 275 GHz
• ITU-R P.676 — Attenuation by atmospheric gases
• ITU-R M.2160 — IMT-2030 framework
• Integrated sensing and communication
• What is 6G?