Antenna-Coupled TeraFETs Overview

• Antenna-coupled TeraFETs are terahertz detectors that combine integrated or external antennas with field-effect transistors to rectify high-frequency signals into DC outputs.
• They exploit both overdamped broadband self-mixing and resonant plasmonic regimes, with performance depending on material choice, gate geometry, and bias conditions.
• Applications include polarization-sensitive detection, spectroscopy, imaging, and array integration across platforms like Si CMOS, graphene, GaN/AlGaN, and p-diamond.

Antenna-coupled field-effect transistors, commonly called TeraFETs, are terahertz detectors in which an integrated or external antenna couples incident radiation into a gated transistor channel, and the channel’s nonlinear transport rectifies the resulting high-frequency excitation into a dc photovoltage or photocurrent. In the cited literature, their operating range extends from roughly $100\,\mathrm{GHz}$ to $10\,\mathrm{THz}$, and the concept encompasses Si CMOS, GaN/AlGaN and GaAs/AlGaAs HEMTs, InGaAs, graphene, and p-diamond platforms. Depending on mobility, gate length, bias, and coupling geometry, TeraFETs operate either in an overdamped broadband self-mixing regime or in a resonant plasmonic regime, and they have been developed not only as single detectors but also as polarization analyzers, spectrometers, high-speed arrays, and cryogenically enhanced receivers (Ludwig et al., 2023, Ludwig et al., 2024, Zhang et al., 2020, Gorbenko et al., 2018, Holstein et al., 21 Jul 2025).

1. Physical basis of TeraFET operation

The canonical physical picture is the Dyakonov–Shur plasma-wave model of a gated two-dimensional electron gas or hole gas. In its compact form, the channel is described by continuity and momentum equations, with nonlinearity arising from convective acceleration and from the product of carrier density and drift velocity. In the gated-channel approximation, the carrier density is tied to the gate swing by $n=(C/e)U$, which converts ungated $2$D plasmon dispersion into a shallow-water-like law $\omega = s k$, with plasma-wave velocity $s=\sqrt{eU_0/m^*}$. Resonant behavior occurs when $\omega\tau \gg 1$, whereas overdamped broadband detection corresponds to $\omega\tau \ll 1$; in the latter limit the standard small-signal rectified response reduces to $\Delta U \approx U_a^2/(4U_0)$ for long-gate operation (Dyakonov, 2011).

Later device-level work reformulated the same physics as plasma-wave-assisted frequency mixing driven by antenna-coupled terminal voltages. In this description, the terahertz field launches charge-density and velocity oscillations in the channel, and the distributed nonlinearity converts them into a dc output. The 2024 two-dimensional hydrodynamic treatment explicitly distinguished overdamped broadband mixing from underdamped resonant mixing and emphasized that realistic devices require self-consistent coupling between transport, electrostatics, and the external circuit rather than a purely local gate-capacitance picture (Ludwig et al., 2024).

For asymmetric double-grating-gate plasmonic detectors, the hydrodynamic equations were written directly for the velocity $V(x,t)$ and electron density $10\,\mathrm{THz}$0,

$10\,\mathrm{THz}$1

with $10\,\mathrm{THz}$2. In that framework, the rectified photocurrent density is

$10\,\mathrm{THz}$3

so both convective nonlinearity and plasmon-driven electron drag contribute to detection (Popov et al., 2011).

2. Antenna coupling and device architectures

Antenna coupling is not an accessory feature but the defining boundary condition of the TeraFET. In some devices the antenna is external to the transistor core, as in graphene FETs coupled to log-periodic antennas or Si CMOS FETs integrated with rectangular patches. In others, the transistor metallization itself becomes the coupler, most prominently in double-grating-gate and periodic multi-gate structures, where the gate geometry both launches plasmons and breaks channel symmetry (Vicarelli et al., 2012, Holstein et al., 2024, Popov et al., 2011, Zhang et al., 2023).

Architecture	Coupling scheme	Representative source
Asymmetric double-grating gate	Slit near-fields launch plasmons; no supplementary antenna	(Popov et al., 2011)
Graphene log-periodic feed	Source and top gate formed by asymmetric antenna lobes	(Vicarelli et al., 2012)
Patch-antenna CMOS	Monolithic microstrip patch above MOSFET	(Holstein et al., 2024)
Dual-antenna phase-asymmetry FET	Equal amplitudes at source and drain with phase shift	(Gorbenko et al., 2019)

The asymmetric double-grating-gate structure is the most explicit example of an antenna-free TeraFET. There the periodic interdigitated metal gate acts as an aerial matched antenna, coupling normally incident terahertz radiation directly into channel plasmons without supplementary bow-tie or slot antennas. The narrower slit produces stronger local fields, ungated sections synchronize oscillations under neighboring fingers, and strong lateral asymmetry enables photovoltaic response at zero drain bias (Popov et al., 2011).

External-antenna implementations use different asymmetry strategies. In the 2012 graphene devices, one lobe of a log-periodic circular-toothed antenna was patterned as the source contact and the second identical lobe as the top-gate contact, while the drain remained a simple metal line. The asymmetry between source and drain feeding was essential for the rectified photovoltage, and the measured polarization dependence showed that responsivity was nearly suppressed when the terahertz polarization was orthogonal to the antenna axis (Vicarelli et al., 2012). Patch-coupled Si CMOS devices instead used monolithic microstrip-like patches realized in the back-end metal stack, with later work extending the concept to $10\,\mathrm{THz}$4 electrically combined arrays for QCL applications (Holstein et al., 2024).

The coupling problem is equally central in polarization-sensitive and interferometric devices. The single-TeraFET spectrometer and the phase-asymmetry detector used identical source and drain antennas so that the useful asymmetry was transferred from amplitude to phase. A plane wave then induced equal-amplitude terminal voltages with a phase difference governed by incidence geometry or helicity, and the transistor rectified their interferometric superposition (Gorbenko et al., 2018, Gorbenko et al., 2019).

3. Detection regimes and rectification mechanisms

A common oversimplification is to identify TeraFET detection with a single universal mechanism. The literature instead shows a family of related nonlinear processes whose relative weights depend strongly on material system, mobility, channel geometry, and coupling asymmetry.

In the standard overdamped regime, the core formula is the resistive self-mixing relation

$10\,\mathrm{THz}$5

or equivalently $10\,\mathrm{THz}$6. This was the central interpretation of room-temperature graphene detectors at $10\,\mathrm{THz}$7, where sign reversals across the charge neutrality point followed directly from ambipolar transport (Vicarelli et al., 2012).

Graphene TeraFETs, however, also exhibit a pronounced photothermoelectric contribution. In epitaxial graphene on SiC, the measured photovoltage was reproduced qualitatively only by superposing an overdamped plasmonic term,

$10\,\mathrm{THz}$8

with a thermoelectric term,

$10\,\mathrm{THz}$9

The finite offset at the charge neutrality point and the small sign reversal at positive gate bias indicated competition between the two mechanisms, although the sign reversal was interpreted as evidence that the plasmonic term remained stronger than the thermoelectric one in that device class (Bianco et al., 2018).

The 2023 CVD graphene study sharpened this distinction by concluding that the photoresponse can be treated as a linear combination of resistive self-mixing and photothermoelectric response, with the photothermoelectric term dominating over resistive self-mixing above $n=(C/e)U$0. Electromagnetic simulations traced that crossover to the redistribution of dissipated power between gated and ungated channel regions as frequency increases, and the analysis associated the photothermoelectric penetration length under the gate with the electronic cooling length (Ludwig et al., 2023).

By contrast, room-temperature Si CMOS TeraFETs up to $n=(C/e)U$1 remained predominantly non-resonant. The ADS-HDM decomposition showed that diffusive and plasmonic effects become more significant with rising frequency but are never dominant; the maximum difference between the full hydrodynamic model and purely resistive self-mixing was about $n=(C/e)U$2 at $n=(C/e)U$3 and $n=(C/e)U$4 (Ludwig et al., 2024).

A distinct branch of the field uses phase rather than amplitude asymmetry. When equal-amplitude terahertz signals are applied at source and drain with a phase shift $n=(C/e)U$5, the dc response acquires a term proportional to $n=(C/e)U$6. In the two-antenna theory of helicity-sensitive detection, this interference term is the origin of the helicity-dependent photovoltage. Experimentally, GaAs/AlGaAs HEMTs showed a strong helicity-dependent response interpreted as interference of plasma oscillations launched from opposite sides of the channel, and the same class of devices was proposed for all-electric determination of terahertz Stokes parameters (Drexler et al., 2012, Romanov et al., 2013).

4. Material platforms and reported performance

The TeraFET concept spans a wide range of materials, and the reported figures of merit reflect different combinations of antenna efficiency, transport regime, and readout definition. Direct comparison must therefore distinguish optical NEP, electrical NEP, cross-sectional responsivity, and intrinsic responsivity.

Platform and device	Reported figure	Conditions
Asymmetric DGG InAlAs/InGaAs/InP	Responsivity exceeds $n=(C/e)U$7	$n=(C/e)U$8, photovoltaic mode (Popov et al., 2011)
SLG/BLG graphene FET	$n=(C/e)U$9–$2$0, minimum optical NEP $2$1	$2$2, room temperature (Vicarelli et al., 2012)
Epitaxial graphene on SiC	$2$3, NEP $2$4	$2$5 and $2$6, room temperature (Bianco et al., 2018)
65-nm Si CMOS patch detector with Si lens	Optical NEP $2$7	Tuned to $2$8 (Krysl et al., 2024)
AlGaN/GaN HEMT	Optical NEP $2$9	$\omega = s k$0, $\omega = s k$1 (Qin et al., 2017)

The 2011 asymmetric double-grating-gate calculation remains one of the most striking results in the plasmonic literature: at room temperature, strong unit-cell asymmetry and depletion under one sub-grating produced predicted photovoltaic responsivity exceeding $\omega = s k$2, an order of magnitude above previously reported uncooled plasmonic detectors, without any supplementary antenna element because the grating gate itself served as the aerial matched coupler (Popov et al., 2011).

Graphene results illustrate the importance of platform-specific limitations. Exfoliated single-layer and bilayer graphene detectors at $\omega = s k$3 achieved room-temperature operation with responsivities of about $\omega = s k$4–$\omega = s k$5 and minimum optical NEP down to approximately $\omega = s k$6, while free-space transmission imaging of macroscopic objects was demonstrated. The same paper emphasized that these responsivities were lower bounds because of impedance mismatch between the high-impedance graphene channel and the antenna output (Vicarelli et al., 2012). Epitaxial graphene on SiC operated in the firmly overdamped regime with $\omega = s k$7, gave lower responsivity and higher NEP, but retained room-temperature functionality and wafer-scale scalability (Bianco et al., 2018).

Si CMOS results have progressively moved from proof-of-principle to system-level utility. Modeling and experiment on 65-nm CMOS patch-coupled devices yielded measured optical NEP minima of $\omega = s k$8, $\omega = s k$9, $s=\sqrt{eU_0/m^*}$0, and $s=\sqrt{eU_0/m^*}$1 for four devices at $s=\sqrt{eU_0/m^*}$2, $s=\sqrt{eU_0/m^*}$3, $s=\sqrt{eU_0/m^*}$4, and $s=\sqrt{eU_0/m^*}$5, while simulated intrinsic electrical NEP under ideal coupling was $s=\sqrt{eU_0/m^*}$6–$s=\sqrt{eU_0/m^*}$7 (Ludwig et al., 2024). Superstrate-lens coupling pushed a $s=\sqrt{eU_0/m^*}$8-class patch TeraFET to optical NEP $s=\sqrt{eU_0/m^*}$9 and demonstrated post-fabrication resonance tuning by more than $\omega\tau \gg 1$0 of the center frequency (Krysl et al., 2024).

Material comparison studies further suggest that p-diamond is unusual among plasmonic platforms. Numerical work found that p-diamond has a relatively low minimum resonant mobility, supports operation in the $\omega\tau \gg 1$1 to $\omega\tau \gg 1$2 window, improves substantially when cooled from $\omega\tau \gg 1$3 to $\omega\tau \gg 1$4, and at $\omega\tau \gg 1$5 channel length exhibits the highest dc response among the modeled p-diamond, n-diamond, Si, GaN, and InGaAs TeraFETs across a large frequency window (Zhang et al., 2020).

5. Polarization analysis, spectroscopy, imaging, and arrays

The maturation of TeraFETs is visible most clearly in applications where the detector is part of an optical or spectroscopic system rather than an isolated transport experiment. One early milestone was the demonstration that AlGaN/GaN HEMT TeraFETs could detect incoherent broadband radiation from hot blackbodies. In that framework the rectified dc output depended on the spectral integral $\omega\tau \gg 1$6, so phase coherence was unnecessary. Using detectors designed around $\omega\tau \gg 1$7, $\omega\tau \gg 1$8, and $\omega\tau \gg 1$9, a Fourier-transform spectrometer covered $\omega\tau \ll 1$0 to $\omega\tau \ll 1$1, and the $\omega\tau \ll 1$2 detector reached optical NEP $\omega\tau \ll 1$3 at $\omega\tau \ll 1$4 (Qin et al., 2017).

Polarization-sensitive detection constitutes another branch of development. GaAs/AlGaAs HEMTs exhibited helicity-dependent photovoltages several orders of magnitude higher than earlier reported all-electric helicity responses, along with response times better than $\omega\tau \ll 1$5. The generalized Dyakonov–Shur interpretation attributed this to interference of plasma oscillations excited through gate–source and gate–drain effective antennas, and the resulting device family was proposed as a route to all-electric terahertz ellipsometry and Stokes-parameter detection (Drexler et al., 2012, Romanov et al., 2013).

Spectrometer concepts followed naturally from phase control and resonance tuning. The single-TeraFET radiation spectrometer connected identical antennas to source and drain so that the relevant asymmetry was the phase difference $\omega\tau \ll 1$6, not the amplitude. The reported dc response scaled as

$\omega\tau \ll 1$7

which simultaneously encoded polarization and plasmonic resonance. In that model, p-diamond TeraFETs with $\omega\tau \ll 1$8–$\omega\tau \ll 1$9 supported room-temperature operation in the $\Delta U \approx U_a^2/(4U_0)$0 to $\Delta U \approx U_a^2/(4U_0)$1 window (Gorbenko et al., 2018). A related 2019 treatment described a homodyne phase-sensitive detector in which a strong local oscillator dramatically enhanced the phase-asymmetric response near plasmonic resonances (Gorbenko et al., 2019).

The current-driven TeraFET extended functionality even further. Numerical and theoretical work on InGaAs/GaAs devices showed that in short channels, a dc drain current can drive the source-to-drain response negative, so that the small-signal drain-side voltage amplitude exceeds the source-side amplitude and the device effectively serves as a terahertz amplifier. The same analysis proposed a current-driven TeraFET spectrometer based on the current at which the response crosses zero (Zhang et al., 2022).

Array integration has turned these concepts into practical instrumentation. A monolithic $\Delta U \approx U_a^2/(4U_0)$2 patch-antenna-coupled Si-CMOS array, electrically combined into a single detector element, reduced output impedance from about $\Delta U \approx U_a^2/(4U_0)$3 for a single pixel to about $\Delta U \approx U_a^2/(4U_0)$4 at $\Delta U \approx U_a^2/(4U_0)$5, enabling $\Delta U \approx U_a^2/(4U_0)$6 modulation bandwidth up to $\Delta U \approx U_a^2/(4U_0)$7. The same system provided high-resolution methanol spectroscopy around $\Delta U \approx U_a^2/(4U_0)$8 with an estimated detection limit of $\Delta U \approx U_a^2/(4U_0)$9 absorbance, or $V(x,t)$0, under optimal coupling (Holstein et al., 2024). A later liquid-nitrogen-cooled $V(x,t)$1 system optimized for $V(x,t)$2 to $V(x,t)$3 demonstrated experimental linear dynamic range exceeding $V(x,t)$4 without saturation at $V(x,t)$5 bandwidth and a $V(x,t)$6 detection bandwidth of $V(x,t)$7, with projected NEP approaching $V(x,t)$8 to $V(x,t)$9 under efficient coupling (Holstein et al., 21 Jul 2025).

6. Modeling, optimization, and unresolved design constraints

The modeling literature reflects a transition from analytical Dyakonov–Shur formulas to multi-scale device–circuit–electromagnetic co-simulation. On the compact side, the TSMC RF foundry model and the physics-based ADS-HDM both predicted responsivity and gate-bias dependence of 65-nm Si CMOS TeraFETs with good accuracy up to $10\,\mathrm{THz}$00, despite the foundry model having been intended originally for much lower frequencies. The same work concluded that antenna/optics coupling is the dominant lever in practical CMOS detectors and that the Gaussian-beam coupling efficiency $10\,\mathrm{THz}$01 is typically $10\,\mathrm{THz}$02 even for larger patches, so optical NEP can remain far from intrinsic electrical NEP unless coupling is improved (Ludwig et al., 2024).

Beyond one-dimensional small-signal models, the two-dimensional Poisson–hydrodynamic framework solved the transient 2D Poisson equation self-consistently with hydrodynamic transport using a well-balanced HLLC Riemann solver. That treatment was designed for transient, large-signal, and ultrahigh-frequency simulations and explicitly went beyond the gradual-channel approximation by resolving gate-edge fringe fields, transverse fields in the oxide, and non-local gate coupling (Ludwig et al., 2024). A complementary compact SPICE model represented the channel as a nonlinear transmission line with segmentwise conductance, capacitance, and Drude inductance $10\,\mathrm{THz}$03, improving agreement with experiment, analytical theory, and multiphysics simulations in the strong resonant regime (Liu et al., 2024).

Optimization studies also show that geometry-induced asymmetry remains a primary design principle. In the asymmetric double-grating-gate TeraFET, responsivity exceeded $10\,\mathrm{THz}$04 only when the asymmetry factor $10\,\mathrm{THz}$05 was driven below $10\,\mathrm{THz}$06, one sub-grating was negatively biased to deplete the underlying channel, and ungated segments were preserved to couple and synchronize neighboring cavities (Popov et al., 2011). More moderate but systematic symmetry breaking can be introduced through non-uniform gate capacitance or threshold voltage. For 130-nm Si TeraFETs, an exponentially varying $10\,\mathrm{THz}$07 produced about $10\,\mathrm{THz}$08 responsivity enhancement, while sawtooth profiles gave the largest tunability (Zhang et al., 2021). Periodic multi-gate plasmonic FETs went further: simulations showed that spatially alternating gate sections create enhanced and suppressed spectral regions, with up to $10\,\mathrm{THz}$09 increase in dc response relative to a uniform-channel device and a simple map locating mountain and valley bands at $10\,\mathrm{THz}$10 and $10\,\mathrm{THz}$11 (Zhang et al., 2023).

Several recurring limitations cut across platforms. Graphene devices are often constrained by antenna–device impedance mismatch, modest on/off ratio, and parasitic thermoelectric or junction-related contributions (Vicarelli et al., 2012, Ludwig et al., 2023). CMOS systems can become readout-limited even when the intrinsic detector is faster or quieter, as shown by the gap between Johnson-limited and experimental NEP under room-temperature electronics in both room-temperature and cryogenic arrays (Holstein et al., 2024, Holstein et al., 21 Jul 2025). Superstrate optics solve one problem while introducing another: silicon lenses on patch TeraFETs strongly improve optical NEP, but the resonance frequency then becomes sensitive to spacer material and thickness, shifting by more than $10\,\mathrm{THz}$12 of center frequency in the reported $10\,\mathrm{THz}$13 devices (Krysl et al., 2024).

Taken together, these results define TeraFETs less as a single detector topology than as a design space organized around antenna-defined boundary conditions, gate-controlled channel nonlinearity, and plasmonic transport. The central technical question is not whether a FET can rectify terahertz radiation, but which asymmetry, coupling network, and transport regime best convert a given field distribution into a useful dc observable. Across that design space, the literature shows stable room-temperature operation, deep sub-THz and multi-THz coverage, polarization and phase sensitivity, imaging and spectroscopy capability, and a continuing convergence between analytic plasma-wave theory, compact circuit models, and full device-level simulation (Dyakonov, 2011, Ludwig et al., 2024).