betaFeedback

WorksPhysics and Astronomy

Hot-Carrier Distribution Spectroscopy by Transconductance in Two-Dimensional Field-Effect Transistors

Abstract

The transconductance gm=dID/dVGg_m = dI_D/dV_G of a field-effect transistor (FET) is conventionally read as a proxy for carrier density. We show that it is instead a spectroscopic probe of the carrier distribution: because gmg_m weights the spectral current j(E)j(E) by the gate-voltage derivative ∂f(E)/∂VG\partial f(E)/\partial V_G and integrates over energy, it is sensitive to the shape of f(E)f(E), not merely its integrated weight nn. We develop an energy-resolved transport framework for two-dimensional (2D) FETs and, within a gate-independent spectral-kernel approximation, derive the decomposition gm=gm(n)+gm(α)g_m = g_m^{(n)} + g_m^{(α)} into the conventional density-modulation term gm(n)g_m^{(n)} and a distribution-shape-driven term gm(α)g_m^{(α)}. The latter, obtained as the residual after subtracting the smooth density-modulation background from the measured gmg_m, exhibits a characteristic anomalous peak at a gate voltage VGpkV_G^{\rm pk}. This peak has no counterpart in equilibrium transport and cannot be explained by carrier density modulation alone. With the spectral kernel calibrated, the peak position and height – extracted from standard DC/lock-in gmg_m sweeps – constrain the hot-carrier energy E0E_0, spectral width σσ, and generation threshold ncn_c, realizing a steady-state, all-electrical spectroscopy of the carrier distribution. An optional time-resolved extension further recovers the carrier relaxation time ττ from the transient response following a pump excitation, establishing the 2D FET as a distribution-function spectrometer that requires no optical readout.

Introduction

Two-dimensional (2D) semiconductors such as transition metal dichalcogenides (TMDs) [1–3] and their van der Waals heterostructures have emerged as a rich arena for exploring electronic transport in reduced dimensions [4, 5]. Field-effect transistors (FETs) fabricated from these materials simultaneously serve as testbeds for fundamental physics and as building blocks of next-generation devices [6–10].

In the standard description of FET transport, the drain current is written as

ID=qWLnμVD,(1)I_D = \frac{qW}{L}n\mu V_D, \tag*{(1)}

where qq is the elementary charge, WW and LL are the channel width and length, nn is the areal carrier density, μ\mu is the mobility, and VDV_D is the source–drain voltage [11]. This relation implicitly assumes that carriers obey an equilibrium or quasi-equilibrium distribution, so that μ\mu is a single scalar that summarizes all microscopic scattering. Under strong source–drain fields, carrier injection, or optical pumping, however, the distribution function acquires a hot-carrier tail whose energy is well above the lattice temperature. In such nonequilibrium states, the shape of the distribution—not merely its integrated weight nn—governs transport, rendering Eq. (1) inadequate [12, 13].

Nonequilibrium carriers in 2D materials have been studied extensively by ultrafast optical spectroscopy [14]. Pump–probe experiments on graphene [15] and MoS2_2 [16–19] reveal hot-carrier relaxation times in the range τ∼0.1\tau\sim0.1–1010 ps, reflecting carrier–phonon scattering dynamics. Yet such optical probes are rarely combined with operando electrical gating, and they do not directly access the transport-relevant, velocity-weighted distribution that governs IDI_D.

Here we develop an energy-resolved transport theory for 2D FETs that connects the shape of f(E)f(E) to measurable device characteristics. Our starting point is a spectroscopic reformulation of transport: within an energy-resolved transport description with a gate-independent spectral current kernel j(E)j(E), the transconductance takes the form

gm=qWL∫0∞j(E)∂f(E;VG)∂VG dE,(2)g_m = \frac{qW}{L}\int_0^\infty j(E)\frac{\partial f(E;V_G)}{\partial V_G}\,dE, \tag*{(2)}

which makes explicit that the transconductance does not merely reflect carrier density modulation but probes the energy distribution of carriers. Equation (2) is itself a natural consequence of writing the current as a spectral integral and differentiating with respect to VGV_G. It is a close relative of the Landauer–Büttiker formula [13, 20] and of the transport distribution function of thermoelectric transport [21]. More broadly, it instantiates a general principle—that a derivative of the conductance resolves spectral structure—which underlies scanning-tunneling [22] and quantum-dot transconductance [23] spectroscopies as well as the electrical measurement of the quasiparticle distribution in mesoscopic wires [24].

Our contribution is to apply this principle to the shape of a nonequilibrium distribution in a 2D FET, where it yields the specific, measurable anomaly developed below. In a related line of work, the same transconductance-as-probe principle is applied to the valley degree of freedom in WSe2_2, using gmg_m to read out valley thermodynamics in multilayer films [25] and to guide valley engineering in bilayer gate-all-around transistors [26]. A dynamical extension of the same idea uses the transconductance to extract the intervalley relaxation time of WSe2_2 transistors [27]. Corrections from a VGV_G-dependent kernel j(E,VG)j(E,V_G) (e.g. barrier lowering, contact modulation, or band bending) are additive and discussed in Sec. IV.

For a Boltzmann distribution, ∂feq/∂VG\partial f_{\mathrm{eq}}/\partial V_G is featureless in energy, and gmg_m is flat—the conventional result. For a nonequilibrium distribution with a hot-carrier peak at E0E_0, ∂fneq/∂VG\partial f_{\mathrm{neq}}/\partial V_G acquires structure at E0E_0 and gmg_m reports it as a measurable anomaly in the shape-resolved residual. The Gaussian hot-carrier peak adopted below is only a minimal model: the mechanism applies to any localized nonthermal excess δf(E)\delta f(E) above the transport onset, not to a specific line shape. Figure 1(a) shows the device geometry, (b) the spectral mechanism, and (c) the resulting measurement principle.

Concept, device, and measurement principle schematic

Figure 1. Concept, device, and measurement principle (schematic). (a) A gated monolayer 2D-semiconductor field-effect transistor: a gate voltage tunes the carrier density in the channel between source and drain contacts, while field-induced hot-carrier generation drives the distribution out of equilibrium. (b) Physical mechanism (energy on the vertical axis). Left: the carrier distribution f(E)f(E) is either thermal (dotted) or non-thermal (red), the latter carrying a localized hot-carrier excess near E0E_0. Right: the transport kernel—an energy-selective transmission T(E)\mathcal{T}(E) that turns on near the onset EbE_b and a velocity v(E)v(E) that grows with energy—weighs high-energy carriers more heavily, so each carrier in the excess near E0>EbE_0 > E_b contributes more current. (c) Measurement principle (schematic): a conventional, density-only response is featureless in VGV_G (blue), and the total non-thermal gmg_m (red) rises monotonically. The non-thermal signature is the shape-driven contribution gm(α)g_m^{(\alpha)} (purple, shaded), which produces an anomalous peak whose position and height read out the spectral parameters of f(E)f(E)—transconductance spectroscopy. (In the representative case shown here the total measured gmg_m itself remains monotone; the peak is isolated in gm(α)g_m^{(\alpha)}, as quantified in Figure 3.) The response turns on at the threshold voltage VthV_{\mathrm{th}} [Eq. (12)], where the channel begins to populate, and the anomaly peaks at VGpkV_G^{\mathrm{pk}} [Eq. (20)].

Our main results are threefold. (i) The drain current takes the compact form ID=(qW/L)nvˉ(α)I_D=(qW/L)n\bar{v}(\alpha), with an effective velocity vˉ(α)\bar{v}(\alpha) set by a single parameter α\alpha that quantifies the hot-carrier population. (ii) Within the same spectral-kernel approximation, gmg_m admits a transparent decomposition gm=gm(n)+gm(α)g_m=g_m^{(n)}+g_m^{(\alpha)}. The new term gm(α)g_m^{(\alpha)} produces an anomalous peak in the shape-resolved residual—obtained after calibrating and subtracting the smooth density-modulation background—that cannot be explained by carrier density modulation and directly constrains the spectral parameters {E0,σ,nc}\{E_0,\sigma,n_c\}. This peak carries a decisive drain-voltage fingerprint—its position shifts to lower VGV_G and its height saturates with VDV_D—that

no separable, quasi-equilibrium mobility-rolloff model can reproduce. (iii) A time-resolved electrical measurement after a pump excitation yields the carrier relaxation time τ\tau, completing an all-electrical spectroscopy of the nonequilibrium distribution.

The remainder of this paper is organized as follows. Section II develops the energy-resolved transport framework and derives the decomposition gm=gm(n)+gm(α)g_m=g_m^{(n)}+g_m^{(\alpha)}. Section III presents the anomalous gmg_m peak, its drain-voltage fingerprint, and the resulting extraction of the spectral parameters {E0,σ,nc}\{E_0,\sigma,n_c\}. Section IV places the framework in the context of prior hot-carrier work and addresses corrections from a gate-dependent kernel, and Sec. V concludes. Appendices A–C present the time-resolved extension, a fit to synthetic data, and the robustness of the anomaly against modeling choices.

Model and Theory

Energy-resolved drain current

For definiteness, we formulate the model for an n-type 2D FET, where E≥0E \ge0 denotes the electron kinetic energy measured from the conduction-band minimum. The same formalism applies to p-type devices by interpreting EE as the hole kinetic energy measured from the valence-band maximum.

For a 2D semiconductor of width WW and length LL, we write the drain current as an integral over carrier energy [12, 13]:

ID=qWL∫0∞D(E) v(E) T(E) f(E) dE,(3)I_D = \frac{qW}{L}\int_{0}^{\infty} D(E)\,v(E)\,T(E)\,f(E)\,\mathrm{d}E, \tag*{(3)}

where D(E)D(E) is the density of states, v(E)v(E) is the group velocity, T(E)T(E) is an energy-selective transmission probability, and f(E)f(E) is the carrier distribution function. For a 2D system with parabolic dispersion, D(E)=D0=gm∗/(πℏ2)D(E)=D_0=g m^*/(\pi\hbar^2) is constant (with gg the spin–valley degeneracy and m∗m^* the effective mass), and the group velocity is

v(E)=vTE/kBT,(4)v(E)=v_T\sqrt{E/k_BT}, \tag*{(4)}

where vT=kBT/m∗v_T=\sqrt{k_BT/m^*} is a reference (thermal) velocity scale.[28] The spectral current j(E)≡v(E)T(E)j(E)\equiv v(E)T(E) acts as an energy filter: it weights high-energy carriers more heavily in proportion to their velocity.

We model the energy selectivity of the transmission by a smooth onset,

T(E)=[1+exp⁡(Eb−EΔ)]−1,(5)T(E)=\left[1+\exp\left(\frac{E_b-E}{\Delta}\right)\right]^{-1}, \tag*{(5)}

a smooth step centered at an effective onset energy EbE_b with a broadening parameter Δ\Delta. We do not identify EbE_b with a unique microscopic barrier height. Rather, EbE_b is an effective crossover energy of the transport kernel that folds contact injection (e.g. a Schottky barrier), electrostatic bottlenecks (channel pinch-off, i.e. the saddle point of the channel potential), and energy-dependent scattering into a single energy-filtering scale. Carriers with E<EbE<E_b contribute weakly to the drain current, while those with E>EbE>E_b are efficiently transmitted [12]. Both EbE_b and Δ\Delta are thus effective transport-kernel parameters, to be calibrated from the equilibrium response vˉeq(T)\bar{v}_{\mathrm{eq}}(T) rather than identified with an independent microscopic barrier. Importantly, none of our central results depend on this specific functional form. They require only that the spectral current j(E)=v(E)T(E)j(E)=v(E)T(E) be a smooth, monotonically increasing filter that favors high-energy carriers, so that a hot-carrier population at E0>EbE_0>E_b is preferentially transmitted. Any transmission with these properties yields the same anomalous gm(α)g_m^{(\alpha)} peak, in the same spirit as the robustness against the choice of α(n)\alpha(n) demonstrated in Appendix C.

Carrier distribution functions

We compare two classes of distribution, both normalized to the same areal carrier density nn via ∫0∞f(E) dE=n\int_0^\infty f(E)\,\mathrm{d}E=n. Here f(E)f(E) is a carrier spectral density (carriers per unit area and energy), not a dimensionless occupation probability; accordingly it carries dimensions of [n]/[energy][n]/[\mathrm{energy}], with the density of states D0D_0 absorbed into the current prefactor as noted below.

a. Equilibrium (Boltzmann).

feq(E)=Aeqe−E/kBT,(6)f_{\mathrm{eq}}(E)=A_{\mathrm{eq}}e^{-E/k_BT}, \tag*{(6)}

where Aeq=n/NeqA_{\mathrm{eq}}=n/N_{\mathrm{eq}} and Neq=∫0∞e−E/kBT dE=kBTN_{\mathrm{eq}}=\int_0^\infty e^{-E/k_BT}\,\mathrm{d}E=k_BT. We adopt the Boltzmann (nondegenerate) form for analytical transparency. It is the leading approximation to the Fermi–Dirac distribution fFD(E)={1+exp⁡[(E−EF)/kBT]}−1f_{\mathrm{FD}}(E)=\{1+\exp[(E-E_F)/k_BT]\}^{-1} when the Fermi level lies below the transport states by more than a few kBTk_BT, i.e. in the moderately doped, nondegenerate regime relevant to the density scale ne∼1012 cm−2n_e\sim10^{12}\ \mathrm{cm}^{-2} considered here. For strongly degenerate channels the Boltzmann factor e−E/kBTe^{-E/k_BT} in Eqs. (6)–(7) should be replaced by fFD(E)f_{\mathrm{FD}}(E). This changes only the numerical values of the spectral moments NeqN_{\mathrm{eq}} and JeqJ_{\mathrm{eq}}, leaving the spectroscopic identity Eq. (15) and the decomposition gm=gm(n)+gm(α)g_m=g_m^{(n)}+g_m^{(\alpha)} unchanged in form. The hot-carrier component, centered well above EFE_F at E0≫kBTE_0\gg k_BT, is nondegenerate in either description.

b. Nonequilibrium (Boltzmann + hot-carrier Gaussian).

fneq(E)=Aneq[e−E/kBT+αe−(E−E0)2/σ2],(7)f_{\mathrm{neq}}(E)=A_{\mathrm{neq}}\left[e^{-E/k_BT}+\alpha e^{-(E-E_0)^2/\sigma^2}\right], \tag*{(7)}

where Aneq=n/(Neq+αNneq)A_{\mathrm{neq}}=n/(N_{\mathrm{eq}}+\alpha N_{\mathrm{neq}}), Nneq=∫0∞e−(E−E0)2/σ2 dE≈σπN_{\mathrm{neq}}=\int_0^\infty e^{-(E-E_0)^2/\sigma^2}\,\mathrm{d}E\approx\sigma\sqrt{\pi} (for E0≫σE_0\gg\sigma), and α≥0\alpha\geq0 controls the amplitude of the Gaussian hot-carrier bump centered at energy E0>EbE_0>E_b with width σ\sigma. The constraint ∫f(E) dE=n\int f(E)\,\mathrm{d}E=n—where the constant D0D_0 is absorbed into the current prefactor, so that it cancels in vˉ\bar{v} and in every gmg_m ratio below—is satisfied by construction via AneqA_{\mathrm{neq}}. The Gaussian is a minimal representation of a localized nonthermal excess at E0E_0.

Physically, such a peaked nonthermal population can be established in several ways: high-field acceleration in the drain-side region of a short-channel FET followed by energy-selective optical-phonon or intervalley scattering; carrier injection over a contact or barrier bottleneck that admits a narrow energy window; impact-ionization-like generation near a characteristic energy; or optical pumping at a fixed photon energy. In each case carriers accumulate around a characteristic energy E0E_0 set by the dominant generation and relaxation bottleneck rather than by the lattice temperature. We do not commit to a specific microscopic mechanism; the Gaussian models only the resulting localized excess. The spectroscopic mechanism developed below relies only on the presence of such an excess well above the transport onset EbE_b. The same gm(α)g_m^{(\alpha)} anomaly therefore arises for any nonthermal excess δf(E)\delta f(E) that is sharply peaked in this energy range, irrespective of its detailed shape.

The physical meaning of α\alpha is the relative number of hot carriers in the Gaussian component: at α=0\alpha=0 the distribution is purely thermal, while increasing α\alpha builds up a high-energy tail at E0≫kBTE_0\gg k_BT.

Effective drift velocity and current decomposition

Substituting Eqs. (6)–(7) into Eq. (3) and using the linearity of the integral, we define spectral current moments

Jeq=∫0∞j(E)e−E/kBT dE,(8)J_{\rm eq}=\int_{0}^{\infty}j(E)e^{-E/k_{\rm B}T}\,dE, \tag*{(8)}
Jneq=∫0∞j(E)e−(E−E0)2/σ2 dE.(9)J_{\rm neq}=\int_{0}^{\infty}j(E)e^{-(E-E_{0})^{2}/\sigma^{2}}\,dE. \tag*{(9)}

The drain current then takes the compact form

ID=qWL n vˉ(α),(10)I_{\rm D}=\frac{qW}{L}\,n\,\bar{v}(\alpha), \tag*{(10)}

where the effective drift velocity is

vˉ(α)=Jeq+αJneqNeq+αNneq.(11)\bar{v}(\alpha)=\frac{J_{\rm eq}+\alpha J_{\rm neq}}{N_{\rm eq}+\alpha N_{\rm neq}}. \tag*{(11)}

Equivalently, vˉ(α)=∫0∞j(E)f(E) dE/∫0∞f(E) dE\bar{v}(\alpha)=\int_{0}^{\infty}j(E)f(E)\,dE/\int_{0}^{\infty}f(E)\,dE is simply the carrier-weighted average of the spectral current j(E)j(E): the normalization AneqA_{\rm neq} and the density nn cancel between numerator and denominator, leaving each moment split into a thermal and an α\alpha-weighted hot-carrier part. Equation (10) preserves the form of the conventional relation ID∝nvdriftI_{\rm D}\propto nv_{\rm drift}, but now vˉ(α)\bar{v}(\alpha) is distribution-shape-sensitive: it interpolates between the pure-Boltzmann limit vˉ(0)=Jeq/Neq≡vˉeq\bar{v}(0)=J_{\rm eq}/N_{\rm eq}\equiv\bar{v}_{\rm eq} and the Gaussian limit vˉ(∞)=Jneq/Nneq≡vˉneq\bar{v}(\infty)=J_{\rm neq}/N_{\rm neq}\equiv\bar{v}_{\rm neq}. Because hot carriers at E0≫EbE_{0}\gg E_{\rm b} are above the barrier (T(E0)≈1T(E_{0})\approx1) and have higher velocity (v(E0)∝E0v(E_{0})\propto\sqrt{E_{0}}), we have vˉneq≫vˉeq\bar{v}_{\rm neq}\gg\bar{v}_{\rm eq}, so even a small α\alpha significantly boosts IDI_{\rm D}.

Gate-controlled carrier density and nonequilibrium parameter

Above threshold the carrier density follows the linear gate relation,

n(VG)=Coxq(VG−Vth),(12)n(V_{\rm G})=\frac{C_{\rm ox}}{q}(V_{\rm G}-V_{\rm th}), \tag*{(12)}

where CoxC_{\rm ox} is the oxide capacitance per unit area and VthV_{\rm th} is the threshold voltage. This is an idealized sharp threshold (n≡0n\equiv0 below VthV_{\rm th}, linear above), so dn/dVGdn/dV_{\rm G} jumps discontinuously at VthV_{\rm th}. Through the decomposition derived below, this produces two turn-on features in gmg_{\rm m} exactly at VthV_{\rm th}: a step in the density term gm(n)∝dn/dVGg_{\rm m}^{(n)}\propto dn/dV_{\rm G} (which jumps from zero to a constant), and—in the non-thermal case—a kink in the shape term gm(α)∝ng_{\rm m}^{(\alpha)}\propto n (whose slope breaks where nn begins to rise from zero). Both are artifacts of the idealized threshold: in a real device, subthreshold conduction (n∼eqVG/kBTn\sim e^{qV_{\rm G}/k_{\rm B}T} below VthV_{\rm th}) smooths them into a gradual turn-on. In any case, these features sit at VthV_{\rm th} and are well separated from—and do not affect—the anomalous peak at VGpkV_{\rm G}^{\rm pk}.

The hot-carrier amplitude α\alpha depends on both VGV_{\rm G} and VDV_{\rm D}. The key qualitative property required for an anomalous peak in gm(α)g_{\rm m}^{(\alpha)} is that α(n)\alpha(n) rises from zero at small nn and saturates as n→∞n\to\infty. A broad class of such monotone-and-saturating functions generically produces a localized peak in the distribution-shape contribution gm(α)g_{\rm m}^{(\alpha)} derived below, because dα/dnd\alpha/dn inherits the bell shape of the derivative of a saturating function. The qualitative peak position, height, and VDV_{\rm D}-dependence are robust fingerprints of this mechanism, only weakly dependent on the precise functional form of α(n)\alpha(n). Alternative analytic forms (hyperbolic-tangent, exponential saturation, and a Hill function in nn) yield quantitatively similar gm(α)(VG)g_{\rm m}^{(\alpha)}(V_{\rm G}) peaks, differing only in the prefactor of the peak density, as verified explicitly in Appendix C (Fig. 8).

For analytical tractability we adopt the representative phenomenological model

α(VG,VD)=αmax⁡VD2VD2+Vc2n(VG)n(VG)+nc,(13)\alpha(V_G,V_D)=\alpha_{\max}\frac{V_D^2}{V_D^2+V_c^2}\frac{n(V_G)}{n(V_G)+n_c}, \tag*{(13)}

which is proportional to the power density (through VDV_D) and to the density of impact-ionizing or inelastically scattered carriers (through nn). Here αmax⁡\alpha_{\max} is the saturation amplitude, VcV_c is the drain-voltage crossover, and ncn_c is the carrier-density crossover for hot-carrier generation. The high-field crossover at VcV_c parallels the onset of velocity saturation observed in monolayer MoS2_2 [29, 30]. In real devices ncn_c is not a universal material constant but an effective crossover density set by the channel length, contact field concentration [31], dielectric environment, heat dissipation, and the dominant inelastic-scattering pathway [32]. The value adopted below should therefore be viewed as a representative effective scale for strongly gated TMD FETs. For n≪ncn \ll n_c, α∝n\alpha\propto n (few carriers, few collisions); for n≫ncn \gg n_c, α→αmax⁡h(VD)\alpha\to\alpha_{\max}h(V_D) (saturation).

We stress that Eq. (13) is a transparent phenomenological interpolation, not a microscopic derivation, and that the two factors are chosen for their limiting behavior rather than for a specific functional form. The drain-voltage factor h(VD)=VD2/(VD2+Vc2)h(V_D)=V_D^2/(V_D^2+V_c^2) represents field-driven hot-carrier generation that switches on with the power input at low bias and saturates once the high-field region spans the channel. Any monotonic function rising from zero and saturating at large VDV_D yields the same qualitative gm(α)(VD)g_m^{(\alpha)}(V_D). The density factor n/(n+nc)n/(n+n_c) expresses that at low density the hot-carrier population grows in proportion to the number of available carriers (and hence collisions), whereas at high density it saturates as phase-space restriction, screening, and energy-relaxation bottlenecks limit further accumulation. What matters for the anomalous peak is only that α(n)\alpha(n) rise from zero and saturate, so that dα/dnd\alpha/dn is bell-shaped. We verify explicitly in Appendix C (Figure 8) that hyperbolic-tangent, exponential, and Hill forms all produce the same peak, with only minor quantitative differences. A first-principles kinetic treatment of α(n,VD)\alpha(n,V_D) from the coupled hot-carrier generation–relaxation balance is beyond the scope of the present work and is left for future study.

Figure 8. Robustness of the gm(α)g_m^{(\alpha)} peak against the two phenomenological modeling choices. Top row—saturation function α(n)\alpha(n). (a) Four saturating forms of g(n)g(n) versus gate voltage: hyperbolic n/(n+nc)n/(n+n_c) (blue), tanh⁡(n/nc)\tanh(n/n_c) (red dashed), 1−e−n/nc1-e^{-n/n_c} (green dash-dot), and the Hill function n2/(n2+nc2)n^2/(n^2+n_c^2) (purple dotted). (b) Corresponding gm(α)(VG)g_m^{(\alpha)}(V_G), each normalized to its own peak: the peak position stays near VG∗V_G^* for every form. Bottom row—transmission T(E)\mathcal{T}(E). (c) Four transmission forms rising from 0 to 1 across EbE_b: Fermi step (blue), error-function (red dashed), linear ramp (green), and the discontinuous hard step Θ(E−Eb)\Theta(E-E_b) (purple dotted); E0E_0 marks the hot-carrier energy. (d) Corresponding gm(α)(VG)g_m^{(\alpha)}(V_G), normalized to the equilibrium plateau gmeqg_m^{\mathrm{eq}}: the peak occurs at the same gate voltage for all forms (position set by α(n)\alpha(n), not T\mathcal{T}); only the height varies.

Transconductance as a spectral probe

Before decomposing gmg_m, we establish its fundamental spectroscopic content. Differentiating Eq. (3) with respect to VGV_G gives, by the product rule, two contributions,

gm=qWL∫0∞[∂j(E;VG)∂VGf(E;VG)+j(E;VG)∂f(E;VG)∂VG] dE,(14)g_m=\frac{qW}{L}\int_0^\infty\left[\frac{\partial j(E;V_G)}{\partial V_G}f(E;V_G)+j(E;V_G)\frac{\partial f(E;V_G)}{\partial V_G}\right]\,dE, \tag*{(14)}

a kernel-modulation term and a distribution-modulation term. Within the present transport description the spectral current kernel j(E)=v(E)T(E)j(E)=v(E)\mathcal{T}(E) [Eqs. (4)–(5)] has no explicit gate dependence, so ∂j/∂VG=0\partial j/\partial V_G=0, the first term vanishes, and we obtain

gm=qWL∫0∞j(E)∂f(E;VG)∂VG dE.(15)g_m=\frac{qW}{L}\int_0^\infty j(E)\frac{\partial f(E;V_G)}{\partial V_G}\,dE. \tag*{(15)}

The kernel-modulation term is nonzero when the gate lowers the barrier or modulates the contacts [T=T(E,VG)\mathcal{T}=\mathcal{T}(E,V_G)], induces band bending, or otherwise renders jj gate-dependent. This term is additive and is estimated in Sec. IV. Equation (15) shows that the distribution-function contribution to gmg_m is exactly the spectral current j(E)j(E) weighted by ∂f/∂VG\partial f/\partial V_G: a spectrally resolved response function of the distribution. For feqf_{\mathrm{eq}}, the derivative ∂feq/∂VG∝e−E/kBT\partial f_{\mathrm{eq}}/\partial V_G\propto e^{-E/k_{\mathrm{B}}T} is featureless, and gmg_m depends only on the total carrier density. For fneqf_{\mathrm{neq}}, ∂fneq/∂VG\partial f_{\mathrm{neq}}/\partial V_G acquires additional structure at E0E_0 whenever E0E_0 depends on VGV_G (i.e. on nn), and gmg_m directly reports this spectral feature—even though no optical measurement is performed. This is the central spectroscopic principle of the paper.

Decomposing gm=dID/dVGg_m=dI_D/dV_G explicitly via Eq. (10) with respect to VGV_G:

gm=WCoxLvˉ(α)⏟gm(n)+WCoxLndvˉdαdαdn⏟gm(α),(16)g_m=\underbrace{\frac{WC_{\mathrm{ox}}}{L}\bar{v}(\alpha)}_{g_m^{(n)}}+\underbrace{\frac{WC_{\mathrm{ox}}}{L}n\frac{d\bar{v}}{d\alpha}\frac{d\alpha}{dn}}_{g_m^{(\alpha)}}, \tag*{(16)}

where we used dn/dVG=Cox/qdn/dV_G=C_{\mathrm{ox}}/q from Eq. (12). The two terms have distinct physical origins:

  • gm(n)=(WCox/L)vˉ(α)g_m^{(n)}=(WC_{\mathrm{ox}}/L)\bar{v}(\alpha): the conventional density-modulation term. At equilibrium (α=0\alpha=0), this equals the constant value (WCox/L)vˉeq(WC_{\mathrm{ox}}/L)\bar{v}_{\mathrm{eq}}, giving a flat gmg_m–VGV_G curve in the linear regime.

  • gm(α)g_m^{(\alpha)}: a new distribution-shape-modulation term that arises because the gate voltage also controls α\alpha through n(VG)n(V_G). Its magnitude is proportional to dvˉ/dα=NeqNneq(vˉneq−vˉeq)/(Neq+αNneq)2d\bar{v}/d\alpha=N_{\mathrm{eq}}N_{\mathrm{neq}}(\bar{v}_{\mathrm{neq}}-\bar{v}_{\mathrm{eq}})/(N_{\mathrm{eq}}+\alpha N_{\mathrm{neq}})^2, which is positive whenever vˉneq>vˉeq\bar{v}_{\mathrm{neq}}>\bar{v}_{\mathrm{eq}}.

The gate-voltage dependence of gm(α)g_m^{(\alpha)} is governed by

dαdn=αmax⁡h(VD)nc(n+nc)2,(17)\frac{d\alpha}{dn}=\frac{\alpha_{\max}h(V_D)n_c}{(n+n_c)^2}, \tag*{(17)}

where h(VD)=VD2/(VD2+Vc2)h(V_D)=V_D^2/(V_D^2+V_c^2). To find the peak of gm(α)g_m^{(\alpha)} analytically, we substitute the α(n)\alpha(n) dependence into the expression for dvˉ/dαd\bar{v}/d\alpha and simplify. Defining β=αmax⁡h(VD)\beta=\alpha_{\max}h(V_D), one finds

gm(α)∝n(Neq+βNneqnn+nc)2(n+nc)2=n(An+B)2,(18)g_m^{(\alpha)}\propto\frac{n}{\left(N_{\mathrm{eq}}+\beta N_{\mathrm{neq}}\frac{n}{n+n_c}\right)^2(n+n_c)^2} =\frac{n}{(An+B)^2}, \tag*{(18)}

where A=Neq+βNneqA=N_{\mathrm{eq}}+\beta N_{\mathrm{neq}} and B=NeqncB=N_{\mathrm{eq}}n_c. Setting d/dn[n/(An+B)2]=0d/dn[n/(An+B)^2]=0 gives the exact peak density

npk=BA=NeqncNeq+αmax⁡h(VD)Nneq,(19)n_{\mathrm{pk}}=\frac{B}{A}=\frac{N_{\mathrm{eq}}n_c}{N_{\mathrm{eq}}+\alpha_{\max}h(V_D)N_{\mathrm{neq}}}, \tag*{(19)}

and the corresponding peak gate voltage

VGpk=Vth+qnpkCox.(20)V_G^{\mathrm{pk}}=V_{\mathrm{th}}+\frac{q n_{\mathrm{pk}}}{C_{\mathrm{ox}}}. \tag*{(20)}

In the limit αmax⁡→0\alpha_{\max}\to0, Eq. (19) gives npk→ncn_{\mathrm{pk}}\to n_c and VGpk→Vth+qnc/Cox≡VG∗V_G^{\mathrm{pk}}\to V_{\mathrm{th}}+q n_c/C_{\mathrm{ox}}\equiv V_G^*. For finite αmax⁡\alpha_{\max}, the peak shifts to lower gate voltages by the factor Neq/(Neq+αmax⁡h(VD)Nneq)N_{\mathrm{eq}}/(N_{\mathrm{eq}}+\alpha_{\max}h(V_D)N_{\mathrm{neq}}). Since Neq=kBTN_{\mathrm{eq}}=k_B T and Nneq≈σπN_{\mathrm{neq}}\approx\sigma\sqrt{\pi}, this shift encodes the hot-carrier energy width σ\sigma. Consequently, gm(α)g_m^{(\alpha)} exhibits a peak at VGpkV_G^{\mathrm{pk}}—an anomalous feature absent in equilibrium transport—and its position jointly reports ncn_c (the density crossover) and σ\sigma (the hot-carrier spectral width).

Results

We evaluate Eqs. (10)–(16) numerically for parameters representative of a monolayer MoS2_2-like 2D semiconductor (m∗=0.5m0m^*=0.5m_0, spin–valley degeneracy g=4g=4, T=300T=300 K). The gate dielectric is taken as 30 nm HfO2_2 (εr≈25\varepsilon_r\approx25), giving Cox/q=1.4×1013 cm−2V−1C_{\mathrm{ox}}/q=1.4\times10^{13}\ \mathrm{cm}^{-2}\mathrm{V}^{-1}. The transport-kernel onset parameters are Eb=50E_b=50 meV and Δ=20\Delta=20 meV. Hot carriers are characterized by E0=250E_0=250 meV, σ=40\sigma=40 meV, αmax⁡=0.80\alpha_{\max}=0.80, nc=2.0×1012 cm−2n_c=2.0\times10^{12}\ \mathrm{cm}^{-2}, and Vc=0.30V_c=0.30 V. Without loss of generality we set Vth=0V_{\mathrm{th}}=0, so that all gate voltages are measured relative to threshold. A finite VthV_{\mathrm{th}} merely translates the VGV_G axis and leaves every result unchanged.

Carrier distributions at fixed density

Figure 2(a) compares feq(E)f_{\mathrm{eq}}(E) and fneq(E)f_{\mathrm{neq}}(E) at the same carrier density nn. The equilibrium distribution decays exponentially with thermal energy kBT=26k_B T=26 meV, while the nonequilibrium distribution is depressed at low energies (due to normalization) but acquires a prominent Gaussian bump at E0=250E_0=250 meV.

Figure 2. (a) Carrier distribution functions f(E)f(E) for the equilibrium (Boltzmann, blue dashed) and nonequilibrium (Boltzmann + Gaussian bump, red solid, α=0.34\alpha=0.34) cases, both normalized to the same carrier density nn. The gray dotted curve on the right axis shows the transmission T(E)T(E); vertical dashed lines mark the onset energy EbE_{\rm b} and the hot-carrier peak energy E0E_{0}. (b) Spectral current integrand j(E)f(E)j(E)f(E) for both distributions (shared energy axis with (a)). The shaded region is the hot-carrier excess contribution to the current. Despite identical carrier densities, the nonequilibrium current is enhanced by IDneq/IDeq≈5.1I_{\rm D}^{\rm neq}/I_{\rm D}^{\rm eq}\approx5.1 at this density, reflecting the larger per-carrier current of the hot-carrier excess near E0E_{0} (vˉneq/vˉeq≈9.4\bar{v}_{\rm neq}/\bar{v}_{\rm eq}\approx9.4 for the pure Gaussian component).

Figure 2(b) shows the spectral current integrand j(E)f(E)j(E)f(E), which determines the contribution to IDI_D at each energy. The equilibrium integrand is concentrated at low energies where most carriers reside, but the barrier T(E)T(E) strongly suppresses the contribution below EbE_b. The nonequilibrium integrand acquires a large additional peak at E0E_0, because those hot carriers have both higher velocity v(E0)=vTE0/kBTv(E_0)=v_T\sqrt{E_0/k_B T} and unit transmission T(E0)≈1T(E_0)\approx1. The ratio of integrated spectral currents is vˉneq/vˉeq≈9.4\bar v_{\mathrm{neq}}/\bar v_{\mathrm{eq}}\approx9.4, meaning each hot carrier contributes 9.4×9.4\times more current than a thermal carrier at T=300T=300 K. This value 9.4 quantifies the per-carrier current contribution in the pure Gaussian component (i.e., vˉneq/vˉeq\bar v_{\mathrm{neq}}/\bar v_{\mathrm{eq}}), whereas the actual mixed-distribution current enhancement at α=α(n0,VD=0.5 V)\alpha=\alpha(n_0,V_D=0.5\ \mathrm{V}), weighted by the relative populations of the two components, is the smaller value IDneq/IDeq≈5.1I_D^{\mathrm{neq}}/I_D^{\mathrm{eq}}\approx5.1 displayed in Figure 2(b).

Steady-state transport prediction

Figure 3(a) shows ID(VG)I_D(V_G) at VD=0.5V_D=0.5 V. Both curves increase linearly with VGV_G, as expected from ID∝nI_D\propto n, but the nonequilibrium current is enhanced by a factor that grows with VGV_G up to n∼ncn\sim n_c and then saturates. This saturation is traced in Figure 3(b): as n≫ncn\gg n_c, α(n)\alpha(n) saturates at αmax⁡h(VD)\alpha_{\max}h(V_D), and vˉ(α)\bar v(\alpha) approaches a constant. The vertical line at VG∗V_G^* marks the crossover density n=ncn=n_c. At this gate voltage, the current enhancement is maximal in a differential sense—the nonequilibrium IDI_D increases faster with VGV_G than the equilibrium baseline, a hallmark of the gm(α)g_m^{(\alpha)} contribution.

Four-panel plots of current–voltage, nonequilibrium amplitude, transconductance, and peak height

Figure 3. Main steady-state prediction, combining the current–voltage and transconductance characteristics. (a) Drain current IDI_D vs gate voltage VGV_G at VD=0.5V_D = 0.5 V (logarithmic scale), normalized to the equilibrium saturation current. The nonequilibrium current (red solid) exceeds the equilibrium baseline (blue dashed) due to the hot-carrier velocity enhancement vˉ(α)\bar{v}(\alpha); the vertical dotted line marks VG∗V_G^* where n=ncn = n_c. (b) Left axis: nonequilibrium amplitude α(VG)\alpha(V_G), showing saturation for VG≫VG∗V_G \gg V_G^*. Right axis: effective velocity ratio vˉ(α)/vˉeq\bar{v}(\alpha)/\bar{v}_{\mathrm{eq}}, which reaches its maximum as α→αmax⁡h(VD)\alpha\to\alpha_{\max}h(V_D). (c) Transconductance gm(VG)g_m(V_G) for the equilibrium (gmeqg_m^{\mathrm{eq}}, blue solid) and nonequilibrium (total, red solid) cases, decomposed into the density-driven term gm(n)g_m^{(n)} (orange dashed) and the distribution-shape-driven term gm(α)g_m^{(\alpha)} (purple dotted, filled). The equilibrium gmg_m is flat (conventional) and the total nonequilibrium gmg_m rises monotonically; the anomalous peak near VGpkV_G^{\mathrm{pk}} resides in the shape contribution gm(α)g_m^{(\alpha)} (purple), isolated from the smooth monotone density background gm(n)g_m^{(n)}. (d) Peak height of gm(α)g_m^{(\alpha)} vs drain voltage VDV_D, following h(VD)=VD2/(VD2+Vc2)h(V_D) = V_D^2/(V_D^2 + V_c^2) (gray dashed); the vertical dotted line marks Vc=0.3V_c = 0.3 V. electric-field scale for hot-carrier generation.

The central experimental prediction is shown in Fig. 3(c), which plots gm(VG)g_m(V_G) together with its decomposition. The equilibrium transconductance is constant at (WCox/L)vˉeq(W C_{\mathrm{ox}}/L)\bar v_{\mathrm{eq}} above threshold, as expected from conventional theory. For the representative parameters used here, the total nonequilibrium gmg_m itself rises monotonically with VGV_G and saturates at a higher plateau ((WCox/L)vˉ(αmax⁡h(VD)))((W C_{\mathrm{ox}}/L)\bar v(\alpha_{\max}h(V_D))) than the equilibrium value, because the density term gm(n)g_m^{(n)} is monotone by construction and dominates the magnitude. The nonthermal signature is therefore carried not by the total gmg_m but by its shape-driven contribution gm(α)=gm−gm(n)g_m^{(\alpha)}=g_m-g_m^{(n)}, which exhibits a pronounced anomalous peak at VGpkV_G^{\mathrm{pk}} [Eq. (20)] and vanishes in any density-only description (where gm(α)=0g_m^{(\alpha)}=0). Because the density background gm(n)=(WCox/L)vˉ(α)g_m^{(n)}=(W C_{\mathrm{ox}}/L)\bar v(\alpha) is smooth and monotone, this localized peak appears as a shoulder on gm(VG)g_m(V_G) and is isolated by subtracting the calibrated monotone background. It is this shape term—the gate-voltage derivative of the distribution shape—that directly encodes spectral information beyond nn. This is the key signature of transconductance spectroscopy. With the present parameters (T=300T=300 K, αmax⁡=0.80\alpha_{\max}=0.80, σ=40\sigma=40 meV, VD=0.5V_D=0.5 V), Eq. (19) gives npk/nc≈0.38n_{\mathrm{pk}}/n_c\approx0.38, so the peak appears at VGpk≈55V_G^{\mathrm{pk}}\approx55 mV, well below the reference scale VG∗=qnc/Cox≈143V_G^*=q n_c/C_{\mathrm{ox}}\approx143 mV.

The same peak decomposes gmg_m into its two components. The density-driven term gm(n)g_m^{(n)} is a smoothly increasing function that inherits the VGV_G dependence of vˉ(α)\bar v(\alpha), which rises monotonically as α\alpha increases. The shape-driven term gm(α)g_m^{(\alpha)} is localized near VG∗V_G^* and vanishes both below threshold (n→0n\to0) and for n≫ncn\gg n_c (dα/dn→0d\alpha/dn\to0). The anomalous peak thus resides entirely in gm(α)g_m^{(\alpha)}, while the total gmg_m remains monotone—underscoring that the signature is a feature of the distribution-shape contribution, isolated from the smooth density background, rather than a maximum of the raw transconductance.

Figure 3(d) shows the peak height of gm(α)g_m^{(\alpha)} as a function of VDV_D. It follows h(VD)=VD2/(VD2+Vc2)h(V_D)=V_D^2/(V_D^2+V_c^2) (gray dashed), confirming Eq. (17). For VD≲VcV_D\lesssim V_c, gm(α)g_m^{(\alpha)} grows quadratically with VDV_D, while for VD≫VcV_D\gg V_c it saturates. This VDV_D-tunability provides an additional experimental signature: measuring gm(VG)g_m(V_G) at several VDV_D values and fitting the peak height to h(VD)h(V_D) extracts VcV_c, which encodes the

Discrimination from quasi-equilibrium models

The anomalous peak in the shape contribution gm(α)g_m^{(\alpha)} is the central signature of the non-thermal mechanism, but it is not, by itself, unique to it. Two quasi-equilibrium descriptions must be excluded before a peak extracted from gm(VG)g_m(V_G) can be attributed to the distribution shape: a gate-dependent field-effect mobility μ(VG)\mu(V_G),

and an effective-temperature (heated-Boltzmann) ensemble. Both fail, and the way they fail provides the decisive, falsifiable signature of the mechanism.

Exclusion of a mobility-rolloff peak

A gate-dependent mobility μ(VG)\mu(V_G)—from surface-roughness or phonon scattering that strengthens with the vertical field [32]—also produces a gmg_m peak. The two are separated by their drain-voltage dependence. In the linear regime any quasi-equilibrium mobility description gives

ID=(qW/L) n μ(VG) VDI_D=(qW/L)\,n\,\mu(V_G)\,V_D, so

gm(VG,VD)=VDqWLd[n μ(VG)]dVG≡VDG(VG),(21)g_m(V_G,V_D)=V_D\frac{qW}{L}\frac{d[n\,\mu(V_G)]}{dV_G}\equiv V_DG(V_G), \tag*{(21)}

which is separable in VGV_G and VDV_D: the peak position is VDV_D-independent and the peak height scales linearly with VDV_D. The shape mechanism violates both.

By Eqs. (19)–(20) the peak position shifts to lower gate voltage as VDV_D grows, while the peak height saturates as h(VD)=VD2/(VD2+Vc2)h(V_D)=V_D^2/(V_D^2+V_c^2) [Figure 3(d)]. For the present parameters the position shifts by ΔVGpk≈70\Delta V_G^{\mathrm{pk}}\approx70 mV between VD=0.1V_D=0.1 and 1.01.0 V [Figure 4(a)], and the height saturates above VD∼VcV_D\sim V_c [Figure 4(b)]. The simultaneous position shift and height saturation are not reproduced by any separable, quasi-equilibrium mobility-rolloff model [Eq. (21)]. We note that a contrived contact or mobility response with strong, nonseparable VDV_D dependence could in principle mimic one of the two signatures, so the discriminating power rests on their joint appearance rather than on either alone.

Figure 4. Discrimination of the non-thermal distribution-shape peak from quasi-equilibrium descriptions. Top row—exclusion of a mobility-rolloff peak. (a) Density map of the shape-driven transconductance gm(α)(VG,VD)g_m^{(\alpha)}(V_G,V_D), normalized to the equilibrium plateau gmeqg_m^{\mathrm{eq}}; the overlaid curve is the peak trajectory VGpk(VD)V_G^{\mathrm{pk}}(V_D), which shifts to lower VGV_G (≈70\approx70 mV from VD=0.1V_D=0.1 to 1.01.0 V) while the ridge intensity saturates at large VDV_D. (b) The two fingerprints versus VDV_D: the peak position VGpkV_G^{\mathrm{pk}} (red, left axis) shifts and saturates, and the peak height (purple, right axis) saturates as h(VD)h(V_D); a separable mobility model [Eq. (21)] instead predicts a VDV_D-independent position (blue dashed) and a height linear in VDV_D (purple dotted). Bottom row—exclusion of an effective-temperature model (VD=0.5V_D=0.5 V). (c) The best-fit heated-Boltzmann Teff≈1680T_{\mathrm{eff}}\approx1680 K (green dash-dot) reproduces ID(VG)I_D(V_G) reasonably well, but only by invoking an electronic temperature far above any quasi-equilibrium heating regime. (d) The same TeffT_{\mathrm{eff}} model gives a flat gm(VG)g_m(V_G), whereas the non-thermal gmg_m rises with VGV_G; the shaded region is the non-thermal excess over the flat TeffT_{\mathrm{eff}} prediction, unaccounted for by any effective-temperature model.

b. Exclusion of an effective-temperature model. Previous treatments of hot carriers in FETs typically introduce an effective electron temperature Teff>TT_{\mathrm{eff}} > T and replace μ\mu by μ(Teff)\mu(T_{\mathrm{eff}}) [11, 12]. Such models assume that the heated distribution remains a displaced Maxwellian, and therefore cannot capture the Gaussian hot-carrier component in Eq. (7), which represents a localized non-thermal excess generated, e.g., by impact ionization, optical pumping at a specific photon energy, or injection over a discrete barrier. We find the optimal TeffT_{\mathrm{eff}} by minimizing the least-squares residual in ID(VG)I_D(V_G). For the present MoS2_2-like parameters at VD=0.5V_D = 0.5 V this gives Teff≈1680T_{\mathrm{eff}} \approx1680 K, far beyond the regime in which a simple quasi-equilibrium heating picture is physically transparent. Yet even at this best-fit TeffT_{\mathrm{eff}}, which reproduces ID(VG)I_D(V_G) reasonably well [Fig. 4(c)], the model fails for gmg_m [Fig. 4(d)]: because gmTeff=(WCox/L) vˉ(Teff)g_m^{T_{\mathrm{eff}}} = (W C_{\mathrm{ox}}/L)\,\bar{v}(T_{\mathrm{eff}}) is a constant in VGV_G, the model predicts a flat transconductance, whereas the non-thermal gmg_m rises with VGV_G and carries the localized shape contribution gm(α)g_m^{(\alpha)} that a heated Boltzmann distribution cannot generate. The shaded region in Figure 4(d) is the resulting non-thermal excess of gmg_m over the flat TeffT_{\mathrm{eff}} prediction—a direct fingerprint of the non-thermal character of fneqf_{\mathrm{neq}} that no effective-temperature description can reproduce.

c. Feasibility. The anomaly is large and the discriminator is undemanding. The shape-driven peak reaches gm(α),pk∼(0.4–1.4) gmeqg_m^{(\alpha),\mathrm{pk}} \sim(0.4\text{--}1.4)\,g_m^{\mathrm{eq}} for VD=0.1V_D = 0.1--1.01.0 V—an order-unity modulation of an already routinely measured quantity, far above the ≲1%\lesssim1\% resolution of standard DC/lock-in transconductance. The diagnostic position shift of several tens of mV is resolved by sub-mV gate steps, and because E0≫kBTE_0 \gg k_B T the signature survives to room temperature. The decisive measurement is thus a set of gm(VG)g_m(V_G) sweeps at several VDV_D: a peak that shifts and saturates identifies the shape mechanism, whereas a fixed peak with VDV_D-linear height identifies conventional mobility rolloff.

Two-time-scale separation

Beyond the steady state, the framework predicts a specific transient response to pulsed excitation. In the simplest case a pump injects hot carriers at t=0t = 0 and only the hot-carrier amplitude relaxes, α(t)=α0e−t/τE\alpha(t) = \alpha_0 e^{-t/\tau_E}, while the carrier density stays fixed. This single-time-scale pump–probe limit, which yields the energy relaxation time τE\tau_E directly from the decay of gm(t)g_m(t), is treated in Appendix A. In a more general scenario—such as a pulsed gate voltage or optical pump that transiently injects both extra carriers and hot-carrier excitations—two distinct transients appear simultaneously:

ID(t)=[n0+Δne−t/τn] vˉ(α0e−t/τE),(22)I_D(t) = [n_0 + \Delta n e^{-t/\tau_n}]\,\bar{v}(\alpha_0 e^{-t/\tau_E}), \tag*{(22)}

where τn\tau_n is the carrier-density relaxation time (set by the device RC time or dielectric relaxation) and τE\tau_E is the energy relaxation time (set by carrier–phonon scattering). For τn≪τE\tau_n \ll\tau_E, these two time scales are separated and can be independently extracted from a single ID(t)I_D(t) transient.

Figure 5(a) shows ID(t)/IDeqI_D(t)/I_D^{\mathrm{eq}} on a linear scale for τn=0.2\tau_n = 0.2 ps and τE=1.0\tau_E = 1.0 ps. Two regimes are apparent: a fast initial decay (duration ∼τn\sim\tau_n) where the excess density Δn\Delta n equilibrates, followed by a pronounced intermediate plateau where the density has returned to n0n_0 but α\alpha is still large. The current then decays slowly from the plateau to the equilibrium baseline over the time scale τE\tau_E. The height of the plateau encodes the amplitude α0\alpha_0 through IDplateau=n0 vˉ(α0)I_D^{\mathrm{plateau}} = n_0\,\bar{v}(\alpha_0), providing an independent determination of the initial hot-carrier injection level.

Figure 5. Two-time-scale separation in the transient drain current after a pulsed excitation (Δn/n0=0.30\Delta n/n_0 = 0.30, α0=0.60\alpha_0 = 0.60). (a) ID(t)/IDeqI_D(t)/I_D^{\mathrm{eq}} on a linear scale (τn=0.2\tau_n = 0.2 ps, τE=1.0\tau_E = 1.0 ps). The fast density decay (τn\tau_n) and slow spectral decay (τE\tau_E) are separated by an intermediate plateau where nn has equilibrated but hot carriers persist. (b) ΔID(t)/ΔID(0)\Delta I_D(t)/\Delta I_D(0) on a semi-logarithmic scale, revealing two linear segments with slopes 1/τn1/\tau_n (fast, blue dashed) and 1/τE1/\tau_E (slow, purple dashed). (c) Normalized anomaly for τn/τE=0.05\tau_n/\tau_E = 0.05–0.500.50. All curves converge to the common e−t/τEe^{-t/\tau_E} behavior (gray dotted) at long times, demonstrating clean extraction of τE\tau_E when τn/τE≲0.5\tau_n/\tau_E \lesssim0.5.

Figure 5(b) presents the same transient on a semi-logarithmic scale, where the two exponential components appear as two linear segments with distinct slopes 1/τn1/\tau_n and 1/τE1/\tau_E. This is the standard fingerprint of a bi-exponential decay, familiar from optical pump–probe studies [33] [15], now accessible by all-electrical measurements. Fitting the long-time slope directly gives τE\tau_E without contamination from the fast density dynamics, provided measurements are taken at t≳3τnt \gtrsim3\tau_n.

Figure 5(c) shows the normalized anomaly ΔID(t)/ΔID(0)\Delta I_D(t)/\Delta I_D(0) for a range of separation ratios τn/τE=0.05\tau_n/\tau_E = 0.05--0.500.50. In all cases the long-time tail converges to a common e−t/τEe^{-t/\tau_E} behavior (gray dotted reference), confirming that τE\tau_E can be extracted cleanly once τn/τE≲0.5\tau_n/\tau_E \lesssim0.5. In optimized high-bandwidth FET geometries the device-limited density relaxation can in principle be made substantially shorter than the carrier–phonon energy relaxation time τE∼1\tau_E \sim1--55 ps expected for TMDs, placing this system within the resolvable regime. The specific bandwidth requirements are discussed in Sec. IV.

Discussion

a. Why transconductance spectroscopy is distinct from previous hot-carrier work. One might ask whether the present results are simply a restatement of known hot-carrier physics in different notation. They are not, for the following reasons. First, Eq. (15) holds for any f(E)f(E), not just the Boltzmann-plus-Gaussian model used here, provided the spectral kernel j(E)j(E) is gate-independent (corrections from a VGV_G-dependent kernel are addressed below). This means that gm(VG)g_m(V_G) curves measured in any device, under any excitation, are spectral transforms of the underlying distribution. Second, the anomalous peak in gm(α)g_m^{(\alpha)} cannot be reproduced by any density-only or quasi-equilibrium description without invoking an ad hoc nonmonotonic VGV_G-dependence in μ(n)\mu(n). or Teff(n)T_{\mathrm{eff}}(n). Indeed, a monotone μ(n)\mu(n) or monotone Teff(n)T_{\mathrm{eff}}(n) necessarily yields a monotone gm∝vˉ(α(n))g_m \propto\bar{v}(\alpha(n)) (Cox/qC_{\mathrm{ox}}/q) in the linear regime. By contrast, the distribution-shape mechanism predicts the peak position, height, and VDV_D-dependence simultaneously from a single unified parameter set {E0,σ,nc}\{E_0,\sigma,n_c\}, providing a tightly constrained, falsifiable prediction. Third, the measurement is all-electrical: it requires only a standard lock-in gmg_m sweep and is compatible with any gated 2D device, with no need for pump lasers, time-resolved optics, or sub-kelvin cooling.

b. Robustness against gate-dependent kernels. The spectroscopic identity Eq. (15) was derived assuming a gate-independent spectral kernel j(E)=v(E)T(E)j(E)=v(E)T(E). In real devices, VGV_G can also modulate EbE_b (barrier lowering), the contact transmission [31, 34, 35], or the band bending, giving an additional kernel-modulation contribution

gm(j)=qWL∫0∞f(E,VG)∂j(E,VG)∂VG dE,(23)g_m^{(j)}=\frac{qW}{L}\int_0^\infty f(E,V_G)\frac{\partial j(E,V_G)}{\partial V_G}\,dE, \tag*{(23)}

so that in general gm=gm(n)+gm(α)+gm(j)g_m=g_m^{(n)}+g_m^{(\alpha)}+g_m^{(j)}. For the Fermi barrier of Eq. (5), ∂j/∂VG=−v(E)(∂T/∂E)(∂Eb/∂VG)\partial j/\partial V_G=-v(E)(\partial T/\partial E)(\partial E_b/\partial V_G). Since ∂T/∂E\partial T/\partial E is sharply peaked at E=EbE=E_b (width Δ\Delta) with unit area, the kernel term reduces to

gm(j)≃−qWL∂Eb∂VGv(Eb)f(Eb;VG).(24)g_m^{(j)}\simeq-\frac{qW}{L}\frac{\partial E_b}{\partial V_G}v(E_b)f(E_b;V_G). \tag*{(24)}

Two consequences follow. First, gm(j)g_m^{(j)} samples the distribution at the barrier edge EbE_b, not at the hot-carrier energy E0≫EbE_0\gg E_b. Because the bump contributes ∝e−(Eb−E0)2/σ2≈e−25\propto e^{-(E_b-E_0)^2/\sigma^2}\approx e^{-25} there, gm(j)g_m^{(j)} carries no peak structure from fneqf_{\mathrm{neq}} and adds only a smooth background to gm(VG)g_m(V_G)—it cannot reproduce the anomalous peak. Second, its magnitude is set by the barrier-lowering rate ∂Eb/∂VG\partial E_b/\partial V_G, an electrostatic/contact property that for a well-gated channel in the on-state is small (∂Eb/∂VG≪1\partial E_b/\partial V_G\ll1). The distinguishing signature of gm(α)g_m^{(\alpha)} from gm(j)g_m^{(j)} is their qualitatively different VDV_D dependence: gm(α)g_m^{(\alpha)} scales as h(VD)=VD2/(VD2+VC2)h(V_D)=V_D^2/(V_D^2+V_C^2) [xref]Fig. 3(d) and saturates at large VDV_D, while gm(j)g_m^{(j)} contributions (barrier-lowering, contact modulation) at fixed VGV_G generally scale linearly or only weakly in VDV_D. Measuring the VDV_D-dependence of the anomalous peak thus provides an in-situ separation of distribution-shape and kernel-modulation mechanisms. In practice, since contact and barrier-modulation effects can themselves exhibit nonlinear VDV_D behavior, the most robust criterion is not a single VDV_D-dependence alone but the simultaneous consistency of peak position, peak height, and the joint fit to IDI_D and gmg_m (Appendix B, Figure 7)—the same over-constrained signature that excludes the TeffT_{\mathrm{eff}} description.

Figure 7. Simultaneous fit of gm(VG)g_m(V_G) and ID(VG)I_D(V_G) using the non-thermal model fneqf_{\mathrm{neq}} (red dashed) versus the best-fit Teff≈1680T_{\mathrm{eff}} \approx1680 K model (green dash-dot). Black circles are synthetic data generated from the full non-thermal model with 3 % Gaussian noise added (realistic noise level for transport measurements). (a) gm(VG)g_m(V_G): the non-thermal fit reproduces the data, including the shape contribution gm(α)g_m^{(\alpha)} (purple dotted) that carries the anomalous peak; the TeffT_{\mathrm{eff}} model gives a flat response with no such shape term. (b) ID(VG)I_D(V_G): both models approximate the current, but only fneqf_{\mathrm{neq}} uses the same four parameters as panel (a). The simultaneous fit over-constrains the model and demonstrates that spectral shape—not density or temperature alone—governs transport.

c. Other device-related mechanisms and the over-constrained criterion. We emphasize that the appearance of a single peak in gm(VG)g_m(V_G) is not, by itself, claimed as a unique fingerprint of hot carriers. Contact resistance and its bias dependence, interface traps, self-heating, field- or density-dependent mobility degradation, and gate-induced barrier modulation can each produce nonmonotonic features in the raw transconductance. What discriminates the distribution-shape mechanism is not the peak alone but the joint, over-constrained behavior of several observables that these parasitic effects do not reproduce together: (i) the peak position VGpkV_G^{\mathrm{pk}} shifts to lower gate voltage with increasing VDV_D [Eqs. (19)–(20)], whereas a separable mobility or trap response gives a VDV_D-independent position; (ii) the peak height saturates as h(VD)h(V_D) rather than growing linearly in VDV_D; (iii) a single spectral parameter set {E0,σ,αmax⁡,nc}\{E_0,\sigma,\alpha_{\max},n_c\} must reproduce ID(VG)I_D(V_G) and gm(VG)g_m(V_G) simultaneously (Appendix B); and (iv) the low-bias limit VD≪VCV_D\ll V_C must recover the flat, calibrated equilibrium transconductance with VGpk→VG∗V_G^{\mathrm{pk}} \to V_G^{*}. Contact, trap, and self-heating artifacts are, moreover, suppressible or separable by standard controls—four-probe or transfer-length contact de-embedding, gate-bias- and frequency-dependent trap characterization, and pulsed (low-duty-cycle) measurements that eliminate steady-state self-heating. The mechanism should therefore be regarded as identified by this diagnostic protocol as a whole, not by any single peak.

d. Role of inelastic scattering and the transmission kernel. The energy-selective transmission T(E)\mathcal{T}(E) in Eq. (3) is an effective, elastic-like energy filter calibrated from the equilibrium response; it does not resolve individual inelastic collisions. Inelastic scattering enters our description not through T(E)\mathcal{T}(E) but through the distribution itself—its steady-state hot-carrier amplitude α\alpha and, in the transient case, its relaxation law α(t)=α0e−t/τE\alpha(t) = \alpha_0 e^{-t/\tau_E}. The steady-state spectroscopy is therefore independent of the microscopic collision integral: it reads out whatever nonthermal f(E)f(E) is present, however generated. The time-domain extension (Sec. III; Appendix A) treats energy relaxation as a slow evolution of this distribution parameter relative to the instantaneous transport response, which is justified when the transit/RC time is short compared with τE\tau_E. It is not a microscopic transport theory of energy-nonconserving events during a single traversal, and it would break down if inelastic mean free paths became comparable to the channel length, so that T\mathcal{T} itself acquired a strong, energy-redistributing bias dependence. That regime lies beyond the present kernel approximation.

e. Over-constrained simultaneous fit. Beyond excluding the effective-temperature description (Sec. III C), the non-thermal model provides an over-constrained demonstration: a single parameter set {E0,σ,αmax⁡,nc}\{E_0,\sigma,\alpha_{\max},n_c\} simultaneously reproduces both gm(VG)g_m(V_G) and ID(VG)I_D(V_G), whereas the TeffT_{\mathrm{eff}} model adds one free parameter yet already fails at the first observable (gmg_m). A quantitative simultaneous fit to synthetic data, confirming that the spectral shape of f(E)f(E)—not its integrated density or effective temperature—governs transport, is given in Appendix B (Figure 7).

f. Spectroscopic information from the gmg_m anomaly. Equations (19)–(20) show that the peak position VGpkV_G^{\mathrm{pk}} constrains two quantities: (i) ncn_c, extracted from VGpkV_G^{\mathrm{pk}} in the weak hot-carrier limit (VD≪VcV_D \ll V_c) where npk→ncn_{\mathrm{pk}} \to n_c; and (ii) the Gaussian width σ\sigma (via Nneq≈σπN_{\mathrm{neq}} \approx\sigma\sqrt{\pi}), extracted from the shift ΔVG∗(VD)=VG∗−VGpk(VD)\Delta V_G^{*}(V_D) = V_G^{*} - V_G^{\mathrm{pk}}(V_D) as a function of VDV_D. Independently, the peak height gm(α),pk∝JneqNneq∝vˉneq−vˉeqg_m^{(\alpha),\mathrm{pk}} \propto J_{\mathrm{neq}}N_{\mathrm{neq}} \propto\bar{v}_{\mathrm{neq}}-\bar{v}_{\mathrm{eq}} is sensitive to E0E_0 through v(E0)=vTE0/kBTv(E_0) = v_T\sqrt{E_0/k_B T}, providing the hot-carrier peak energy. By performing gm(VG)g_m(V_G) measurements at multiple VDV_D values one can further extract VcV_c, and temperature-dependent sweeps give EbE_b and Δ\Delta from vˉeq(T)\bar{v}_{\mathrm{eq}}(T). Taken together, a single set of gmg_m–VGV_G–VDV_D measurements at varied TT constrains the parameters {nc,σ,E0,Vc,Eb,Δ}\{n_c,\sigma,E_0,V_c,E_b,\Delta\}, complementary to optical techniques that typically access only one or two of these simultaneously. These extractions are model-assisted: they presuppose the calibrated spectral-kernel form (Step 1) and a localized excess, so the inferred values are best regarded as effective parameters of that description rather than uniquely model-independent quantities. In particular, E0E_0 enters the peak height only together with αmax⁡\alpha_{\max} and the kernel, and is not fixed by the height alone.

g. Two-time-scale separation as the third experimental pillar. Equation (22) establishes a hierarchy of electrical measurements for complete spectroscopy of f(E)f(E). The steady-state gm(VG)g_m(V_G) sweep (Pillar I: Figure 3) constrains the spectral parameters {E0,σ,nc}\{E_0,\sigma,n_c\} through the peak position and height. The over-constrained consistency between gmg_m and IDI_D (Pillar II) excludes any effective-temperature description in the main text [Figure 4(c,Figure 4d)]. A quantitative simultaneous fit to synthetic data, which verifies that the non-thermal model is strongly constrained by the joint gmg_m–IDI_D data, is provided as a supplementary demonstration in Appendix B (Figure 7). The two-time-scale transient (Pillar III: Figure 5) then isolates the energy relaxation time τE\tau_E from the density relaxation time τn\tau_n in a single ID(t)I_D(t) measurement by exploiting the bi-exponential structure of Eq. (22). Together these three observables constrain the full parameter set {E0,σ,nc,τE}\{E_0,\sigma,n_c,\tau_E\} within the calibrated model from electrical measurements alone, without relying on optical spectroscopy for readout. Whereas Pillars I and II rely on standard DC/lock-in sweeps, Pillar III requires GHz-to-THz-bandwidth time-domain instrumentation and is therefore best viewed as a specialized extension that complements the steady-state observables rather than replacing them.

h. Experimental measurement protocol. We outline a self-contained protocol by which the spectroscopic parameters can be extracted sequentially from standard electrical measurements. The minimum experimental dataset for verifying the non-thermal mechanism and extracting {nc,σ,Vc,E0,αmax⁡}\{n_c,\sigma,V_c,E_0,\alpha_{\max}\} is a set of gm(VG)g_m(V_G) traces at several VDV_D values (Steps 2–4 below), which requires only a standard DC/lock-in FET sweep. The time-domain extension (Step 5) is optional and yields the additional parameter τE\tau_E.

Step 1 — equilibrium calibration. At low drain voltage VD≪VcV_D \ll V_c, measure ID(VG)I_D(V_G) and gm(VG)g_m(V_G) at several temperatures T=200T = 200–350350 K. The transconductance is flat and equals T(WCoxv‾eq(T))\mathcal{T}(W C_{\mathrm{ox}} \overline{v}_{\mathrm{eq}}(T)), giving the equilibrium velocity vˉeq(T)\bar{v}_{\mathrm{eq}}(T), which follows the thermally averaged spectral kernel and increases with temperature. Fitting this temperature dependence to Eqs. (8)–(11) with α=0\alpha= 0 extracts the transport-kernel onset parameters EbE_b and Δ\Delta.

Step 2 — hot-carrier signature. Increase VDV_D to values comparable to or exceeding VcV_c. Subtract the smooth equilibrium-calibrated density background gm(n)g_m^{(n)} from the measured gm(VG)g_m(V_G). If the residual shape contribution gm(α)g_m^{(\alpha)} develops a localized peak, the effective-temperature description has broken down and the non-thermal model is required. The absence of such a peak at the same VDV_D and TT would instead confirm quasi-equilibrium conditions, providing a built-in null test. Repeat at several VDV_D values (VD=0.2V_D = 0.2–0.70.7 V recommended) to track the peak height as a function of VDV_D.

Step 3 — VDV_D-dependent peak height. Plot the peak height gm(α),pk(VD)g_m^{(\alpha),\mathrm{pk}}(V_D) against VDV_D. A fit to h(VD)=VD2/(VD2+Vc2)h(V_D) = V_D^2/(V_D^2 + V_c^2) [Eq. (17)] extracts VcV_c, the electric-field crossover for hot-carrier generation. The peak position VGpk(VD)V_G^{\mathrm{pk}}(V_D) simultaneously yields npk(VD)n_{\mathrm{pk}}(V_D) via Eq. (12); fitting to Eq. (19) with VcV_c known then extracts ncn_c and the spectral width σ\sigma jointly.

Step 4 — simultaneous fit and falsification. Using the parameters {Eb,Δ,Vc,nc,σ}\{E_b,\Delta,V_c,n_c,\sigma\} from Steps 1–3, compute ID(VG)I_D(V_G) and gm(VG)g_m(V_G) at all measured VDV_D values. The non-thermal model should reproduce both observables with strongly reduced parameter freedom: only E0E_0 and αmax⁡\alpha_{\max} remain as global fit parameters, both already constrained by Step 2. As a falsification test, confirm that the best-fit TeffT_{\mathrm{eff}} model gives a flat gm(VG)g_m(V_G) and fails to capture the shape contribution gm(α)g_m^{(\alpha)} [Figure 4(d)].

Step 5 — time-domain extraction of τE\tau_E. Apply a fast electrical pulse (gate or drain) with a sub-nanosecond to picosecond-scale rise time and record ID(t)I_D(t) with high-bandwidth (GHz-to-THz) instrumentation. Fit the bi-exponential transient [Eq. (22)] to extract τn\tau_n (the fast, density-relaxation component, set by device RC) and τE\tau_E (the slow, spectral-relaxation component, set by carrier–phonon scattering). The full parameter set {E0,σ,nc,Vc,Eb,Δ,τE}\{E_0,\sigma,n_c,V_c,E_b,\Delta,\tau_E\} is then constrained within the calibrated model from all-electrical measurements.

i. Applicability and experimental scales. The method applies whenever a localized nonthermal excess is separated from the transport onset by more than the thermal broadening, E0−Eb≳E_0 - E_b \gtrsim few ×kBT\times k_B T, so that the hot-carrier feature is not washed out by the Boltzmann tail. For the representative TMD parameters used here (Eb=50E_b = 50 meV, E0=250E_0 = 250 meV) this is comfortably satisfied at room temperature (kBT≈26k_B T \approx26 meV), and the signature therefore survives to 300 K and above. Lowering the temperature sharpens the spectral resolution but can enhance competing trap and contact effects; raising it broadens both the thermal background and the hot-carrier excess. In gate voltage, the relevant density scale is n∼nc∼1012n \sim n_c \sim10^{12} cm−2^{-2}, reached routinely in strongly gated, high-κ\kappa TMD FETs, and short channels (L∼10L \sim10–100100 nm) maximize the high-field region that generates the excess. The steady-state observables (Pillars I and II) require only standard DC/lock-in instrumentation, whereas the optional time-domain extraction of τE\tau_E (Pillar III) assumes energy-relaxation times τE∼0.1\tau_E \sim0.1–1010 ps typical of TMDs and hence GHz-to-THz measurement bandwidth. Outside these ranges—strongly degenerate channels, E0E_0 comparable to EbE_b, or dominant elastic disorder or spatial inhomogeneity (e.g. threshold variation or contact nonuniformity) that smears the onset—the shape contribution persists in principle but its peak may be too weak or too broad to resolve.

j. Material and geometry considerations. The signal-to-noise ratio for detecting gm(α)g_m^{(\alpha)} is maximized when vˉneq/vˉeq\bar{v}_{\mathrm{neq}}/\bar{v}_{\mathrm{eq}} is large, which requires E0≫EbE_0 \gg E_b and E0≫kBTE_0 \gg k_B T. TMDs with high-κ\kappa gate dielectrics (large CoxC_{\mathrm{ox}}) shift VG∗V_G^* to more accessible gate voltages. Short-channel devices (L∼10L \sim10–5050 nm) increase the high-field region and hence α\alpha at a given VDV_D. For the pump-probe time-domain experiment, the bandwidth requirement is ∼1/τ∼1\sim1/\tau\sim1 THz for TMD materials [36], suggesting that THz near-field probing or GHz on-chip circuitry could realize this measurement.

Conclusion

We have established that the transconductance gmg_m of a 2D FET is fundamentally a spectroscopic quantity. It is the spectral current j(E)j(E) weighted by the gate-voltage derivative of the carrier distribution f(E)f(E) and integrated over energy. As such, it encodes the shape of f(E)f(E), not merely its integrated density. This spectroscopic identity [Eq. (15)] is exact for a gate-independent spectral kernel. It gives rise to an anomalous peak in the distribution-shape contribution gm(α)g_m^{(\alpha)}, isolated from the smooth monotone density background of the measured gmg_m. This peak (i) has no counterpart in equilibrium transport, (ii) cannot be reproduced by any quasi-equilibrium description without invoking ad hoc nonmonotonic VGV_G-dependence in transport coefficients, and (iii) constrains the spectral parameters of the hot-carrier distribution through its position, height, and drain-voltage dependence.

Two steady-state observables, accessible with standard DC/lock-in instrumentation, provide the primary experimental verification of the non-thermal character of the distribution. Pillar I is the anomalous peak in the shape contribution gm(α)g_m^{(\alpha)} (Figure 3). Its decisive drain-voltage fingerprint—a peak that shifts to lower VGV_G and saturates in height with VDV_D—distinguishes it from a conventional mobility-rolloff peak, which would instead remain fixed in position and grow linearly with VDV_D [Figure 4(a,Figure 4b)]. Pillar II is the exclusion of any effective-temperature description, which gives a flat gm(VG)g_m(V_G) and fails to reproduce the shape contribution gm(α)g_m^{(\alpha)} even at its best fit [Figure 4(c,Figure 4d)]. The over-constrained simultaneous fit of both IDI_D and gmg_m with a single parameter set {E0,σ,αmax⁡,nc}\{E_0,\sigma,\alpha_{\max},n_c\} provides a supplementary quantitative confirmation (Appendix B, Figure 7). Together these constrain the effective spectral parameters {E0,σ,nc,Vc,Eb,Δ}\{E_0,\sigma,n_c,V_c,E_b,\Delta\} without optical spectroscopic readout. The energy relaxation time τE\tau_E becomes accessible through a more demanding time-domain extension—the bi-exponential transient ID(t)I_D(t) (Pillar III, Figure 5)—which separates the energy relaxation from the density relaxation τn\tau_n but requires GHz-to-THz-bandwidth instrumentation.

The transconductance provides a spectroscopic probe of carrier distributions, revealing non-thermal features that lie beyond the reach of any quasi-equilibrium description without ad hoc nonmonotonic gate-voltage dependence. This capability positions gated 2D semiconductors as on-chip distribution-function spectrometers, opening pathways to (i) systematic mapping of hot-carrier dynamics across phonon channels, valley degeneracies, and temperatures in van der Waals monolayers; (ii) engineering of gm(α)g_m^{(\alpha)} peaks via dielectric and channel geometry for energy-selective carrier amplification; and (iii) extending transconductance spectroscopy to multi-band and spin-orbit-coupled 2D systems where distribution-shape effects couple to valley and spin degrees of freedom.

Acknowledgments

The author thanks S. Adhikary and T. Kameda for useful discussions. This work was supported by JSPS KAKENHI (Grants No. JP25K01609, No. JP22H05473, and No. JP21H01019).

Data availability

The data that support the findings of this study, and the code used to generate the figures, are available from the corresponding author upon reasonable request.

Pump–probe transient at fixed carrier density

The two-time-scale analysis of the main text reduces, in the limit of a fixed carrier density, to a single-exponential pump–probe transient that isolates the energy relaxation time τE\tau_E. Consider a pump (optical or electrical) that injects hot carriers at t=0t=0, establishing α(t=0)=α0\alpha(t=0)=\alpha_0. For t>0t>0 the hot carriers relax via carrier–phonon scattering, α(t)=α0e−t/τE\alpha(t)=\alpha_0 e^{-t/\tau_E}, with τE\tau_E the energy relaxation time. In a pump–probe geometry, the gate is held fixed at VG=VG∗V_G=V_G^* and a small AC modulation δVG\delta V_G probes the instantaneous transconductance. Because the pump-generated α\alpha is externally imposed and does not depend on VGV_G, only the density-driven term survives:

gm(t)=WCoxL vˉ(α0e−t/τE).(25)g_m(t)=\frac{W C_{\mathrm{ox}}}{L}\,\bar{v}\left(\alpha_0 e^{-t/\tau_E}\right). \tag*{(25)}

Figures 6(a) and (b) show the resulting transients in ID(t)I_D(t) and gm(t)g_m(t) for τE=1\tau_E=1 ps, normalized to their equilibrium values. Both decay monotonically from their pump-excited values back to equilibrium. The decay is exponential in α(t)\alpha(t) but not strictly exponential in IDI_D or gmg_m, because vˉ(α)\bar{v}(\alpha) is nonlinear in α\alpha. Fitting the transient to Eq. (25) therefore provides τE\tau_E and simultaneously vˉneq/vˉeq\bar{v}_{\mathrm{neq}}/\bar{v}_{\mathrm{eq}}, a ratio that encodes E0E_0 (the hot-carrier energy) via Eqs. (8)–(11).

Figure 6(c) shows normalized transients Δgm(t)/Δgm(0)\Delta g_m(t)/\Delta g_m(0) for several relaxation times τE\tau_E.

Figure 6. Time-resolved response after pump excitation that injects hot carriers with α0=0.6\alpha_0 = 0.6 at t=0t = 0, for τE=1\tau_E = 1 ps. (a) Drain current transient ID(t)/IDeqI_D(t)/I_D^{\mathrm{eq}} at VG=VG∗V_G = V_G^*. (b) Transconductance transient gm(t)/gmeqg_m(t)/g_m^{\mathrm{eq}}, which decays from the nonequilibrium value back to equilibrium as hot carriers relax. (c) Normalized transconductance anomaly Δgm(t)/Δgm(0)\Delta g_m(t)/\Delta g_m(0) for relaxation times τE=0.5\tau_E = 0.5, 1.0, 2.0, and 5.0 ps (blue, green, red, purple). The gray dotted line shows the pure exponential e−t/τEe^{-t/\tau_E} for comparison, illustrating the slight nonlinearity due to vˉ(α)\bar{v}(\alpha).

The curves are distinguishable over a decade in τE\tau_E within a 6 ps window, demonstrating that the measurement discriminates phonon-scattering channels with sub-picosecond resolution, consistent with time-resolved Raman measurements of longitudinal-optical (LO) phonon populations in III–V semiconductors [33]. Compared to optical pump–probe techniques [15, 16] and recent electrical photoresponse measurements in van der Waals heterostructures [37], the present approach directly accesses the transport-weighted carrier distribution rather than the optical spectral weight or the transient photocurrent, and can be combined with electrostatic gating to tune nn and VG∗V_G^* in situ.

Simultaneous fit to synthetic data

Figure 7 demonstrates quantitatively that the non-thermal model fneqf_{\mathrm{neq}} [Eq. (7)] with the single parameter set {E0,σ,αmax⁡,nc}\{E_0,\sigma,\alpha_{\max},n_c\} simultaneously reproduces both gm(VG)g_m(V_G) [panel (a)] and ID(VG)I_D(V_G) [panel (b)]. The two observables probe fundamentally different moments of f(E)f(E): gmg_m is sensitive to ∂f/∂VG\partial f/\partial V_G and therefore to the shape of ff, while IDI_D tracks the total spectral weight. A model that fits both with a single parameter set is therefore over-constrained, providing unambiguous evidence that the spectral shape of f(E)f(E) governs transport, not merely its integrated density or effective temperature. The TeffT_{\mathrm{eff}} model cannot achieve this: it adds only one free parameter yet already fails at the first observable (gmg_m).

Robustness against modeling choices

The two phenomenological ingredients of the model—the saturation function g(n)g(n) in the hot-carrier amplitude α(n)\alpha(n) and the energy-selective transmission T(E)\mathcal{T}(E)—are not derived from first principles. We show here that the anomalous gm(α)g_m^{(\alpha)} peak is insensitive to both choices.

Saturation function (n)

The phenomenological amplitude of Eq. (13) uses the saturating form g(n)=n/(n+nc)g(n)=n/(n+n_c). The anomalous peak in the shape-driven transconductance does not rely on this specific choice: any monotonically saturating g(n)g(n) gives a bell-shaped dα/dnd\alpha/dn and hence a localized gm(α)g_m^{(\alpha)} peak. Figure 8(a,b) compares four representative forms—the baseline hyperbola n/(n+nc)n/(n+n_c), a hyperbolic tangent tanh⁡(n/nc)\tanh(n/n_c), an exponential saturation 1−e−n/nc1-e^{-n/n_c}, and a Hill function n2/(n2+nc2)n^2/(n^2+n_c^2)—all normalized to rise from zero and saturate on the scale n∼ncn\sim n_c [Figure 8(a)]. The resulting gm(α)(VG)g_m^{(\alpha)}(V_G) curves [Figure 8(b)], each normalized to its own peak, all exhibit a localized peak near VG∗V_G^* with comparable width. The forms with a steeper onset (tanh, Hill) place the peak slightly lower in VGV_G and decay faster at large VGV_G, but the qualitative signature—a localized gm(α)g_m^{(\alpha)} peak driven purely by the distribution shape—is common to all of them. The peak position is therefore a robust fingerprint of the saturating hot-carrier generation mechanism rather than an artifact of the particular analytic model adopted in the main text.

Transmission T(E)T(E)

The main text models the transmission by the Fermi-function barrier of Eq. (5). The anomaly does not rely on this form either. The shape-driven term enters through the spectral current moments Jeq,neq=∫j(E)feq,neq(E) dEJ_{\mathrm{eq,neq}} = \int j(E)f_{\mathrm{eq,neq}}(E)\,dE with j(E)=v(E)T(E)j(E) = v(E)T(E), and its sign and localization require only that the hot-carrier component at E0>EbE_0 > E_b be transmitted more efficiently per carrier than the thermal component, i.e. vˉneq>vˉeq\bar{v}_{\mathrm{neq}}>\bar{v}_{\mathrm{eq}}. Figure 8(c,d) compares four transmission forms that all rise from 0 to 1 across EbE_b—the baseline Fermi step [1+e(Eb−E)/Δ]−1[1+e^{(E_b-E)/\Delta}]^{-1}, an error-function step, a linear ramp, and the discontinuous hard step Θ(E−Eb)\Theta(E-E_b) [Fig. 8(c)]. The resulting gm(α)(VG)g_m^{(\alpha)}(V_G), normalized to the equilibrium plateau gmeqg_m^{\mathrm{eq}} of the same form [Fig. 8(d)], peaks at the same gate voltage for every transmission—the peak position is fixed by α(n)\alpha(n), not by T(E)\mathcal{T}(E)—while its height varies only mildly, remaining of order the equilibrium response (1.3(1.3–1.8 gmeq)1.8\,g_m^{\mathrm{eq}}) even for the discontinuous hard step. The transmission shape thus sets the magnitude of the anomaly but neither its existence nor its location. Taken together, the two robustness checks are complementary: the peak position is set by α(n)\alpha(n) and is insensitive to T(E)\mathcal{T}(E), while the peak height is set by T(E)\mathcal{T}(E) and the hot-carrier energy—so neither phenomenological input controls the existence of the signature.

Extension to linear (Dirac) dispersion

The framework does not rely on the constant density of states of a parabolic 2D band. For a linear (Dirac) dispersion E=ℏvF∣k∣E = \hbar v_{\mathrm{F}}|\mathbf{k}|, the density of states is D(E)=gE2πℏ2vF2∝∣E∣D(E) = \frac{gE}{2\pi\hbar^{2}v_{\mathrm{F}}^{2}} \propto|E| and the band velocity is energy-independent, v(E)=vFv(E) = v_{\mathrm{F}}. Writing the drain current in the same energy-resolved form Eq. (3), the transport now enters through the spectral kernel

jD(E)=D(E)vFT(E)∝ET(E),(26)j_{\mathrm{D}}(E) = D(E)v_{\mathrm{F}}T(E) \propto ET(E), \tag*{(26)}

which replaces the parabolic-band kernel j(E)=v(E)T(E)∝ET(E)j(E) = v(E)T(E) \propto\sqrt{E}T(E). In both cases j(E)j(E) is a smooth, monotonically increasing energy filter that favors high-energy carriers, so a localized nonthermal excess at E0>EbE_{0} > E_{\mathrm{b}} is still weighted more heavily than the thermal population. Consequently the transconductance retains the spectroscopic form Eq. (15) and the decomposition gm=gm(n)+gm(α)g_{\mathrm{m}} = g_{\mathrm{m}}^{(n)} + g_{\mathrm{m}}^{(\alpha)}, and the anomalous gm(α)g_{\mathrm{m}}^{(\alpha)} peak persists. Only the quantitative details change. Because the Dirac kernel grows linearly rather than as E\sqrt{E}, it weights the hot-carrier excess even more strongly, so the per-carrier enhancement vˉneq/vˉeq\bar{v}_{\mathrm{neq}}/\bar{v}_{\mathrm{eq}}, and hence the peak height, is generally larger. The peak position, set by α(n)\alpha(n) (Appendix C), is unchanged. The same reasoning applies to any monotonic product D(E)v(E)D(E)v(E), so the mechanism is generic to gapped and gapless 2D materials alike.

References

  1. [26]K. Wakabayashi, S. Adhikary, and K. Tsukagoshi, Valley-landscape engineering in bilayer WSe₂ gate-all-around transistors (2026), arXiv:2606.08955 [cond-mat.mes-hall].arxiv.org/abs/2606.08955
  2. [27]K. Wakabayashi, Electrical spectroscopy of intervalley relaxation in WSe₂ transistors, Phys. Rev. B (2026), in press, arXiv:2606.10470 [cond-mat.mes-hall].DOI
  3. [28]The exact group velocity of the 2D parabolic band is v(E) = √2E/m*; here we absorb the O(1) numerical prefactor (the √2 and the angular average of the transport-direction component) into the definition of vT, since only the √E energy dependence enters the spectral shape—overall constants cancel in the velocity ratios v̄(α) and in gm.
  4. [29]K. K. H. Smithe, C. D. English, S. V. Suryavanshi, and E. Pop, High-field transport and velocity saturation in synthetic monolayer MoS₂, Nano Lett. 18, 4516 (2018).DOI
  5. [30]M. A. Wang and E. Pop, Monte carlo simulation of electrical transport with joule heating and strain in monolayer MoS₂ devices, Nano Lett. 25, 6841 (2025).DOI
  6. [31]A. Allain, J. Kang, K. Banerjee, and A. Kis, Electrical contacts to two-dimensional semiconductors, Nat. Mater. 14, 1195 (2015).DOI
  7. [32]K. Kaasbjerg, K. S. Thygesen, and K. W. Jacobsen, Phonon-limited mobility in n-type single-layer MoS₂ from first principles, Phys. Rev. B 85, 115317 (2012).DOI
  8. [33]J. A. Kash, J. C. Tsang, and J. M. Hვან, Subpicosecond time-resolved Raman spectroscopy of LO phonons in GaAs, Phys. Rev. Lett. 54, 2151 (1985).pubmed.ncbi.nlm.nih.gov/10031242
  9. [34]P.-C. Shen, C. Su, Y. Lin, A.-S. Chou, C.-C. Cheng, J.-H. Park, M.-H. Chiu, A.-Y. Lu, H.-L. Tang, M. M. Tavakoli, G. Pitner, X. Ji, Z. Cai, N. Mao, J. Wang, V. Tung, J. Li, J. Bokor, A. Zettl, C.-I. Wu, T. Palacios, L.-J. Li, and J. Kong, Ultralow contact resistance between semimetal and monolayer semiconductors, Nature 593, 211 (2021).DOI
  10. [35]M. Du, W. Li, G. Xiong, C. Zhao, F. Hou, W. Gan, X. Gong, N. Zou, L. Liu, X. Zou, T. Li, W. Sun, D. Fan, Z. Yu, X. Tu, Y. Gao, H. Shen, H. Qiu, L. Ma, J. Wang, Y. Nie, L. Tao, J.-B. Xu, J. Lin, J. Xu, Y. Shi, and X. Wang, Scaled crystalline antimony on ohmic contacts for two-dimensional transistors, Nat. Electron. 8, 1191 (2025).
  11. [36]N. Kumar, J. He, D. He, Y. Wang, and H. Zhao, Charge carrier dynamics in bulk MoS₂ crystal studied by transient absorption microscopy, J. Appl. Phys. 113, 133702 (2013).DOI
  12. [37]M. Massicotte, P. Schmidt, F. Vialla, K. G. Schädler, A. Reserbat-Plantey, K. Watanabe, T. Taniguchi, K. J. Tielrooij, and F. H. L. Koppens, Picosecond photoresponse in van der Waals heterostructures, Nat. Nanotechnol. 11, 42 (2016).DOI

Paper details

vlnc:W_bgbddb23
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 11

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents