Coupled harmonic oscillators are ubiquitous in physics and an exemplary platform to observe them is with cavity exciton-polaritons1,2. Pioneering work reported planar cavity photon and quantum well exciton-polariton formation3,4. Since then a wide range of optical resonances have been coupled to excitonic resonances5,6. Studies have ranged from studying polariton amplification, condensation and interactions with a two-dimensional electron gas7,8,9,10,11,16 to leveraging these quantum matter coherences for light generation17,18,19. But, in almost all the previous work, the focus has been on polaritons formed out of neutral excitons and photons, with little work on other exciton complexes20,21.

The past decade has witnessed the emergence of atomically thin transition metal dichalcogenides (TMDCs) that support strongly bound neutral22, charged exciton resonances23 and defect excitons24. TMDC materials exhibit large exciton binding energies, stabilizing them at room temperature25,26. Recent work demonstrated polariton formation27, and shortly thereafter negatively charged excitons (trions) and polarons were coupled to cavity photons28,29. In this work we study cavity polariton physics with a device that supports dispersing and mutually interacting, out-of-equilibrium, cavity photons, neutral excitons and negatively charged excitons.

Figure 1a presents an illustration of our device architecture (see Methods) containing the single-layer TMDC, molybdenum diselenide (MoSe2), at the cavity anti-node. Natural doping levels are such that the device supports both neutral excitons and trions that can couple with the cavity photon. A conventional approach to describe the above excitations in the system would involve modelling them in terms of bosonic oscillators, as shown in Fig. 1b. We would expect to observe three polariton branches, an upper, middle and lower polariton branch (UPB, MPB, LPB). Such theoretically expected dispersions are presented in Fig. 1c using known energy detunings from our experiments, and this was indeed observed in recent experiments21.

Figure 1: Device architecture and cavity polariton physics.
figure 1

a, Schematic of the optical cavity. A monolayer of MoSe2 is embedded between the top and bottom distributed Bragg reflectors (DBRs). b, A coupled spring and mass system represents the three-coupled-oscillator problem involving the cavity photon (Ph), exciton and trion. g1 and g2 are the coupling constants. c, Dispersion relation for both the uncoupled cavity photon, exciton and trion (dashed lines) and for the coupled cavity photon, exciton and trion (solid lines) assuming three coupled oscillators as illustrated in b. Notice the middle polariton branch (MPB) and lower polariton branch (LPB) exhibit positive dispersion. This is true irrespective of detuning between the cavity photon, exciton and trion. d, A single-shot angle-resolved photoluminescence (PL) measurement of our device exhibiting negative dispersion for the LPB. The solid line is the fit to our theoretical model. The flat dashed lines are the dispersion of the bare MoSe2 exciton and trion PL obtained at 6 K (no optical cavity) and the blue parabola is the bare cavity dispersion. Error bars denote resolution of the experimental set-up.

The previous are in contrast to our observation shown in Fig. 1d. Figure 1d is a measurement of our device’s polariton dispersion. The dispersion of the lower trion-polariton branch is completely inverted (compare to Fig. 1c), which translates to a negative mass for the trion-polaritons in the neighbourhood of kǁ = 0. This is the central observation of our work. In what follows we describe the physics resulting in this spectrum and find it depends on the fermionic nature of the trions and how they interact with cavity photons, the strong interaction between the trions, the MPB and electrons, and the out-of-equilibrium regime of the experiment. Note that the fundamental difference between the earlier observation30 of negative polariton mass and this work is that many-body interactions result in an anomalous trion-polariton dispersion, resulting in charged cavity polaritons exhibiting negative mass in the neighbourhood of kǁ = 0.

In our system, the constituent resonances are the MoSe2 exciton, trion and the cavity photon. The bottom panel of Fig. 2a presents the measured photoluminescence (PL) spectrum of the single-layer MoSe2 flake on a SiO2 substrate. We observe a stress-induced redshift of 8 meV for both the neutral and charged excitons when the monolayer is embedded inside the microcavity. The exciton (trion) linewidth is 7 meV (9 meV). Reflectance spectroscopy of the unloaded cavity (top panel, Fig. 2a) identifies the spectral location of the cavity resonance and reveals a quality factor of 600 and a linewidth of 3 meV. A PL spectrum of the loaded cavity is presented in Fig. 2b.

Figure 2: Optical characterization of the cavity, MoSe2 flake and the polariton device.
figure 2

The same experimental set-up has been used to image the sample plane and to obtain the reflectance of the cavity at normal incidence without the MoSe2 flake using a low NA 0.13 objective. a, Top: Reflectance (R) spectrum obtained at 5 K (NA 0.13). The cavity full-width at half-maximum (FWHM) is 3 meV. Bottom: PL from a bare MoSe2 sample at 5 K exhibiting both the neutral exciton and trion peaks (NA 0.7). b, The PL spectrum of the MoSe2 sample embedded in the cavity at 5 K shows the appearance of the upper (1.66 eV), middle (1.634 eV) and lower (1.621 eV) polariton branches (NA 0.7).

To unmask the polariton dispersion an angle resolving measurement set-up was constructed that images the objective (numerical aperture, NA 0.7) Fourier plane onto a spectrometer (Fig. 3a). The raw angle-resolved spectrum is shown in Fig. 3b, where the colour represents the differentiated intensity of PL in log (the raw PL data are shown in the Supplementary Section 2). As shown in Fig. 3c, it is possible to measure PL spectra at specific values of in-plane momentum. Fits to each of the polariton resonances in the spectra of Fig. 3c map the full polariton dispersion relation as reported in Fig. 1d. We observe a large Rabi splitting of 25 meV at zero in-plane momenta when the cavity photon and exciton are resonant (see Supplementary Section 1a).

Figure 3: Measuring the device dispersion relation.
figure 3

a, Schematic of the optical microscope set-up for the angle-resolved measurements. A beam splitter is used to excite the sample through a high NA (0.7) objective lens and a tube lens is placed to image the Fourier plane of the objective directly on to the imaging spectrometer. b, A colour scale plot of the angle-resolved PL spectrum obtained at 6 K; the logarithm of the raw data is differentiated to enhance the contrast. The solid lines are the fitting obtained from our model as discussed in the text. c, PL spectrum as taken from the line cuts at five different kǁ obtained from the raw data which are presented in Supplementary Section 2.

To understand the microscopic mechanism responsible for the observed inverted dispersion relation and the negative trion mass, we introduce the following Hamiltonian written in momentum space

where ck (ck), ak (ak), ψk (ψk), and Γl, k (Γl, k) are the electron, photon, trion, upper (l = 1) and middle (l = 2) polariton creation (annihilation) operators, respectively; Δkel, ωktr, Ek1, Ek2, are the bare electron, trion, upper, and middle polariton resonance energies. v2 quantifies the interaction strength of the cavity photon, electron and trion and v3 quantifies the interaction strength of the middle polariton, the electron and the trion (see Supplementary Fig. 1). In equation (1) we have neglected the interaction between the upper polariton and the trion as they are detuned from resonance. In this experiment, we have a small relative detuning between the cavity photon and the uncharged exciton and the trion is detuned by 30 meV below the cavity resonance.

Assuming that the free electron concentration is such that the majority of free electrons form trions with the optically excited excitons31, we integrate out the electronic contribution in the model. To the leading order in perturbation theory in v2 and v3 (see Supplementary Section 1b), this generates interaction between the photons and the trions, between the middle polariton branch and the trions, as well as mutual interaction among the photons, middle polariton branch and the trions. In addition, it also renormalizes the bare dispersion of the trion. Now, to account only for the dispersion, we make a mean-field approximation by introducing mean-field occupation numbers for the trions and the polaritons: 〈ψkψk〉 = δkkntk and 〈ΓakΓbk〉 = δkkδabnΓa, k, where ntk and nΓa, k are the respective out-of-equilibrium occupation numbers for the trion and upper (a = 1) and middle (a = 2) polariton modes. Since such out-of-equilibrium occupancy is expected to be very different from the equilibrium Bose–Einstein (for polaritons) or Fermi–Dirac (for trions) distributions, within mean-field theory we take the ratio of occupancies of the MPB and UPB as a k-independent fitting parameter. The data in Fig. 1d are fitted by a ratio of 1.3 in accordance with the PL data (see Fig. 2b).

The mean-field Hamiltonian (see Supplementary Section 1d) can be diagonalized to yield the dispersion relation. The data in Fig. 1d are fitted by our model (seen Supplementary Table 1). In particular, for the trion-polariton branch our mean-field calculations yield the following dispersion

where αk and βk (see Supplementary Section 1c) are two matrix elements that transform the exciton–photon basis to the polariton basis, Etr0 is the bare trion energy and E(αk, βk) captures the many-body effects present in our device. The sign of E(αk, βk) is determined by the relative sign of the coupling constants v2 and v3, and we find the relative sign of v2 and v3 has to be opposite to give the hybridization.

We can understand the above inverted trion-polariton dispersion as resulting from an electron-mediated interaction between the exciton-polariton branch and the trion-polariton that overwhelms the weak trion cavity repulsion, leading to level attraction. The origin of such an interaction may be understood in terms of exchange of an electron between a neutral exciton and the negatively charged trion, and thus gaining delocalization energy of the trion’s electron. Note in ref. 32 that the combination of increased sample doping and zero cavity–trion detuning overwhelms the many-body-interaction-induced inverted dispersion observed in Fig. 1d.

To demonstrate the level attraction is controllable we vary the device temperature. Figure 4a shows data for the temperature-dependent trion-polariton dispersion branch (T = 6 K, 25 K and 45 K). In this temperature range the bare exciton is first red-detuned from the cavity (T = 6 K), brought into resonance near 25 K, and then becomes blue-detuned from the cavity for higher temperatures (above 55 K the trion becomes unstable). The tunability of the LPB negative mass, determined via meff−1 = −2(d2E/dk2), in the vicinity of low kǁ for different temperatures is presented in Fig. 4b (see Supplementary Section 3). In Fig. 4c, the trion-polariton zero momentum energy Etr(0) is plotted. The solid (open) circles present the raw data (model). As the temperature is increased, the exciton is brought into resonance with the optical cavity and the level hybridization between the exciton-polariton and the trion-polariton is enhanced. This is again seen in Fig. 4d, where the energy difference at zero in-plane momentum between the UPB and MPB (ΔEUPB–MPB) and MPB and LPB (ΔEMPB–LPB) is plotted as a function of temperature.

Figure 4: Temperature dependence of the interactions.
figure 4

a, Lower polariton branch dispersion at three temperature points in the range for 6 K to 45 K. The curvature and position of each trion-polariton branch is a function of the detuning between the exciton resonance and cavity photon, which changes with the temperature as discussed in the text. b, LPB effective mass near zero in-plane momentum values for the same temperature points as in a. c, Plot of the trion-polariton resonance at kǁ = 0 as a function of temperature. Solid (open) circles represent data points obtained from the raw data (fitting of the model). Note that the resonance energy is blue shifted and reaches its maximum near 25–30 K when the exciton–cavity photon detuning is minimum. d, Plot of the energy difference between the upper polariton and middle polariton branch at kǁ = 0 (ΔEUPB–MPB) (red) with maximum level repulsion near 25–30 K, leading to maximum level attraction for the energy difference between the middle polariton and lower polariton branch at kǁ = 0 (ΔEMPB–LPB) (black) near the same temperature. Dashed lines are guides to the eye. Error bars denote resolution of the experimental set-up.

The ease with which atomically thin materials can be doped and incorporated in photonic devices will provide many opportunities to study Fermi polaritons in the solid state. Future work will involve controlled in situ doping and resonant, angle-sensitive polariton pumping. Further, the combination of the magnitude and sign of the lower polariton branch effective mass and its extra charge provide a unique opportunity to build responsive current-carrying polaritonic circuitry that exhibits anomalous dissipative behaviour. We intend to explore how the negative mass influences the effect of disorder on the flow of the observed negative mass charged polaritons.


We deposit silicon dioxide (SiO2) and tantalum pentoxide (Ta2O5) as two dielectrics with refractive indices of 1.45 and 2.10, respectively, by electron beam evaporation to fabricate the distributed Bragg reflector (DBR). The thickness of each layer of oxide film (controlled by the thickness monitor) is λ/4n, where λ and n are the resonance wavelength of the cavity and the refractive index of the dielectric. First, a λ/4n thick (half of the total cavity thickness) SiO2 film is deposited on top of the bottom mirror, then we deposit a mechanically exfoliated monolayer of MoSe2 layer by a dry transfer technique. Subsequently, we deposit the other half of the cavity and the rest of the top mirror. A 675 nm (1.836 eV) continuous wave laser was used to excite the sample from the top mirror in a microscopy cryostat. An empty cavity was used to calibrate the charge-coupled device (CCD) pixel to sample momentum in-plane via the relation kǁ = k0 sin(θ). The reflectivities of the top and bottom mirrors were designed to have a slight asymmetry by choosing a different number of pairs for the top (7.5) and bottom (10.5) mirrors.

Data availability.

The data that support the plots within this paper and other findings of this study are available from the corresponding author upon reasonable request.

Additional Information

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.