# Fileset

[s41467-025-61607-2.pdf](https://mdr.nims.go.jp/filesets/26abeed0-6b50-4188-a672-3c34078b8ced/download)

## Creator

Armando Genco, Charalambos Louca, Cristina Cruciano, Kok Wee Song, Chiara Trovatello, Giuseppe Di Blasio, Giacomo Sansone, Sam A. Randerson, Peter Claronino, Kyriacos Georgiou, Rahul Jayaprakash, [Kenji Watanabe](https://orcid.org/0000-0003-3701-8119), [Takashi Taniguchi](https://orcid.org/0000-0002-1467-3105), David G. Lidzey, Oleksandr Kyriienko, Stefano Dal Conte, Alexander I. Tartakovskii, Giulio Cerullo

## Rights

[Creative Commons BY-NC-ND Attribution-NonCommercial-NoDerivs 4.0 International](https://creativecommons.org/licenses/by-nc-nd/4.0/)

## Other metadata

[Femtosecond switching of strong light-matter interactions in microcavities with two-dimensional semiconductors](https://mdr.nims.go.jp/datasets/6f7c527f-9b2e-47ad-a734-0278f5cc1b3a)

## Fulltext

Femtosecond switching of strong light-matter interactions in microcavities with two-dimensional semiconductorsArticle https://doi.org/10.1038/s41467-025-61607-2Femtosecond switching of strong light-matter interactions in microcavities withtwo-dimensional semiconductorsArmando Genco 1,11, Charalambos Louca 1,2,11, Cristina Cruciano1,Kok Wee Song 3,4, Chiara Trovatello 1,5, Giuseppe Di Blasio1,Giacomo Sansone 6, Sam A. Randerson 7, Peter Claronino7,Kyriacos Georgiou 8, Rahul Jayaprakash 7, Kenji Watanabe 9,Takashi Taniguchi 9, David G. Lidzey 7, Oleksandr Kyriienko 3,7,Stefano Dal Conte 1, Alexander I. Tartakovskii 7 & Giulio Cerullo 1,10Ultrafast all-optical logic devices based on nonlinear light-matter interactionshold thepromise to overcome the speed limitations of conventional electronicdevices. Strong coupling of excitons and photons inside an optical resonatorenhances such interactions and generates new polariton states which giveaccess to unique nonlinear phenomena, such as Bose-Einstein condensation,used for all-optical ultrafast polariton transistors. However, to reach thethreshold for condensation high quality factors and high pulse energies arerequired. Here we demonstrate all-optical switching exploiting the ultrafasttransition from the strong to the weak coupling regime in low-Qmicrocavitiesembedding bilayers of transition metal dichalcogenides with high opticalnonlinearities and fast exciton relaxation times. We observe a collapse ofpolariton gaps as large as 55 meV, and their revival, lowering the threshold foroptical switching below 4 pJ per pulse, while retaining ultrahigh switchingfrequencies. As an additional degree of freedom, the switching can be trig-gered pumping either the intra- or the interlayer excitons of the bilayers atdifferent wavelengths, speeding up the polariton dynamics, owing to uniqueinterspecies excitonic interactions. Our approach will enable the developmentof compact ultrafast all-optical logical circuits and neural networks, show-casing a new platform for polaritonic information processing based onmanipulating the light-matter coupling.All-optical switches based on nonlinear optical materials have beenextensively investigated to overcome the speed limitations of elec-tronic circuits, thanks to their potential to work at much higher fre-quencies owing to the inherently fast light-matter interactionsunderlying their operation1. Demonstrations of ultrafast all-opticallogic gates have been achieved in a plethora of solid state platforms,exploiting optical nonlinearities (χ(2) and χ(3))2,3 and saturableabsorption4. Notable examples employed microring resonators5,plasmonic nanostructures6, photonic crystals7, metasurfaces8, 2Dmaterials9 or even single molecules10. Such devices showcasedswitching times down to tens of femtoseconds, but usually at theexpense of the switching energy or the on/off contrast11,12. Morerecently, an optimal combination of femtosecond switching times andfemtojoule operating energies has been obtained in an all-opticalReceived: 14 March 2025Accepted: 26 June 2025Check for updatesA full list of affiliations appears at the end of the paper. e-mail: a.tartakovskii@sheffield.ac.uk; giulio.cerullo@polimi.itNature Communications |         (2025) 16:6490 11234567890():,;1234567890():,;http://orcid.org/0000-0002-1292-2614http://orcid.org/0000-0002-1292-2614http://orcid.org/0000-0002-1292-2614http://orcid.org/0000-0002-1292-2614http://orcid.org/0000-0002-1292-2614http://orcid.org/0000-0002-1122-3133http://orcid.org/0000-0002-1122-3133http://orcid.org/0000-0002-1122-3133http://orcid.org/0000-0002-1122-3133http://orcid.org/0000-0002-1122-3133http://orcid.org/0000-0002-5867-5452http://orcid.org/0000-0002-5867-5452http://orcid.org/0000-0002-5867-5452http://orcid.org/0000-0002-5867-5452http://orcid.org/0000-0002-5867-5452http://orcid.org/0000-0002-8150-9743http://orcid.org/0000-0002-8150-9743http://orcid.org/0000-0002-8150-9743http://orcid.org/0000-0002-8150-9743http://orcid.org/0000-0002-8150-9743http://orcid.org/0000-0001-8361-3555http://orcid.org/0000-0001-8361-3555http://orcid.org/0000-0001-8361-3555http://orcid.org/0000-0001-8361-3555http://orcid.org/0000-0001-8361-3555http://orcid.org/0000-0001-7993-8218http://orcid.org/0000-0001-7993-8218http://orcid.org/0000-0001-7993-8218http://orcid.org/0000-0001-7993-8218http://orcid.org/0000-0001-7993-8218http://orcid.org/0000-0001-5744-7127http://orcid.org/0000-0001-5744-7127http://orcid.org/0000-0001-5744-7127http://orcid.org/0000-0001-5744-7127http://orcid.org/0000-0001-5744-7127http://orcid.org/0000-0002-2021-1601http://orcid.org/0000-0002-2021-1601http://orcid.org/0000-0002-2021-1601http://orcid.org/0000-0002-2021-1601http://orcid.org/0000-0002-2021-1601http://orcid.org/0000-0003-3701-8119http://orcid.org/0000-0003-3701-8119http://orcid.org/0000-0003-3701-8119http://orcid.org/0000-0003-3701-8119http://orcid.org/0000-0003-3701-8119http://orcid.org/0000-0002-1467-3105http://orcid.org/0000-0002-1467-3105http://orcid.org/0000-0002-1467-3105http://orcid.org/0000-0002-1467-3105http://orcid.org/0000-0002-1467-3105http://orcid.org/0000-0002-8558-1160http://orcid.org/0000-0002-8558-1160http://orcid.org/0000-0002-8558-1160http://orcid.org/0000-0002-8558-1160http://orcid.org/0000-0002-8558-1160http://orcid.org/0000-0002-6259-6570http://orcid.org/0000-0002-6259-6570http://orcid.org/0000-0002-6259-6570http://orcid.org/0000-0002-6259-6570http://orcid.org/0000-0002-6259-6570http://orcid.org/0000-0001-8582-3185http://orcid.org/0000-0001-8582-3185http://orcid.org/0000-0001-8582-3185http://orcid.org/0000-0001-8582-3185http://orcid.org/0000-0001-8582-3185http://orcid.org/0000-0002-4169-5510http://orcid.org/0000-0002-4169-5510http://orcid.org/0000-0002-4169-5510http://orcid.org/0000-0002-4169-5510http://orcid.org/0000-0002-4169-5510http://orcid.org/0000-0002-9534-2702http://orcid.org/0000-0002-9534-2702http://orcid.org/0000-0002-9534-2702http://orcid.org/0000-0002-9534-2702http://orcid.org/0000-0002-9534-2702http://crossmark.crossref.org/dialog/?doi=10.1038/s41467-025-61607-2&domain=pdfhttp://crossmark.crossref.org/dialog/?doi=10.1038/s41467-025-61607-2&domain=pdfhttp://crossmark.crossref.org/dialog/?doi=10.1038/s41467-025-61607-2&domain=pdfhttp://crossmark.crossref.org/dialog/?doi=10.1038/s41467-025-61607-2&domain=pdfmailto:a.tartakovskii@sheffield.ac.ukmailto:giulio.cerullo@polimi.itwww.nature.com/naturecommunicationsnonlinear device based on a lithium niobate waveguide13, but withmillimeter-scale lengths hindering the production of densely on-chipintegrated circuits. Therefore, achieving high performances in all-optical switches is still an open challenge, with the future perspectiveof integrating them in a compact photonic processor.Excitons in semiconducting materials embedded in optical reso-nators can be used for all-optical switching, as they show a highlynonlinear response when they are in strong coupling (SC) with reso-nant photons confined in the structure. In such a regime, the rate ofcoherent energy transfer between the energy-degenerate excitons andphotons is higher than the loss rate, and new hybrid light-matterquasiparticles arise, called polaritons14. The energy splitting of thepolariton states (Rabi splitting) is a direct measure of their couplingstrength, which is proportional to the quality factor (Q factor) of theresonator and to the exciton absorption cross section.Harnessing polaritonic nonlinear interactions is of key impor-tance for a broad range of phenomena and applications, such aslasing15, optical parametric amplification16, Bose-Einstein condensation(BEC)17, or for quantum effects (polariton blockade)18,19. A combinationof blueshift of the polariton states and gain in photoluminescenceintensity occurring above the threshold of BEC has been used asoperational principle of all-optical polariton logic devices and neuralnetworks20–22. Ultrafast optical switching (with switching times ≤1 ps)relying on BEC has been demonstrated even at room temperature23–25,but typically using high Q factor architectures and high pulse energiesto reach the condensation threshold, i.e. from tens to hundreds pJ perpulse26,27, although the energy/pulse thresholds for polariton non-linearities have been recently brought down from ~10 nJ28 to ~1 pJ29.Alternative strategies for all-optical polariton switching have beenalso shown30–33, used, for example, for optical spin switches34,35. Apromising approach to achieve high performances in polaritonswitching relies on the modulation of the light-matter couplingstrength, leveraging on the exciton absorption saturation, which pro-duces a spectral shift of the polariton states, acting as a gate for thelight transmitted/reflected by the device. Varying strongly the cou-pling strength would eventually lead to a complete transition fromstrong to weak coupling regime or viceversa36. SC can be switched offin strongly coupled optical microcavities comprising GaAs quantumwells throughoptical saturation of excitons36,37 or via electrically-tunedcharge build-up38. Alternatively, it can be switched on by opticallyinduced absorption, i.e. for inter sub-band transitions39. However,achieving complete on/off switching cycle below 1 ps acting on thecoupling strength in these material platforms is not possible due thelong excitons and excited carriers lifetime40.Atomically thin transition metal dichalcogenides (TMDs) are pro-mising nonlinear optical materials41, where excitons remain stable up toroom temperature due to large binding energies and oscillatorstrength42,43. Owing to these properties, TMDs can easily enter the SCregime,when integrated inoptical resonators44,45. Recent studies ofTMDpolaritons aim tomaximize nonlinear interactions going beyond the useof 1s neutral excitons, exploring higher Rydberg excitonic states,charged excitonic complexes, moiré or dipolar excitons46–50. Moreover,the fast dynamics of TMD excitons51 makes these materials very pro-mising for ultrafast logic gates. Optically pumping TMD monolayerscoupled tooptical resonatorswithultrashort laser pulses canmodulate52or completely quench the Rabi splitting53 increasing the pump fluence,owing to strong exciton nonlinear interactions. However, a time-resolved study of the complete strong-to-weak coupling transition inmicrocavities with atomically thin TMDs and the demonstration of itsuse for high performance all-optical switching have never been shown.Here we use MoS2 bilayers embedded in low-Q factor micro-cavities to produce an ultrafast collapse and revival of the SC regimeusing very low pulse energies ( <4 pJ). Compared to monolayers,bilayers offer a unique combination of crucial properties to obtainsuch effect, such as (i) ultrafast and efficient exciton relaxation, (ii)strong nonlinearities, i.e. Coulomb dipole-dipole interactions andphase spacefilling, enhancedby the reduceddielectric screening54, (iii)hybridized interlayer excitons with high oscillator strength55, leadingto distinctive interspecies intra-interlayer exciton interactions54. Weemploy femtosecond transient reflectivity (TR) spectroscopy todemonstrate the ultrafast switching of the SC regime in compactdevices at both cryogenic and room temperature (RT). We show thefull tunability of this process through different degrees of freedom,such as pump wavelength, pulse power and cavity detuning. The SCswitching leads to a strongmodulation of the polariton peaks splitting,reducing the initial energy separation from 42 meV to less than theirlinewidth, making them indistinguishable from a single peak. The Rabisplitting modulation is further enhanced by placing a stack of twobilayers separated by hBN in the cavity, going from 55 meV to a com-plete collapse, resulting in an effective extinction ratio of about 7.5 dBworking in reflection. We further demonstrate an on/off SC switchingfrequency as high as 250 GHz, which can be extended up to 1 THz.ResultsStatic and dynamic optical behavior of MoS2 bilayersThe TMD structures used in our work are made of monolayers (MLs)and bilayers (BLs) of MoS2 encapsulated in hBN, placed on distributedBragg reflectors (DBRs), for the subsequent fabrication of opticalmicrocavities. Unless specified, all the spectroscopy experiments inthis work are performed at T = 8K. Figure 1a shows the ReflectanceContrast (RC) spectra of ML and BL MoS2 outside the cavity. Theabsorption of the intralayer A exciton in the BL (XA−BL) is higher than inthe ML, due to presence of the additional layer. In the BL a new exci-tonic resonance appears at ≈2 eV, which is attributed to dipolarhybridized interlayer excitons (hIX) with a high oscillator strength,resulting from the coherent tunneling of holes between the valencebands of the two layers (Fig. 1a inset)55,56.We study the ultrafast response of MoS2 excitons by ultrafast TRmicro-spectroscopy. We deliver two collinear pulses, a narrow-bandpump and a broad-band probe, focused on the sample using amicroscope objective (Fig. 1b). We then vary the delay time τ betweenthem andmonitor the changes in the broadband reflectivity spectrumof the probe (see Methods for experimental details). Figure 1c showsthe differential reflectivity (ΔR/R) map measured as a function of theprobe photon energy and pump-probe delay for a BLMoS2, tuning theenergy of the pump pulses at ≈ 1.94 eV, in resonance with the XA−BL.Generally, the shape of the TR spectra in TMD MLs is a result ofmultiple effects, such as optical saturation (photo-bleaching), linebroadening and spectral shift of the exciton peaks, leading to positiveand negative TR signals around the exciton energies57. To show moreclearly the temporal evolutionof the exciton features inour system,weperform an analysis of the transient ΔR/R response based on theTransfer Matrix Method (TMM) to extract the time-dependent RCspectra of thematerial (seeMethods for details)58. Figure 1d shows theMoS2 BL dynamic RC spectrum at a delay of 0.3 ps (purple curve),compared to the one before the pump pulse ( − 0.3 ps, orange curve).We fit the dynamic RC with three Lorentzians (solid lines in Fig. 1d) toextract the time-varying intensity and energy shift of each excitonicmode. After excitation, the XA−BL peak is quenched and slightly blue-shifted. The small shoulder at 1.91 eV appearing at positive delays isinstead related to a photo-induced absorption of the trion (X*A�BL)59–61.Since the excitation pulses are in resonance with XA−BL (shaded yellowarea in Fig. 1d), at lower energies compared to hIX, we would expectnegligible optical saturation of the latter if the two excitonic specieswere totally uncoupled. On the contrary, we observe a photo-bleaching of the hIX absorption, although less intense than in the XA−BL case. This indicates their hybridization with intralayer excitons dueto the coherent hole tunneling between the valence bands of the twolayers and to the fermionic interactions between holes of XA−BL and hIXsharing the same valence band (see inset of Fig. 1a)54.Article https://doi.org/10.1038/s41467-025-61607-2Nature Communications |         (2025) 16:6490 2www.nature.com/naturecommunicationsTracking the RC peak intensity as a function of the delay time,we can extract the ultrafast dynamics of the excitonic species(Fig. 1e). The transient behavior of exciton energies and linewidthsis shown in Supplementary Note S1. For both XA−BL and hIX, theexciton population rises instantaneously (within the ≈ 100 fs tem-poral resolution of our setup), then decays exponentially, withabout 50% of the initial population already relaxed within 2 ps(a comparison with MoS2 ML exciton dynamics is reported in Sup-plementary Note S2). The fast exciton decay time is similar betweenXA−BL and hIX.The ultrafast nonlinear optical response of mono and few-layersTMDs has been studied extensively in the past62–64. Transient excitonline shifts in TMDs are usually ascribed to Coulomb interactions atshort time scales (few ps)58,65, or bandgap renormalization66, and totransient heating effects67 at longer times (from tens to hundreds ofps). Exciting TMD monolayers close or below the exciton energy alsoleads to strong and instantaneous (within the pump pulse duration)line shifts due to the optical Stark effect68,69. High exciton densities inTMDs lead to optical saturation, due to phase-space filling (i.e. Pauliblocking)67, and line broadening caused by excitation-induceddephasing57. Tracking the time-dependent exciton saturation in ultra-fast pump-probe experiments allows monitoring the exciton popula-tion dynamics.In MoS2 BLs we observe a bi-exponential population decay with afast and a slow component. While in MLs the fast decay is usuallyattributed to radiative and non-radiative relaxationprocesses of brightexcitons70, in BLs it is more probably related to electron-phonon inter-valley scattering processes from the K points to the lowest energypoint of the Brillouin zone71–73. The slow decay component can berelated to phonon-assisted recombination fromdark states70 or defect-mediated non-radiative recombination74.Femtosecond switching of the strong coupling regimeWe exploit the highly nonlinear exciton interactions in MoS2 BL todrastically modify the light-matter coupling strength in microcavitieson ultrafast time-scales. The microcavity samples are fabricated bycovering the hBN-encapsulated MoS2 heterostructures placed onDBRs with a transparent polymeric spacer (polymethylmethacrylate,PMMA) and a top silver (Ag)mirror, as illustrated in Fig. 2a.Weperformk-space (Fourier) spectroscopy to image the angular dispersion of themonolithic cavity embedding the MoS2 BL (Fig. 2b). Two distinctanticrossings appear when the cavity mode is in resonance with XA−BLand hIX energies, a clear signature of the SC regime, resulting in upper,middle and lower polariton branches (UPB, MPB, LPB). Fitting thedispersion with a three coupled oscillators model, we extract Rabisplittings of ΩABL= 42 meV and ΩhIX= 23 meV for XA−BL and hIX,a cb1.85 1.9 1.95 2Energy (eV)0123456-6-4-202468101214ΔR/R x10-2τ (ps)Obj.lenshBNMoS2BLT=8KτPumpProbed e1.85 1.9 1.95 2Energy (eV)00.10.20.30.40.5Dynamic RC (arb. units)τ = -0.30 psτ = 0.30 psPumpbandwidthXA-BLhIXX*A-BL1.9 1.95 2 2.05 2.1 2.15Energy (eV)00.10.20.30.40.5Reflectance Contrast (arb. units)MoS2 MLMoS2 BLhIXXA-BLhXBXA-MLXB-MLLayer 1 Layer 2IXXA XB0 1 2 3 4 5 6Delay (ps)00.20.40.60.81ΔAexc (norm.)XA-BLhIXX*A-BLFig. 1 | Optical characterizationof theBLMoS2. a StaticRC spectraof aMLandBLMoS2 encapsulated in hBN. RC= (Rsub−RTMD)/Rsub, whereRTMD is the reflectance ofthe sample,while Rsub is taken on the substrate. Inset: sketchof the banddiagram inMoS2 BL; the dashed red line indicates the coherent tunneling of holes. b Sketch ofthe MoS2 BL encapsulated in hBN (out of scale) measured by pump-probe micro-spectroscopy. c Transient differential reflectivity map as a function of delay time τand probe photon energy measured for MoS2 BL. d Dynamic RC of the MoS2 BL atnegative (before pulsed excitation) and positive (after pulsed excitation) delaytimes, extracted from the differential reflectivity map in c. Solid curves show theLorentzian fit of the dynamic RC. Black arrows show optical saturation and energyshift of the XA−BL and hIX transitions, and the photo-induced absorption of theX*A�BL trion. The shaded yellow areadisplays the energy andbandwidthof thepumppulses. e Normalized exciton peak amplitude variation (ΔAexc) of XA−BL and hIX,extracted from the dynamic RC at different time delays.Article https://doi.org/10.1038/s41467-025-61607-2Nature Communications |         (2025) 16:6490 3www.nature.com/naturecommunicationsrespectively. We also fabricated a microcavity with a similar structureembedding a MoS2 ML, which shows an anticrossing between thecavity mode and intralayer excitons with a Rabi splitting of ΩAML=28 meV. The latter is reduced compared to the BL cavity due to thelower absorption (see Supplementary Note S3 for the static analysis ofthe ML cavity).We use ultrafast TR spectroscopy to excite the MoS2 BL-basedmicrocavities with narrowband ultrashort pulses tuned at the energyof XA−BL. To better visualize the dynamic behavior of polariton spec-trum, we plot directly the reflectance (1-R) spectra measured on thecavity as a function of delay time and probe photon energy (Fig. 2c, d),while we include the TR data of the same measurement in Supple-mentaryNote S4. Considering that in the spectral regionof interest thereflectance of the cavity without the TMD is close to 1, plotting 1-R as afunction of time is equivalent to showing the dynamic RC. We focusour analysis on incidence angles close to normal, on the anticrossingbetween the cavity mode and the XA−BL, resulting in the MPB and LPB.In a stark contrast to the out-of-cavity experiments, we do not observea direct reduction of the exciton absorption in this measurement, butwe monitor it indirectly through huge shifts of the polariton states. Atnegative delays, MPB and LPB are clearly separated, located at 1.953 eVand 1.911 eV respectively. When the pump and probe pulses are syn-chronous, the twopolaritonpeaks collapse symmetrically in onebroadcentral peak at ≈ 1.94 eV (purple line in Fig. 2c). Already after 2 ps, thetwo polariton branches start to reappear, while after 100 ps they havealmost completely recovered. The 1-R map as a function of the timedelay (Fig. 2d) showsmore clearly the complete collapse and revival ofpolaritons, which can be only explained as a reversible transition fromthe strong to the weak coupling regime. In our system, the collapse ofSC is mostly related to a large density of uncoupled excitons whichsaturates the optical transition. The SC recovery is consequent to therelaxation of such excitons, leading to a regaining of oscillatorstrength. In fact, we note that the SC recovery followswell the dynamicabsorptionof the excitonic speciesmeasuredoutside the cavity (Fig. 1eand Supplementary Note S5), being a direct consequence of density-dependent optical saturation of excitons. The two polariton branchesshow different recovery times depending on their Hopfield coeffi-cients, and in particular on their photonic component. In fact, apolariton branch with a larger photonic character will be closer inenergy to theweakly coupled cavitymode, leading to a faster recovery.Therefore, a positive detuning benefits theMPB recovery over the LPBone, as shown in Fig. 2d, while the opposite happens for negativedetunings (see Fig. S13).We performed a quantitative analysis of the ultrafast behavior ofMoS2 BL polaritons by fitting the experimental 1-R peaks with Gaussianfunctions. The results are shown in Fig. 2e, where the extracted peakenergies and linewidths are plotted against the time delay up to 1.5 ps.Within few hundreds of femtoseconds from the zero-delay, the LPBshows a blueshift of about 27meV, while theMPB redshifts by about 14meV, merging in a single peak at about 250 fs. Such huge shifts cannotbe explained just taking into account the bare exciton energy varia-tions, which are in the order of only a few meVs (Fig. 1d). When thea bdceObj.lensBottomDBRAgPMMAhBN/MoS2 BL/hBNT=8K τEnergy (eV)Angle (°)Normalized RC (arb. units)MPBLPBUPBCavXA BLhIXBL1.9 1.91 1.92 1.93 1.94 1.95 1.96 1.97Energy (eV)1-Rpump on(arb. units)τ = -350 fsτ = 350 fsτ = 2000 fsτ = 100000 fsMPBLPB Cav0 2000 4000 6000Delay (fs)1.91.911.921.931.941.951.961.97Energy (eV)104 105Delay (fs)1-R (arb. units)LPBMPB0 500 1000 1500Delay (fs)1.91.911.921.931.941.951.96Energy (eV)MPBLPB0.18-0.0401Fig. 2 | Ultrafast switchingof strong coupling inaMoS2BLmicrocavity. aSketchof the MoS2 BL microcavity structure measured by pump-probe spectro-micro-scopy. bColor map of the angle-resolved RC spectra of a microcavity embedding aBLMoS2 in strong coupling regime, showing two distinct anticrossings around theXA−BL and hIX energies (black dashed lines) respectively. The coupled oscillatorsmodel fit (white dashed lines) and the cavity mode dispersion (blue line) are shownin overlay. c 1-R spectraof the BLmicrocavity taken atdifferent pump-probe delays,pumping the system at 1.94 eV with 3.75 pJ. Immediately after excitation, thepolariton branches collapse in a central weakly coupled cavity mode. d Color mapof the 1-R spectra of the BL microcavity as a function of the pump-probe delayshowing the ultrafast collapse and later revival of the MPB and LPB (white dashedlines). e Results of Gaussian fits of the polariton/cavity modes dynamic spectraextracted fromFig. 2d. Theblue (orange) trace refers to theMPB (LPB) peakenergy,while the shaded areas depict the linewidth of the modes (Full Width Half Max-imum, FWHM). Under the shaded yellow area only a weakly coupled cavity modecan be fitted, with the red star highlighting the crossing region between strong andweak coupling regimes.Article https://doi.org/10.1038/s41467-025-61607-2Nature Communications |         (2025) 16:6490 4www.nature.com/naturecommunicationsenergy separation of the polariton states is lower than the linewidth ofthe cavity mode or the exciton, the anticrossing is not visible anymoreand the system falls into the weak coupling regime (red star in Fig. 2e).Already after ≈ 500 fs, the SC is recovered. TheQ factor of our cavity isabout 190, leading to a photon lifetime of ~ 65fs, being much fasterthan the observed recovery dynamics. This suggests that such beha-vior is dominated by incoherent excitonic processes. We note that inthe weak coupling, the cavity mode is strongly broadened by thebackground absorption of the excitons, already broad due toexcitation-induced dephasing57. Such broadening also affects thepolariton peaks after the collapse, as shown in the color bars of Fig. 2e,which become more discernible only after 2 ps. The polariton line-widths narrow down even more after 10 ps, when the effects of exci-tation induced dephasing fade away, as shown in SupplementaryNote S11. To a first approximation, we can consider that the strong toweak coupling full transition is reachedwhen the Rabi splitting is equalor below the unperturbed exciton linewidth (the FWHMof XA−BL, γ0exc,is ~ 20meV in static conditions). Amore precise definition of strong toweak coupling threshold implies that the energy exchange betweencavity andexciton resonances is larger than thedifferencebetween theloss rates75,76. On the other hand, considering in our case the excitonline broadening caused by excitation-induced dephasing, thisbecomes a less stringent criterion, as discussed later in this sec-tion (Fig. 3c).Leveraging on the large binding energy of excitons in TMDs, wefabricated an additional BL MoS2 microcavity in SC regime at RT. Weperformed a full SC switching also in this device at ambient conditions,shifting the LP by about 20 meV using a pump pulse energy of ~ 1.8 pJ(see Supplementary Note S11).Finally, we observed a similar SC collapse also in the microcavityembedding a ML of MoS2, but in that case the longer exciton lifetimesled to a much slower SC recovery, while the smaller Rabi splittingworsened the on/off contrast, i.e. the signal intensity ratio between the1-R spectra of the cavity in the unperturbed SC and weak couplingconditions respectively (see Supplementary Note S6). The switchingcontrast is influencedby anumber of factors. Themost important onesare the visibility of the polariton modes, controlled by the detuning,the maximum achievable dynamic energy shift, directly proportionalto the Rabi splitting, and the exciton and polariton linewidth broad-ening. The latter can significantly worsen the switching contrast and isalso dependent on the exciton density and the pump fluence. Redu-cing the static and dynamic exciton and polariton broadening orenlarging the Rabi splitting will increase the on/off contrast.The pump pulse energy plays a major role in the SC switchingdynamics, as shown in Fig. 3a where MoS2 BL cavity spectra taken at adelay of 250 fs for different excitation pulse energies demonstrate thegradual quenching of the Rabi splitting. The SC collapse in TMD cav-ities is a direct consequence of exciton nonlinear interactions, whichscale proportionally to their density54. We demonstrate this effect bycarrying out theoretical simulations of the cavity 1-R spectra using theTMM (Fig. 3b), employing the MoS2 BL optical constants calculatedfrom the exciton nonlinear absorption as a function of the density (seeSupplementary Note S7). The match between experiments and simu-lations proves that the main cause behind the observed femtosecondswitching of the SC regime is the optical saturation and broadening ofMoS2 BL excitons at high excitation densities, which recovers veryrapidly due to the fast radiative and non-radiative exciton relaxationmechanisms in this system.1.89 1.9 1.91 1.92 1.93 1.94 1.95 1.96Energy (eV)Simulated 1-R (arb. units)a cb d1.9 1.91 1.92 1.93 1.94 1.95 1.96 1.97Energy (eV)1-Rpump on(arb. units) 0.62 pJ3.75 pJ250 fsExcitondensitye0 2000 4000 6000 8000 10000Delay (fs)1.91.911.921.931.941.951.961.97Energy (eV)1-R (arb.units)0.21-0.05ObjectivelensτMicrocavitysampleProbePumpMBSPol.YVO4T02.5 pJ1.87 pJ1.25 pJPulse energy0 500 1000 1500 2000 2500Delay (fs)0.30.40.50.60.70.80.91ΔEPol/ ΔEPol MAX104 105Delay (fs)Weak CouplingStrong Coupling2.5 pJ3.75 pJ7.5 pJ γ0 exc/ΔEPol MAX |γexc(t) - γcav|ΔEPol MAXFig. 3 | Control of strong coupling switching. a BL cavity 1-R spectra taken at adelay time of 250 fs, pumping the system at increasingly higher pulse energies.b Simulated cavity spectra for increasing exciton-polaritons densities in the MoS2BL. c MPB-LPB energy difference as a function of delay time for ultrafast SCswitching experiments in BL cavities at different pump pulse energies, normalizedto the value before excitation. Error bars are drawn from the confidence interval ofthe double peak fits. The dashed black curve shows the trend of the exciton-cavitylinewidth difference, while the red dashed-dotted line shows the linewidth of theunperturbed exciton, both normalized by the polariton splitting value beforeexcitation. The weak coupling time window duration can be tuned by changing theexcitation pulse energy. d Sketch of the experimental configuration used to pro-duce delayed double pump pulses. M mirror, BS beam splitter, Pol polarizer.eColormap of the 1-R spectra of the BLmicrocavity as a function of the delay time,excited by double pump pulses delayed by ≈4 ps.Article https://doi.org/10.1038/s41467-025-61607-2Nature Communications |         (2025) 16:6490 5www.nature.com/naturecommunicationsFigure 3c reports the MPB-LPB energy difference against the timedelay, normalized with respect to its value before excitation. In thisfigure, the blue dots are related to the experiment reported in Fig. 2,performed at 3.75 pJ (pump fluence: 212 μJ cm−2), and show therecovery of SC occurring on twodifferent time-scales, a fast onewithin1 ps and a slowonewhich is concluded after ≈ 100ps.We ascribe thosetwo recovery steps to the population decay dynamics of the bareexcitons (see Fig. 1e and Supplementary Note S5). The red dashedhorizontal line in Fig. 3c represents the threshold when the polaritonsplitting is smaller than the unperturbed exciton linewidth (γ0exc),while the black dashed curve shows the time-dependent normalizeddifference between exciton and cavity linewidths. While the cavitylinewidth remains approximately the same in all the experiments( ~ 10 meV), the exciton linewidth changes with fluence and timebecause of excitation-induced dephasing. We extract the transientXA−BL linewidth, γexc(t), analysing the time-dependent reflectivity of theout-of-cavity sample, excited with a pump fluence comparable to theones used in the cavity experiments (see Supplementary Note S5).Using the exciton-cavity linewidth difference to set the threshold forSC, the switching is not as sharp, but it still occurs in a sub-picosecondtime window, between ~ 50 fs and ~ 700 fs, pumping with 3.75 pJ. Onthe other hand, using such a definition for the strong to weak couplingtransition, the pulse energy to induce the SC collapse will decrease. Atthe SC switching pump energy threshold, we estimated a peak polar-iton density of about 105μm−2. Increasing or decreasing by few pico-joules the excitation energy, we can extend the temporal window ofweak coupling regime (7.5 pJ, red dots in Fig. 3c) or suppress thetransition (2.5 pJ, yellow dots in Fig. 3c).We provide a theoretical explanation on the energy dependentdynamic SC switching upondirect excitation of the intralayer excitons,considering several possible contributions to the dynamics. We notethat the ≈ 1 ps timescale for the fast recovery coincides with the bareXA−BL exciton fast decay. However, this short timescale cannot accountfor the later slower recovery ( ~ 100 ps) of the Rabi splitting, indicatingthat a significant exciton population remains in the sample and non-linear phase space filling impacts the spectrum. One plausible expla-nation is the existence of long-lived dark excitonic states at lowerenergies77,78. In this case photoexcited excitons can be transferred tosuch states which interact with light weakly, forming a reservoir thatcontributes to the nonlinear phase space filling. In a bilayer MoS2, low-energy states are represented by spin-forbidden states due to spin-orbit coupling, or momentum-forbidden states77 due to indirectbandgap78. These states possess a very long lifetime and can explainthe TR dynamics. The corresponding model is summarized in theMethods, while we providemore details about the simulated polaritondynamics in Supplementary Note S8.Leveraging on the ultrafast recovery times of SC in our samples,we demonstrate the possibility to modulate light-matter interactionsat very high frequencies, illuminating the cavity with two subsequentpump pulses at 1.94 eV delayed by only ≈ 4 ps. To produce such pulsepair, we used a birefringent YVO4 crystal with optical axis rotated by45° with respect to the polarization of the incoming pump pulse, fol-lowed by a linear polarizer (see Fig. 3d and Methods for more details).The first pulse energy was tuned to be slightly lower than in theexperiment of Fig. 2 in order to get a faster SC recovery, while thesecond pulse energy was adjusted to take into account the residualexcitonpopulation after the first pulse. The resulting transient 1-Rmapshows two reversible on/off cycles (Fig. 3e), proving a very fastswitching frequency of ≈ 250 GHz. We note that this value was limitedby technical constraints (the fixed delay between the pulse pairdetermined by the thickness of the available YVO4 crystal), while thetheoretical limit is given by the recovery time of the SC.Ultrafast SC switching by interspecies interactionsWe exploit the interspecies exciton interactions specific ofMoS2 BL togenerate optical saturation of XA−BL acting on the hIX, exciting selec-tively the latter and probing the quenching of the Rabi splitting on XA−BL (Fig. 4a). Thisprocess relies on nonlinear fermionic interactions (i.e.involving a single charge carrier constituting the exciton) between thetwo excitonic species: the XA−BL valence band is shared with hIX,therefore exciting the latter causes optical quenching of the former,due to Pauli blocking of holes for XA−BL54. Figure 4b shows the transient1-R map of the MPB-LPB, pumping the hIX of the BL: the femtosecondswitching of SC regime occurs very clearly also in this case. Comparingthis result with the previous case of resonant XA−BL pumping (Fig. 2c),the fast SC recovery is evenmoredistinct, with the twopolariton peaksbeing clearly visible and well separated already after 1 ps, as shownin Fig. 4c.The dynamics of the nonlinear response when pumping in reso-nance with the hIX is also consistent with the developed model basedon the nonlinear saturation and phase space filling from the long-livedstates (see Methods). In this case, we considered similar lifetimes forhIX compared to XA−BL, but we assumed the rate for transferring thepumped hIX to the long-lived reservoir states contributing to thephase space filling to be smaller than in the XA−BL case. This may beunderstood as the result of the hIX’s wavefunction spreading in theout-of-plane direction. This implies that hIX is less 2D than XA−BL,leading to a weaker scattering effect with disorder and to a smallera1.9 1.92 1.94 1.96 1.98Energy (eV)1-Rpump on(arb.units)τ = -350 fsτ = 300 fsτ = 1000 fs0 1000 2000 3000 4000 5000Delay (fs)1.91.921.941.961.98Energy (eV) 1-R (arb. units)bhIXXAProbePumpFermionicinteractionscLPBMPBCav0.18-0.03Fig. 4 | Ultrafast strong coupling switching by interspecies interactions.a Sketch of the MoS2 BL cavity measured in pump-probe, exciting the hIX spectralregion and probing the polariton states formed around XA−BL. b Color map of the1-R spectra of the BL microcavity as a function of the delay time, exciting the hIXand probing the MPB-LPB spectral region. c 1-R spectra of the BL microcavitypumping the hIX, taken at different pump-probe delay times. Already after 1 ps, thepolariton peaks are clearly recovered.Article https://doi.org/10.1038/s41467-025-61607-2Nature Communications |         (2025) 16:6490 6www.nature.com/naturecommunicationstransfer rate. Furthermore, thermalization through exciton-phononscattering is another important mechanism converting the brightstates into momentum-dark states which may also gives smallertransfer rate upon hIX pumping. These scattering effects yield a fasterrecovery of SC, in agreement with our experimental observation (seeSupplementary Note S8).Another effect leading to faster recovery in the hIX pumpingscheme is the mitigation of the Pauli blockade. In contrast toXA−BL pumping, the pumped exciton only shares holes with theprobed exciton but not the electrons54, see Fig. 4a, leading to aweaker saturation effect. This can also result in a faster recoveryof SC if similar conditions as in XA−BL case are used (except thepump photon energy). Combining the effects of a smaller bright-to-dark exciton transfer rate and weaker Pauli blockade, theinfluence of the reservoir long-lived states is less significantexciting the hIX. Therefore, a faster recovery time of SC is easierto achieve in this case. Such fast recovery would allow to furtherincrease the switching frequency, up to ≈1 THz. We underlinethat to achieve such hIX-induced Rabi quenching we use anexcitation energy of 4.37 pJ, only moderately higher compared tothe resonant excitation case. Higher pulse energies will increasethe recovery time and the weak coupling time window.SC switching in a double BL microcavityFinally, we fabricated a device comprising two vertically stackedbilayer MoS2 separated by an hBN spacer of 40 nm. We placed thisstructure in amicrocavitymadeof the sameDBRand silvermirror usedin the single BL cavity, with a PMMAspacer between theTMDstack andthe top mirror (Fig. 5a). Similarly to microcavities with multiplequantumwells79, the SC is enhanced in this sample, as theRabi splittingis increased to 55 meV due to the additional BL unit (see Supplemen-tary Note S9 for the coupled oscillators model fit of the stronglycoupled cavity dispersion). Figure 5b shows a comparison between thestatic 1-R spectra of a single BL (blue line) and a double BL (red line)cavity, taken at small angles. It clearly appears that theMPB-LPB peaksare more separated in the double BL cavity compared to the single BLsample, being also redshifted, as an effect of the more negativedetuning of the former. The negative detuning also leads to a decreaseof the LPB linewidth in the double BL cavity, being more cavity-like.By exciting the double BL cavity with pump pulses resonant withXA−BL, we observed again an ultrafast collapse of the MPB-LPB polar-iton peaks into a weakly coupled cavity mode, followed by their laterrecovery (Fig. 5c). We note that the pulse energy used for thisexperiment (8.25 pJ) was not adjusted to obtain a sub-ps SC recovery,but just to demonstrate the SC switching. Figure 5d shows the fulldynamics of the SC collapse and recovery in the double BL cavity,where the energy separation betweenMPB and LPB (blue and red linesrespectively) drops from ≈ 55meV to zero immediately after the pumppulse. Considering the unperturbed SC condition as the on state of theoptical switch and the weak coupling as the off state, we calculated thespectral power extinction ratio (ER) from the 1 − R spectra before andafter the pump pulse, taken at − 350 fs and 350 fs respectively, whereER(dB)= 10 logðð1� RonÞ=ð1� Rof f ÞÞ. The ER is increased significantly inthe double BL cavity compared to the single BL device, over a broadenergy range, as shown in Fig. 5e. For both the devices, the maximumER in absolute value is reached around the energy of the weakly cou-pled cavitymode, between the LPB andMPB,which transmits (reflects)more during the off (on) state. While for the single BL the ER absolutea bdceBottomDBRAgPMMA0 2000 4000 6000Delay (fs)1.861.881.91.921.941.961.98Energy (eV)104 105Delay (fs)1-R (arb. units)1.86 1.88 1.9 1.92 1.94 1.96 1.98Energy (eV)1-R pump on(arb. units)τ = -350 fsτ = 350 fsτ = 100000 fsLPBMPBCavDouble BL cavity1.88 1.9 1.92 1.94 1.96 1.98Energy (eV)1-R(arb. units)BL cav.Double BL cav.LPBMPBLPBMPBhBNMoS2hBNhBNMoS21.86 1.88 1.9 1.92 1.94 1.96Energy (eV)-10-50510On-off extinction ratio (dB) BL cav.Double BL cav.0.270Fig. 5 | Ultrafast switching of a double BL microcavity. a Sketch of the micro-cavity embedding a double BL of MoS2. b Static 1-R spectra of the single BL com-pared to the double BL microcavity, showing a redshift of the MPB-LPB and anincreased polariton splitting in the latter. c 1-R spectra of the double BLmicrocavityexcited with pump pulses at 1.91 eV and 8.25 pJ, taken at different pump-probedelays. d Color map of the 1-R spectra of the double BL microcavity versus pump-probe delay showing the ultrafast collapse and later revival of the SC. The orange(blue) line in overlay displays the fitted peak energy of the LPB (MPB) or the weaklycoupled cavity mode. e Effective on/off extinction ratio calculated from the 1-Rspectra taken at −350 fs and 350 fs, for the single (blue line) and double BL cavity(red line).Article https://doi.org/10.1038/s41467-025-61607-2Nature Communications |         (2025) 16:6490 7www.nature.com/naturecommunicationsvalue reaches 3.2 dB at 1.932eV, it is enhancedup to 7.5dB at 1.915 eV inthe double BL. It is worth to mention that the ER is also high in thespectral regions of the LPB and MPB, where it shows opposite sign,meaning that the optical switch can be used in direct or reverse modejust by changing the operational wavelength. We also tested the SCswitching in a single BL cavity with a slightly negative detuning, similarto that of the double BL cavity, as shown in Supplementary Note S12.The negative detuning in this sample leads to a maximum ER of ~ 7 dB(Fig. S13c), higher compared to that of the single BL cavity shown inFig. 5. Such value is similar to themaximum observed in the double BLsample, pointing out the importance of the detuning for a high con-trast. However, in a single BL device the ER is maximized only in anarrow spectral region around 1.93 eV. A double BL cavity insteadensures maximum contrast in a broader energy range owing to theincreased Rabi splitting, and consequently the larger energy separa-tion between LPB and the cavity mode.We note that working at high frequencies, the effective extinc-tion ratio for a second switching event decreases due to the residualexciton population after the first pulse excitation (e.g. by about fourtimes in the double pump pulse experiment shown in Fig. 3). Weforesee that this drawback can be mitigated by reducing the longexciton decay component, for example suppressing the excitonscattering to dark states and using a different optical resonatorwith higher Q factor and smaller mode volume, to induce a strongPurcell effect.DiscussionIn summary, we exploit the transient behavior of MoS2 exciton-polaritons to demonstrate ultrafast optical switches using low pulseenergies ( < 4 pJ), whose operational principle is based on the instan-taneous transition from the strong to the weak coupling regime due tooptical saturation of excitons, which can recover on the sub-picosecond timescale. The MoS2 BL system uniquely combines nano-metric thickness, large nonlinearities and large Rabi splitting withshort lifetimes, with the latter enabling observation of the exception-ally fast recovery. SC switching can be performed in this platform evenat ambient conditions, still using very low pulse energies, below 2 pJ.Furthermore, we show that by increasing slightly the pump pulseenergies above the threshold for the SC collapse, the weak couplingtime window can be significantly extended and deterministically con-trolled, being sensitive to energy variations of hundreds of femtojoulesand below, crucially important for sensing and low light applications80.Such strongly fluence-dependent switching dynamics can be alsoexploited to emulate spiking neurons in novel neuromorphic com-puting architectures81. This system, also, offers additional degrees offreedom. It can operate at different excitation energies, for example inresonance with either intra- or interlayer excitons, leveraging on thestrong interspecies interactions between these excitonic species,unique to the MoS2 BLs. We foresee this property to be particularlyuseful formultiplexed logic operations82. Owing to the fast recovery ofSC, wewere able to perform subsequent switching events delayed by 4ps, demonstrating anoperational frequency of ≈250GHz. Consideringthe sub-ps SC switching time, this frequency can be pushed up to 1THz, surpassing even the fastest electronic transistors demonstratedso far83.We alsodemonstrate that using amicrocavitywith twostackedMoS2 BLs can boost the Rabi splitting and greatly enhance the on/offextinction ratio, reaching a maximum of 7.5 dB in a single switchingevent. Compared to BEC-based polariton switching, our system doesnot need a high Q factor to obtain the switching effect. A furtherimprovement of the optical resonator Q factor and a shorter excitonlifetime will lead to even greater on/off contrast for high frequencyswitching. For example, increasing the Q factor even by one order ofmagnitudewill still result in a polariton lifetime below 1 ps, resulting atthe same time in a polariton linewidth of few meV, hence obtaining amuch higher on/off contrast.Our work highlights TMD bilayers as a flexible system with richphysics in which sub-ps all-optical switching can be achieved and finelycontrolled. Such platform shows clear advantages compared to othermaterials in SC regime or even to TMD monolayers (see Supplemen-tary Note S13 for a detailed comparisons with other systems).The insights provided canbepivotal for the development of TMD-based high speed all-optical circuits. Moreover, the developed ultra-fast nonlinear switching unit can improve the performance of opticalneural networks84,85 acting as an all-optical nonlinear activation func-tion. Considering also the nanometric thickness of each hBN/BL/hBNstack, a microcavity could be filled by many TMD units, greatlyincreasing the Rabi splitting and subsequently the spectral shifts whenused as ultrafast switches, which will lead to enhanced on/off extinc-tion ratio. Moreover, the integration of electrical contacts in themicrocavity structures86 would enable the fabrication of electro-optical interfaces by tuning the electrostatic doping and electric field,which can provide giant shifts of the hIX energy in MoS2 BLs56, alsoenhancing their nonlinear interactions.Our SC switching approach canbe extended also to other types of TMDhomobilayers or even tomoiréheterobilayers. In the latter case, the exciton confinement within themoiré potential will foster polariton nonlinear interactions49, leadingto optical saturation and SC quenching at lower exciton-polaritondensities. Owing to the low pulse energies used, we observed nodegradation of the devices after several switching experiments, evenunder ambient conditions, ensuring good long-term switching stabi-lity. Identifying strategies to suppress the slow exciton decay compo-nent will also ensure a high on/off extinction ratio for multipleswitching events while working at very high frequencies. Optimizingthe coupling of the TMD with a different optical resonator, e.g.waveguide resonances or nanophotonic structures, will enable the on-chip integration of multiple switching nodes within in-plane opticalnetworks. Very small mode volumes and strongly localized light fieldstypical of such structureswill also decrease thepulse energies requiredfor the switching. Nanophotonic devices embedding TMD MLs thathost quasi-bound states in the continuum modes with high Q factorhave been recently demonstrated87,88. In such systems, the reflectivity(transmissivity) in the spectral regions around the uncoupled excitonenergies can bemoderately low (high). Hence, increasing the Q-factorwill improve the on/off contrast by reducing the polariton linewidth,while still ensuring optical access to the excitons.Developing ultrafast all-optical switches based on the transitionfrom the strong to the weak coupling regime would be crucial also tounveil more exotic physical phenomena. The transition between thestrong and weak coupling regime is linked to the observation ofexceptional points, where the eigenvalues and eigenfunctions of thecoupled systemscoalesce, enfoldingexoticphysics arising fromthenon-Hermitian Hamiltonian describing such condition89,90. The encirclementof exceptional points in microcavities, controlling the detuning and thecoupling strength, has been recently demonstrated91,92, paving the wayfor the investigation of non-Hermitian physical phenomena, such asanomalous topological phases93 and dissipative phase transitions94,95.Our platform offers a new approach to tune the system parameters forencircling the exceptional point on ultrafast timescales.MethodsSample fabricationThe hBN/MoS2/hBN heterostructures were assembled using a poly-dimethylsiloxane (PDMS) polymer stamp method. The PMMA spacerfor the monolithic cavity was deposited using a spin-coating techni-que, while a silver mirror of 45 nm thickness was thermally evaporatedon top of it.Optical measurementsFor the transient reflectivity measurements 100-fs pulses from anamplified Ti:Sapphire laser at 2 kHz repetition rate are used. The laserArticle https://doi.org/10.1038/s41467-025-61607-2Nature Communications |         (2025) 16:6490 8www.nature.com/naturecommunicationsoutput is split in two beams. A portion of the laser output is utilized todrive a non-collinear optical parametric amplifier (NOPA),whichallowstuning the pumpwavelength. The rest is used for the generation of thebroadbandwhite light probe pulse by focusing the beamon a sapphireplate. The delay between pump and probe pulses is controlled by amechanical delay line. The pulses are combined collinearly andfocused on the sample using a 50x objective, resulting in a spot size of≈ 1.5 μm. The sample is kept in a helium cryostat at 8 K. The differentialreflectivity (ΔR/R) spectra are recorded at various timedelays τ to trackthe changes induced by the pump. Specifically, the reflectivity spec-trum of the probe with the pump on, RPumpOn, is compared at eachdelay with a reference spectrum obtained when the pump is off,RPumpOff. These areused to calculate ΔRR =RPumpOn�RPumpOffRPumpOff, shown in theTRmaps. The pump is orthogonally polarized with respect to the probeand it is filtered out by using a polarizer in the detection path. Toremove any residual pump signal we also subtract a backgroundspectrum taken without the probe to all the differential reflectivityspectra. The dynamic probemaps are extracted froma combination ofthe measured RC spectrum with the pump off and the ΔR/R maps, asRPumpOn =RPumpOff ð1 + ΔRR Þ. For thedouble pumppulses experiments,weuse a thick YVO4 birefringent crystal with optical axis rotated at 45∘with respect to the vertical pump polarization, which produces areplica of the pulse with horizontal polarization delayed by ≈ 4 ps. Therotation of a subsequent polarizer is changed to finely adjust theenergy of each pulse in order to ensure the SC recovery after eachexcitation pulse.Transfer matrix method analysisIn order to extract the spectral and temporal evolutionof the excitonicoptical properties from theΔR/Rmaps,we follow a procedure recentlyreported in refs. 58,96. The transient reflectivity R(ω, τ) of MoS2 BL ateach delay time is determined from the equilibrium reflectivity R(ω),which is reconstructed from theTMMfit of the static RC spectrum, andthe transient reflectivity ΔRR ðω, τÞ, following this relation:Rðω, τÞ=RðωÞ ΔRRðω, τÞ+ 1� �ð1ÞThen, the dynamic RC is obtained by applying the formula: RC(ω,τ) = 1 − (R(ω, τ)/Rsub), where the substrate reflectivity Rsub is simulatedwith the TMM. See Supplementary Note S7 for more details on theTMM simulations.Theoretical model for the polariton dynamicsTo gain further insight into the dynamics in the system, we develop amean-field model that captures the main trends in our experimentover different excitation regimes. The Hamiltonian corresponds to thecoupled cavity-photon system, where the XA−BL mode (being the pro-bed A exciton of a homobilayer) hybridizes with the cavity mode. TheHamiltonian readsH =Ec + iκ12 gðnX ÞΩABL12 gðnX ÞΩABLEABL+ iγ" #, ð2Þwhere Ec and κ are the energy and linewidth of the cavity photon, andEABLand γ are the energy and linewidth of the probed exciton. In theHamiltonian above ΩABLis the Rabi splitting at weak pumping, andgðnX Þ= e�αnX is the dimensionless nonlinear coupling, with α being thenonlinear phase space filling (saturation) coefficient97. The magnitudeof the Rabi splitting is dependent on the total number of excitons inthe system, nX. In general nX(t) = np(t) + nR(t) is time-dependent, andincludes excitons (electron-hole pairs) from different states. Specifi-cally, we separate the two fractions corresponding to np and nR beingthe population of the pumped exciton and the long-lived excitons inthe reservoir. Crucially, both contribute to the nonlinear saturationeffect. The dynamics of excitonic fractions can be described by rateequations defining the transfer and population redistribution, whichreaddnpdt= � γpnp � rnp +ΘðtÞ, ð3ÞdnRdt= � γRnR + rnp, ð4Þwhere γp is the pumped exciton decay rate, γR is the decay rate of thelong-lived exciton in the reservoir, and r is the rate constant fortransferring the pumped excitons into the reservoir. The functionΘ(t)depends on the pump laser profile in time. For example, Θ(t) can be aHeaviside step function tomodel thepumpas anon-off switchingfield.In this model, we assume γR ≈ γp/100 for the long-lived statescorresponding to 100 ps decay time, and consider the decay timescalebeing similar to that of spin-forbidden dark states98. In fact, the decaytimemay be different, but this does not change our later conclusion ina qualitative way. Furthermore, when considering a spin-conservingprocess, we let the transfer rate be comparable to the pumped excitondecay rate99, r ≈ γp, such that this allows the pumped exciton transferinto the reservoir. With this, we find a good qualitative agreementbetween the theoretical spectrum and the experimental measurement(see Supplementary Note S8). Particularly, the theory demonstratedthe excitation pulse energy dependence of the recovery time of SC.Inclusion and ethics statementAll collaborators of this study that have fulfilled the criteria forauthorship required by Nature Portfolio journals have been includedas authors, as their participation was essential for the design andimplementation of the study. Roles and responsibilities were agreedamong collaborators ahead of the research. This work includes find-ings that are locally relevant, which have been determined in colla-boration with local partners. This research was not severely restrictedor prohibited in the setting of the researchers, and does not result instigmatization, incrimination, discrimination or personal risk to par-ticipants. Local and regional research relevant to our study was takeninto account in citations.Data availabilityThe data generated in this study are available on Zenodo publicrepository, with the https://doi.org/10.5281/zenodo.15716409 (2025).Code availabilityThe computer codes and algorithms used to process the data includedin this study are available from the corresponding authors uponrequest.References1. Miller, D. A. Are optical transistors the logical next step? Nat. Pho-tonics 4, 3 (2010).2. Grinblat, G. et al. Ultrafast sub–30-fs all-optical switching based ongallium phosphide. Sci. Adv. 5, eaaw3262 (2019).3. Assanto, G., Stegeman, G., Sheik-Bahae, M. & Van Stryland, E. All-optical switching devices based on large nonlinear phase shiftsfrom second harmonic generation.Appl. Phys. Lett. 62, 1323 (1993).4. Iizuka, N., Kaneko, K. & Suzuki, N. All-optical switch utilizing inter-subband transition inGaNquantumwells. IEEE J.QuantumElectron.42, 765 (2006).5. Almeida, V. R., Barrios, C. A., Panepucci, R. R. & Lipson, M. All-optical control of light on a silicon chip. Nature 431, 1081 (2004).6. Fu, Y. et al. All-optical logic gates based on nanoscale plasmonicslot waveguides. Nano Lett. 12, 5784 (2012).Article https://doi.org/10.1038/s41467-025-61607-2Nature Communications |         (2025) 16:6490 9https://doi.org/10.5281/zenodo.15716409www.nature.com/naturecommunications7. Nozaki, K. et al. Sub-femtojoule all-optical switching using aphotonic-crystal nanocavity. Nat. Photonics 4, 477 (2010).8. Mann, S. A. et al. Ultrafast optical switching and power limiting inintersubband polaritonic metasurfaces. Optica 8, 606 (2021).9. Ono, M. et al. Ultrafast and energy-efficient all-optical switchingwith graphene-loaded deep-subwavelength plasmonic wave-guides. Nat. Photonics 14, 37 (2020).10. Hwang, J. et al. A single-molecule optical transistor. Nature 460,76 (2009).11. Qiet, H. et al. All-optical switch based on novel physics effects. J.Appl. Phys. 129 (2021).12. Chai, Z. et al. Ultrafast all-optical switching. Adv. Opt. Mater. 5,1600665 (2017).13. Guo, Q. et al. Femtojoule femtosecond all-optical switching inlithium niobate nanophotonics. Nat. Photonics 16, 625 (2022).14. Kavokin, A. V., Baumberg, J. J., Malpuech, G., Laussy, F. P. Micro-cavities, Vol. 21 (Oxford University Press, 2017).15. Kéna-Cohen, S. & Forrest, S. Room-temperature polariton lasing inan organic single-crystal microcavity. Nat. Photonics 4, 371 (2010).16. Saba, M. et al. High-temperature ultrafast polariton parametricamplification in semiconductor microcavities. Nature 414, 731(2001).17. Deng, H., Haug, H. & Yamamoto, Y. Exciton-polariton bose-einsteincondensation. Rev. Mod. Phys. 82, 1489 (2010).18. Delteil, A. et al. Towards polariton blockade of confinedexciton–polaritons. Nat. Mater. 18, 219 (2019).19. Muñoz-Matutano,G. et al. Emergenceofquantumcorrelations frominteracting fibre-cavity polaritons. Nat. Mater. 18, 213 (2019).20. Ballarini, D. et al. All-optical polariton transistor. Nat. Commun. 4,1778 (2013).21. Ballarini, D. et al. Polaritonic neuromorphic computing outperformslinear classifiers. Nano Lett. 20, 3506 (2020).22. Mirek, R. et al. Neuromorphic binarized polariton networks. NanoLett. 21, 3715 (2021).23. Feng, J. et al. All-optical switching based on interacting excitonpolaritons in self-assembled perovskite microwires. Sci. Adv. 7,eabj6627 (2021).24. Chen, F. et al. Optically controlled femtosecond polariton switch atroom temperature. Phys. Rev. Lett. 129, 057402 (2022).25. Zasedatelev, A. V. et al. A room-temperature organic polaritontransistor. Nat. Photonics 13, 378 (2019).26. Sannikov, D. A. et al. Room temperature, cascadable, all-opticalpolariton universal gates. Nat. Commun. 15, 5362 (2024).27. Li, H. et al. All-optical temporal logic gates in localized excitonpolaritons. Nat. Photonics 18, 864 (2024).28. Yadav, R. K. et al. Direct writing of room temperature polaritoncondensate lattice. Nano Lett. 24, 4945 (2024).29. Le Roux, F., Mischok, A., Bradley, D. D. & Gather, M. C. Efficientanisotropic polariton lasing using molecular conformation andorientation in organic microcavities. Adv. Funct. Mater. 32,2209241 (2022).30. Marsaultet, F. et al. Realization of an all optical exciton-polaritonrouter. Appl. Phys. Lett. 107, 201115 (2015).31. Nguyen, H. S. et al. Realization of a double-barrier resonant tun-neling diode for cavity polaritons. Phys. Rev. Lett. 110, 236601(2013).32. Gao, T. et al. Polariton condensate transistor switch. Phys. Rev. BCondens. Matter Mater. Phys. 85, 235102 (2012).33. Sturm, C. et al. All-optical phase modulation in a cavity-polaritonMach–Zehnder interferometer. Nat. Commun. 5, 3278 (2014).34. Amo, A. et al. Exciton–polariton spin switches. Nat. Photonics 4,361 (2010).35. Zhao, J. et al. Room temperature polariton spin switches based onVan der Waals superlattices. Nat. Commun. 15, 7601 (2024).36. Butté, R. et al. Transition from strong to weak coupling and theonset of lasing in semiconductor microcavities. Phys. Rev. B 65,205310 (2002).37. Takemura, N. et al. Dephasing effects on coherent exciton-polaritons and the breakdown of the strong coupling regime. Phys.Rev. B 92, 235305 (2015).38. Tsotsis, P. et al. Tuning the energy of a polariton condensate viabias-controlled Rabi splitting. Phys. Rev. Appl. 2, 014002 (2014).39. Günter, G. et al. Sub-cycle switch-on of ultrastrong light–matterinteraction. Nature 458, 178 (2009).40. Sermage, B. et al. Radiative recombination of free excitons in gaasquantum wells. Superlattices Microstruct. 13, 271 (1993).41. Wen, X., Gong, Z. & Li, D. Nonlinear optics of two-dimensionaltransition metal dichalcogenides. InfoMat 1, 317 (2019).42. Wang, G. et al. Colloquium: Excitons in atomically thin transitionmetal dichalcogenides. Rev. Mod. Phys. 90, 021001 (2018).43. Novoselov, K., Mishchenko, A., Carvalho, A. & Castro Neto, A.2d materials and van der Waals heterostructures. Science 353,aac9439 (2016).44. Schneider, C., Glazov, M. M., Korn, T., Höfling, S. & Urbaszek, B.Two-dimensional semiconductors in the regime of strong light-matter coupling. Nat. Commun. 9, 2695 (2018).45. Kang, H., Ma, J., Li, J., Zhang, X. & Liu, X. Exciton polaritons inemergent two-dimensional semiconductors. ACS Nano 17, 24449(2023).46. Gu, J. et al. Enhanced nonlinear interaction of polaritons via exci-tonic rydberg states in monolayer WSe2. Nat. Commun. 12, 2269(2021).47. Emmanuele, R. et al. Highly nonlinear trion-polaritons in a mono-layer semiconductor. Nat. Commun. 11, 3589 (2020).48. Lyons, T. et al. Giant effective zeeman splitting in a monolayersemiconductor realized by spin-selective strong light–matter cou-pling. Nat. Photonics 16, 632 (2022).49. Zhang, L. et al. Vanderwaals heterostructure polaritonswithmoiré-induced nonlinearity. Nature 591, 61 (2021).50. Datta, B. et al. Highly nonlinear dipolar exciton-polaritons in bilayerMoS2. Nat. Commun. 13, 6341 (2022).51. Moody, G., Schaibley, J. & Xu, X. Exciton dynamics in monolayertransition metal dichalcogenides. JOSA B 33, C39 (2016).52. Tang, Y. et al. Interacting plexcitons for designed ultrafast opticalnonlinearity in a monolayer semiconductor. Light. Sci. Appl. 11,94 (2022).53. Zhao, J. et al. Exciton polariton interactions in Van der Waalssuperlattices at room temperature. Nat. Commun. 14, 1512 (2023).54. Louca, C. et al. Interspecies exciton interactions lead to enhancednonlinearity of dipolar excitons and polaritons in mos2 homo-bilayers. Nat. Commun. 14, 3818 (2023).55. Gerber, I. C. et al. Interlayer excitons in bilayer MoS2 with strongoscillator strength up to room temperature. Phys. Rev. B 99, 1(2019).56. Leisgang, N. et al. Giant Stark splitting of an exciton in bilayerMoS2.Nat. Nanotechnol. 15, 901 (2020).57. Katsch, F., Selig, M. & Knorr, A. Exciton-scattering-induceddephasing in two-dimensional semiconductors. Phys. Rev. Lett. 124,257402 (2020).58. Trovatello, C. et al. Disentangling many-body effects in the coher-ent optical response of 2d semiconductors. Nano Lett. 22, 5322(2022).59. Genco, A. et al. Ultrafast dynamics of rydberg excitons and theiroptically induced charged complexes in encapsulated wse2monolayers. Nano Lett. 25, 7673 (2025).60. Genco, A. et al. Ultrafast exciton and trion dynamics in high-qualityencapsulated MoS2 monolayers. Phys. Status Solidi (B) 260,2200376 (2023).Article https://doi.org/10.1038/s41467-025-61607-2Nature Communications |         (2025) 16:6490 10www.nature.com/naturecommunications61. Jeong, T. Y., Lee, S.-Y., Jung, S. & Yee, K. J. Photoinduced trionabsorption in monolayer WSe2. Curr. Appl. Phys. 20, 272 (2020).62. Aivazian, G. et al. Many-body effects in nonlinear optical responsesof 2d layered semiconductors. 2D Mater. 4, 025024 (2017).63. Singh, A. et al. Coherent electronic coupling in atomically thinMoSe2. Phys. Rev. Lett. 112, 216804 (2014).64. Dal Conte, S., Trovatello, C., Gadermaier, C. & Cerullo, G. Ultrafastphotophysics of 2d semiconductors and related heterostructures.Trends Chem. 2, 28 (2020).65. Shahnazaryan, V., Iorsh, I., Shelykh, I. A. & Kyriienko, O. Exciton-exciton interaction in transition-metal dichalcogenide monolayers.Phys. Rev. B 96, 115409 (2017).66. Pogna, E. A. et al. Photo-induced bandgap renormalization governsthe ultrafast response of single-layer MoS2. ACS Nano 10, 1182(2016).67. Ruppert, C., Chernikov, A., Hill, H. M., Rigosi, A. F. & Heinz, T. F.The role of electronic and phononic excitation in the opticalresponse of monolayer WS2 after ultrafast excitation. Nano Lett.17, 644 (2017).68. Cunningham, P. D., Hanbicki, A. T., Reinecke, T. L.,McCreary, K.M.&Jonker, B. T. Resonant optical stark effect in monolayer WS2. Nat.Commun. 10, 5539 (2019).69. LaMountain, T. et al. Valley-selective optical stark effect of exciton-polaritons in a monolayer semiconductor. Nat. Commun. 12,4530 (2021).70. Pöllmann, C. et al. Resonant internal quantum transitions andfemtosecond radiative decay of excitons in monolayer WSe2. Nat.Mater. 14, 889 (2015).71. Nie, Z. et al. Ultrafast electron and hole relaxation pathways in few-layer MoS2. J. Phys. Chem. C. 119, 20698 (2015).72. Din, N. U., Turkowski, V. & Rahman, T. S. Ultrafast charge dynamicsand photoluminescence in bilayer MoS2. 2D Mater. 8, 025018(2021).73. Nie, Z. et al. Ultrafast carrier thermalization and cooling dynamics infew-layer MoS2. ACS Nano 8, 10931 (2014).74. Wang, H., Zhang,C. &Rana, F. Ultrafast dynamics of defect-assistedelectron–hole recombination in monolayer MoS2. Nano Lett. 15,339 (2015).75. Rodriguez, S. R.-K. Classical and quantum distinctions betweenweak and strong coupling. Eur. J. Phys. 37, 025802 (2016).76. Wurdack, M. et al. Negative-mass exciton polaritons induced bydissipative light-matter coupling in an atomically thin semi-conductor. Nat. Commun. 14, 1026 (2023).77. Robert, C. et al. Measurement of the spin-forbidden darkexcitons in MoS2 and MoSe2 monolayers. Nat. Commun. 11,4037 (2020).78. Pandey, S. K., Das, R. & Mahadevan, P. Layer-dependent electronicstructure changes in transition metal dichalcogenides: The micro-scopic origin. ACS Omega 5, 15169 (2020).79. Christmann, G. et al. Large vacuum rabi splitting in a multiplequantum well gan-based microcavity in the strong-couplingregime. Phys. Rev. B 77, 085310 (2008).80. Zasedatelev, A. V. et al. Single-photon nonlinearity at room tem-perature. Nature 597, 493 (2021).81. Tyszka, K. et al. Leaky integrate-and-fire mechanism inexciton–polariton condensates for photonic spiking neurons. LaserPhotonics Rev. 17, 2100660 (2023).82. Stern, B. et al. On-chipmode-divisionmultiplexing switch.Optica 2,530 (2015).83. Chakraborty, P. S. et al. A 0.8 THz fMAX Si Ge HBT operating at 4.3 K.IEEE Electron Device Lett. 35, 151 (2014).84. Lin, X. et al. All-optical machine learning using diffractive deepneural networks. Science 361, 1004 (2018).85. Wang, T. et al. Image sensing with multilayer nonlinear opticalneural networks. Nat. Photonics 17, 408 (2023).86. Del Pozo-Zamudio, O. et al. Electrically pumped WSe2-based light-emitting van der waals heterostructures embedded in monolithicdielectric microcavities. 2D Mater. 7, 031006 (2020).87. Maggiolini, E. et al. Strongly enhanced light–matter coupling ofmonolayer WS2 from a bound state in the continuum. Nat. Mater.22, 964 (2023).88. Weber, T. et al. Intrinsic strong light-matter coupling with self-hybridized bound states in the continuum in van der waals meta-surfaces. Nat. Mater. 22, 970 (2023).89. Miri, M.-A. & Alu, A. Exceptional points in optics and photonics.Science 363, eaar7709 (2019).90. Khurgin, J. B. Exceptional points in polaritonic cavities and sub-threshold fabry–perot lasers. Optica 7, 1015 (2020).91. Opala, A. et al. Natural exceptional points in theexcitation spectrumof a light–matter system. Optica 10, 1111 (2023).92. Kedziora, M. et al. Non-hermitian polariton–photon coupling in aperovskite open microcavity. Nanophotonics 13, 2491 (2024).93. Gong, Z. et al. Topological phases of non-hermitian systems. Phys.Rev. X 8, 031079 (2018).94. Hanai, R., Edelman, A., Ohashi, Y. & Littlewood, P. B. Non-hermitianphase transition from a polariton bose-einstein condensate to aphoton laser. Phys. Rev. Lett. 122, 185301 (2019).95. Rahmani, A., Opala, A. & Matuszewski, M. Exceptional points andphase transitions in non-hermitian nonlinear binary systems. Phys.Rev. B 109, 085311 (2024).96. Raja, A. et al. Dielectric disorder in two-dimensional materials. Nat.Nanotechnol. 14, 832 (2019).97. Song, K. W., Chiavazzo, S. & Kyriienko, O. Microscopic theory ofnonlinear phase space filling in polaritonic lattices. Phys. Rev. Res.6, 023033 (2024).98. Robert, C. et al. Fine structure and lifetime of dark excitons intransition metal dichalcogenide monolayers. Phys. Rev. B 96,155423 (2017).99. Selig, M. et al. Excitonic linewidth and coherence lifetime inmonolayer transition metal dichalcogenides. Nat. Commun. 7,13279 (2016).AcknowledgementsCT acknowledges the European Union’s Horizon Europe research andinnovation program under the Marie Skłodowska-Curie PIONEER HOR-IZON-MSCA-2021-PF-GF grant agreement No 101066108. AG, CT, SDCandGCacknowledge funding from the EuropeanHorizon EIC PathfinderOpen program under grant agreement no. 101130384 (QUONDEN-SATE). This work reflects only authors’ view and the European Com-mission is not responsible for any use that may be made of theinformation it contains. AIT and SR acknowledgefinancial support of theEPSRC grants EP/V006975/1, EP/V026496/1 and EP/S030751/1. OK andKWS acknowledge the support fromUK EPSRCgrant EP/X017222/1. DGLthanks the EPSRC for support via ProgrammeGrant ‘Hybrid Polaritonics’EP/M025330/1.Author contributionsAG, CL, CC and CT carried out the optical spectroscopy experimentswith contribution from GDB and GS. CL and SR fabricated the 2D sam-ples. KW and TT synthesized the high quality hBN. KG, PC, RJ and DGLfabricated themicrocavities. AG, CL, CC, GDB and GS analyzed the datawith contribution from SDC, AIT and GC. AG performed the transfermatrix simulations. KWS and OK developed the theory on dynamicoptical saturation inmicrocavities. AG, CL and GCwrote themanuscriptwith contribution from all other co-authors. DGL, OK, SDC, AIT and GCmanaged various aspects of theproject. SDC,AIT andGCsupervised theproject.Competing interestsThe authors declare no competing interests.Article https://doi.org/10.1038/s41467-025-61607-2Nature Communications |         (2025) 16:6490 11www.nature.com/naturecommunicationsAdditional informationSupplementary information The online version containssupplementary material available athttps://doi.org/10.1038/s41467-025-61607-2.Correspondence and requests for materials should be addressed toAlexander I. Tartakovskii or Giulio Cerullo.Peer review information Nature Communications thanks the anon-ymous reviewers for their contribution to the peer review of this work. Apeer review file is available.Reprints and permissions information is available athttp://www.nature.com/reprintsPublisher’s note Springer Nature remains neutral with regard to jur-isdictional claims in published maps and institutional affiliations.Open Access This article is licensed under a Creative CommonsAttribution-NonCommercial-NoDerivatives 4.0 International License,which permits any non-commercial use, sharing, distribution andreproduction in any medium or format, as long as you give appropriatecredit to the original author(s) and the source, provide a link to theCreative Commons licence, and indicate if you modified the licensedmaterial. Youdonot havepermissionunder this licence toshare adaptedmaterial derived from this article or parts of it. The images or other thirdparty material in this article are included in the article’s CreativeCommons licence, unless indicated otherwise in a credit line to thematerial. If material is not included in the article’s Creative Commonslicence and your intended use is not permitted by statutory regulation orexceeds the permitted use, you will need to obtain permission directlyfrom the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.© The Author(s) 20251Dipartimento di Fisica, Politecnico di Milano, Piazza LeonardoDa Vinci 32, 20133Milano, Italy. 2NanoPhotonics Centre, Cavendish Laboratory, Department ofPhysics, JJ Thompson Ave, University of Cambridge, Cambridge, UK. 3Department of Physics, University of Exeter, Stocker Road, EX4 4PY Exeter, UK.4Department of Physics, Xiamen University Malaysia, 49300 Sepang, Malaysia. 5Department of Mechanical Engineering, Columbia University, New York10027 NY, USA. 6Dipartimento di ScienzeMatematiche, Fisiche e Informatiche, Università di Parma, Parco Area delle Scienze 7/A, 43124 Parma, Italy. 7Schoolof Mathematical and Physical Sciences, University of Sheffield, Sheffield S10 2TN, UK. 8Department of Physics, University of Cyprus, 1 Panepistimiou Avenue,2109 Aglantzia, Nicosia, Cyprus. 9Advanced Materials Laboratory, National Institute for Materials Science, 1-1 Namiki, Tsukuba 305-0044, Japan. 10CNR-IFN,Piazza Leonardo da Vinci 32, Milano 20133, Italy. 11These authors contributed equally: Armando Genco, Charalambos Louca.e-mail: a.tartakovskii@sheffield.ac.uk; giulio.cerullo@polimi.itArticle https://doi.org/10.1038/s41467-025-61607-2Nature Communications |         (2025) 16:6490 12https://doi.org/10.1038/s41467-025-61607-2http://www.nature.com/reprintshttp://creativecommons.org/licenses/by-nc-nd/4.0/http://creativecommons.org/licenses/by-nc-nd/4.0/mailto:a.tartakovskii@sheffield.ac.ukmailto:giulio.cerullo@polimi.itwww.nature.com/naturecommunications Femtosecond switching of strong light-matter interactions in microcavities with two-dimensional semiconductors Results Static and dynamic optical behavior of MoS2 bilayers Femtosecond switching of the strong coupling regime Ultrafast SC switching by interspecies interactions SC switching in a double BL microcavity Discussion Methods Sample fabrication Optical measurements Transfer matrix method analysis Theoretical model for the polariton dynamics Inclusion and ethics statement Data availability Code availability References Acknowledgements Author contributions Competing interests Additional information