moiré superlattice. It is well known that the two neighboring layers of 2-H TMDC flakes, due to the intralayer inversion symmetry breaking, possess a layer degree of freedom that acts as pseudospins alternating in odd and even layers 2,21 . In an angle-aligned trilayer WSe 2 /monolayer WS 2 heterojunction (3L WSe 2 /1L WS 2 ), new types of interlayer excitons emerge, with holes residing in the first WSe 2 layer either trapped in moiré A or B site (Fig. 1c) 5 , and electrons with the same pseudospin residing in the third WSe 2 layer. In particular, we find that these two interlayer excitons can hybridize through coupling with the intralayer excitons in the third WSe 2 layer. The resulting hybridized exciton inherits both the large oscillator strength from the intralayer excitons and the sensitive electric field dependence from the moiré interlayer excitons 7,[22][23][24] . More interestingly, by applying an electric field, we can drive the transition between the two interlayer moiré excitons' hybridization with the intralayer exciton in the third WSe 2 layer, enabling the continuous tuning of hopping of the interlayer exciton from 100% at one moiré site to 100% at the other, which is otherwise suppressed. In between the transition points, we obtain an excitonic complex that is the superimposition of the interlayer excitons that are otherwise localized at moiré A and B sites.

The schematic of the 3L WSe 2 /1L WS 2 moiré heterojunction is shown in Fig. 1a, which is fabricated into a dual-gated device structure in which the doping and electric field can be independently controlled. We also fabricated a device of a dual-gated 2-H phase trilayer WSe 2 (3L WSe 2 ) (schematically shown in Fig. 1b) for the control study.

The doping-dependent optical reflectance contrast spectra of the 3L WSe 2 /1L WS 2 heterojunction device are shown in Fig. 2e, which is evidently different from that of the natural trilayer (3L) WSe 2 device (Fig. 2f). The most pronounced resonance for the natural trilayer WSe 2 (Fig. 2f) is the intralayer exciton resonance X A , which is at ~1.70 eV at zero doping, redshifted compared to the A exciton resonance in monolayer WSe 2 (~1.73 eV) 25 . X A is redshifted linearly for both n and p doping in a symmetric fashion, with a slope of ~1.3 meV/10 12 cm -2 . IX 3L are the interlayer excitons with the hole and electron separated in the first and third WSe 2 layer, which have two degenerate modes as schematically shown in Fig. 2c and are named as IX + 3L and IX À 3L ("+" and "-" denote the direction of the dipole moment in the sample coordinate. The direction of the positive electric field or dipole moment is defined as from the top gate to the back gate in 3 L WSe 2 , and from WSe 2 to WS 2 in 3L WSe 2 / 1 L WS 2 .). IX 2s 3L is the 2s state of the IX 3L . The natures of IX 3L and IX 2s 3L become obvious in our later discussion of the electric field dependent reflectance contrast spectra. Zoom-in of Fig. 2f with enhanced contrast is plotted in Fig. S2 to show IX 3L and IX 2s 3L more clearly. Accompanying X A is a less pronounced resonance X A ' with a larger slope (2.7 meV/10 12 cm -2 ). X A ' is likely the exciton resonance of the middle (second) layer WSe 2 and is not the focus of this work (see detailed discussion in Supplementary Section 14).

In the optical reflectance contrast spectra of 3L WSe 2 /1L WS 2 moiré heterojunction (Fig. 2e), there is an exciton resonance ðX I M Þ located at the lower energy side of X A (~1.667 eV), which is only observable in angle-aligned 3L WSe 2 /1L WS 2 heterojunctions but absent in heterojunctions with large twist angles (See supplementary section 10 for detailed discussion). X I M is the previously discovered moiré intralayer exciton peak in the first layer WSe 2 interfacing WS 2 , with the exciton trapped at the moiré A site. The doping dependence of X I M clearly show the signature of the correlated insulating states at the filling factor of 1 and -1, corresponding to one electron and one hole per moiré superlattice, which was discussed in our previous publication 26 . On the p-doping side, the exciton resonances of X A and X A ' are labeled as such due to their similar behaviors compared with that from the trilayer WSe 2 (Fig. 2f), with a redshift slope of 1.0 and 2.1 meV/10 12 cm -2 , respectively. The n-doping side is different because the electrostatically introduced electrons are in the WS 2 layer instead of the WSe 2 layers due to the type II alignment, leaving the WSe 2 layers charge-neutral. We identify the X A and X A ' in the n-doping side through their slopes as well, 1.0 and 2.1 meV/10 12 cm -2 , respectively, the same as those in the p-doping region. The abrupt blueshift of the X A in the n-doping side (starts at n > 1 and resonant energy around 1.725 eV) is likely due to the built-in electric field on WSe 2 layers arising from the electron accumulation in WS 2 . We leave the related discussion in Supplementary Information Section 15. The focus of our work here is on the interlayer excitons within the 3L WSe 2 of the 3L WSe 2 /1L WS 2 moiré heterojunction, with their schematics shown in Fig. 2a, b. The IX + 3L branch is visible and pronounced at the blue arrow in Fig. 2e, partially because it hybridizes with intralayer excitons and gains some oscillator strength but also because it retains the extended nature of interlayer excitons, hence sensing dielectric environment change associated with the Mott insulator transition at filling of one electron per moiré superlattice 20 , which we also call as A, B and C moiré sites for convenience. A and B sites are energy minima for holes and behave as quantum dots that confine carriers and excitons. We use the holes for illustration in c, but the trapping of electrons and excitons will be similar.

(n = 1). The nature of both resonances are revealed in our later discussion of the electric field dependence study.

The electric field-dependent reflectance contrast spectra of the trilayer WSe 2 device is shown in Fig. 3a, which is symmetric about the electric field due to its symmetric structure. The most noticeable feature is the "cross" pattern originating from the electric field evolution from interlayer exciton IX 3L . The slope of each branch of the cross is roughly the same. These are arising from the Stark shift of the interlayer exciton IX 3L , with the two degenerate modes (IX À 3L and IX + 3L ) shifting oppositely under an electric field due to the dipole moment of opposite polarity. The Stark energy shift can be expressed as ΔE = À edF, where F is the local electric field, e is the electron charge, and d is the electron and hole separation. We extract the value of d to be about 1.26 nm for both IX À 3L and IX + 3L , which is about twice that of interlayer exciton dipole moment in WSe 2 /WS 2 (0.7 nm) 7 , confirming that the electron and hole of interlayer exciton occupy the two outside WSe 2 layers in a natural trilayer WSe 2 .

The level avoiding at the intralayer exciton A (~1.70 eV) in Fig. 3a arises from the hybridization of the interlayer exciton and intralayer exciton. In the 2H trilayer WSe 2 , there is significant tunneling of holes between the first and third layer WSe 2 as they have the same valleylayer pseudo spin, allowing the hybridization of the interlayer excitons with the intralayer excitons in either the first or third WSe 2 layer 27 , as schematically shown in Fig. 3d. This hybridization can be well captured by a coupled two-level system, which is given by the following Hamiltonian in the basis of intralayer exciton and interlayer exciton:

where X a is the energy of the intralayer exciton, X i ðFÞ is the energy of the interlayer exciton at a given electric field F, Δ is the coupling strength (see Supplementary Information Section 13 for details). Take the positive electric field (direction defined in Fig. 3d) scenario as an example (Fig. 3c): a linearly dispersed interlayer exciton IX + 3L (white dotted line in Fig. 3c) and a non-dispersed intralayer exciton X A (black dotted line) can be used to well fit the observed hybridized spectra (red and blue dashed lines). From the fitting, we extract the coupling strength to be 10.7 ± 0.3 meV, larger than the linewidth of the hybridized exciton (~9.0 ± 0.3 meV). The scenario of the negative electric field is similar, where the other interlayer exciton mode, IX À 3L , hybridizes with the intralayer exciton (X A ) when the energy of the two excitons is tuned to resonance via the electric field. It is worth noting that we ignore the conduction band hybridization of the first and third layer WSe 2 , which is theoretically predicted to be nonzero but orders of magnitude smaller than the holes 27 . The neglection of the conduction band hybridization is also justified by the electric-field-dependent reflectance contrast spectra of 3 L WSe 2 / 1 L WS 2 , which is asymmetric about positive and negative electric fields (later discussion of Fig. 4).

The additional level avoiding at the energy around 1.79 eV in Fig. 3a is due to the hybridization of the interlayer exciton ðIX 3L Þ with the 2s state of intralayer A exciton (Fig. 3a and Fig. S4a, b). The second level avoiding at higher energy (~1.80 eV) is due to the hybridization of the excited state of the interlayer exciton ðIX 2s 3L Þ and 2s of the A exciton, which we enhance the contrast and show in Fig. S4a, b. It is interesting to note that the energy difference between the ground state and 2s of interlayer exciton IX 3L is about 51 meV, smaller but at the same order of magnitude compared with the energy difference between 2s and 1s of A exciton for trilayer WSe 2 (~95 meV, Fig. S4a, b), suggesting the strongly bound nature of the interlayer exciton IX 3L . All these hybridization features are absent in a dual-gated nature bilayer WSe 2 (Fig. S8), which is AB stacked with two layers of different layer pseudospin, further confirming our interpretation. The electric-fielddependent reflectance contrast spectra of a 4 L WSe 2 device (Fig. S7) show similar hybridization features but with two "crosses" slightly shifted in energy, about 10 meV. According to the interpretation of the 3 L WSe 2 data, these two crosses are the two types of interlayer excitons from the 1st and 3rd layer WSe 2 and the 2nd and 4th layer WSe 2 , which slightly shift in energy due to possible dielectric environment differences 28,29 .

We now turn to the study of the electric field-dependent reflectance contrast spectra of the 3 L WSe 2 / 1 L WS 2 moiré heterojunction, shown in Fig. 4c. The negative electric field side has some similarity compared with that from trilayer WSe 2 , while the positive electric field side is significantly different. More specifically, the hybridized spectrum on the positive electric field side involves three exciton branches: two dispersive (white dotted lines in Fig. 4e) and one non-dispersive (black dotted line in Fig. 4e) branch. The necessity of involving three exciton branches is also obvious from the derivative of Fig. 4e with respect to the electric field, as shown in Fig. S3d. The two dispersive excitons have a similar slope for the Stark shift, translating to electron and hole separations of 1.313 ± 0.004 nm and 1.609 ± 0.004 nm (fitting details in Supplementary Information Section 13). Therefore, they are the interlayer excitons, similar to IX + 3L in the natural trilayer WSe 2 , with the hole in the first WSe 2 layer interfacing WS 2 and the electron in the third WSe 2 layer away from the interface. The two different interlayer excitons stem from the moiré coupling modified valence band of the first WSe 2 layer. As schematically shown in Fig. 4a and b, the moiré modulation folds the valence band of the first WSe 2 layer into moiré minibands. The two interlayer excitons correspond to holes occupying the two moiré minibands located at different moiré sites 20 , which effectively behave as two spatially separated quantum dots. Each of them is located at an energy minimum at a high symmetry point within the moiré unit cell, which we call moiré A and B sites, respectively. We thus label these two interlayer excitons as IX + ðAÞ 3L and IX + ðBÞ 3L . Since the WS 2 and first WSe 2 layer are aligned at 60 degrees (H stacked) as determined by the second harmonic generation (SHG) spectra (Fig. S5), the moiré A and B sites correspond to the H h h and H X h stacking configurations shown in Fig. 4b. The energy separation of the two interlayer excitons at zero electric field, 69 meV, represents the energy difference between the top two moiré minibands, if we ignore the difference in exciton binding energy. This value is consistent with the energy difference between the intralayer excitons trapped at moiré A and B sites in the WSe 2 /WS 2 moiré superlattice, ~53 meV 26 . The remaining non-dispersive branch corresponds to the intralayer exciton, X A , with both hole and electron in the third WSe 2 layer. Therefore, IX + ðAÞ 3L , IX + ðBÞ 3L , and X A hybridize by sharing the electron in the third WSe 2 layer. The above picture of hybridization involving two moiré interlayer excitons are confirmed by a control device (D4) of 3 L WSe 2 / 1 L WS 2 in the dual-gate configuration, with an intentionally misaligned angle (20-degree) between WSe 2 and WS 2 layers. The electric-fielddependent reflectance contrast spectra (Fig. S11) indeed become symmetric about the electric field and similar to that of natural 3 L WSe 2 , and they show no signs of interlayer moiré excitons (IX + ðAÞ 3L and IX + ðBÞ 3L ). It is worth noting that direct between moiré A and B sites is suppressed due to the energy barrier, their spatial separation, and different stacking symmetry. Therefore, a direct hybridization between these two sites is difficult to achieve, unlike the degenerate valley-spin bands in TMDCs. However, with the assistance from the mobile intralayer exciton in the third WSe 2 layer, hybridization of moiré A and B sites is realized, and we can controllably tune the interlayer excitons IX + 3L between moiré A and B sites. In fact, the hybridized exciton notated with the cyan dashed line is a mixture of interlayer excitons localized at the moiré A site and B site, with the probability tunable from 100% at A to 100% at B site by controlling the electric field (Fig. 4f).

On the negative electric field side, the interlayer exciton involved in the hybridization is IX À 3L , with the hole in the third layer WSe 2 not experiencing the moiré modulation. Meanwhile, the intralayer exciton in the 1 st WSe 2 layer is modified by the moiré potential to have a lower energy of ~1.667 eV and is trapped at the moiré A site, which is labeled as X I M . As a result, hybridization occurs between IX À 3L and X I M . Their coupling strength is extracted to be 11.4 ± 0.1 meV. The interlayer exciton IX À 3L can also couple to the other moiré excitons from the 1 st layer WSe 2 , which contributes to the weak features in Fig. 4 and are shown with enhanced contrast in Fig. S4c, d.

The asymmetry of Fig. 4c between the n-and p-doping sides further justifies our neglection of conduction band hybridization: IX À observe the hybridization of two interlayer excitons (due to moiré modulated conduction bands) and X A . Similarly, the interlayer exciton from region e (IX + 3L ) in Fig. 4c should hybridize with X I M . We include a detailed discussion in Supplementary Information Section 16.

In summary, we have demonstrated a strategy to realize continuous tuning of interlayer exciton hopping between different moiré sites in 3 L WSe 2 / 1 L WS 2 moiré superlattices. These additional degrees of freedom enable the formation of a tunable honeycomb lattice of excitons with exciting opportunities for engineering new quantum states. For example, considering the large spin-orbit coupling in TMDCs, the continuous tuning of the hopping can be potentially exploited for constructing Dirac and Weyl modes of excitons, as well as the topologically protected edge states connecting these modes 19 . Our demonstration of the superposition of excitons across the different moiré sites also inspires new venues of quantum information processing and harnessing the new moiré site degree of freedom for twistronics.

We used the same dry pick-up method 30 as reported in our earlier work to fabricate TMDC heterostructures 17,26 . The gold electrodes are prepatterned on the Si/SiO 2 substrate. The monolayer TMDC flakes, BN flakes, and few-layer graphene (FLG) flakes are exfoliated on silicon chips with 285 nm thermal oxide. The thickness of BN flakes was determined by atomic force microscopy (AFM). The layer numbers of WSe 2 flakes were identified by optical contrast with the assistance of second-harmonic generation (SHG). Top BN and bottom flakes with equal thickness were intentionally used for devices D1, D2, and D3. The polycarbonate (PC)/ polydimethylsiloxane (PDMS) stamp was used to pick up TMDC monolayer and other flakes sequentially. The alignment of each layer is achieved under a home-built microscope transfer stage with the rotation controlled with an accuracy of 0.02 degrees. The PC is then removed in the chloroform/isopropanol sequence and dried with nitrogen gas. The final constructed devices were annealed in a vacuum (<10 -6 torr) at 250 °C for 8 hours.

During the optical measurements, a home-built confocal imaging system was used to focus the laser onto the sample (with a beam spot diameter ~2 µm) and collect the optical signal into a spectrometer (Princeton Instruments). The reflectance contrast measurement performed using a supercontinuum laser source (YSL photonics). A relative flat reflectance background R 0 was obtained by fitting the reflectance spectrum at high hole-doping level with a polynomial function for each measured spot (see Supplementary Section 11 for details). The reflectance contrast is defined as dR R = RÀ R 0 R 0 . The reflectance contrast from device D1 and D2 are added by 0.3 and -0.3 for better presentation in the log scale. All optical spectroscopy measurements were performed at the temperature of 10 K with a Montana cryostat. The polarized SHG measurements were performed with a pulsed laser excitation centered at 900 nm (Ti: Sapphire; Coherent Chameleon) with a repetition rate of 80 MHz and a power of 80 mW. The crystal axes of the sample were fixed. A half-waveplate was placed between the beam splitter and the objective and was rotated to change the polarization angles of both the excitation laser and the SHG signal.

The density of carriers introduced by the electrostatic gating is given by n e ðn p Þ = C tg ðV tg À V 0 tg Þ + C bg ðV bg À V 0 bg Þ, where C tg ðC bg Þ are the geometry capacitance of the top gate (back gate) and V tg ðV bg Þ are the top gate (back gate) voltage. V 0 tg and V 0 bg are the onset gate voltages of the top gate and the back gate, determined experimentally from the regions where the 2s peaks remain visible. The electrical field in the TMDC is given by F = ε BN =ε TMDC ðV tg =2d 1 À V bg =2d 2 Þ, where d 1 ðd 2 Þ is the thickness of the top (bottom) BN determined by atomic force microscopy, ε BN = 3:5 and ε TMDC = 7:2 are the relative dielectric constants of h-BN and TMDC, respectively 31,32 .