The optical properties of atomically thin semiconductors are dominated by excitons, tightly bound electron-hole pairs, which give rise to particularly rich and remarkable physics. Despite their importance, the microscopic formation mechanisms of excitons remain very poorly understood due to the complex interplay of concurrent phenomena occurring on an ultrafast timescale. Here, we investigate the exciton formation processes in 2D materials based on transition-metal dichalcogenide (TMD) monolayers using a technique based on the control of excitation light polarization. It allows us to distinguish between the two competing models of exciton formation: geminate and bimolecular formation. The geminate process is the direct formation of the exciton from the initially photogenerated electron-hole pair before the loss of correlation between them, whereas the bimolecular process corresponds to the random binding of free-electron-hole pairs from the initially photogenerated plasma. These processes control the exciton formation time. Our findings reveal that the luminescence intensity is higher by up to 40% for circularly polarized excitation compared to linearly polarized excitation for laser energy above the free-carrier gap. We show that this spin-dependent exciton emission is a fingerprint of the bimolecular formation process. Importantly, we observe that exciton linear polarization (valley coherence) persists even for laser excitation energies exceeding the gap. We demonstrate that it is the result of a fraction of excitons formed by a geminate process. This shows that two formation processes coexist for excitation energies above the gap, where both mechanisms operate concurrently. Similar results obtained on the two most emblematic materials of the TMD semiconductor family, WSe2 and MoS2 monolayers, confirm this dual formation mechanism. These findings overturn the prevailing simplistic view of purely geminate or bimolecular exciton formation and provide crucial insights into exciton physics and means to preserve spin or valley coherence. Our methodology can be applied to a broader range of semiconductor for investigating the intricate interplay between nonresonant excitation conditions, Coulomb interactions, electron-phonon couplings, excitonic dynamics, and quantum coherence.
Mourzidis et al. (Mon,) studied this question.