We scrutinize the real-frequency structure of the self-energy in the superconducting state of the attractive Hubbard model within the dynamical mean-field theory. Within the strong-coupling superconducting phase which has been understood in terms of the Bose-Einstein condensation in the literature, we find two qualitatively different regions crossing over each other. In one region close to zero temperature, the self-energy depends on the frequency only weakly at low energy. On the other hand, in the region close to the critical temperature, the self-energy shows a pole structure. The latter region becomes more dominant as the interaction becomes stronger. We reveal that the self-energy pole in the latter region is generated by a coupling to a hidden fermionic excitation. The hidden fermion persists in the normal state, where it yields a pseudogap. We compare these properties with those of the repulsive Hubbard model relevant for high-temperature cuprate superconductors, showing that hidden fermions are a key common ingredient in strongly correlated superconductivity.PACS numbers: 67.85. Lm, 71.10.Fd A range of metals show superconductivity below a critical temperature (T c ), at which paired electrons acquire a spatial coherence. In conventional superconductors, the pairing mechanism is well described by the Bardeen-CooperSchrieffer (BCS) theory [1]. However, the BCS theory does not straightforwardly apply when the attractive interaction between electrons is strong. In this case, the electron pairing occurs at a temperature higher than T c and superconductivity arises when the electron pairs (regarded as composite bosons) go through the Bose-Einstein condensation (BEC) at a lower temperature [2]. Such superconductivity in the BEC regime has been explored for ultracold fermionic atom systems [3][4][5]. In fact, tightly-bound pairs [6] and an associated pseudogap behavior [7,8] have been observed for 40 K gas in the strongly-interacting region. A recent experiment also suggested that the superconductivity in FeSe is in the BCS-BEC crossover regime [9]. The preformed pairing has also been proposed in the context of cuprate high-T c superconductors [10,11]. Although it looks unlikely that the preformed pairing solely can explain the whole pseudogap behavior in the cuprates [12][13][14][15], it may be relevant in a region close to T c around the optimal doping [16,17].On the theoretical side, the crossover from BCS to BEC [5,[18][19][20] has been intensively studied with continuous [21][22][23][24][25][26][27] or lattice [28-41] fermion models, both of which give a similar phase diagram. The latter is, however, more tractable with numerical simulations and allows us to employ nonperturbative methods to study this problem. In particular, the dynamical mean-field theory (DMFT) [42], which becomes exact in infinite spatial dimensions, is a suitable tool to study dynamical properties of the lattice models. In fact, the DMFT and its extensions have been extensively applied to the attractive Hubbard model [43][44][45][4...