The transition from quasi-static slip growth to dynamic rupture propagation constitutes one possible scenario to describe earthquake nucleation. It occurs when a region of critical size Lc of the fault, called the nucleation length, is slipping. If this transition is rather well understood for homogeneous faults, how the friction properties of multiscale asperities may influence the overall stability of seismogenic faults remains largely unclear. Combining the nucleation theory of [Uenishi and Rice, JGR, 2003] and concepts borrowed from condensed matter physics, we propose a comprehensive analytical framework that predicts the influence of heterogeneities of weakening rate/fracture energy on the nucleation length Lc for linear slip-dependent friction laws. Model predictions are compared to nucleation lengths measured from two-dimensional dynamic simulations of earthquake nucleation along heterogeneous faults. Our results show that the interplay between frictional properties and the asperity size gives birth to three instability regimes: (i) a local one, where the nucleation length is controlled by the local weakening rate, (ii) an extremal one, where it is governed by the weakest asperities, and (iii) a homogenized regime, where the instability threshold is set by the average frictional properties. We also show that heterogeneities at a scale far lower than the nucleation length do not participate in the fault stability. The precise knowledge of the nucleation length Lc ultimately allows to predict the loading levels (i.e. when) as well as the position (i.e. where) at which earthquakes may nucleate along heterogeneous faults, where both the weakening rate, the peak and residual frictions evolve continuously along the fault length (see Figure 1). Figure 1 (a) Nucleation dynamics along a heterogeneous coplanar fault: (b) under the influence of the steadily increasing loading, several regions of the slip-dependent frictional fault start slipping where the shear stress locally exceeds the fault strength. (c) Slip growth develops quasi-statically without any significant velocity burst, until one slip patch reaches a critical length that leads to the dynamic rupture of the whole fault interface. Our work aims at predicting how, when, and where such an instability nucleates.