Injecting spins into a semiconductor channel and transforming the spin information into a significant electrical output signal is a long-standing problem in spintronics. This is the prerequisite of several concepts of spin transistor. In this paper, we discuss the general problem of spin transport in a nonmagnetic channel between source and drain. Two problems must be mastered: 1) In diffusive regime, the injection/extraction of a spin-polarized current into/from a semiconductor beyond the ballistic zone at the interface with a magnetic metal requires the insertion of a spin-dependent and large enough interface resistance. 2) In both the diffusive and ballistic regimes and whatever the metallic or semiconducting character of the source/drain, a small enough interface resistance is the condition to keep the dwell time shorter than the spin lifetime and, thus, to conserve the spin-accumulation-induced output signal at an optimum level (Delta R/R approximate to 1 or larger). Practically, the main difficulties come from the second condition. In our presentation of experimental results, we show why the transformation of spin information into a large electrical signal has been more easily achieved with carbon nanotubes than with semiconductors, and we discuss how the situation could be improved in the later case.