Spin injection and extraction are at the core of semiconductor spintronics. Electrical injection is one method of choice for the creation of a sizeable spin polarization in a semiconductor, requiring especially tailored tunnel or Schottky barriers. Alternatively, optical orientation can be used to generate spins in semiconductors with significant spin-orbit interaction, if optical selection rules are obeyed, typically by using circularly polarized light at a well-defined wavelength. Here we introduce a novel concept for spin injection/extraction that combines the principle of a solar cell with the creation of spin accumulation. We demonstrate that efficient optical spin injection can be achieved with unpolarized light by illuminating a p-n junction where the p-type region consists of a ferromagnet. The discovered mechanism opens the window for the optical generation of a sizeable spin accumulation also in semiconductors without direct band gap such as Si or Ge.

Spin injection from Fe(001) and (Ga,Mn)As(001) into n-GaAs(001) was investigated using a method which provides two-dimensional cross-sectional images of the spin polarization in GaAs. While the distribution of the spin polarization below the injecting contact is nearly uniform for (Ga,Mn)As, a strong confinement near the contact edge is observed for Fe and FeCo. The spin polarization in GaAs changes sign when the injected current is reversed. Multiple sign reversals as a function of bias voltage as reported previously for Fe injectors are not observed with (Ga,Mn)As and Fe contacts grown on clean n þþ ÀGaAs in agreement with earlier results for an epitaxial FeCo injector.

The spin accumulation in an n-GaAs channel produced by spin extraction into a (Ga,Mn)As contact is measured by cross-sectional imaging of the spin polarization in GaAs. The spin polarization is observed in a 1 µm thick n-GaAs channel with the maximum polarization near the contact edge opposite to the maximum current density. The one-dimensional model of electron drift and spin diffusion frequently used cannot explain this observation. It also leads to incorrect spin lifetimes from Hanle curves with a strong bias and distance dependence. Numerical simulations based on a two-dimensional drift-diffusion model, however, reproduce the observed spin distribution quite well and lead to realistic spin lifetimes.