Field studies and seismic data show that semi-brittle flow of fault rocks probably is the dominant deformation mechanism at the base of the seismogenic zone at the so-called frictional - viscous transition. To understand the physical and chemical processes accommodating semi-brittle flow, we have performed an experimental study on synthetic granitoid fault rocks exploring a broad parameter space (temperature, T = 300, 400, 500 and 600°C, confining pressure, Pc ≈ 300, 500, 1000 and 1500 MPa, shear strain rate, γ ≈ 10-3, 10-4, 10-5 and 10-6 s-1, to finite shear strains, = 0 - 5). The experiments have been carried out using a granular material with grain size smaller than 200 μm with a little H2O added (0.2wt%). Only two experiments (performed at the fastest strain rates and lowest temperatures) have failed abruptly right after reaching peak strength ( ~ 1400 MPa). All other samples reach high shear stresses ( ~ 570 – 1600 MPa), then weaken slightly (by Δ ~ 10 – 190 MPa) and continue to deform at a more or less steady state stress level. Clear temperature dependence and a weak strain rate dependence of the peak as well as steady state stress levels is observed. In order to express this relationship, the strain rate – stress sensitivity has been fit with a stress exponent, assuming γ  n and yields high stress exponents (n ≈ 10 - 140), which decrease with increasing temperature. The microstructures show widespread comminution, strain partitioning and localization into slip zones containing, at first, nanocrystalline and partly amorphous material, later, during continued deformation, fully amorphous material in slip zones. Despite the mechanical steady state conditions, the fabrics in the slip zones and outside continue to evolve and do not reach a steady state microstructure below γ = 5. Within the slip zones, the fault rock material progressively transforms from a crystalline solid to an amorphous material. We present and interpret the experimental results both in terms of sliding friction and viscous flow, and we discuss the possible effect that the formation of nanocrystalline and amorphous layers may have on earthquake nucleation.