Hafnium oxide based ferroelectric FETs (FeFETs) are highly suitable for in-memory computing applications like neuromorphic hardware due to their CMOS compatibility, high dynamic range, low power consumption and good linearity. Device-to-device and die-to-die variability play an important role, especially due to the polycrystalline nature of ferroelectric hafnium oxide. Here, the variability of FeFET based synapses integrated in 300 mm wafers is investigated, showing low drain current variability for up to 32 states per cell. Furthermore, Si doping of HfO2 enables lower voltage amplitudes for learning compared to Zr. Finally, simulation of current percolation paths in these devices reveals more insight in the parameters affecting variability.