The drive towards device scaling and large output power in millimeter and sub-millimeter wave power amplifiers results in a highly non-linear, out-of-equilibrium charge transport regime. Particle-based Full Band Monte Carlo device simulators allow an accurate description of this carrier dynamics at the nanoscale. This work initially compares GaN high electron mobility transistors (HEMTs) based on the established Ga-face technology and the emerging N-face technology, through a modeling approach that allows a fair comparison, indicating that the N-face devices exhibit improved performance with respect to Ga-face ones due to the natural back-barrier confinement that mitigates short-channel-effects. An investigation is then carried out on the minimum aspect ratio (i.e. gate length to gate-to-channel-distance ratio) that limits short channel effects in ultra-scaled GaN and InP HEMTs, indicating that this value in GaN devices is 15 while in InP devices is 7.5. This difference is believed to be related to the different dielectric properties of the two materials, and the corresponding different electric field distributions. The dielectric effects of the passivation layer in millimeter-wave, high-power GaN HEMTs are also investigated, finding that the effective gate length is increased by fringing capacitances, enhanced by the dielectrics in regions adjacent to the gate for layers thicker than 5 nm, strongly affecting the frequency performance of deep sub-micron devices. Lastly, efficient Full Band Monte Carlo particle-based device simulations of the large-signal performance of mm-wave transistor power amplifiers with high-Q matching networks are reported for the first time. In particular, a CellularMonte Carlo (CMC) code is self-consistently coupled with a Harmonic Balance (HB) frequency domain circuit solver. Due to the iterative nature of the HB algorithm, this simulation approach is possible only due to the computational efficiency of the CMC, which uses pre-computed scattering tables. On the other hand, HB allows the direct simulation of the steady-state behavior of circuits with long transient time. This work provides an accurate and efficient tool for the device early-stage design, which allows a computerbased performance evaluation in lieu of the extremely time-consuming and expensive iterations of prototyping and experimental large-signal characterization.