Monte Carlo analysis of Gunn Oscillations and thermal effects in GaN-Based devices

Abstract

In recent years, the development of GaN technology has made significant inroads into high-power and high-frequency applications with respect to other semiconductor competitors such as GaAs or InP. In this dissertation, by means of an in-house Monte Carlo tool, we study the possibility of generating Gunn oscillations through vertical n+nn+ (without notch) and n+n-nn+ (notched) diodes based on InP and GaN with different lengths of the active region and two doping profiles. In general, when the notch accomplishes its role of fixing the onset of charge accumulation near the cathode, the oscillations are of lower frequency and power. For InP-based diodes, the fundamental frequency reaches 140 GHz (notched, L=1.2 µm) and 400 GHz (without notch, L=0.75 µm). For the GaN-based diode with an active length L=1.5 µm, despite the fact that the fundamental harmonic is around 100 GHz, the power spectral density for the 10th harmonic ( 1 THz) is still significant. InP diodes with L=0.9 µm offer an efficiency (¿) of up to 5.5 % for frequencies around 225 GHz. The generation bands in GaN diodes appear at higher frequencies (up to 675 GHz with ¿=0.1 %) than in InP. When circuits work at high powers, thermal models become essential to determine temperature limits with a view to preventing device failure, thus reducing manufacturing costs. In order to include thermal effects in our Monte Carlo code, two techniques have been implemented: (i) a thermal resistance method (TRM), and (ii) an advanced electro-thermal model that solves the steady-state heat diffusion equation, called HDEM. We calibrate/validate our simulator by comparison with experimental measurements of an AlGaN/GaN diode. For the TRM, several thermal resistance values are employed, and for the HDEM different substrates (polycrystalline diamond, diamond, carbide silicon, silicon and sapphire), thicknesses and die lengths are tested. In addition, we include the effect of thermal boundary resistance. Using temperature-independent thermal conductivity in the HDEM allows us to extract an equivalent thermal resistance, Rth, for each geometry and substrate material. The extracted Rth values are constant with the dissipated power, Pdiss. However, when a more real temperature-dependent thermal conductivity is employed, Rth exhibits a strong dependence on Pdiss. As final test device, we analyse for an HEMT, the effect of (i) the heat-sink temperature and (ii) gate-length, through electro-thermal simulations.