High-frequency response and thermal effects in gan diodes and transistorsmodeling and experimental characterization

Abstract

GaN-based self-switching diodes (SSDs) and high-electron-mobility transistors (HEMTs) have been analyzed in DC and AC regimes both from the point of view of experiments and simulations. The non-linearities present in their current-voltage curves allow their operation as zero-bias microwave detectors. Despite of the good properties of GaN, technological problems often related to defects, traps and heating are still issues that need to be investigated to boost the performance of future power electronics. Pulsed and transient measurements performed in SSDs reveal the inuence of surface and bulk traps on the DC characteristic and AC impedance. Surface trapping eects become relevant in narrow-channel SSDs as the surface-to-volume ratio of the device is increased, while in wider diodes bulk trapping eects prevail. Measurements show an anomalous enhancement of the microwave detection at low temperature, while the detected voltage exhibits a roll-o in frequency, which can be attributed to the presence of surface and bulk traps. Virgin AlGaN/AlN/GaN HEMTs exhibit strong low-frequency dispersion in the microwave range both in the transconductance and output conductance, attributed to the presence of traps and defects both in the volume of the GaN channel and in the source and drain contacts. These eects have been modeled by means of a modied small-signal equivalent circuit (SSEC), achieving an excellent agreement with the measured S-parameters. The device geometry aects the values of the SSEC elements and hence the cuto frequencies, the gate length being the most determinant geometrical parameter. For LG = 75 nm, ft and fmax are 72 and 89 GHz, respectively, in the HEMTs under analysis. In the SSDs, a noise equivalent power (NEP) of 100-500 pW/Hz1/2 and a responsivity of tens of V/W was observed with a 50 Ω source. A cuto frequency of about 200 GHz, along with a square-law response up to 20 dBm of input power, have been demonstrated. At low frequency, RF measurements exhibit a responsivity that agrees well with the calculations performed by means of a quasi-static (QS) model based on the slope and curvature of the current-voltage curve. Biasing the devices increases the detected voltage with the disadvantage of the power consumption and the excitation of 1/f noise. The QS model predicts that the reduction of the channel width improves the responsivity, what was conrmed by experiments. The increase of the number of diodes in parallel reduces the device impedance; when it coincides with 3 times that of the transmission line (or antenna) to which they are connected, the NEP reaches a minimum value. Diodes with a top-gate electrode, called gated SSDs (G-SSDs), exhibit, in free-space measurements at 300 GHz, a responsivity around 600 V/W and a NEP around 50 pW/Hz1/2 when the gate bias approaches the threshold voltage. Again, a good agreement is found, but only above sub-threshold gate bias, between the results coming from the QS model and those obtained at low frequency (900 MHz) and in free space at 300 GHz. The NEP value can be improved by increasing the number of channels in parallel. A comparison between the injection of the RF signal at the drain (DCS) or the gate (GCS) electrode in the HEMTs operating as detectors is performed up to 40 GHz. For DCS, a responsitivy around 400 V/W and a NEP around 30 pW/Hz1/2 were obtained in a HEMT with LG = 150 nm at room temperature under zero drain current and when the gate is biased near pinch-o conditions. On the other hand, the responsivity is strongly enhanced in GCS, up to 1.4 kV/W, but with the drawback that it is necessary to apply a supplementary drain bias of ID = 1.2 mA. Both congurations show a similar cuto frequency, with a -3 dB roll-o at about 40 GHz. Interestingly, in GCS, at a frequency high enough for the gate-to-drain branch to eectively short the RF signal to the non-linearity, a non-zero detected voltage has been recorded at zero drain current. When devices work at high-power conditions, the study of self-heating becomes relevant. Simulations were done by means of an in-house Monte Carlo (MC) tool coupled with two thermal models: (i) a thermal resistance model (TRM) and (ii) an advanced electrothermal model (ETM) based on the self-consistent solution of the steady-state heat conduction equation. At room-temperature, the MC tool was rst calibrated by comparison with experimental results in TLMs (transfer length measurement), so that sheet-carrier density and mobility values were replicated. Including the eects of the contact resistance, the Schottky barrier and the thermal boundary resistance, our results are validated with experimental measurements of a HEMT with LDS = 1.5 µm and LG = 150 nm, nding a reasonably good agreement. The TRM with well-calibrated values of thermal resistances (RTH) provides a similar behavior to ETM simulations. The advantage of the ETM is that it provides the temperature map inside the channel and allows to identify the hotspot location. In addition, the SSEC was obtained with MC simulations, nding a good correspondence with the experimental values of the parameters. Impact of the biasing on the SSEC elements and discrepancies between TRM and ETM calculations are discussed. Pulsed measurements up to 500 K are used to estimate the channel temperature and the value of the RTH. For T < 250 K, the responsivity in the DCS decreases abruptly in the sub-threshold region after reaching a maximum, while it remains practically constant for T > 250 K.