2.4.1 Reports by Research Groups

The evolution of MBE growth technique and modulation doping together with a vivid interest in the behavior of quantum well structures in the late 70s made the demonstration of the first AlGaAs/GaAs HEMT by Mimura et al. [3] possible. The potential of the technology was quickly realized and several designs (AlGaAs/InGaAs PHEMT and AlInAs/InGaAs/InP HEMT) were proposed in order to counter various problems. The first AlGaN/GaN based HEMTs were demonstrated in the early 90s [8] after methods for deposition of GaN on sapphire by MOCVD were developed.

However, the main driving force for the continuous improvements in the material growth technology [55] were Nitride-based light-emitting diodes (LEDs). There is a strongly growing demand for the latter for home electronics but also lighting and back-lighting, where the market in 2010 amounts to over 10 billions USD. For comparison the market for GaN RF transistors in 2010 is only 100 millions USD.

The very high values for the electron sheet charge density were the reason for a shift of the research interest from AlGaAs/GaAs to AlGaN/GaN based devices. Consequently the characteristics of GaN based HEMTs have been improved steadily in the last decade. While the first HEMTs exhibited a cut-off frequency $ f_\ensuremath {\mathrm {t}}$ and a maximum oscillation frequency $ f_\ensuremath{\mathrm{max}}$ of 11 GHz and 35 GHz ( $ L_{\mathrm{g}}$=0.25 $ \mu$m) [56] later devices reached 50 GHz and 120 GHz, respectively, in 2002 [57]. Currently the highest reported $ f_\ensuremath {\mathrm {t}}$ and $ f_\ensuremath{\mathrm{max}}$ values are 190 GHz and 240 GHz, respectively, for $ L_{\mathrm{g}} =60 \ensuremath{\mathrm{nm}}$ devices with high Al-composition and thin barrier layers [58]. Another notable achievement is reported by Shinohara et al., who measured a cut-off frequency of 153 GHz of a DH-HEMT with $ L_{\mathrm{g}} =60 \ensuremath{\mathrm{nm}}$ [59], and also Chung et al., who produced a device with the same gate length and a maximum frequency of 300 GHz [60]. However, such a performance is near the limit of the AlGaN/GaN technology, imposed by the limited polarization-induced electric fields and current collapse. Fig. 2.3 shows a steady increase of the measured $ f_\ensuremath {\mathrm {t}}$ over the years, however such a illustration does not account only for the technology improvement, but also for the down-scaling of the gate lengths.

Figure 2.3: Cut-off frequency of GaN HEMTs over time.
\includegraphics[height=0.43\textheight]{figures/state/Ft.eps}

This is avoided in Fig. 2.4, where the product $ f_\ensuremath {\mathrm {t}}$$ \times $ $ L_{\mathrm{g}}$ is depicted. There is a clear limit of roughly 20 GHz$ \times\mu$m, which was rarely exceeded. This value was also recently reached with AlGaN/GaN HEMTs [61]. First proposed by Kuzmik in 2001 [62], the InAlN/GaN interface posseses a higher polarization-induced sheet charge density as AlGaN/GaN. As the InAlN layer can be grown lattice-matched to GaN, possible strain relaxation problems are significantly reduced. Consequently, this potential was quickly realized and the focused research of such structures is yielding excellent results: e.g. cut-off frequencies of 144 GHz for a $ L_{\mathrm{g}} =100 \ensuremath{\mathrm{nm}}$ device [63].

Figure 2.4: Cut-off frequency $ \times $ gate length of GaN HEMTs.
\includegraphics[height=0.43\textheight]{figures/state/Ftxlg.eps}

Another optimization goal is the maximum power density (Fig. 2.5). The first HEMTs exhibited barely 1.1 W/mm at 2 GHz [64] (Fig. 2.5). Employing multiple field-plates the power density was raised up to 40 W/mm at 4 GHz [65]. By using internally matched amplifier technology the limits have been pushed up to 550 W at 3.5 GHz [66]. Because of the large band gap, GaN based HEMTs are also considered for high power operations.

Figure 2.5: Power density vs. frequency of GaN HEMTs.
\includegraphics[height=0.43\textheight]{figures/state/PvsF.eps}

Early samples demonstrated impressive breakdown voltages in the range of 230 V, but also poor subthreshold behavior [67]. Those issues were addressed and breakdown voltages $ V_\ensuremath {\mathrm {BR}}$=570 V were reached by gate geometry optimization [68]. By using a ``slant-field-plate'' technology a $ V_\ensuremath {\mathrm {BR}}$=1900 V could be achieved at the cost of drain current degradation [69]. Another way of increasing $ V_\ensuremath {\mathrm {BR}}$ is by using a thick buffer layer, leading to a maximum value of $ V_\ensuremath {\mathrm {BR}}$$ >$1800 V [70]. While the breakdown voltage increases steadily over the years (Fig. 2.6), a more significant characteristic is the product $ f_\ensuremath {\mathrm {t}}$$ \times $ $ V_\ensuremath {\mathrm {BR}}$. Some of the achieved very high voltages are due to exceptionally large devices. The cut-off frequency of the latter is quite low, which translates in a low $ f_\ensuremath {\mathrm {t}}$$ \times $ $ V_\ensuremath {\mathrm {BR}}$ (Fig. 2.7).

Figure 2.6: Breakdown voltage $ V_\ensuremath {\mathrm {BR}}$ of GaN HEMTs over time.
\includegraphics[height=0.43\textheight]{figures/state/VB.eps}

Figure 2.7: Cut-off frequency $ \times $ breakdown voltage of GaN HEMTs.
\includegraphics[height=0.43\textheight]{figures/state/FtxVB.eps}


S. Vitanov: Simulation of High Electron Mobility Transistors