4.1.3 The Real Device

Five 400x400 $\mu$m$^2$ Si$_{1-x}$Ge$_{x}$ HBTs have been fabricated with the same geometry and layer specification as shown in Fig. 4.9. The only difference is the Ge content $x$ in the base, which is constant for a given device and varies between 16% and 28% for the respective devices. The following layer sequence applies for all devices: The emitter consists of an antimony doped n$^+$=2.10$^{20}$ cm$^{-3}$ contact region and n=10$^{18}$ cm$^{-3}$ region. The base is boron doped (p$^+$ = 5.10$^{19}$ cm$^{-3}$) and has thin undoped spacers on the top and beneath of it, respectively. The collector is antimony doped (n$^-$ = 2.10$^{16}$ cm$^{-3}$) and is followed by an intrinsic buffer and the substrate.

The device with the lowest Ge fraction in the base $x$ = 16% is analyzed first. In Fig. 4.10 results of several simulations are presented in their consecutive order compared to measured data at V$_\mathrm {BC}$ =0 V (red symbols). They are presented as an example of calibration of forward Gummel plot characteristics. The first simulation is performed with the default models and the device is defined according to the specification (green dashed lines). Thus, important effects, such as BGN and SRH recombination are still not taken into account. Note the significant disagreement between simulation and measurement, especially in the high-field region. However, the slope in the collector current density at low and middle voltages is correct, which shows that the conduction band discontinuity at base-to-emitter junction is modeled correctly. In the next simulation the BGN is switched on (black dot-dashed lines) and comparatively good agreement can be observed for I$_\mathrm C$. In the high-field region still a different behavior is observed. This can be explained by the complete depletion of the two spacers in the simulation which is not the case for the real device. As stated in [197] the spacers are used to prevent outdiffusion of boron from the base to the emitter in the following thermal processing. In the final device the SiGe spacers contain boron and are parts of real the base layer. In the next simulation already gradual boron doping profiles varying from 5.10$^{13}$ cm$^{-3}$ at the spacer-to-emitter and spacer-to-collector interfaces, respectively, to 5.10$^{19}$ cm$^{-3}$ at the base-to spacer interfaces are included. The only remaining step is to include SRH recombination in order to match the base current. Finally, good agreement for the complete bias range is achieved (solid red lines).

The next simulations are performed with the remaining four devices, which have higher Ge content in the base. In Fig. 4.11 a comparison between the simulated and measured Gummel plots for the devices with $x$ = 16%, 22%, and 28% Ge are shown. All simulations are performed with the same set of models and no adjusting of model parameters is performed. The only exception is done for the concentration of traps in the base, because with the high Ge content also more traps are introduced. Thus, quite good agreement is finally achieved.

Finally, the assumption for the doping is confirmed to some extend by SIMS measurements, which were not initially available. As can be seen in Fig. 4.12 it is true that in the real device outdiffusion of boron has taken place. However, the effect appears to be even more pronounced than initially assumed.

Figure 4.9: Simulated device structure of five SiGe HBTs

Figure 4.10: Forward Gummel plots at V$_\mathrm {BC}$ = 0 V: Study of different effects in a Si$_{0.84}$Ge$_{0.16}$ HBT

Figure 4.11: Forward Gummel plots at V$_\mathrm {BC}$ = 0 V: Comparison between simulation and measurement for different material contents

Figure 4.12: Boron profile in the base region: Comparison between specification and SIMS data