1. Introduction
Much work has been done on the mechanism of band-to-band tunneling (BTBT), which is the basic principle for tunneling field effect transistors (TFET), especially in indirect semiconductors[1-4], such as Si and Ge. However, because the L and
In this paper, we will mainly focus on the comparison among three widely-used BTBT models when they are employed in Si homo-junction and Si–Ge hetero-junction simulation at reverse biases. At the same time, the performance of our recently developed two-dimensional (2D) full-band Monte Carlo (FBMC) simulator with BTBT models is also evaluated (by comparison with Sentaurus). This FBMC simulator is then used to study the BTBT mechanism in Si–Ge hetero-junctions with diverse doping concentrations.
2. Physical models
There are three BTBT tunneling models that can be chosen in Sentaurus: Hurkx's model, Schenk's model and the nonlocal path model.
The most widely used, Hurkx's model, gives[10, 11],
RBTBT=ADεpe−B/ε. |
(1) |
This model is based on a two-band model.
D=np−n2i,,eff(n+ni,,eff)(p+ni,,eff). |
(2) |
The effect of
In the device with a steep p–n junction and high field, exceeding (approximately) 8
RBTBT=ASDε[(F∓C)−3/2exp(−F∓Cε)exphωkt−1+,(F∓C)−3/2exp(−F∓Cε)1−exphωkt], |
(3) |
with
F±C=expBS(Eg±hω)3/2. |
(4) |
The upper sign refers to the generation process (
For a more accurate calculation, the nonlocal BTBT models in the Sentaurus TCAD should be switched on in the device simulation. This model allows the simulator to dynamically search for the tunneling path that has a direction which is opposite to the gradient of the valence band. To compute the BTBT rate of such a method is very complex and the method has been discussed in Ref.[14], but in the uniform electric-field limit and one-dimensional case, these equations can be reduced to a form the same as Eq. (1)[15].
For both Si and Ge, the values of
For Si, where the indirect transitions are dominant in the BTBT process, the theoretical values of A and B are 3.29
3. Simulation method
The 2D FBMC simulator has been discussed in detail in Refs. [18, 19]. The band structures, including four conductions bands and three valence bands, were calculated from the local empirical pseudo-potential method. Scattering mechanisms, including acoustic and optical phonon scattering, and impact ionization scattering for both Ge and Si were considered.
The non-self-consistent mode was used in our Monte Carlo (MC) simulator as follows. Firstly, we obtained the electrostatic potential for the whole device using the drift–diffusion method, and then this potential was kept frozen during the simulation. For each step, the program calculated the BTBT generation or recombination rate over the whole device using Eq. (1) or Eq. (3), and the generated particles with charge
Pc=RBTBTΔSΔt, |
(5) |
where
As a comparison with the MC method, in Sentaurus we used the Fermi-Dirac statistics model (Fermi), the drift-diffusion carrier transport model, the doping–dependent mobility model (DopingDep), and the high-field velocity saturation model (HighField saturation). The doping-dependent band-gap-narrowing model (BandGap narrowing) was also switched on.
4. Results and discussions
In order to verify the validity of the MC simulator, the results of an Si diode at a series of forward biases (0.15–1 V) were obtained (Fig. 2). At a forward bias, BTBT can hardly occur since the conduction band edge is located above the valence band edge (see the inset of Fig. 2). It is found that the forward currents computed by the two simulators agree well with each other when the reverse bias is above 0.5 V. That is to say, there is no difference between the results calculated from them if BTBT does not occur. These results prove the reliability of our MC simulator, as well as the reasonability of the comparisons in this paper.
Figure 3 plots the BTBT currents of an Si homo-junction at various reverse biases, from 1 to 4 V. It is worth noting that the current density obtained from Schenk's model is one order of magnitude less than that from Hurkx's model in both Sentaurus and the MC simulator in the range 1.5–4 V (reverse bias). This can also be verified from the difference of

On the other hand, at very low reverse biases (0–1.5 V), the current obtained from the MC simulator is much smaller than that from Sentaurus. In Fig. 5, it is found that at the peak of the BTBT rates,

However, in the Si–Ge hetero-structure, the situation is very different. The
For Ge devices, it has been reported that the direct BTBT process dominates over the indirect BTBT process[6, 14]. Similarly, in the Si–Ge hetero-junction used in Fig. 7, the direct tunneling current is about four orders of magnitude larger than the indirect one at the same bias, which is attributed to the band structures of Ge. Figure 9 is a comparison of
5. Summary
In this paper, we compared three different BTBT models with both a Sentaurus TCAD and 2D FBMC hetero-junction simulator in diverse Si homo-junctions and Si–Ge hetero-junctions. It was shown that in Si homo-junctions, these models achieved the same results in predicting the BTBT rates and currents at reverse biases. However, for the Si–Ge hetero-junctions, there were some contradictions in these models, especially at high reverse biases, and an underestimation of the tunneling rates and spatial range by local models could be seen from our comparisons.
It was also found that our improved MC simulator with BTBT models has good computation accuracy compared with Sentaurus, as well as good convergence for the hetero-junctions. In addition, we utilized our MC simulator to invest the BTBT phenomenon in three different abrupt Si–Ge hetero-junctions, and interactions of the direct and indirect transitions near the interface of the Si and Ge regions were observed for the first time.