2216 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 48, NO. 10, OCTOBER 2001
Comparative Study of Electron Transit Times Evaluated
by DD, HD, and MC Device Simulation for a
SiGe HBT
Christoph Jungemann, Member, IEEE, Burkhard Neinhüs, and Bernd Meinerzhagen, Member, IEEE
Abstract—Transit times of a silicon/germanium heterojunc-
tion bipolar transistor (HBT) with a base width of 24 nm are
investigated in the quasi–stationary limit for the first time by
consistent drift-diffusion (DD), hydrodynamic (HD), and fullband
Monte Carlo (MC) simulations. The quasi-ballistic transport in
the base and collector leading to a strong velocity overshoot is
well described by the HD model and corresponding transit times
are in good agreement with the MC results. On the other hand,
the DD model fails in this region and substantially overestimates
the base transit time bearing the possibility of wrong guidelines
for transistor design optimization. However, since the base transit
time is no longer dominating the cutoff frequency of high-speed
HBTs, the failure of the DD model leads to an underestimation
of the peak cutoff frequency by only 10%. Close to high injection
differences in the emitter transit times of the HD and MC model
are observed which are mainly related to small differences in the
Gummel plot.
Index Terms—Bipolar transistor, cutoff frequency, device simu-
lation, silicon.
I. INTRODUCTION
T
HE silicon/germanium (SiGe) alloy has made
bandgap–engineering possible based on more or less
mainstream Si technologies retaining the cost advantage of Si
over III/V materials [1], [2]. SiGe heterojunction bipolar tran-
sistors (HBT) have been fabricated with a maximum oscillation
frequency in the excess of 100 GHz [3], [4]. Due to the small
base widths of less than 50 nm, the electron transport becomes
more and more ballistic. This calls into question the validity
of the standard simulation tools like the drift-diffusion (DD)
or hydrodynamic (HD) model used to design high-frequency
HBTs. In addition, the strain caused by the pseudomorphic
growth of the epilayers results in a strong anisotropy of the
band structure only captured in full detail by the fullband
Monte Carlo (MC) model.
In this paper, the impact of the quasi-ballistic transport and
the anisotropic band structure on the accuracy of the DD and HD
models are investigated by a comparison with the more accurate
MC model for transit times and cutoff frequencies of an HBT.
First, the simulation details, and then results for the HBT are
presented.
Manuscript received February 27, 2001; revised June 25, 2001. This work
was supported in part by the Bundesministerium für Bildung, Wissenschaft,
Forschung und Technologie under Contract 01M2416A. The review of this
paper was arranged by Joachim Burghartz.
The authors are with the University of Bremen, 28334 Bremen, Germany
(e-mail: junge@item.uni-bremen.de).
Publisher Item Identifier S 0018-9383(01)08357-5.
II. THEORY
The details of the MC model can be found in [5]. The band
structure is calculated with the nonlocal empirical-pseudopoten-
tial method [6]. The conduction and valence band edge are mod-
eled similar to [1], [7], [19] and are the same for all three models.
Apparent bandgap narrowing due to heavy doping is included as
described in [8]. Further details of the DD and HD models are
given in [9], [10]. All transport parameters of the DD and HD
models are generated by MC bulk simulations to ensure consis-
tency of the simulation models. The heat flux of the HD model
is reduced to 25% compared to [9] as described in [11].
Using the three simulation models, transit times are evaluated
for an HBT. In order to reduce the CPU time of the MC simu-
lations, a one-dimensional (1-D)-device approximation is used.
The missing base contact is simulated by setting the quasi-fermi
potential of the holes in the center of the base to the value of the
base voltage. The 1-D-device approximation works well as long
as the device is not operated too far in the high-injection regime.
In addition, generation/recombination processes are neglected,
because they are difficult to include into an MC simulation. In
the case of the MC model, no holes are simulated and their con-
tribution to the space charge is calculated based on the DD hole
quasi-fermi potential in conjunction with a nonlinear Poisson
equation. These restrictions make it difficult to compare the re-
sults to experimental data, but this is not the objective of this
paper, which is the investigation of the approximations inherent
to the DD and HD models by comparison with the more fun-
damental MC model. Thus, a high degree of consistency of the
models is more important here than a comparison with experi-
ments.
In the case of an 1-D-device approximation, the electron
transit time for a region extending from to reads [12]
(1)
where
electron charge;
electron density;
collector current per device area;
collector/emitter voltage.
The derivative is calculated by varying the base/emitter voltage.
The total transit time is inversely proportional to the common
emitter cutoff frequency: , where is the
device length. The transit time and cutoff frequency are calcu-
lated in the quasi–stationary approximation in accordance with
0018–9383/01$10.00 © 2001 IEEE