Modeling of carrier transport in modern semiconductor devices has reached a
point where the classic drift-diffusion model is no longer sufficient. While
in the last decade reasonably good results could be obtained with the
drift-diffusion model, a plethora of new effects needs to be considered in any
accurate model. These effects include various quantum mechanical phenomena
like carrier quantization in the channel, tunneling through oxides, and
resonant tunneling, in addition to ballistic transport phenomena. Various
models have been proposed to capture one or the other effect. In general, this
involves a considerable amount of additional computational effort. For
instance, to account for quantization in the channel, the Schrödinger equation
has to be solved, while the quasi-ballistic transport regime is best captured
by a Monte Carlo solution of the Boltzmann equation. However, for very small
devices, Boltzmann's equation is no longer valid and has to be replaced, for
instance, by the Wigner-Boltzmann equation. If device scales are of the order
of a few nanometers, scattering effects might loose their importance and a
coherent transport model based on Schrödinger's equation can be used.
Up to now, no established 'standard TCAD model' for end-of-the-roadmap
semiconductor devices is available, leaving the device designer with many
uncertainties regarding the best modeling approach. In particular, the best
choice depends on the application at hand, where a good balance between accuracy
and efficiency needs to be found. Since this often cannot be decided in
advance, the most promising approaches need to be tried and the results
compared. Therefore, a modern device simulator has to offer at least the most
established approaches. However, it is essential that these approaches are
implemented as consistently as possible, because, otherwise, one would not
be able to perform a meaningful comparison.
To achieve this goal we have decided to provide all additional simulation
modes, as a first step a Monte Carlo and a Schrödinger module, via Minimos-NT
as a generalized interface. The Monte Carlo and Schrödinger solvers have been
designed as libraries which obtain their data directly from Minimos-NT. For
this, several regions can be defined in Minimos-NT. In these regions the
original mesh can be used or a new mesh can be generated which is suitable for the
specialized module. This new mesh can be of lower dimensionality, and a
generalized interface is provided to extend the calculated result to the
original mesh.
In addition, this approach has the advantage that all modules use the
Minimos-NT material database and, when possible, the same material models.
Inconsistencies can thus be avoided and a comparison of the models is made
possible. In addition, the transport parameters required by Minimos-NT for the
Monte Carlo table-based transport models can be generated directly by
Minimos-NT.
|