|
|
As devices are scaled to characteristic sizes of one-tenth-micron and below,
special care must be taken to insure these devices are modeled
appropriately. The underlying physics in such small devices
can have a profound effect on device behavior and performance, yet
modeling tools developed for past device generations
rarely possess the
required physical basis
to correctly predict
device behavior at these small dimensions.
This problem is well recognized by many research groups world-wide.
As a result, research activities to improve the
physical basis of device simulation programs divide into three approaches:
- extensions to present
"drift-diffusion" analyses, by adding
additional equations governing carrier momentum and energy. Such
modeling is often referred to as
"energy-transport" or
"hydrodynamic" device modeling.
- solutions based upon the
Boltzmann Transport Equation, often incorporating
detailed descriptions of semiconductor properties such as
band structure and scattering rates.
- full quantum-mechanical treatments, emphasizing the
wave-like nature and interference properties of carriers at these
small dimensions.
(The preceeding hot-links point to the detailed pages by the
Computational Electronics Group,
University of Illinois at Urbana-Champaign).
There is activity at IBM in all three of these areas. The first
approach, being evolutionary in nature, has readily found a
following among device engineers.
Unfortunately, for sufficiently small devices we have seen
that a more rigorous physical basis is demanded.
The third approach above address the requirements for a sophisticated
and rigorous physical basis; however, such full quantum approaches
suffer from daunting implementation and theoretical issues.
Namely,
it is easy to view electrons as billiard balls as they collide.
It is relatively easy to treat them as waves freely moving
across the device. But an accurate description of electrons as
traveling waves "colliding" among themselves or scattering
off the vibrations of the ions requires computing (and brain)
power exceeding what is available today.
The second alternative seeks a middle ground between
too little and too much physics, and is fast becoming
recognized as a sound theoretical approach.
In 1987, a project was undertaken to create a device modeling computer
program that would contain the
necessary physics but allow
modeling of a broad class of "realistic" device structures.
The result is a program called DAMOCLES, which is mnemonic for
Device Analysis Using
Monte Carlo et Poisson solver.
This program combines a self-consistent solution,
via a
Monte Carlo sampling technique,
to the
Boltzmann transport equation (BTE)
and the Poisson equation. Conditionally, the Schrödinger equation (see the
details
provided by the Computational Electronics Group, University of Illinois, Urbana-Champaign)
can also be coupled into the self-consistent solution, which allows
DAMOCLES to model quantization in inversion layers and quantum wells.
DAMOCLES uses the full
band structure of the semiconductor with
consistently calculated
scattering rates in pursuit of
physical accuracy and rigor. DAMOCLES solves the BTE in three wavevector-space
dimensions, and the Poisson equation in two real-space dimensions.
Adding time as a final state variable, DAMOCLES conducts a six-dimensional
calculation. The program is computationally intensive, but
delivers a wealth of detailed information about the internal
behavior of a semiconductor device.
Today, after 16 man-years of development, the DAMOCLES
program represents the state-of-the-art in this type of
device simulation program. DAMOCLES was written by M.V. Fischetti
and S.E. Laux.
damoclesNO-SPAM@watson.ibm.com
(last updated: January 26, 1999)
|
|