Ba. Biegel et Jd. Plummer, COMPARISON OF SELF-CONSISTENCY ITERATION OPTIONS FOR THE WIGNER FUNCTION-METHOD OF QUANTUM DEVICE SIMULATION, Physical review. B, Condensed matter, 54(11), 1996, pp. 8070-8082
In the present work, we compare the efficiency, accuracy, and robustne
ss of four basic iteration methods for implementing self-consistency i
n Wigner function-based quantum device simulation. These methods inclu
de steady-stale Gummel, transient Gummel, steady-stale Newton, and tra
nsient Newton. Ln a single mathematical framework and notation, we pre
sent the numerical implementation of each of these self-consistency it
eration methods. As a lest case to compare the iteration methods, we s
imulate the current-voltage (I-V) curve of a resonant tunneling diode.
Standard practice ibr this task has been to rely solely on either st
steady-state or a transient iteration method. We illustrate the danger
s of this practice, and show how to take advantage of the complimentar
y strengths of both steady-state and transient iteration methods where
appropriate. Thus, because the steady-state methods are vastly more e
fficient (i.e., have a much lower computational cost), and are usually
equal in accuracy to the transient methods, the former are preferable
for wide-ranging initial device investigations such as tracing the I-
V curve. Implementation difficulties which we address here may have re
duced the use of the steady-stale methods in practice. On the other ha
nd, the transient methods are inherently more robust and accurate (i.e
., they reliably and correctly reproduce device;physics), However, the
high computational cost of the transient methods makes them more appr
opriate for a narrower range of directed investigations where transien
t effects are inherent or suspected, rather than for full I-V curve tr
acts. Finally, we found the two Gummel methods to be generally prefera
ble to their (theoretically more accurate) Newton counterparts, since
the Gummel methods are: equally accurate in practice, while having a l
ower computational cost.