.

Sebastian:

.

I am not sure I can help, since I've never used the

particular simulator you refer to only in the header.

The confusing results might be due to an inherent

math problem in your simulator.

.

But as one who has conducted hundreds of ultra-

high resolution transient simulations using our in-

house simulation suite of programs I can suggest

a few things to look out for, The first is to use the

tightest possible transient tolerances. If you state

– in frank detail – what are the convergence limits

you have set up for these experiments, we may

be able to suggest some changes.

'

A second thought occurs to me: do the transistor

models include thermal modeling? Anyway, what

is the peak junction current in experiments? If it's

(as a suspect) very high and thermal modeling is

included you would expect different results on the

fine scale, if a succession of ramps in applied to

courageous, but long-suffering, device.

.

Thirdly, you might consider putting your transistor

collector junction into DC avalanche operation, by

shorting the base to the emitter and applying a DC

current of the appropriate polarity into this junction.

You can still apply transient tests to this transistor.

For example, using a pulse voltage applied to the

base, you may find that a shorted stub (s few cms

of 50-ohm cable, say) will discharge very rapidly -

with sub-nanosecond rise-times.  

.

But, for this to work as advertised, one must hand-

select devices whose punch-through voltage (that

value of the reverse collector bias that causes the

acitive base-width is exactly zero) is slightly more

than the collector breakdown (avalanche) voltage.

In the 1960's I used discrete germanium transistors

in this mode, to generate sub-nanosecond pulses 

in a sampling oscilloscope (see attachment).

.i

Finally, if you have to use an applied ramp, it may

be useful to limit its swing, from slightly under the

breakdown of these junctions to slightly over.

.

Barrie

.

 Mr. Gilbert, Mr. Dubey, let me thank you first for your kind responses. I'm addressing each suggestion/question as inline comments:

(1) vary the load capacitor value

I’m not including a load capacitor in the simulation. I’m simply ramping quasistationary bias values to the Anode and then simulating the Transient Ion Impact under these conditions.

(2) change the initial and final time and time steps

I’ve tried this, and actually I’m seeing strange behavior at the beginning of the transient analysis: current starts in very high values and tends to its qusistationary value in the first picosecond of simulation (0.8ps). See attachment.

(3) minimize the step size

Also tried this, but with no positive result: RHS remains high even for error-converging solutions. (see data density in attachment)

"The first is to use the tightest possible transient tolerances."

For what I understand from the software Manual, Transient tolerances are fixed through a couple of variables: TransientDigits, TransientError and Transient ErrRef, that relate through the following expressions, where x is the updated variable:

RelativeTransientError =  10-TransientDigits

TransientErrRef =( AbsoluteTransientError / RelativeTransientError ) * x

Transient simulation method is TRBDF (trapezoidal rule/backward differentiation formula), in which I’m definitely no expert. These parameters control the Time Step. Default values are 3 for digits and Error parameters depend on the solved equation (for example, 0.1*VT for Poisson’s equation,  or 0.01*300K for Thermal equations).

I’ve tried reducing the relative error (i. e., increasing the number of digits), but with no improvement. I still could try reducing the absolute error, but I’m having very long simulation times (3 days on a dense but 2D mesh without very complex transport models) and I should first try to reduce the simulation time maintaining current convergence in order to make some trial runs. I’ll start with this right away.

Once again, thanks for your help.

Best regards.

Sebastian