What is the maximum signal frequency SPICE can handle ?

blue perhaps.

how good are your parts models?

That depends on how low you can set the step size and how well your models match real parts.

There's a theoretical limit where you run out of mantissa in the floating point but you'd be simulating gamma ray frequencies before you hit that limit.

I ran a fimple LPF sim in ltspice with 10THz sine wave and it gave what appeared to be consistent results.

At the end of the day it's just arithmentic.

YES! separate math limit from simulation effectiveness limit.

IME, for discrete components at around 10MHz, PSpice seems to no longer represent reality. I've always attributed that to the 'weakness' in the modeling, but never bothered to pursue further. Inside components, translate that to SMALL geometry, at around 100MHz.

Mathematically I constantly use to beyond 1GHz, however, more for 'learning' what's going on than 'describing' what's going on.

At 1+ THz, EVERYTHING must be acting like a radiating, transmission line. Can't imagine the complexity of a useful model. Even the wavefronts through conduction must have some effects, right? Be like trying to model the acoustics of a concert hall with all that wave/ reflections, etc.

Below about 10 THz, the method of moments (MOM) works okay. Above there, material dispersion and limited conductivity make full-wave simulations de rigueur--FEM or FDTD, usually.

The conductivity limit comes in because the effective dielectric constant of a normal conductor is

epsilon = epsilon.re + j sigma/omega

so conductors look steadily less conductive as the frequency goes up. Antenna simulations using MOM at 28 THz (CO2 laser territory) look vaguely like the fullwave ones, and vaguely like the experiments, but if you go any higher, all resemblance is lost. My 200 THz stuff was all fullwave.

Cheers

Phil Hobbs

at

spice small documentary

I have used PSpice to model RF stages at 5GHZ, no problem. If you don't set a "Maximum Time-step" small enough (my rule-of-thumb is 1/32 of the period of the highest frequency... or smaller), you will get a jagged sampling output.

That might be with behaviorally-modeled devices.

Finite element analysis? ...Jim Thompson

There should be lots of floating-point range to get to THz. There are a few Spice parameters that might matter. Minimum time step is one. LT Spice has "chgtol" and "cshunt" and "cshuntintern" that will matter at super speeds, possibly others. There is a "minimum inductance damping" thing, too.

The default Spice is probably optimized/compromised for speed and convergence with practical circuit values and speeds. After all, you can't really make a 150 THz LC filter.

all spice knows is time steps and equations

it doesn't really know the physical concept of frequency

-Lasse