|
How TSTEP Affects a Transient Simulation
How does the value of TSTEP affect my transient simulation?
- s: l. y+ S7 S+ `1 B P- m( m1 mTSTEP is specified in the transient analysis netlist command. P% p3 D$ G* g
.TRAN TSTEP1 TSTOP1 TSTEP2 TSTOP2 ... TSTEPN TSTOPN
- b4 k% j$ P, r8 ]# A9 ZThe most common usage is a single TSTEP/TSTOP pair. For example:6 u3 w: G! y- _
.TRAN 0.1ps 100ns
) f O3 j; s4 H; _TSTEP has a variety of effects on the operation of HSPICE.
; l7 ]4 ]7 ~: A' T1. When the RUNLVL algorithm is in use:
# @0 G$ v* J0 | - RUNLVL "in use" is defined as HSPICE version >= 2004.09 and OPTION RUNLVL > 0.
% v3 q+ j2 P9 B F6 q; n For HSPICE version >= 2006.09, RUNLVL=3 by default.
3 A- ^9 N S+ Q+ g6 E: w! n( u - The minimum timestep is determined by the RUNLVL algorithm. The RMIN option is ignored.7 m9 _+ L. m( j- ?
- The maximum timestep has a "soft" limit of RMAX*TSTEP. The RUNLVL algorithm2 I( l% Q8 R% F) N4 h$ |
allows the timestep to exceed RMAX*TSTEP as long as tighter Newton-Raphson5 v( Z" N8 L; {3 N8 Y1 f' Q4 w
tolerances are met (voltage, current, charge).The default value of RMAX is 5.
- _: f) x+ @* S - The DELMAX option can be used to override the RUNLVL algorithm's choice of
( e" v- J+ U* _5 L- f4 P maximum timestep. When DELMAX is used, TSTEP has no impact on simulation accuracy (except
9 i* `6 g& t) a( s/ a when used in conjunction with .OPTION INTERP, see #3 below).; _4 N+ S: h" |5 R7 m+ K/ f) E; P
2. When RUNLVL=0 (discouraged):" T, F+ f% u' v0 f! ]
- The minimum timestep is defined as TSTEP*RMIN. The default value of RMIN is 1e-9.7 X2 k! y, c: B6 ] Z4 y
- The maximum timestep is defined as TSTEP*RMAX. The default value of RMAX is 5.
9 |1 M" U$ l2 U0 e+ v u e. ~8 O - These limits are "hard" limits.3 V6 c" V2 O) x
- If the simulation engine attempts to choose a timestep < TSTEP*RMIN, a2 |' ?) f+ w. {0 W: j) @
Timestep Too Small Error will result.3 M( X$ l5 t* V! N9 Q
- The simulation engine is never allowed to take a larger timestep than
& n* N0 m5 P. I* Y TSTEP*RMAX.
/ \4 b# f& ]1 l$ k5 O Summary
4 C4 |% l( B3 f3 H6 WWith the introduction of the RUNLVL algorithm, the impact of TSTEP has been
* C& k! o8 Z+ H7 |5 x7 l/ hgreatly reduced to the point that it can be ignored except for niche applications. @' \5 H4 G2 H8 E4 D: s
like INTERP. One should still be careful not to set TSTEP too small for long,
; G& L# _& Y( Z9 |low frequency simulations in which the soft timestep upper limit of RMAX*TSTEP
" V+ y1 B8 S' N5 }8 Imay still force more timestep solutions than is necessary which will cause* \; g' h9 E) }
excessive simulation time. |
|