Speed baby, speed. As far as accuracy - I did not see too much difference between spectre, Turbo and APS. Also the difference is number of tokens (licenses) it takes.
I recommend to use brain when using APS though. For smaller circuits(bellow 100 - 250 FETS) it does not make much difference to use APS. Just few seconds.
Also APS is really good if you have machine with multiple CPUs.
But main difference is speed. On larger circuits orr on ascillators it is quite amazing.
Speed is the main difference.. but watch out if you are using Turbo, u may want to specify the max time step if u are doing really accurate sims. spectre turbo reduces the tolerance to converge faster which can introduce errors. APS is pretty good esp if u have a multi core system.
Speed is the main difference.. but watch out if you are using Turbo, u may want to specify the max time step if u are doing really accurate sims. spectre turbo reduces the tolerance to converge faster which can introduce errors. APS is pretty good esp if u have a multi core system.
What about the simplifications that APS makes to the MOS models?
I'm not too worried about the parasitic reduction and 0V source removal, but the removal of source and drain resistors worries me quite a lot... </understatement>
Well, if you calculate real resistance of SD with contacts and vias based on the layout you will find out that it does not really make sense to use the estimated values anyway,. Moreover submicron MOS proceses are silicided
so the resistance is very miniskule.
The model simplification - I believe (since cadence realy does not say) that fet models are tabulized and then during simulation are used as look up table. So yes there will be small difference. But accuracy with APS is really good. Never had problem with it.
APS does not simply MOS models. It works with the same models as spectre, but it is written in a way it utilizes multi core processors efficiently.
ultrasim from cadence is a table simulator which is used for chip-levels sims. when u try to use it on transistor level it adjust models to run quicker.
as Mr.Teddy said, source and drain resistance are minuscule in sub-micron technologies. if u are so worried add 1-2ohm resistor in ur source and drains and simulate.