I often parameter mismatches when doing LVS with Assura under Linux (Fedora core 2). It keeps reporting parameter mismatch errors of a very tiny percent (like 1e-7 percent). I don't think it's a problem of my layout. Instead, it should be the problem of Assura handling floating-point number incorrectly under linux.

Any ideas?
Thanks.

I see your point. But the same version of cadence works all right under unix without this kind of error. Also we have never met this kind of error before in Diva.

[quote author=Andrew Beckett  link=1108324028/0#1 date=1108359273]Most rule decks would do LVS comparison with a tolerance to avoid this sort of thing (not just for Assura, but for all physical verification tools). Floating point rounding errors are inevitable with IEEE floating point arithmetic, and so doing exact comparison of calculations done from different source data is asking for trouble.

Andrew.
[/quote]

I guess the right way to do is to modify the rule to add a little tolerance. I checked the rule and it seems I have to modify many lines, like resistance tolerance, mos tolerance, capacitor tolerance, etc. Is there any general tolerance which controls all the devices?

[quote author=Andrew Beckett  link=1108324028/0#3 date=1108401131]Either way, I wouldn't consider that a tool bug. There are quite likely to be differences in floating point calculations between different hardware platforms. By luck you've not hit this problem before (probably because all the testing was done on another platform), but it is dangerous to rely on exact matching of floating point numbers, IMHO.

I've not seen any reports of similar problems with Assura, so I doubt it's a tool bug.

Regards,

Andrew.
[/quote]