quantized
Member level 2
- Joined
- Jul 6, 2012
- Messages
- 51
- Helped
- 1
- Reputation
- 2
- Reaction score
- 1
- Trophy points
- 1,288
- Activity points
- 1,887
Hi, this isn't a question but rather an answer -- I'm posting it here because it took me a long time to find it, it's almost completely undocumented, and google is totally unhelpful. Since a lot of google searches lead to edaboard this sounded like a good place to post it.
I had a circuit that works correctly from schematics in HSIM, passes LVS, and also works correctly when simulating from the RCX-generated SPICE deck (i.e. just using the RCX output .spi as direct input to HSIM).
HSIM has a better way to interact with RC extraction tools: it can use your schematic SPICE netlist and "copy over" the resistors and capacitors from an SPF-annotated RCX netlist. The huge advantage here is that the hierarchy doesn't get flattened at all. The other big advantage is that RCX tools strip out any SPICE directives that aren't standard or which deal with hierarchy, and this avoids that. You definitely want to use this mode if you can.
The problem I encountered is that for some weird reason when I used the SPF-style back annotation rather than direct RCX input, my circuit's behavior changed wildly and I would see, in the waveform viewer, what looked like huge amounts of ground bounce. My first instinct was to assume my layout was bad (insufficiently wide power/ground delivery lines, etc), but if that were the case I would have seen the same problems using the RCX netlist directly. What was weird was that the problem only occurred when reading the RCX netlist in SPF mode, but didn't occur when reading it in SPICE mode.
Anyways, after struggling for a long long time and trying a lot of random stuff, I came across "HSIMPOSTLMANY2M=1". There are only two sentences in the manual about it, and they're extremely vague, saying only that it enables a "more conservative reduction algorithm". FWIW 95% of the HSIM options (there are hundreds of them) deal with varying levels of conservativeness, so this description is totally unhelpful…
But it fixed the problem, right quick.
So, I definitely recommend trying this.
I had a circuit that works correctly from schematics in HSIM, passes LVS, and also works correctly when simulating from the RCX-generated SPICE deck (i.e. just using the RCX output .spi as direct input to HSIM).
HSIM has a better way to interact with RC extraction tools: it can use your schematic SPICE netlist and "copy over" the resistors and capacitors from an SPF-annotated RCX netlist. The huge advantage here is that the hierarchy doesn't get flattened at all. The other big advantage is that RCX tools strip out any SPICE directives that aren't standard or which deal with hierarchy, and this avoids that. You definitely want to use this mode if you can.
The problem I encountered is that for some weird reason when I used the SPF-style back annotation rather than direct RCX input, my circuit's behavior changed wildly and I would see, in the waveform viewer, what looked like huge amounts of ground bounce. My first instinct was to assume my layout was bad (insufficiently wide power/ground delivery lines, etc), but if that were the case I would have seen the same problems using the RCX netlist directly. What was weird was that the problem only occurred when reading the RCX netlist in SPF mode, but didn't occur when reading it in SPICE mode.
Anyways, after struggling for a long long time and trying a lot of random stuff, I came across "HSIMPOSTLMANY2M=1". There are only two sentences in the manual about it, and they're extremely vague, saying only that it enables a "more conservative reduction algorithm". FWIW 95% of the HSIM options (there are hundreds of them) deal with varying levels of conservativeness, so this description is totally unhelpful…
But it fixed the problem, right quick.
So, I definitely recommend trying this.