electronrancher
Advanced Member level 1
well install and general debugging have their threads, so i think this topic needs a different one to keep everything focused.
i got simulation working tonight with spectreS on the NCSU models of the AMI 1.6u process. no replies to my other post about how to get the models running, so i guess that most others have these troubles. first my howto.
1) New library "test" made with lib manager, attached to AMI 1.6u tech library since NCSU CDK is up & running. This process is the cheapest at MOSIS, so I will probably send my "hobby" designs here once this is all running smoothly..
2) New cellview "nsweep" created, and schematic created to virtuoso. Added one nmos from the NCSU Lib (default was 4um/1.6um). Added 2 dc sources, one for drain (called Vcc) and one for gate (Vgate). Threw in the GND arrow and Vcc bar, and wired it all up. This is just a simple test bench to characterize a FET.
3) Saved schematic and ran Analog Env. Changed simulator from hspice (NCSU default) to spectreS (likes the models), since I dont have hspice yet.. Added a new "output to be plotted", and chose the drain of the nmos. This will give us current into the drain. Choosing the wire will give you voltage at that node. Set parameters of Vcc supply to be DC=>5 and Vgate supply to be DC=>2.5 .
4) Analysis options lets us choose dc sweep, and we decide to sweep Vcc. (choose the supply from schematic, just click & go). Sweep from 0-5v while our gate stays at 2.5v. You all know this plot of Idd/Vdd I'm sure.
5) NOW ADD MODELS FROM NCSU KIT! This was my old mistake, I assumed IC5 is smart enough to link from TechLib. Nope!
"Setup model path" from menu, and you can delete the two generic paths it gives you - mine were empty so it complained about it. Now add the NCSU kit from /local/models/spectre/nom. This means the nominal process run, an average device.
Note:You can also choose fast (hot transistors, thin gates or heavy doping makes k' big) or slow (the opposite, low k') directories to see how your IC works over all process corners.
5) Run the sim. I see about 141uA flat when i run transient analysis from 0-15ns, Vg=2.5, Vd=5. Not bad for a little 4u/1.6u nmos. The DC sweep we were setting up before gives a nice little plot so you can see the classic mosfet sweep. Look at the "flat" part (saturation) to estimate output resistance and, well you know your physics already i guess..
Again, looks pretty nice, I like AMI's process so far.
Now for the bad part. Sweeping 1 NMOS took 9 seconds on an athlon 1800 with 1/2GB ram and 2GB swap! Setting VGate to a variable, and running paramtric sweep to give me the whole set of nmos curves took almost 10 minutes for 50 sweeps!! This is garbage in my opinion!
Has anyone else gotten results like this? How will I simulate top-level 5k transistor designs? The spectre manual suggests 256MB ram for 1M transistor simulations, or something very impressive - I didn't think they meant it would take 3 days!
I am wondering how long it takes to sweep the NCSU nmos on everyone's system. I am using SuSe 8.2 and KDE. I closed the schematic window and only had open - icms, Analog Environment, and... that's all i think
I would like your feedback, everyone - I will take some more experiments and let you know if i am mistaken, and spectre is not garbage. Maybe i'll try hspice, but I used that one before and decided it pretty slow a long time ago.
i got simulation working tonight with spectreS on the NCSU models of the AMI 1.6u process. no replies to my other post about how to get the models running, so i guess that most others have these troubles. first my howto.
1) New library "test" made with lib manager, attached to AMI 1.6u tech library since NCSU CDK is up & running. This process is the cheapest at MOSIS, so I will probably send my "hobby" designs here once this is all running smoothly..
2) New cellview "nsweep" created, and schematic created to virtuoso. Added one nmos from the NCSU Lib (default was 4um/1.6um). Added 2 dc sources, one for drain (called Vcc) and one for gate (Vgate). Threw in the GND arrow and Vcc bar, and wired it all up. This is just a simple test bench to characterize a FET.
3) Saved schematic and ran Analog Env. Changed simulator from hspice (NCSU default) to spectreS (likes the models), since I dont have hspice yet.. Added a new "output to be plotted", and chose the drain of the nmos. This will give us current into the drain. Choosing the wire will give you voltage at that node. Set parameters of Vcc supply to be DC=>5 and Vgate supply to be DC=>2.5 .
4) Analysis options lets us choose dc sweep, and we decide to sweep Vcc. (choose the supply from schematic, just click & go). Sweep from 0-5v while our gate stays at 2.5v. You all know this plot of Idd/Vdd I'm sure.
5) NOW ADD MODELS FROM NCSU KIT! This was my old mistake, I assumed IC5 is smart enough to link from TechLib. Nope!
"Setup model path" from menu, and you can delete the two generic paths it gives you - mine were empty so it complained about it. Now add the NCSU kit from /local/models/spectre/nom. This means the nominal process run, an average device.
Note:You can also choose fast (hot transistors, thin gates or heavy doping makes k' big) or slow (the opposite, low k') directories to see how your IC works over all process corners.
5) Run the sim. I see about 141uA flat when i run transient analysis from 0-15ns, Vg=2.5, Vd=5. Not bad for a little 4u/1.6u nmos. The DC sweep we were setting up before gives a nice little plot so you can see the classic mosfet sweep. Look at the "flat" part (saturation) to estimate output resistance and, well you know your physics already i guess..
Again, looks pretty nice, I like AMI's process so far.
Now for the bad part. Sweeping 1 NMOS took 9 seconds on an athlon 1800 with 1/2GB ram and 2GB swap! Setting VGate to a variable, and running paramtric sweep to give me the whole set of nmos curves took almost 10 minutes for 50 sweeps!! This is garbage in my opinion!
Has anyone else gotten results like this? How will I simulate top-level 5k transistor designs? The spectre manual suggests 256MB ram for 1M transistor simulations, or something very impressive - I didn't think they meant it would take 3 days!
I am wondering how long it takes to sweep the NCSU nmos on everyone's system. I am using SuSe 8.2 and KDE. I closed the schematic window and only had open - icms, Analog Environment, and... that's all i think
I would like your feedback, everyone - I will take some more experiments and let you know if i am mistaken, and spectre is not garbage. Maybe i'll try hspice, but I used that one before and decided it pretty slow a long time ago.
