https://open-source-silicon.dev logo
#analog-design
Title
# analog-design
t

Tim Edwards

05/06/2022, 12:45 PM
<!channel>: To everyone: I'm going to try again today to get Tim Ansell to merge the continuous device models into the SkyWater PDK. I already sent him a diff file (last week), after having spent the better part of a month carefully integrating the new models into the open PDK and vetting them with both ngspice and Xyce. The continuous models replace the discrete BSIM models with their hundreds of bins, and ngspice loads them a lot faster. And, of course, the continuous models are continuous over W and L, so they clear up the problems with simulating devices with W and L outside of the narrow characterized specs. Needless to say, Tim Ansell is a digital guy and does not appreciate how desperately we analog designers need those models. Do not hesitate to drop him a line, and be vocal about it, as that will help ensure that Tim gets it done in a timely manner. I am literally one
git push
away from delivering the continuous models to everyone through open_pdks. But SkyWater did not give me, specifically, permission to post the models, they gave Google permission to post the models, so only Tim Ansell can initiate the process.
Also in the next push of open_pdks (along with a corresponding update to magic), I have sorted through all of the layout-specific device models (mainly the vertical parallel plate capacitors and the RF MOSFETs) and annotated the layouts in magic so that they will extract the correct layout-specific device model. Note that the new continuous models do not have any RF device models, so the RF devices will still be discrete, and tied to the specific RF layouts in the PDK.
👍 1
🌍 1
a

Arman Avetisyan

05/06/2022, 12:56 PM
Can confirm that reduction of runtime of ngspice is super important
💯 2
👍 3
t

Tim Edwards

05/06/2022, 1:00 PM
Not having device models that have parameter discontinuities across bin boundaries is also super important.
👍 3
h

Harald Pretl

05/06/2022, 1:08 PM
@User please provide us the continuous SPICE models, they will be a BIG help!
👍 6
@User we appreciate the support of the RF models. We will use them in a 7GHz UWB Transmitter to be designed in the next months.
p

Philipp Gühring

05/06/2022, 1:33 PM
@User The continuous models are badly needed, even for the standard cell characterization that is then used by digital design
l

Leonardo Gomes

05/06/2022, 1:35 PM
@User please, provide us the continuous MOSFET models. That would improve A LOT the design flow and accuracy of our designs
(Also, having continuous RF MOSFET models would be nice at some point in the near future. Or at least some way to choose a high-frequency accurate model while extracting the netlist)
r

Ryan Wans

05/06/2022, 2:51 PM
@User hi
😀 3
k

Keno Fischer

05/06/2022, 2:57 PM
👍 From me on this being hugely helpful. I've been doing a lot of work working around the lack of continuity, so this will help enormously
👍 1
k

Kunal

05/07/2022, 5:56 AM
@User - you might want to follow this thread
s

SUMANTO KAR

05/07/2022, 6:20 AM
@User thanks for adding me here. It looks like it can solve our problem if we can have continuous models.
p

Pepijn de Vos

05/07/2022, 8:40 AM
Pretty pretty please :))))
No doubt every software developer or designer knows that a short feedback loop greatly speeds up design. This change could bring the feedback loop of lets say the operating point of a small analog design from a minute of just loading the models, to a few seconds. Game changer.
💯 1
👍 1
h

Harald Pretl

05/20/2022, 7:55 AM
@Tim Edwards @Tim 'mithro' Ansell Just wanted to checkin with you both to see how the continuous model release is going. Needless to say, this release is highly anticipated by the analog design community… 🙂
c

Chris

05/18/2023, 7:15 PM
Would like to know the progress of this so far too 🙂
t

Tim Edwards

05/26/2023, 11:01 PM
@Harald Pretl: It currently exists in a branch created by @proppy so in principle it can be used right now. I had been waiting for a merge into the main branch before pushing the code I have in open_pdks to integrate the continuous models into the installed PDK. The part open_pdks does is to combine the set of continuous models, which is not a complete set, with the original models of the rest of the devices, then create a new sky130.lib.spice file that has
.include
statements pointing variously to the new continuous models or to the old models, depending on the device. In theory, I could have open_pdks merge the branch and work from there. But so far, everyone who has investigated the continuous models isn't finding any significant different from the original models. The same underlying characterization data set was used to create the continuous models, so the quality of the new models depends entirely on the software used to do the conversion and whether it is smoothing parameters in a meaningful way.
👍 1
h

Harald Pretl

05/27/2023, 8:44 AM
@Tim Edwards Even if the models are technically not better, the model files would be a lot smaller, and sweeping MOSFET
W
and
L
would likely be a bit better, as not traversing model bins this way. So I would prefer to move to continuous models in SKY130, but I am not strongly pushing. I would like to move our analog/mixed-signal development anyway to IHP’s SG13G2 as soon as I can 🙂
p

proppy

05/31/2023, 6:40 AM
@Harald Pretl any luck with SG13G2 ngspice simulation? I bumped into the following issue: https://github.com/IHP-GmbH/IHP-Open-PDK/issues/3
h

Harald Pretl

05/31/2023, 6:47 AM
I haven’t tried to work with SG13G2 yet, I am waiting until a bit more pieces of the PDK are released by IHP.