Discrepancies between local OpenRAM results and Published

Chris Kjellqvist

unread,

Feb 21, 2022, 12:45:10 AM2/21/22

to openram-u...@ucsc.edu

Hi,

I’m using OpenRAM to look at some bank configurations and I’ve been having a bit of trouble. I’ve tried running some configurations published here (https://vlsida.github.io/OpenRAMLibrary/) and my local runs are not giving me the same results. I’m unsure what I could be doing wrong here.

My current git commit is f66aac3264598eeae31225c62b6a4af52412d407 and I’m using the exact configuration file used here (https://vlsida.github.io/OpenRAMLibrary/deliverables/freepdk45/2019-02-21/sram_16b_1024_1rw_freepdk45/sram_16b_1024_1rw_freepdk45.html) .

The published result says operating frequency is ~2GHz, I get ~800MHz – this is the most striking difference. Power numbers are also different by around 10%.

I’m unable to run the commit that the published result says it came from because it says I’m lacking DRC + LVC tools. I don’t get these warning in newer commits.

Any thoughts on why this may be?

Thanks,

Chris Kjellqvist

Matthew Guthaus

unread,

Feb 21, 2022, 1:33:19 AM2/21/22

to openram-u...@ucsc.edu

Hi Chris,

It's a mistake that that web site is still public. We made the repo private because we aren't updating it but evidently the GitHub pages are still there.

There have been many updates since then so I expect the results won't be the same. Most especially, those results are analytical models which haven't been accurately calibrated.

Matt

--
https://openram.soe.ucsc.edu
---
You received this message because you are subscribed to the Google Groups "OpenRAM User" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openram-user-gr...@ucsc.edu.
To view this discussion on the web visit https://groups.google.com/a/ucsc.edu/d/msgid/openram-user-group/SN6PR05MB531275F89FE1F270E363E856EB3A9%40SN6PR05MB5312.namprd05.prod.outlook.com.

Chris Kjellqvist

unread,

Feb 21, 2022, 6:58:42 AM2/21/22

to OpenRAM User, Matthew Guthaus

Ah I see. My original analysis has used CACTI and I've been been using results that I think indicate I should be able to access at a multiple GHz frequency (access times in around .1-.3ns with the .04um technology node) whereas OpenRAM says even the smallest models can hardly achieve over 1GHz operating frequency. Is this realistic? I've tried increasing the maximum voltage (supply_voltages=[3.3] for example) but the results are the same.

Chris

Matthew Guthaus

unread,

Feb 21, 2022, 9:20:29 AM2/21/22

to Chris Kjellqvist, OpenRAM User

CACTI aren't real memories, so I wouldn't believe those models.

Also, Freepdk45 isn't a real technology node either.

So, I'm not sure I would trust the numbers you are seeing...

Matt

Reply all

Reply to author

Forward