Hi All. I'm running openlane2 on a basic 4b counter and have encountered an error. In dir ~/mydesigns/counter_4b/, I have created the attached config.json and counter_4b.v files, as per the below link: https://openlane2.readthedocs.io/en/latest/getting_started/newcomers/index.html#configuration The flow errors out with the following: --------------------------------------------------------------------------- [065447] ERROR The following error was encountered while running the flow: main.py:163 OpenROAD.GeneratePDN failed with the following errors: [PDN-0185] Insufficient width (18.86 um) to add straps on layer met4 in grid "stdcell_grid" with total strap width 4.9 um and offset 16.32 um. [065447] ERROR OpenLane will now quit. --------------------------------------------------------------------------- Being a newbie to openlane Im not so good at interpreting these errors yet. It appears to have errored out at step 20 - "20-openroad-generatepdn". Any ideas what it is and how to overcome it? Thanks, Diarmuid

Yeah, basically your module is too small and there isn't space to put power rails on it.

@Diarmuid Collins you might try adding

"FP_SIZING": "absolute",
"DIE_AREA": "0 0 300 300",

Thanks @tnt @Mitch Bailey. So that brings me to my next error: ERROR [GPL-0307] RePlAce divergence detected. Re-run with a smaller max_phi_cof value. What is "max_phi_cof"? Also, w.r.t my previous question - how does "DIE_AREA": "0 0 300 300" read? Is x = 300um = y? Thanks,

I believe the

DIE_AREA

is lower-left-x, lower-left-y, upper-right-x, upper-right-y. I don’t know about max_phi_cof. You might check the expanded

config.tcl

file in the

runs/timestamp

directory.

Thanks @Mitch Bailey. Regarding the DIE_AREA, so is the die area for the Efabless openframe = 300x300um2? Regarding my error, when I look in my runs lib I don't see a timestamp lib, as shown. However I went in to the last step (#23) and looked at the config.json file in that. I found the shown lines with MAX_PHI_CO in them. Don't know if that helps with any debug. Be great if it did!

@Diarmuid Collins the

DIE_AREA

for

user_project_wrapper

is fixed. You can find it in

caravel_user_project/openlane/user_project_wrapper/config.json

. If you’re just creating and arbitrary cell, then you can set the

DIE_AREA

to whatever you want as long as it’s tall/wide enough to have a power grid.

@Mitch Bailey: I am using openlane2 for SW130. Is there some resource for me to dig a bit on this or how does the community go about debugging openlane error messages in general?

That error is because the area is too big for the amount of logic you have and it messes with the placement algorithm. Try setting PL_TARGET_DENSITY to say 30 instead of letting it pick automatically.

@donn This is an openlane2 question. I couldn’t find a default value in any

config.tcl

file or documentation for

PL_MAX_PHI_COEFFICIENT

. Could you point us in the right direction?

See https://openroad.readthedocs.io/en/latest/main/src/gpl/README.html

Thanks @tnt! @Diarmuid Collins what about adding the defaults to your local

config.json

file? (from the link above)

-min_phi_coef

Set pcof_min (
 Lower Bound). The default value is 0.95. Allowed values are floats [0.95, 1.05].

-max_phi_coef

Set pcof_max (
 Upper Bound). Default value is 1.05. Allowed values are [1.00-1.20, float].

like this

"PL_MIN_PHI_COEFFICIENT": 0.95,
  "PL_MAX_PHI_COEFFICIENT": 1.05,

I still think setting an explicit place density is a better fix. IIRC when you don't set one, it will basically take used_area / available_area but if you have very little area used and forced a large die area, it will be a very low place density. It's trying to spread the logic over the whole available area but that's not really what you want. here, it'd be better to have a small cluster with all the logic and the rest of the die empty.

@Mitch Bailey https://openlane2.readthedocs.io/en/latest/reference/step_config_vars.html#var-openroad-globalplacement-pl-max-phi-coefficient

@donn Thanks for the update. The documentation that @tnt referenced seemed to indicate a default, but it didn’t look like that was getting set (or maybe “none” overrides the default.). Do you think this should be added to the pdk config, or that the placement program should be updated to substitute the default value for “none”, or that openlane should not set a “none” value to pass to the placer?

@Mitch Bailey

None

doesn't pass the flag to OpenROAD from OpenLane, which uses the OpenROAD default. Maybe this should be made clearer in the documentation.

Thanks all for this amazing support! I'll try out the above and report back tomorrow. Cheers.

The color difference on the above pics is more likely due to selected layer than anything else. To see logic density, looking only at poly or met1 is a bit better.

Thanks @tnt. Attached is the layout setting target density at 0.1 (which I presume means 10%). I show poly/M1/2/3.

Heh 🤷 There is so little logic that I doubt this make any difference.

Sorry, I can’t give any input on the conclusions. As for opening a specific run, would changing

--run-RUN

to

--run=RUN

make a difference or is that just a typo above.

@Mitch Bailey Yeah, tried --run=RUN but again no luck. Also tried specifying the config.json file of the final run with the below: openlane --flow openinklayout ~/mydesigns/counter_4b/runs/RUN_2024-11-02_12-01-54/75-klayout-opengui/config.json --manual-pdk That didnt work either. It would be useful to know how since if I want to compare 2 layouts it avoids me always having to run the flow (to enable me to open the "last" run). All the examples I see (e.g. https://openlane2.readthedocs.io/en/latest/usage/caravel/index.html) use --last-run.

@Diarmuid Collins @tnt Hey. I am getting the same issue with max_cof_value. i have set the PL_TARGET_DENSITY to 0.08 and if i set it below it the floe fails. and also reduced the value for max_phi_cof. but it still gives the same error. I am not sure why.

use a higher target density

but for him u said use smaller target density?