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

Teo Ene

11/19/2021, 11:43 PM
@User It is a not-significant amount of effort to make 9-track versions as well. I see tasks related to this listed on our Jira. I would say it would be only a week and a half of focused work to get 9-track cells all the way through design, testing, and characterization in select target corners. I am not sure about a 7.4-track version matching hd; I don't see any internal documentation about its feasibility, so it is unknown how difficult it would be to create.
m

mehdi

11/19/2021, 11:45 PM
I actually have just a few aux. cells (a handful) that I would need to make larger which is even better! Just curious do you have the FO4 delay of all these libs?
t

Teo Ene

11/19/2021, 11:48 PM
I have characterization datasheets (like this HTML file: https://foss-eda-tools.googlesource.com/skywater-pdk/libs/sky130_osu_sc_t18/+/refs/heads/mai[…]18T_ms_TT_1P8_25C.ccs/SKY130_OSU_SC_18T_MS__INVx.html ), but I don't have FO4 off-hand. I can run sims over the weekend to get those specs.
🙌 2
t

Tim Edwards

11/23/2021, 1:32 AM
My observation is that it is a lot of trouble to use the HD cells, was a lot of work to get the OpenROAD tools to insert the tap cells properly, is a constant pain with people not used to the tools not knowing that they need tap cells, or putting the tap cells at the wrong distance. . . And in the end, the design has to be padded with 60% fill to get it to route. So what was the point of the high density? Just causes a lot of pain and trouble.
🎯 1
💯 1
m

mehdi

12/02/2021, 1:03 AM
@User Hi! were you able to simulate the FO4 delay and compare?
t

Teo Ene

12/02/2021, 5:38 PM
Sorry @User, last week of the semester has been crazy. I did a FO4 set-up with just the x1 drive cells from each library, The SkW cells, I connected their VNB and VPB pins directly to the rails, which may be a little unrealistic, but probably not too much. This is the table I got out: Cell library Averaged FO4 delay (ns) sky130_fd_sc_hd 0.03457 sky130_fd_sc_hs 0.031975 sky130_osu_sc_18T_hs 0.051365 sky130_osu_sc_18T_ms 0.05408 sky130_osu_sc_18T_ls 0.055905 sky130_osu_sc_15T_hs 0.055215 sky130_osu_sc_15T_ms 0.056715 sky130_osu_sc_15T_ls 0.05753 sky130_osu_sc_12T_hs 0.05413 sky130_osu_sc_12T_ms 0.057105 sky130_osu_sc_12T_ls 0.059195
Averaged refers to the fact that I just averaged fall delay one one clock edge and rise delay on the next
I just realized, one big difference is that the OSU ones were run with PEX, the scs8 ones were run without. My mistake. Let me re-run and see if that changes anything.
m

mehdi

12/03/2021, 3:07 PM
Thanks! @User
The SkW cells, I connected their VNB and VPB pins directly to the rails, which may be a little unrealistic, but probably not too much.
They need to be right? you can't leave them hanging.
I just realized, one big difference is that the OSU ones were run with PEX, the scs8 ones were run without. My mistake. Let me re-run and see if that changes anything.
Thanks! I am curious to see the difference you see after running with PEX. Thanks a lot for taking the time to do this!
t

Teo Ene

12/03/2021, 3:21 PM
They need to be connected, yeah. I just imagine that simulating the effect of the tap cells, with some wire delay artificially added, may be more effective. I'll run the PEX alternatives after I get out of my current class. Sorry that I got so wrapped up this week. Project deadlines.
💪 1
m

mehdi

12/03/2021, 3:23 PM
sounds good! and good luck with that! You just need to connect VNB/VPB on the extracted spice to vdd/gnd. No tapecell is needed for the fo4 delay 🙂
t

Teo Ene

12/03/2021, 3:24 PM
Yeah, that's the way I have it done.
1
@User See the following: Cell library Averaged FO4 delay (ns) sky130_fd_sc_hd PEX 0.046175 sky130_fd_sc_hs PEX 0.046595 sky130_osu_sc_18T_hs 0.051365 sky130_osu_sc_18T_ms 0.05408 sky130_osu_sc_18T_ls 0.055905 sky130_osu_sc_15T_hs 0.055215 sky130_osu_sc_15T_ms 0.056715 sky130_osu_sc_15T_ls 0.05753 sky130_osu_sc_12T_hs 0.05413 sky130_osu_sc_12T_ms 0.057105 sky130_osu_sc_12T_ls 0.059195
🙌 1
m

mehdi

12/03/2021, 5:04 PM
can you share more details on the simulator etc.. and it is weird that hd is tiny bit faster than hs
t

Teo Ene

12/03/2021, 7:12 PM
Absolutely! Let me private message you the SPICE decks and the spreadsheet.