I've been able to improve the litho simulation of ...
# sky130m
Matthew Guthaus
12/14/2021, 10:36 PMI've been able to improve the litho simulation of the SRAM cells. My initial results were not actually aerial intensity and resist. They were actually just an smoothing of the mask from one of their papers. This is actually modeling the light and resist of the dual port bitcell:
❤️ 3
m
Matt Liberty
12/14/2021, 11:01 PMCould you summarize the process to produce these results?
m
Matthew Guthaus
12/14/2021, 11:03 PMNot right now!
m
Matt Liberty
12/14/2021, 11:03 PMNo rush, just interested
Matt Liberty
12/14/2021, 11:04 PMI agree it is pretty cool!
m
Matthew Guthaus
12/14/2021, 11:04 PMEventually, I will. I'm still getting some unexpected results. The layout with OPC is worse than the layout without OPC, for one!
Matthew Guthaus
12/14/2021, 11:05 PMThe mirroring is wrong on that one too. It assumes that the same cell is tiled like that, but in reality, it is flipped in X and Y each row/column
Matthew Guthaus
12/14/2021, 11:07 PMThis one is with the single port bitcell with surrounding cells to get rid of that problem. I've also played with the focus and dose to see how it affects the printing.
This shows the mask, aerial image, resist exposed and an "error plot" of the difference between the resist and mask.
Matthew Guthaus
12/14/2021, 11:08 PMI see almost no difference between the two focuses
Matthew Guthaus
12/14/2021, 11:09 PMsky130_bitcell_array_opc_poly_f-50_d1.png
m
Matt Liberty
12/14/2021, 11:11 PMhow long does it take to generate a result like that?
m
Matthew Guthaus
12/14/2021, 11:12 PMA few seconds since it is pretty small
Matthew Guthaus
12/14/2021, 11:12 PMIf I had a bigger array, it ran out of memory during the SVD computation
Matthew Guthaus
12/14/2021, 11:13 PMI think that the pyfftw library uses some multithreading. I see ~12 threads fire up. I was suspecting that might be related to the memory usage, but I just shrunk the example instead. (And those likely shared the same memory image)
Matthew Guthaus
12/14/2021, 11:18 PMThe biggest challenge was getting the scale right. The original GDS parser ignored the gds scaling. Then there is a "pixelization" step in the litho tool. Together, I think there was a bug in the original code where sizes were actually off 10x...
a
Amro Tork
12/15/2021, 6:06 AM@User That looks great. Which tool you use for doing Lithography simulation?
m
Matthew Guthaus
12/15/2021, 6:22 AM@User Dimmiolitho with plenty of patches
❤️ 1
💯 1
a
Amro Tork
12/15/2021, 7:37 AMThanks @User
💯 1
Amro Tork
12/15/2021, 7:37 AMCould you please share a link?
m
Matthew Guthaus
12/15/2021, 2:19 PMhttps://github.com/mguthaus/DimmiLitho
But not all the changes are there yet
a
Amro Tork
12/15/2021, 5:27 PM@User Does that simulate the proper lighting?
m
Matthew Guthaus
12/15/2021, 5:27 PM@User what do you mean "proper"?
a
Amro Tork
12/15/2021, 5:28 PMI mean we don't know which lighting system is used by skywaters in there Litho process
Amro Tork
12/15/2021, 5:28 PMWavelength and configuration
Amro Tork
12/15/2021, 5:28 PMI would assume it's 192 nm
Amro Tork
12/15/2021, 5:28 PMBut for the configuration I don't know
Amro Tork
12/15/2021, 5:29 PM@User Any chance you could push the updates to a public repo?
m
Matthew Guthaus
12/15/2021, 5:29 PMThat is all configurable. I'm not sure about some of the parameters for the resist and lens for example. Yes, it's 192nm.
Matthew Guthaus
12/15/2021, 5:29 PMI'll push when it's working. It's in flux right now.
a
Amro Tork
12/15/2021, 5:29 PMThanks @User
m
Matthew Guthaus
12/15/2021, 5:30 PMThat repo is public already
a
Amro Tork
12/15/2021, 5:30 PMNo but I mean your changes. Yours looks nicer
m
Matthew Guthaus
12/15/2021, 5:31 PMah, yeah. I had to write a new top-level script. DimmiLitho is more of a set of tools for inverse lithography but we're just using the simulation stuff
a
Amro Tork
12/15/2021, 5:31 PMDo we also know if Skywaters apply OPC or any kind of RET ?
m
Matthew Guthaus
12/15/2021, 5:31 PMI also had to write a script to use the add/drop OPC layers from Skywater
Matthew Guthaus
12/15/2021, 5:31 PMIt's already in the SRAM cells
a
Amro Tork
12/15/2021, 5:31 PMNice
Amro Tork
12/15/2021, 5:32 PMThanks Dr. @User
Amro Tork
12/15/2021, 5:34 PMWe might need to apply that for all other designs to improve yield
Amro Tork
12/15/2021, 5:34 PMNot just SRAM
m
Matthew Guthaus
12/15/2021, 5:34 PMWe don't have bit masks though so the ILT approach doesn't work. There would need to be some other way of generating the OPC
7 Views