[Libre-silicon-devel] Design rules

[David Lanzendörfer]

Hi Hagen
I just had the enlightenment on how to fix it.

Thanks for your input! It's exactly what I realized myself as well, some 
moments ago :D

-lev

On Thursday, 1 August 2019 4:24:45 PM HKT Hagen SANKOWSKI wrote:
> Hello David.
> 
> On 8/1/19 7:42 AM, David Lanzendörfer wrote:
> > Hi everyone
> > Magic uses the MOSIS design rules by default (on which we've built our
> > rules). The minimum width and length for contacts (including poly
> > contact) is 4 lambda (2 microns side length).
> > So either I hack Magic in a way that it requires less height/length for
> > the
> > vias of LibreSilicon or we try to use 2 microns side length.
> 
> Yes, in Magic the Tool like to gets rectangles which are at least 4 by 4
> lambdas (2 by 2 micron)
> 
> I think, we are free to export them as 2 by 2 lambdas (1 by 1 micron),
> 'cause we also like to get the overlapping area.
> 
> I ascii art a contact on metal1 would would look like
> 
> --------+_______+-------
> 
>  met1   | cont  | met1
> 
>         |_______|
> 
> --------+       +-------
>          2u by 2u
> 
> The contact itself would 'break', or 'interrupt' the metal1 stripe. As
> long as the contact is developed fine, no problem. But
> 
> 
> --------+-------+-------
> 
>         :  ___  :
>  met1   : |   | : met1
> 
>         : |___|	:
> --------+-------+-------
>          1u by 1u
> 
> exporting the contact hole smaller than the metal stripe ensures that
> the metal1 stripe is still conductive even if the contact itself is broken.
> 
> Same, when contacts are longer, as in the following ascii art.
> 
> --------+-------+-------+-------
> 
>         :  ___  :  ___  :
>  met1   : |   | : |   | : met1
> 
>         : |___|	: |___| :
> --------+-------+-------+-------
>          2 times 1u by 1u
> 
> This contact would have a size of 8 lambdas (4 micron) length and the 4
> lambdas (2 micron) width in Magic. In reality, after exporting them from
> Magic, there are two holes with 1 by 1 micron each on our masks.
> 
> So my suggestion is, we use the 4 by 4 lambda rule for Magic and
> exporting them (with help of our scripts) by
> - splitting up longer / wider holes into 4 by 4 lambda areas
> - leave 1 lambda margin around contact holes
> - every contact hole has to be 2 by 2 lambdas (1 by 1 micron) to satisfy
> etching issues
> 
> BTW, this design pattern of dozens of smaller contact holes instead of
> one bigger one is quite common in industry. Now, with our etching
> issues, I figured out why :-)
> 
> Regards,
> Hagen.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 195 bytes
Desc: This is a digitally signed message part.
URL: <http://list.libresilicon.com/pipermail/libresilicon-developers/attachments/20190801/e5535a73/attachment.sig>