[comp.lsi] Problems with magic reading scmos labels

davis@unc.UUCP (Mark Davis) (01/19/87)

We at UNC-CH are using magic (from Berkeley) as part of a silicon
compiler project.  We have encountered a problem with the way that
magic handles labels when reading scmos CIF.  Magic has trouble
assigning labels when a magic layer corresponds to more than one CIF
layer.  The standard cmos technology files for some time have
guarenteed that no labels will be assigned directly to ndiff.  Instead
they go to the pwell by default after being assigned to space.  This
makes it tough to label drains of transistors.

Does anyone have a technology file for scmos that properly handles
labels on ndiff?  Is this a problem with magic for which there is no
easy work around?

Thanks for your help - Mark Davis (davis@unc.csnet-relay)

wss@solaria..ARPA (Walter S. Scott) (01/21/87)

There are two solutions.  First, CIF "94" labels can have a final field
giving the CIF layer to which the label is attached.  You can use the
CIF layername for diffusion (e.g., CDIF) to force the labels to be
attached to {n,p}diffusion by Magic:

    94 Cr:A$ -22 80 CDIF;

Second, the order in which layers are picked for labels whose layers
AREN'T specified in the CIF file is the opposite from the order in which
those layers are defined in the technology file.  If you want {n,p}diff
to be preferable to pwell when both cover a label, define the diffusion
layers LATER in the "types" section of the technology file than pwell.

						Walter Scott