GDS Data Base

We now have the GDS data base.  The first step is to determine which layers are which.  A Freescale engineer provided an old document (though newer than the data base) which gave the Freescale layer numbers.  Using it, we were able to identify the significant layers.  The data base is a 3u layout, which was optically shrunk to 1u dimensions for fabrication.  It has a single metal layer, and a single metal poly layer.  It also has buried contacts and n-channel depletion devices.

For those under 30, a buried contact was a way of attaching poly directly to diffusion without using contacts or metal.  It involved the stripping of gate oxide and a subsequent implant prior to poly deposition.  It was a good way to increase interconnect density prior to the development of multilayer metals.  Buried contacts also mean that I would have a terrible time finding a fab to make the data base, should I want to proceed in that direction.

Next, I wrote a technology table for our LVS program.  This allows the software to extract a transistor level netlist from the layout.  There are a number of LVS programs out there, running from free to very expensive.  We use a public domain program from ICED that works quite well. 
 
Our first LVS run ran in a few minutes, and extracted about 160K transistors.  It also had about 300 devices that it didn't like.  It is really important that every transistor be accounted for, and so I had to determine what made a bad transistor bad.

In the layout, large clock drivers had been created using a grid-shaped gate.  Every other square in the grid was either a source or a drain.  This works, but the transistor at the corners of the grid is an odd structure that is hard to define in terms of traditional source/drains.  I resolved this by adding a CAD layer that "cut" the diffusion from that spot.  This left a structure that could be deciphered by LVS, and took care of most of my errors.

The remaining errors turned out to be depletion mode capacitors.  I modified my technology table to include them.  The next run produced a netlist, which when fed back into the LVS run produced a perfect match. A perfect match with an unusable netlist.

Date

2011-07-27