( ESNUG 465 Item 6 ) -------------------------------------------- [06/28/07]
Subject: ( ESNUG 464 #10 ) Yup, I see Synopsys DFTMax benchmarks well, too
> We recently came across what seems to be very encouraging results of a
> Synopsys DFTMax benchmarking exercise (12X scan pattern compression and
> an area overhead of less than 0.02%).
>
> We performed our tests on a single design, which may not yet allow for
> conclusive answers, but I am interested in knowing if other people have
> tried DFTMax yet, and if they are seeing good results, too.
>
> - Erik Comparini
> AMI Semiconductor Bangalore, India
From: Mark Wroblewski <mw1=user domain=agere bot calm>
Hi, John,
We are using DFTMax on a project right now. We've set it up to get ~19X
compression, with an area overhead similar to what Erik's seen.
Two things we've found that need fixing:
a) SCANDEF files for hierarchical scan
DFT Compiler does a nice job working with lower-level blocks that are
scan stitched, and brought into a compression network at the top level.
However, the SCANDEF files it writes for now are only available at the
lower-level blocks. This can cause issues if you want to lay out a
section of the design that includes more than one of the lower-level
blocks. You need to combine the SCANDEF files, and adjust for
hierarchy.
The main issue we saw here is START PIN and STOP PIN statements which
refer to ports on the lower-level blocks. When you adjust hierarchy,
those ports are no longer valid, and ICC's SCANDEF reader coughs.
Synopsys R&D is working on a solution.
Once the SCANDEF files get into ICC, they seem to work well to drive
the cell swapping and other scan optimizations.
b) scan wrapper combined with scan compression
This is highly desired for our design, but the tools don't support
this natively today. For now, we work around this by creating the
scan wrapper completely outside the scan compression core. However,
a paper at SNUG San Jose 2007 describes one way to make this happen
using the current crop of tools, and I've heard that Synopsys R&D
are working to see if they can bring the native support for this to
the tools by late 2007.
Overall, the tools (DFT Compiler, TetraMax, and ICC for most of the physical
work) have worked very well together to craft the compression logic without
burdening our mission-mode logic. That is, they pretty much automatically
get us to within 2% of completing the work required -- it's closing out on
the last 2% that keeps me off the streets at night. ;-)
- Mark Wroblewski
Agere Systems Longmont, CO
Index
Next->Item
|
|
