( ESNUG 477 Item 5 ) -------------------------------------------- [11/20/08]

From: [ Moon Unit Zappa ]
Subject: We switched from Hercules to Calibre for Equation-based DRCs

Hi, John,

I gotta be anon on this email.

We just switched from Hercules to Calibre because of it's equation-based
DRCs and its GUI.  Our chip has over 100 K gates of logic, plus multiple
memories and an array.

Before using Calibre, we used Hercules for over 10 years.  Two years ago,
Synopsys switched Hercules' graphical debugger from Explorer to VUE; we
could never understand why because it reduced functionality.   Explorer
wasn't fancy, but it was reliable. The change to VUE, plus the fact that we
were looking at a new process node, got us to look at Mentor Calibre.
Hercules was OK at 130 nm and we weren't really having problems, but we
were looking at moving to a 90 nm process.

We evaluated Calibre for over a year before purchasing it.  What I initially
noticed for LVS was that Calibre's RVE debugging GUI was better than
Synopsys' VUE.  We could probe results better with Calibre because it was
better connected with the Cadence Virtuoso.

Another big difference is that Calibre is edge-based, and Hercules is
polygon-based.  With more advanced DRC's you need to use the edges more
than the shape.  Although there is a way to use edges in Hercules, it is
not very usable -- for complex checks it takes 5X as many lines to create
proper rules in Hercules vs Calibre.

Calibre Equation-Based DRC (eqDRC):

The standard way to write DRCs is to create rules for a set of discreet
points.  With Calibre equation-based DRC's you can define those same rules
using an equation.  This equation based method is continuous (versus
discontinuous).  If you are using Calibre, eqDRC is just like another rule
type and doesn't change the design flow.

Let's say we are using DRCs to check a space.  The fab gives us spacing
criteria and polynomial type information:

   - With standard DRCs (not eqDRC), we must sample specific discreet
     points on the polynomial to test and write DRCs at those
     sampling points.  We can get finer and finer resolution by sampling
     more discreet points, but it takes more and more lines of code so
     no one can afford to do it.  We would tend to select 4 points; so
     our layout designers were forced to implement the design according
     to what fits in those buckets, which wastes a lot of space. 

   - With Calibre eqDRC we can just write an equation to define the shape.
     Our designers can do maximum optimization of the design so they don't
     waste real estate.

Manufacturing variability control.  eqDRC basically extends the life of
Calibre to deal with DFM issues before needing to go to a litho-type
checking tool such as Mentor's YieldAnalyzer, where you generate an aerial
image of design and then run DRCs. We were writing standard DRC to improve
the manufacturing variability especially on metal.  What killed us is that
we definitely had to be coarse on the steps (buckets) and the designers
complained a lot that is was too restrictive -- so we kept creating design
rule waivers for them.  For our next design we want to use equation-based
DRC so we can deal with DFM issues but won't need to over restrict design.

Types of equations.  We have used Calibre's mathematical equations primarily
for metal, as it is the most advanced level as far as processing for our
product.  We use edges and polygons as inputs -- Mentor can also deal with
edge cluster types, but we haven't used them.  Mentor has a set of commonly
used equations with predefined functions that you can invoke, but we haven't
looked at them either as we are still focused on is defining our own.

Error reporting.  A nice thing with Calibre's equation based DRC is that not
only are we checking the rule in the manner we would like, but we can also
report errors back in a way that pushes the design in the right direction.
Standard DRC might give us an error that says we failed the rule, for
example "you failed space" -- but with eqDRC it is clearer what we need to
do to rectify the error, adjust the line or the space.

Debugging.  Calibre equation-based DRC resolves to some value for a
particular area of the design -- such as 2 lines and a space for a spacing
equation.  We measured the width of line and the length of edges that run in
parallel.  Equation-based DRC takes this input and produces a number, and
then we use Calibre's RVE to view the results.  Mentor does a good job of
bolting things on to Calibre in this way, where everything is integrated.

Good and Bad:

The best thing about using Calibre's equation based DRC is that I don't have
to break up the design rule check into buckets and over-constrain my design.
This is because the most important thing for us is to have the best design.

We're still new to using Calibre in general and would like better pictures
in the User's manual to help explain how a function works.

Also, we were a little surprised initially about Calibre's conditional
statements, since SVRF language didn't have any conditional statements
(e.g. for loops and while loops).  This was bad because I was used to
using conditional statements in Hercules.  To be clear there are not
runtime conditional statements in SVRF.  I dislike this because when
converting code it would have been simpler.  Mentor did show us how to
replace them in SVRF with other logic and you can use TVF (Tickle
Verification Format).

Overall, we've been happy with eqDRC.  If you are using Calibre, eqDRC is
just like another rule type and doesn't change your design flow.  It
initially took us a couple of weeks to get comfortable writing the rules in
equation form.  Once we got familiar with using equations it took much less
time to write them, but we find that we end up taking advantage of the
extra time to write even more equations to better deal with the design
complexity.  D'oh!

Mentor's technical support was way beyond what we are used to with Hercules.
With Synopsys, we had to always to go to their help desk phone number.  (You
know, their general help desk person who would not be able to answer our
complicated questions, but they made you go through them anyway, even though
they called us a "high priority customer".)

Mentor's people came onsite and we know exactly who to call for help.  And
their support didn't go down once we bought Calibre.

    - [ Moon Unit Zappa ]
Index    Next->Item








   
 Sign up for the DeepChip newsletter.
Email
 Read what EDA tool users really think.


Feedback About Wiretaps ESNUGs SIGN UP! Downloads Trip Reports Advertise

"Relax. This is a discussion. Anything said here is just one engineer's opinion. Email in your dissenting letter and it'll be published, too."
This Web Site Is Modified Every 2-3 Days
Copyright 1991-2024 John Cooley.  All Rights Reserved.
| Contact John Cooley | Webmaster | Legal | Feedback Form |

   !!!     "It's not a BUG,
  /o o\  /  it's a FEATURE!"
 (  >  )
  \ - / 
  _] [_     (jcooley 1991)