!!! "It's not a BUG, 
/o o\ / it's a FEATURE!" (508) 429-4357
( > )
\ - / INDUSTRY GADFLY: "My Big Sloppy Catena Apology"
_] [_
by John Cooley
Holliston Poor Farm, P.O. Box 6222, Holliston, MA 01746-6222
I, John Cooley, being of unsound mind and overweight body, do hearby confess
the most woeful of sins of writing a wrongfully mocking column dissing the
honest and hard earned works of my technology brethren (and sistren), the
Cadence Catena engineers. It was due to personal poor judgement and weak
character on my part that I fell susceptible to the journalistic sorcery and
evil machinations of The Santarini. I, being sadly inept, I did not check
the story myself, but instead fell unto temptation. I was irresponsible.
Woe is me! I once was lost, but now I'm found; was blind, but now I see.
It was amazing grace, how sweet the sound, that saved a wretch like me. All
glory goes to Cadence (engineering). My bad.
There. Now that my big sloppy self-effacing apology is done, you might be
interested in the dirt I stumbled across while investigating Catena.
Mind you, I got these stories and rumors from all sorts of screwball people
and it's a nightmare to track exactly how many sources said exactly what.
Some of them were: Catena employees, ex-Cadence employees, users, the angry
archrivals to the mild indirect competitors of Cadence, IBM employees, the
press, start-ups praying that Cadence buys them, industry watchers, and
some of those Wall Street hangers on. Caveat Lector. That said:
- I found out why the Cadence Precision Router (CPR) was so quickly
renamed within a week of it being first announced; Cadence got a nasty
"cease-and-desist" letter from the Mentor Legal Department. Clearly
none of the geniuses in Cadence Corporate Marketing had ever heard of
Mentor's Precision RTL synthesis for FPGAs. "It shows you how amateur
these Intel execs are that Fister's bringing in," said one informant.
"Not knowing the industry's product names is a mistake that not even a
high school dropout would make." Inside Catena, "Finale" was the
project code name for CPR; now Cadence Corporate Marketing is trying
to get the name "Cadence Space-Based Router" (CSBR) to stick.
- Contrary to the "official" Catena story you get from the Cadence
press releases, multiple people told me that Catena was created as
a refugee camp in 2001 for the old CCT R&D gurus who had failed to
get the Integration Ensemble floorplanner out the door. Because of
this flop, Cadence had to fork out $110 million to buy the Silicon
Perspectives (SPC) floorplanner to cover this disaster. Unexpectedly
the new VP of Cadence's IC tools at the time, Lavi Lev, put Plato
and SPC guys into most of his key digital R&D slots while he slated
the failed Integration Ensemble R&D guys to be laid off.
- Fearful of losing these rockstar CCT developers, Jim Hogan (who had
just had Lavi Lev's job right before Lavi had joined Cadence) and
Jake Buurma, then the head of Cadence engineering, convinced the
non-technical CEO of Cadence, Ray Bingham, it would be better to
somehow keep these guys "in" Cadence. Otherwise they'd be "outside"
in a start-up that might be eventually working "against" Cadence; so
the idea for the so-called "Catena Incubator" was born.
- To help isolate it from the then malevolent Cadence mothership,
Catena started out in swank Los Gatos with 10 kickass R&D guys but
it grew to: 4 managers, ~20 R&D, and 9 apps guys. Some of the later
additional engineers were also rumored to be quite good but were
out of political favor at Cadence Central for one reason or another.
To model a start-up, some of the Catena refugees took a pay cut,
but all were given pseudo-stock they could "cash in" if Catena was
successful. (There was talk of Catena actually spinning off from
Cadence as a real start-up, but some sort of ruling from the SEC or
the FTC in the 1996-97 CCT antitrust litigation said that Cadence
could not buy another routing company without Synopsys and Mentor
having equal access to bid on it -- hence Catena could only "play"
start-up on paper and not be a start-up in real life -- otherwise
SNPS or MENT could buy it if it ever was actually worth anything.)
- Not too soon after Catena was created in 2001, Ted Vucurevich, the
CTO of Cadence took over. My hat goes off to him because it wasn't
a paper change of power; Ted actually got an office inside Catena and
spends many days a week there. In addition, over the past 5 years
Ted somehow managed to fend off the many attacks Catena got from the
SPC/Plato NanoRoute and Virtuoso groups who were still trapped inside
the massive 5,000 employee Cadence Borg mothership. They desperately
wanted to either kill or take over Catena. "Why are you laying off
our own people when you could cut all of those Catena misfits over in
Los Gatos?" How Ted protected the Catena refugees for 5 years in that
sea of increasingly hostile Big Company politics amazes me. Keep in
mind, Catena is from ye goode King Bingham's reign; His Highness, the
new King Fister has no creator's loyalty to it whatsoever.
- One measure of Catena being poltically outside Cadence is evident in
its first product, Chip Optimzer, launched in January 2006. While
every press release from Cadence Corporate Marketing overhypes how
let's-all-sing-Kumbaya tightly integrated the Cadence tools are with
each other -- one line from the Chip Optimzer briefing that was given
to Richard Goering speaks volumes:
"Chip Optimizer is a standalone product that can work with any
vendor's place and route tools, Brashears said." (EET 01/30/06)
It's a bolt-on post-processing layout tool that's NOT an integral
part of either Encounter or Virtuoso! That is, it's a sub-100 nm tool
that reportedly gets 10 to 20 percent better timing and better yields
for Synopsys and Magma backend users that does NOT force you to switch
to a Cadence P&R environment. Huh? King Fister might not like this!
- Catena's second product was the controversial Cadence Precision Router
(CPR) which was announced right before DAC. It consists of ~1,000,000
lines of C++ and OpenGL source code broken into 14 CVS directories
which both the Catena and IBM R&D teams access on a daily basis. IBM
is a weird CPR "user"; they only use parts of CPR that they built into
their own inhouse IBM tools. In rough terms, the Catena people wrote
900,000 lines, while 6 to 10 IBM people wrote 100,000 lines. (This
also includes counting the Neolinear NeoCell platform inside CPR as a
"Catena" part.) Basically the IBM folks wrote CPR's wire spreader and
CAA, while the Catena folks either wrote (or borrowed internally from
Cadence) everything else.
- For traditional customers (i.e. whole CPR tool users, not weird parts-
of-source-code users), I found Infineon, STmicroelectronics, ATI, and
Azul Systems. There was a press release with ATI cheerily yarping
about how they did the Xbox 360 chips using Chip Optimzer, so it's no
surprise that CPR's there, too. The other 3 are unannounced users. I
heard the Azul people actually had a CPR tapeout 9 months ago. They
ran into a problem where the DFM patches CPR did couldn't be read back
into NanoRoute. Data goes from NanoRoute to CPR OK, but from CPR to
NanoRoute not OK. They also had some crosstalk issues from the wire
spreading, and metal fill was not ready yet, but in the end they felt
they got better yield for their chip using CPR.
- On the technical side, CPR does some neat DFM stuff: fat single vias,
normal double vias, fat double vias, extra overlap, wire spreading,
line end pullback, fattening up jogs, wider wires for PWR/GND/clocks,
(eventually) metal fill, and (currently only for IBM customers) CAA.
- Supposedly Pyxis is a competitor to CPR, but so far not one user has
validated whether the Pyxis stuff works. I haven't even heard of who
any of the rumored hands-on Pyxis "users" are supposed to be.
- At present Cadence Corporate Marketing doesn't seem to know exactly
what to do with Catena now. Before DAC, they did a big push for the
new Catena CPR being the new heart of Virtuoso; yet now 3 months later
at their CDN Live sales conference, they yarp on how beefed up the
"new" Virtuoso is with no mention of anything Catena whatsoever.
- A number of the Catena people were iffy about re-integrating into the
Cadence Borg -- specifically if it's with the NanoRoute or Virtuoso
teams that were so eager to either take over or kill Catena. What's
murky for me is that I don't know if Cadence HR still implements its
olde "Thunderdome Policy" for situations like this. Late at night HR
used to put up a giant steel cage in the Cadence Bldg. 5 cafeteria
and have people with the same jobs from different companies pair off
to fight to the death. "Two men enter, one man leaves!" (This is how
Penny Hersher rose so quickly in the ranks of Cadence management when
she left Synopsys.) If they're no longer doing Thunderdome, it'll
still be scary for the Catena and Virtuoso engineers. Who will be on
top? Who's on bottom? Who's now "pitching"? Who's now "catching"?
Either way, everyone knows that merging these two R&D teams will NOT
be a pleasant experience for all those involved. God help them.
So, anyway, the moment I found out that the Catena guys wrote 90% and the
IBM guys wrote 10% of the source code on CPR, I knew I owed that apology
I wrote at the beginning of this column. It was a matter of honor.
-----
John Cooley runs the E-mail Synopsys Users Group (ESNUG), is a
contract ASIC designer, and loves hearing from engineers at
or (508) 429-4357.
|
|
