( ESNUG 583 Item 5 ) -------------------------------------------- [06/06/18]
Subject: Hogan and Anirudh on Ansys ANSS Redhawk/SeaScape vs. CDNS Voltus
DAC'17 Troublemakers Panel in Austin, TX
Cooley: So, Joe, speaking of IR drop, you and Anirudh last year talked
about Apache having problems with IR drop and their Redhawk,
Seahawk, SeaScape tool... (ESNUG 563 #10) I don't know...
they keep changing the names of it now. Just recently they
called it "Redhawk-SC" or something like that. So, what's the
status that you guys are seeing now in the Ansys tool? Last
year, it had accuracy and repeatability problems.
Hogan: Well, you know, this goes back to... Someone just made a
comment about ICC II, (ESNUG 563 #6) how trying to replace the
engine on the plane as it's flying is a hard thing to do. So
Ansys Redhawk had been leading the market for a while. But...
Cooley: Yeah Redhawk was doing really well for something like 15 years.
Hogan: But Redhawk just didn't have the scale to be handling some of the
larger SoC designs, so they just started doing things -- it's
very analogous to what Cadence Dracula started doing 20 years ago.
They started having to break up designs, and then try to do
some kind of higher level thing where you use abstractions.
And as soon as that starts happening, there's an opportunity
there.
John Lee with Gear had a tool he was able to bring in and show a
couple of places where it ran really, really fast. But the delta
between showing it running really fast at a couple of places,
versus a production tool that's trying to fit into the same
infrastructure as your existing tool -- it's a hard job, and I
don't think they're done. So, I think that remains in a space
that has some opportunities. It could be interesting over time.
Cooley: Anirudh?
Anirudh: So, I think in this space, since it's a chip sign-off tool,
accuracy is paramount. The question is what drives accuracy in
an IR drop tool. And that also drives how we would structure
the R&D team.
Cooley: Well, but the first thing is, are you seeing the same thing?
Are your guys seeing the same thing from last year to this year?
[ Editor's Note: Then, no matter what, Anridh refused to say the
words "Ansys", "Apache", "Redhawk", nor "SeaScape" -- so I had
to edit them in to understand what was being said. - John ]
Anirudh: So, what we are consistently seeing, especially at lower nodes,
is Voltus and Voltus-Fi is much more accurate than other tools.
Cooley: Other tools meaning what?
Anirudh: There is one other tool, right?
Cooley: That other tool being?
Anirudh: John, there is one other tool [Ansys Redhawk]. So, the thing
is that... and there are a lot of reasons for it... and I
don't have enough time to explain all of that. But in IR-drop,
there are the gate-level tools [Redhawk/Voltus] and there are
transistor-level tools [Totem/Voltus-Fi].
So, in the gate-level tool you do a lot of stuff. You do power
analysis, extraction... But the key thing is, you have to solve a
very big matrix.
So, what we decided to do is put SPICE simulation guys with a
lot of experience in SPICE simulation to be in charge of our
Voltus tool. Therefore, it's very accurate when you're run
these big full chip SPICE simulations.
The other part, which is equally important, is you have a make
a model of memory or IP and feed it to Voltus -- which is
Voltus-Fi. Because of my previous experience I did a lot of
collaboration with Tom Beckley, you know, our Spectre team.
And Voltus-Fi is coupled with Spectre in the way it solves the
transistor level. So, it is much more accurate. So Voltus-Fi is
much more accurate than the IP modeling tool [Ansys Totem] from
other companies.
Cooley: From other companies meaning?
Anirudh: I'm saying you have to look at accuracy of the transistor-level
tool, [Totem/Voltus-Fi] and the gate-level tool [Redhawk/Voltus].
Both are important, and for some engagements our wins were driven
at the transistor level as well, which I think you didn't capture
last time. So, it's very important to look at Voltus-Fi, which
is a transistor level tool and Voltus, which is a gate level
tool. And I think accuracy is paramount.
Cooley: So, you're saying this mysterious other tool [Ansys Totem]
has transistor-level problems, too?
Anirudh: It's even worse on this level problems than gate-level in my
opinion. Because to really solve the transistor-level problem
you need true integration with SPICE. And that's the reason we
integrated Voltus-Fi with Spectre and the Spectre APS.
Otherwise, it's very difficult to solve, because what happens
at lower geometries, is there is an interaction between the
IR-drop or the power grid and the transistors. Unless you solve
it in a couple ways it's impossible to get an accurate answer.
Cooley: Jim, what's this other tool? What's going on here?
Hogan: Well, let's go back to Apache acquisition. So, Apache was doing
about $40 million; they filed to go IPO. They got taken out
right before the IPO for a really nice premium. Within a year or
so that $43 million grew to about $100 million in revenue.
Then Apache revenues decreased a little bit with the introduction
of his [points to Anirudh] Voltus product to about $90 million.
They [Ansys] are up around $100 million now, so they've come
back a little bit, but certainly not experienced growth.
Let's ratchet up one more step. If you look at ANSYS, they had a
very high multiple on revenue. But the challenge has been they
haven't been able to experience growth without a margin decrease.
As a result of that, they [Ansys] shouldn't see the multiple
they're seeing. So, they switched the [Ansys] CEO.
I would say the real action on Apache [points to Anirudh] and
the decrease wasn't due to John Lee, but was laid at the feet
of the CEO. So, what are they doing? They [Ansys] hired
Rick Mahoney, whom we all know -- Rick was a former enterprise
sales guy at Cadence. They also hired a couple guys from Magma,
Keith Morton is the other guy, he's a major accounts guy.
They [Ansys] are going after enterprise sales.
That cost is a lot lower there, for higher margins, higher top
line. They're all about growth. So, can the [SeaScape] tool
they've been working on achieve that growth in an Enterprise
sale?
Cooley: You're saying the [Gear/SeaScape] tool Ansys has been working on
is what you're saying.
Hogan: Yes, so John Lee, of course, is trying to introduce his deep
learning product [SeaScape]. He's also trying to change the
engine under the hood with all the [Redhawk] R&D guys who've
been around Apache forever.
So, I agree with Joe Sawicki -- [Ansys] is a work in progress.
However, it's a work in progress while they got to increase sales
and increase margin. That's very difficult to do.
---- ---- ---- ---- ---- ---- ----
Related Articles
Cadence Voltus, Ansys RedHawk, SeaHawk, Tempus, Primetime, Dorado
Anirudh and Sawicki on iffy Apache IR-drop #'s vs. Voltus/Innovus
A second CDNS Voltus-DP vs. ANSS Gear RedHawk-DMP user benchmark
The Nvidia stealth benchmark of CDNS Voltus vs. ANSS Gear RedHawk
Join
Index
Next->Item
|
|
