( ESNUG 563 Item 2 ) -------------------------------------------- [11/01/16]
Subject: Anirudh defends JasperGold's rep and the CDNS-RocketSim purchase
DAC'16 Troublemakers Panel in Austin, TX
Cooley: Anirudh, is Jasper a failure? Katherine sold it to you for a
$145 million. After that JasperGold just fell off the map.
There was Jasper, Jasper everywhere, and then it suddenly got
acquired and I've not seen any news about Jasper anywhere.
Anirudh: So Jasper, I think formal is an important area. Just to give
you some numbers, over the last 12 months, the Jasper business
has more than doubled for Cadence. I think the number of
customers has gone up by 3X. And Jasper, from what I know,
is by far the dominant formal solution in the market.
So yes, Katherine left CDNS after the acquisition; and you
know these things happen in M&A, but more than 90% of the
Jasper team is here, and it's a very talented team. I think
we are still the leaders in the formal market and I think
it will continue to be that way.
The other key thing is what we're trying to do is also combine
formal with other parts of the verification spectrum, like
simulation. There's a lot of interplay between formal and
simulation that needs to happen.
Cooley: Raik?
Raik: Well, I have to give credit to Jasper, they're our biggest
competitor, that's true. And I to praise their work on
educating the market on the importance of formal. But I
haven't seen too many new things from Jasper in the last
2 years.
Cooley: So you're saying after the Cadence acquisition of Jasper,
not much happened.
Raik: Well, I can't comment on Anirudh's numbers.
Cooley: Anirudh?
Anirudh: Well, I think Jasper is by far multiple times bigger than
our next competitor. I think there are a lot of new engines
coming out. The last time I checked, 17 out of the top 20
customers use Jasper. So this whole notion of Jasper not
doing well is a fantasy. Jasper is doing fabulously, in
terms of profit margin, revenue, customer share, technology.
And I think there is a lot of possibility of formal being
more and more pervasively used in the rest of verification.
---- ---- ---- ---- ---- ---- ----
Cooley: Anirudh, why did you by Rocketick?
Anirudh: I think Rocketick is a chance to redefine logic simulation
for the next 10 years. That's the reason we did it. The way
I look at logic simulation: Of course is a critical area;
it's very big -- even as an undergrad, the first EDA tool
I wrote was a logic simulator.
I think we are in the 2nd phase of logic simulation. The
1st phase is what I would call "interpreted simulators".
The 2nd phase is "compiled simulators" (like SNPS VCS),
which are now almost 15-20 years old -- which is almost
like 140 years in EDA years, right?
These are old technology. And so there's a chance to go
to the 3rd generation, which is "parallel simulation". I'm
a big believer in parallel simulation.
And the RocketSim team solved an amazingly hard problem -- to
parallelize logic simulation is very, very difficult, because
you have a lot of events, but each event is very, very small.
So there's not much computation that you can easily parallelize.
Rocketick's a breakthrough technology, and the team is fabulous.
And the third key thing is that they were already production
proven at some really market shaping companies -- in production
use. So Rocketick's the best of 3 worlds; you have a great
team, you have a great technology (that hasn't been solved in
about 20 years) and you have big customers using it. So I
think we should have bought Rocketick even earlier; but we're
glad that they're a part of Cadence now.
Cooley: Why didn't you grow your own? Synopsys is doing that with its
Cheetah flavor of VCS. They're doing basically GPU-based
parallel simulation. If Aart's doing it, why didn't you have
your own R&D guys do it?
Anirudh: We had an internal CDNS R&D effort. I think parallel Incisive
and parallel simulation. But Rocketick is just so far ahead
of whatever else I have seen. Also RocketSim's a very general
purpose approach.
So of course I get involved in all the algorithms, what you have
to make sure of -- the previous approaches were always very
specific. If the circuit has lot of repeated structures, it
would work - or these kind of anomalies. But the RocketSim
algorithm is general purpose. It works at gate-level, it works
at RTL-level, and it works for all kinds of circuits. It's a
phenomenal technology, and I think here's a chance to redefine
the whole area, and not just the design part but also the test
bench part and the whole we have a different methodology on
first spec and other kind of test benches. So I think logic
simulation is going to get a lot more exciting going forward.
Cooley: Joe!
Sawicki: Yes sir.
Cooley: Why didn't Mentor buy Rocketick? According to Anirudh, tt seems
like such a great deal.
Sawicki: We looked at RocketSim. And you know when you're looking at
early technology you come to different opinions of whether
or not it's God's gift to creation? Our view was it wasn't.
Cooley: Are you developing your own?
Sawicki: Everyone is working on parallelism, looking at GPU's, looking
for every bit of performance you can get out of this stuff.
If you look at the amount of cycles you have to run on compute
platforms to do verification, it's just immense now.
And so everyone's looking at all these techniques. They're
critical to make sure you've got some level of development
effort on. The challenge for RTL/gate simulation is that it's
a lot of stuff where you really hard against Amdahl's Law.
And if Rocketick found a way around Amdahl's Law that we
didn't see, then that's going to work well -- but we didn't
see that when we looked at them.
Anirudh: One thing I want to clarify, you asked me, and also Joe
mentioned. RocketSim works on general purpose CPU's, not
on GPU's. I believe that the general purpose CPU is a much
better architecture.
Cooley: Wait a minute, I thought it was a... is it GPU? Oh wait a
minute, they switched over to Xeons, yeah, the 686.
Anirudh: Rocketick Incisive works on a general purpose CPU. You're
current simulation farm, it will run on that. You don't
need any special purpose hardware like GPU's. It works on
Intel CPU's.
Cooley: Ok. That makes a difference alright.
---- ---- ---- ---- ---- ---- ----
Related Articles
Sawicki on how MENT Calibre DRC/LVS "breaks at almost every node"
Anirudh defends JasperGold's rep and the CDNS-RocketSim purchase
Dean on his new peer-to-peer chip design workspace accelerator
Amit on 55nm non-Gaussian variation starts, plus IoT low VDD issues
Raik on it's only OneSpin vs. Jasper in formal; not SNPS nor MENT
Join
Index
Next->Item
|
|
