Advertisement

Zibb

EDA GRAFFITI, WITH EDA VETERAN PAUL MCLELLAN, DIGS INTO THE WORLD OF DESIGN TO FIND OUT HOW WE GOT HERE, WHERE WE ARE GOING, AND WHY EDA IS DIFFERENT.



   Advertisement

Tuesday, July 28, 2009

DAC, day 1

Jul 28 2009 12:00AM | Permalink |Comments (4) |


I went to the CEO panel today. But it wasn’t very illuminating since everyone was being too nice. When someone asked the question about what advice there was for the roughly 250 EDA startup CEOs in the audience, Wally said “Make sure you don’t run out of cash.” That was the nearest anyone would come to “most of you are screwed.” Lip Bu pointed out that the investment needed for a fabless semiconductor company has got too high for VCs to be able to see a return. For both EDA and fabless semi, the only possible hope is to keep burn rate so low that the money in is tiny by the time the company starts to generate cash. That’s a clever trick, however, in 45nm process where the cost of the masks alone could consume all the money.

All 3 CEOs said that they were meeting with CEOs of customer companies in a way that they didn’t used to. It was always a major criticism of EDA that they didn’t have boardroom access. SAP and Oracle were hobnobbing with the CEO, Cadence and Synopsys were talking to the CAD manager. But now customers are apparently much more open to wide-ranging discussions on what EDA can do for their customers beyond cutting their prices. If this is true, it is potentially a genuine opportunity, analogous to when IT in the late 1990s became regarded as a strategic weapon not just an expense line to be managed down. Companies like Walmart, never mind companies like Ebay and Amazon, use IT to outcompete their rivals. Semiconductor companies, increasingly fab lite or fabless, need to outcompete their rivals on design.

The most mysterious piece of the puzzle is software. Where does embedded software fit into EDA? Gary Smith pointed out that EDA would double if it could raise the software investment per embedded software engineer by just $15K. He reckons this is the key year, but outside of virtual platform technology it’s not clear what EDA has on the shelf or in development to achieve that type of uplift. But as more and more of a system is in software, and as semiconductor companies continue to have more software engineers than designers, it is clearly the critical piece of the puzzle to understand.

One bit of news today was that Magma’s auditors raised doubt about whether Magma was a going concern. Basically, Magma needs to succeed in re-negotiating its debt or else it may be insolvent. I guess people are basically optimistic they’ll do this since the stock closed up slightly.


Reader Comments



at 7/28/2009 10:07:50 PM, garydpdx said:
Software still needs target hardware to run on. An system design will need to map from ESL to lower levels of abstraction in software or hardware (RTL) implementations. The key for EDA will be to attract software development tools into its orbit. There have been examples of interaction across the hardware-software interface, Seamless CVE from Mentor is one example. So it can be done.

[Disclaimers: I was a former Seamless customer and then joined that organization within Mentor before moving on to IP-XACT oriented work.]



at 7/29/2009 1:53:11 PM, garydpdx said:
An interesting tweet sequence by Peggy Aycinena from William Daily's keynote at 46th DAC:

- Keynt.NVIDIA - Quest'n from audience - "It's all the fault of the software guys - how do we get them to write parallel code?"

- Keynt.NVIDIA - Ansr to SW guys not writing parallel code - B.Dally: Give SW guys a carrot - better performance if they do!

Dr. Daily's thesis is about utilizing coprocessors to accelerate parallel code implementing EDA tools ... coprocessors like, say, GPU's from NVIDIA! (:

Yes, the EDA platforms will need to execute fast enough to run large, meaningful codes. Seamless CVE has claimed to boot Linux but the models have to use optimization features in that tool.


(Thanks to Jeremy Ralph for pointing out the Twitter feed to me and others ...)



at 7/29/2009 1:55:14 PM, garydpdx said:
P.S.: William Daily is a professor at Stanford and executive at NVIDIA.




at 7/30/2009 3:45:45 PM, garydpdx said:
Okay, I stand corrected on the Professor's actual presentation (thanks to various reporters, traditional and blog) versus the advance abstract. It was more of a computer architecture talk but on the throughput CPU's (a la NVIDIA) versus conventional CPU's, it sounds as if he were proposing a SIMD machine architecture to create next generation CPU's. Versus MIMD, which is what multicore products are trying to bring to the PC.

[another disclaimer: I worked on Nonvon in my early days at Columbia, for D.E. Shaw. Yes, that D.E. Shaw!]

Post a comment



Display Name

Change Image
Before submitting this form, please type the characters displayed above.
Note the letters are NOT case sensitive.


ADVERTISEMENT

©1997-2010 Reed Business Information, a division of Reed Elsevier Inc. All rights reserved.
Use of this Web site is subject to its Terms of Use | Privacy Policy