Siliconcowboy's Blog

Technology, journalism, social media and social responsibility

Innovation or Obsession? A Retrospective on Electronic Design

As the traditional playing of the bagpipes draws the 47th annual Design Automation Conference to a close, it’s tempting to write a retrospective of the show.

But this year, I feel it’s far more appropriate to write a retrospective on the industry itself.

I’ve spent the last 25 years in the electronics industry, moving from a technical writer at Intel Scientific and Tektronix in Oregon, to a business and technology reporter in Silicon Valley, and then from a PR agency writer to a corporate-side flak at Altera, TSMC and Cadence.  Now, appropriately, I’ve moved on to blogging and social media optimization.

In that time, I’ve seen a lot of transitions unfold.

Real floppy discs were replaced by diskettes, and then by CDs and flash drives.  Xenix documents were replaced by  Wordperfect and then by Word. Vax machines were replaced by PCs, Macs, laptops and netbooks. Analog telephones were replaced by digital wireless phones, and then by Blackberries and iPhones. Television networks were replaced by cable, satellite, and then Internet-based HDTV.  Newspapers and television news programs are in transition as bloggers and Internet-savvy stations move in on their territory. Only textbooks have managed to hold their ground, valiantly fighting against the incursion of iPads and netbooks, but Wikipedia and Google are re-writing history anyway, and perhaps with greater accuracy.

And the changes enabled by semiconductor technology have cast an even wider net.

Medicine, for example, has gone from a near-art to applied science, largely due to advances in electronics. The Human Genome Project mapped the entire human DNA genome. Tomographic scanning, which involves the real-time crunching of complex algorithms, has enabled three dimensional views inside the human body and other structures. And for less than two dollars a day, the once deadly AIDS and HIV viruses can be controlled in men, women and children throughout Africa – a breakthrough enabled, in part, by computer-based analysis of new drug combinations.

Throughout these and many other semiconductor-enabled transitions, the one great constant of the electronic engineering world has been Moore’s Law. Collectively, the entire electronic engineering workforce has made a conscious pact with itself, to uphold this quest for as far as it will take us.  Indeed, many of the transitions I’ve outlined above are due, in great part, to the advances in computing power enabled by the steady march of Moore’s Law, which started out as nothing more than an observation that circuit density appeared to be doubling every 18 months or so (and that costs could also be halved at the same time).

Can we continue this course? And just as important, should we?

During a lunchtime panel on issues in lithography, moderator James Hogan suggested that there are no physical barriers to 22- or even 15-nanometer technology, but he hinted that cost could be a limiting factor.  Still, when the panel dove into just exactly what lithographic techniques would be available – let alone cost-effective – for 22 nanometer circuits, the experts could only give a skeptical nod to EUV and a slightly higher degree of confidence for direct-write lithography. When asked what, if any technology might be around as a backup, the panelists were silent. Even more telling, when they were asked what, if anything, EDA could do to solve this problem, they simply turned the question to the EDA industry. Some suggested that stacked die would move in. Others pointed to an innovation in process technology, like a star falling from the sky.

In fairness, lithography experts Aki Fujimura of D2S and Xin Wu of Xilinx shouldn’t be expected to know how EDA can solve this problem, if at all. And in fact, EDA has always been on the trailing edge of semiconductor innovation, with its value proposition being the automation of design techniques that exploit emerging manufacturing process innovations.

So the notion that the end of Moore’s Law may be in sight seems incredibly real this year. Hogan himself tried a save, by suggesting that “one company [Intel] has already announced that it would have 22 nanometer chips next year.”

“Still,” he admitted, “I don’t know how many other companies there are with $120 billion in the bank to bring 22nm to the commercial semiconductor marketplace.”

In other words, he’s not so sure even megafabs like TSMC or GLOBALFOUNDRIES can afford to do it. And 15 nanometers seems to be a complete wall.

There was another perspective, represented by Aaron Thean of Qualcomm: every innovation comes at a price, and if the price of scaling is too high — both in terms of raw dollars and in market delay — it won’t be pursued by OEMs.

And this, really, is where the buck stops. The pure drive for scaling of semiconductor technology may at last be prohibitively expensive.

The good news is, everybody in EDA seems to know that already.

Perhaps because they’ve been nickled and dimed to death by customers and systematically devalued by foundries, the major EDA players have always had cost-consciousness in mind.  Automation has always been about efficiency, and efficiency has always been about saving money and time.  After all, designers aren’t building chips with 5,000 transistors. They’re building system-level devices with 500 million transistors or more.  It’s simply not possible to do a leading-edge semiconductor design today with anything like the same methodology that was  used 25 or 10 or even 5 years ago.

And as these methodologies have changed, they’ve acquired more intelligence. Intellectual property cores have expanded from basic standard cells to complex, fully functional processor and system blocks. Analog, digital and memory all reside on the same chip. Power management takes place across the chip and within individual blocks, and it can be turned on and off at the system level.

To further enable this system-level design shift, Cadence recently purchased the memory IP maker Denali, and Synopsys acquired semiconductor IP developer Virage Logic. Both moves were superseded by Mosys, which last year turned the tables by acquiring a SerDes IP company and transforming itself into a fabless chip maker.   To give you some idea of the importance of these moves, the Synopsys/Virage Logic acquisition builds an impressive semiconductor IP staff of over 18oo engineers — far more, perhaps, than may ultimately be sustainable, but also far more than any other single commercial entity has on hand.

So what does this mean?

It means that irrespective of whether scaling as defined by Moore’s Law is achievable, it’s no longer the ultimate obsession of the entire semiconductor industry.

Irrespective of whether scaling as defined by Moore’s Law is achievable, it’s no longer the ultimate obsession of the entire semiconductor industry.

It means many engineers may be faced with the prospect of withdrawal from the drug that is Moore’s Law, as it becomes less and less attainable by them individually. Denial of this fact will only make them decreasingly employable.

And anyway, it’s time to kick that habit.

Pure research and development may still be important, but in truth, the resources of most EDA R&D teams are already stressed trying to patch up problems with existing software. So semiconductor innovation has to come from other sources, and we all know that funding for startups has all but disappeared. Only Intel, IBM and academia seem to be deeply focused on this challenge any more, and that may well be enough.

Chips are now systems, and systems are becoming increasingly innovative and powerful in their own right.  If we want this industry to continue to grow, we’ve got to put down the syringe, immerse ourselves in the real-world uses of modern electronics, and apply that insight to the next generation of system-level design.

There are plenty of examples of companies that have broken free of their cubicles long enough to see this, and adapt admirably. Microchip Technology, for example, is a great company that rakes in over a billion dollars a year making touch-sensitive circuits and other system-level control devices. Invensense, Panasonic and more recently STMicroelectronics are all high flyers in the MEMS-based gyroscope market for cell phones, enabling motion control for games and location analysis for GPS systems.  These are just a few examples of great break-out strategies by successful semiconductor companies, and there’s room for many more. Automobiles, for instance, contain hundreds of electronic devices, and not one of them is a bleeding-edge processor.

So the next time someone tells you they are in pursuit of Moore’s Law, gently remind them that the world of electronics is expanding, not contracting.

About these ads

7 comments on “Innovation or Obsession? A Retrospective on Electronic Design

  1. Freelance Website Design
    June 17, 2010

    Great article.. Keep it up!

  2. Pingback: SKMurphy » DAC 2010 Blog Coverage Roundup

  3. Dinesh Harjani
    June 18, 2010

    I completely agree; awesome article. I have a question though, isn’t graphene being held as the successor of silicon? Do you think it’ll arrive in time or we’ll have to stack chips in the meantime? Thanks.

  4. siliconcowboy
    June 18, 2010

    Dinesh thanks for your comment. Having spent some time on both the foundry and design sides, I can say that even minor tweaks have a tremendous rippling effect that goes all the way back to front-end design. So a change as significant as a new substrate is monumental, and will likely require a complete reset of the industry’s manufacturing and design strategies. All of this would have to be quality controlled and validated to extremely high levels of confidence, and then ramped to production volumes, which could take years. At the same time, graphene itself requires a level of quality control that has never been encountered with silicon substrates. A good article on this can be found here http://www.dailytech.com/Researchers+Prove+Graphene+Edge+Structure+Affects+Conducting+Properties/article14278.htm. My guess is that stacked die will have to do for some interim period.

  5. Dinesh Harjani
    June 21, 2010

    Thanks to you for taking your time and explaining me the situation; I believe it’s one of those things you find yourself repeating again and again. I’ll read your link right away; blogs like yours remind me that there’s a lot of stuff we take for granted everyday, when there’s actually an epic effort behind nobody ever sees.

  6. Pingback: The Semiconductor Industry Needs an X Prize « Siliconcowboy's Blog

  7. All of this would have to be quality controlled and validated to extremely high levels of confidence, and then ramped to production volumes, which could take years.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Follow

Get every new post delivered to your Inbox.

Join 3,620 other followers

%d bloggers like this: