Silicon Insider

by Jim Turley
Silicon Insider #11, February 2004

Sometimes discretion is the better part of valor. Sometimes tech companies need to downplay their technology in order to grow a business.

Configurable processors are one example. Over the past few years, firms like ARC, Elixent, Improv, MIPS, Quicksilver, and Tensilica (plus other unannounced companies still operating in stealth mode) have developed wonderful technology that allows their chips to adapt to changing workloads. The details are different but the concept is the same: chips don't have to be generic, one-size-fits-all designs. We've progressed to the point where chips can be customized, tailored for individual applications.

There's plenty of evidence to show this really works. Customizable chips can be much faster, or cheaper, or low-power than traditional fixed chips. The trouble is, customers are scared to death of it. Customizing their own chips appeals about as much as customizing their own skeleton. Customers don't want to tinker with their chip designs, even if the benefits will be tangible and compelling.

So configurable-chip vendors need to hide their light under a bushel. Don't tell customers how those benefits are achieved. Lie. Like the Wizard of Oz, urge customers to "pay no attention to that man behind the curtain."

This is easier for the chip makers than for the IP vendors. Chip companies like Toshiba (which uses Elixent's configurable technology in its MeP processor chip) can simply wave their hands and dispel fears about strange and unusual technology. Datasheet specifications don't lie; how they're achieved isn't the customer's concern.

Technology is driven by innovation and inertia in equal measure. We thrive on innovation, yet we're strangely reluctant to adopt it ourselves. We look for leading-edge performance with trailing-edge reliability. Companies with the best technology sometimes need to present it in terms adapted to the meanest understanding. Attracting customers should be the first goal; bragging about the technology can come later.

Comment on this article

by Jim Turley
Silicon Insider #10, January 2004

"Don't hate me because I'm beautiful." That might be the coy marketing come-on for Itanium, Opteron, or many other high-end processor chips. Choosing a microprocessor has gradually become a fashion statement, a decision made from the gut rather than from the head. Hype, myth, and perception play an increasing role in what used to be a nerdy engineering decision. Processors have gone prime time.

Who woulda' thought that Superbowl ads would be flogging million-transistor microprocessors? The Intel Inside campaign has shown how branding can affect sales, increasing demand for a product that customers don't understand.

The innovators' current dilemma -- Intel Itanium or AMD Athlon 64? -- is occupying minds at PC makers and server vendors around the world. The question of which chip is better? is now less important than which chip is likely to succeed? Customers want to back the winning candidate, so they gauge the breeze and measure momentum in daily news articles and usenet postings. Itanium and Athlon 64 are on the tipping point; a tiny nudge one way or the other could start the snowball rolling.

PC processors are the most obvious example of this, but embedded chips are at the whim of fashion trends as well. PowerPC, MIPS and — most especially — ARM vendors are caught up in a battle of perception. What's the coolest architecture this season? What's haute down the the engineering lab? That's the 64-bit question.

This trend is neither good nor bad. It is amoral; neutral. If anything, it's a symptom of maturation in the industry. When branding and marketing overtake content in importance, you've got a mature market. Customers (that is, engineers and buyers) would do well to recognize this, lest they become fashion victims. And CPU makers that haven't already recognized this change will wonder why their superior technology was swept aside by competitors with sexier ads.

Comment on this article

by Jim Turley
Silicon Insider #9, December 2003

Analog Devices' newest SHARC processor includes software to handle MP3, Dolby ProLogic, AAC, Windows Media Audio, DTS, and a gaggle of other audio standards -- but you can't get at it.

It's right there on the chip, burned into mask ROM at the factory. As a SHARC customer, you get all this valuable software whether you ask for it or not. What's the catch?

The software is all licensed, and Analog Devices supplies the key only if you pay extra for it. With the key, the chip decodes and executes the licensed audio-processing software at runtime. Without the key, the chip refuses to run the licensed code. Purchasing the key involves paying royalties to Dolby Labs, Microsoft, DTS, or whomever holds the intellectual property rights to the particular audio algorithm you want to enable. ADI simply passes the royalty payment along to the license holder. The chip acts as safe deposit box.

An unannounced CPU coming next year will do the same thing: include licensed software that's enabled only with a run-time key. This has all the makings of a trend.

With CPU die sizes getting smaller and memory arrays getting larger, it's inevitable that software would constitute a larger portion of the value of a device. (Indeed, I was asked by a client last week to quantify the value of embedded software in a certain microprocessor.) But that software value was always added by the customer, not the semiconductor vendor. Now the circuits-and-software genie is squeezing out of the bottle.

Several other chips have made half-steps in this direction. IDT's new 32434 (above), among others, includes secure SRAM that erases itself if it thinks it detects tampering. Tamper-proof SRAM isn't large enough to hold significant code, but it's fine for encryption keys, access codes, or passwords. Those, in turn, can control access to licensed media (read: legal MP3 files and DVDs) or secure mail servers.

Microprocessors are the new delivery mechanism for licensed code, rather than just the platform on which it runs. Keys and codecs will make up a larger part of new chips -- physically and fiscally. Algorithm providers, such as Dolby Labs, might soon form more lasting and lopsided partnerships with silicon vendors. And then, buying Brand X chips might be the only way to use Brand Y software, keys, or codecs.

Comment on this article

by Jim Turley
Silicon Insider #8, November 2003

IBM computers powering video games? This is not what anyone expected in 1960. Yet with Microsoft's recent announcement that its next-generation Xbox game console will use IBM technology, it looks like Big Blue has scored a big win for the top three game machines. Oh, my.

Neither side has said so, but it appears an IBM PowerPC chip will be at the heart of the next Xbox. There's still some room for doubt, since both firms refer only to "IBM technology," not PowerPC specifically. Still, I think it's a foregone conclusion that IBM will manufacture a custom ASIC for Microsoft and embed at least one PowerPC into it.

Sony's next-generation PlayStation will also use "IBM technology," in the form of the much-anticipated "Cell" microprocessor, which the two companies are developing jointly with Fujitsu. Although details are sketchy, Cell should be partially PowerPC-flavored as well.

Finally, Nintendo's current GameCube is definitely PowerPC powered. With only a little fudging, it's easy to say that all three game consoles are PowerPC-based (albeit one current and two future machines). Can this be what Thomas J. Watson had in mind?

MIPS Technologies was in a similar situation a few years ago. That company's processors powered both Sony and Nintendo game consoles, the two biggest sellers in the 1990s. MIPS (briefly) made a huge pile of money on its video game business, at one point receiving two-thirds of its corporate revenue solely from Nintendo. Video games can be hugely lucrative. The video game industry is larger than the movie industry; hit games make more money than Oscar-winning films, and there are no union hands or millionaire actors to pay.

On the other hand, the game consoles themselves aren't making the money. On the contrary, they're typically sold at a loss. Royalties on game titles make up the difference -- royalties that aren't likely to flow to IBM. (Nintendo's deal with MIPS was an exception.) Although IBM may be famous as the world's "toy processor" vendor, it's less clear the fame will translate into fortune.

We'll never know the financial details of the Xbox (or Sony, or Nintendo) deal, so let's assume that IBM will neither make nor lose a substantial sum. Instead, I think IBM's biggest benefit will be the publicity. PowerPC will be "cool" again, a reputation it hasn't enjoyed since the mid-1990s. PowerPC sales rank well behind sales of its 32-bit competitors. If game systems can get customers thinking PowerPC, it's a win for IBM overall. It could even become the company's new mantra: Think. IBM.

Comment on this article