e.Digital Corporation (fka EDIG)

[Tinroad]

OT: The age of erasable hardware
By Dean Takahashi
Red Herring
May 9, 2001
http://www.redherring.com/index.asp?layout=story&channel=10000001&doc_id=880019288

This article appears in the May 1 and 15, 2001, issue of Red Herring magazine.

One day, someone will make a chip that does everything for the ultimate consumer device. The chip will be smart enough to be the brains of a cell phone that can transmit or receive calls anywhere in the world. If the reception is poor, the phone will automatically adjust so that the quality improves. At the same time, the device will also serve as a handheld organizer and a player for music, videos, or games.

Unfortunately, that chip doesn't exist today. It would require flexibility, high performance, low power, and low cost, and it would need to get to the market quickly, before the multiple features it supports become outdated. Today, designing such a chip crosses too many architectural boundaries in the rigid rules of silicon. Every engineer knows that chip design involves trade-offs. And so far, nobody has figured out a way to get a chip to meet all the criteria for the ultimate consumer device.

But we might be getting closer. Now a new kind of chip may reshape the semiconductor landscape. The chip adapts to any programming task by effectively erasing its hardware design and regenerating new hardware that is perfectly suited to run the software at hand. These chips, referred to as reconfigurable processors, could tilt the balance of power that has preserved a decade-long standoff between programmable chips and hard-wired custom chips. These new chips are able to rewire themselves on the fly to create the exact hardware needed to run a piece of software at the utmost speed.

Reconfigurable computing goes a step beyond programmable chips in the matter of flexibility. It is not only possible but relatively commonplace to "rewrite" the silicon so that it can perform new functions in a split second. Reconfigurable chips are simply the extreme end of programmability.

If these adaptable chips can reach a cost-performance parity with hard-wired chips, customers will chuck the static hard-wired solutions. And if silicon can indeed become dynamic, then so will the gadgets of the information age. No longer will you have to buy a camera and a tape recorder. You could just buy one gadget, and then download a new function for it when you want to take some pictures or make a recording. Just think of the possibilities for the fickle consumer. Those possibilities might be available within the next year.

"We call it providing a chip on demand," says Charles Fox, president and CEO of Chameleon Systems in San Jose, California, a maker of reconfigurable chips that could streamline and add a degree of flexibility to wireless infrastructure.

CHIP TIDE
Hard-wired custom chips, the fastest but least flexible chips, are a $17 billion industry dominated by companies like LSI Logic (NYSE: LSI). Digital signal processors (DSPs) -- high-performance programmable chips that proliferate in such disparate places as cell phones, automobiles, and music players -- comprise a $6.1 billion market led by Texas Instruments (NYSE: TXN). Programmable logic chips, which are arrays of memory cells that can be programmed to perform hardware functions using software tools, are more flexible than DSP chips but slower and more expensive. They make up a $7.4 billion market led by Xilinx (Nasdaq: XLNX) and Altera (Nasdaq: ALTR), both of San Jose. Will Strauss, president and founder of Forward Concepts, a market research firm in Tempe, Arizona, thinks reconfigurables will nibble away at each of those markets and grow from $330 million in 2000 to $1.3 billion in 2004.

"In theory, these reconfigurable chips are pretty ideal," says Mr. Strauss. "If there is an easier way of doing the same thing here, then big companies are in trouble. The reconfigurables are starting in the wireless space, and they should expand from there."

Reconfigurable computing is no slam dunk. The Defense Advanced Research Projects Agency, the Pentagon's central research agency, poured $125 million into reconfigurable computing eight years ago and has little to show for it. It takes careful scheduling for the chip to configure the hardware on exactly the right timetable to keep the software from stalling. And no one has been taught to program this way.

"Reconfigurability is not just a chip problem," admits Jaime Cummins, cofounder, president, and CEO of QuickSilver Technology, a San Jose maker of reconfigurable chips for mobile handsets and one of Red Herring's "Ten to watch" this year. "It's also a tools problem, and it takes years to do the tools right. That's one of the reasons it's been a hard problem for a decade."

Several startups in reconfigurable computing have chosen the next-generation wireless market as the key battleground. Besides QuickSilver and Chameleon, Morphics Technology in Campbell, California, is also targeting the wireless market. One sign that the technology might have a promising future: Broadcom (Nasdaq: BRCM) last year ponied up $1 billion to buy reconfigurable chip startup Silicon Spice.

But such DSP makers as Texas Instruments aren't quite ready to give up this key market. TI contends it can improve its DSPs at a fast enough rate to provide the combination of speed and flexibility that universal cell phones will require (see "Texas changers"). Moreover, TI executive vice president and chief operating officer Richard Templeton contends that his company's chips are designed to accommodate many more applications than reconfigurables.

TIME FOR A CHANGE
Inertia might be the worst problem facing reconfigurable computing. Engineers are slow to change, and they're comfortable designing things the old way, which offered them a spectrum of programmable or hard-wired options. Indeed, there are more than two dozen existing microprocessor architectures, and it isn't clear how many more will be needed in the communications or consumer electronics markets. Already the makers of programmable logic chips say they fill the demand for flexible chips, if people are willing to wait 100 milliseconds to change applications, says Wallace Westfeldt, a product manager at Xilinx. Adds Pat Gelsinger, vice president and chief technology officer at Intel (Nasdaq: INTC): "We don't see the need for that much flexibility."

But Internet time has changed everything, and all chip makers are trying to change with the times. Life cycles for products like digital cameras now are as short as six months. As a result, time to market matters more than ever. Custom chip makers, who usually are burdened with the longest lead time, are working on hybrids that make their chips more flexible. Meanwhile, programmable chip makers are shooting for higher performance.

Programmable logic chips from Altera and Xilinx now offer millions of gates -- the basic logic building blocks of chips that consist of multiple transistors -- at $10 per chip, thanks to advances in semiconductor miniaturization, enabling those chips to catch up with the vast majority of hard-wired custom chips. Even hard-wired custom chips, made by companies like LSI Logic, with features written in stone, now offer limited degrees of programmability. A host of startups, from Triscend in Mountain View, California, to Tensilica in Santa Clara, have begun to deploy "configurable" microprocessors, which users can quickly modify for their own needs. But the degree of flexibility in these chips is limited. The result, says Mr. Fox, is that "all the different types of chip makers will converge on the same battleground."

Big companies from Dell Computer (Nasdaq: DELL) to Cisco Systems (Nasdaq: CSCO) have made bets on the startups -- both have invested in Morphics's recent $60 million round. Four major telecommunications companies are testing the Chameleon chips now and could deploy them in 2002. That prospect helped the company raise a round of $47 million in venture capital from 3i Group and others that gave it a post-money valuation of $220 million. For its part, QuickSilver has raised $20 million and is seeking additional funding.

"I think all of the signs are clear that reconfigurable chips will spell trouble for the hard-wired companies," says Vinod Dham, who runs the Silicon Spice division of Broadcom.

CHIP SHOT
Reconfigurable chips are a long-shot technology to some, but even the less-radical programmable chips are gaining ground on custom chips. Atiq Raza, founder, chairman, and CEO of Raza Foundries in San Jose, says that hard-wired chips can still feature the highest performance and therefore will be needed for years in the performance-crazy network core. But he anticipates programmable chips breaking into the "edge" where consumers wire into the network.

In the broadest scheme, programmability has been moving into many semiconductor markets. Custom chips are being replaced with programmable "network processors" made by dozens of new startups who argue that programmability is a must because processing protocols and standards are changing so often. Nvidia's (Nasdaq: NVDA) newest graphics chips shifted from hard-wired to a programmable architecture. So-called "media processors," which can process multiple types of data from sound to video, are making a comeback, and Transmeta (Nasdaq: TMTA) has designed a microprocessor that can adapt to the program at hand and thereby conserve battery life in a laptop.

Fans of reconfigurability, like Nick Tredennick, a technical adviser to QuickSilver and editor of the Dynamic Silicon newsletter for Gilder Publishing, foresee reconfigurable chips "invading everything that today hosts a DSP or a microprocessor." For consumers, this means that the day isn't far away when a cell phone can be used to talk, transmit video images, connect to the Internet, maintain a calendar, and serve as entertainment during travel delays -- without the need to plug in adapter hardware. QuickSilver plans to begin shipping such chips to cellular phone makers late this year or early next year.

Mr. Cummins calls his company's reconfigurable chip an "adaptive computing machine" (ACM), a kind of programmable logic device with hardware that can be rewritten hundreds of times a second. It has two parts: one that serves as a quickly accessible library, or cache, for hardware components, and another that is like a blank chalkboard.

As needed, the chip takes a hardware component from the library and slaps it into the blank chalkboard. There, the component executes the software running at the moment. When it's finished, the hardware component is erased and a new component is slapped in to process the next piece of software. It takes complex scheduling to map the right piece of hardware into the chalkboard at exactly the right time. But the advantages are potentially huge.

The chip can be smaller because its chalkboard allows it to fetch hardware components from memory, meaning that it doesn't use valuable chip area to store the entire library of hardware components, as a microprocessor does. Without such a chalkboard, a microprocessor has the whole library in place and drawing electricity at all times, even though only 1 to 5 percent is being used at any given time.

By contrast, a QuickSilver chip uses only the piece of hardware that it needs at any one time, and it uses power only for the active function. As an example, if the QuickSilver system needs to process a video image, the chip will tap its hardware library for video-processing components. In a split second it will put the components onto the blank chalkboard. And the hardware will be ready to process the software as it arrives for execution. When the video processing is done and 3D-graphics components are needed, the chip erases the chalkboard, returns the video-processing components to the library, and then pulls out 3D-processing components and puts them on the chalkboard.

Mr. Cummins estimates that an ACM will be able to process data as fast as a custom chip, and 10 or 100 times faster than a DSP, with just a tenth of the power and at a twentieth of the size. That translates into savings on battery life and lower chip costs. The overall performance of the ACM can surpass the DSP because the ACM only constructs the actual hardware needed to execute the software, whereas DSPs and microprocessors force the software to fit its given architecture.

A universal cell phone could certainly use the QuickSilver chip's features. So could a high-end, multifunction personal digital assistant, as well as any of the so-called information appliances that make it simple to hook up to the Internet. But whether designers will use these chips will depend on a lot of things.

Execution will determine whether QuickSilver or others will lead the pack in reconfigurable computing. Rivals at Xilinx and Altera are working on their own reconfigurable technologies. Mr. Cummins believes his company has the right hardware architecture to make chips with as few transistors as possible, compared to his rivals.

Even if QuickSilver, Morphics, Silicon Spice, or Chameleon don't surmount the obstacles as they try to create a new computing infrastructure, they will certainly force the entire chip industry to rethink its strategy and assumptions.

"The dinosaurs of the chip industry dominate the scene now," says chip veteran Gordon Campbell, founder and general partner of incubator TechFarm in Mountain View, which backed QuickSilver. "But there is this new mammal out there, the reconfigurable processor, and the question is whether the dinosaurs are going to adapt."

That all depends on whether the reconfigurable crowd can field its products faster than the dinosaur crowd can breed its next generation of chips.