Java moves to silicon for better phones

A host of companies are working on Java accelerator chips likely to find a welcome home in cell phones. The resulting speed-up could improve cell phones in several ways: by allowing more elaborate Java programs such as games, easing cell phone manufacturers' programming woes, or increasing battery power.

Sun, Java's inventor, has long wanted special-purpose chips to speed up Java. Though its own effort flopped, companies including Nazomi, Aurora VLSI, InSilicon, ARM Holdings, Zucotto Wireless and Parthus have come up with a new method. Instead of taking Sun's approach and creating an entirely new processor, these companies' products add Java acceleration to existing processors.

"I used to not believe in Java because of the performance issues, but now I see it coming. It's just a matter of time," said MicroDesign Resources analyst Markus Levy. "There's no doubt you will see three to 10 times more performance over software only."

The companies' plans--and chips, too, in Zucotto's case--were on display last week at Sun's JavaOne conference.

Java is software invented by Sun to shield programmers from the difficulties of writing programs for different computing devices. Java programs run inside a special piece of software called a "Java virtual machine" that handles all the details of communicating with the underlying hardware. The virtual machine software can make it much faster and cheaper to develop software that runs on multiple devices.

But this added virtual machine layer has a penalty: It takes a lot more processor horsepower to run the virtual machine as well as the Java program itself, so Java programs tend to run more slowly.

The Java accelerator chips take over some of the virtual machine's duties, executing Java instructions more speedily than the software.

Reversal of fortune
The Java accelerator chips fulfill Sun's glaringly premature predictions that Java and Java chips would spread to the world of gadgets such as mobile phones, handheld computers, car navigation systems and TV set-top boxes.

"With the Java programming language positioned to be the platform of choice for next-generation information appliances, Java processor technology, developed to optimize Java applications, will be at the center of these devices in 1998," Sun trumpeted in a 1997 news release. The company's PicoJava chip never caught on, despite being licensed to IBM, NEC, Fujitsu and Rockwell.

Sun offered to license the PicoJava designs to interested parties, but the new generation of chips doesn't use Sun's design, said Curtis Sasaki, director of technology advocacy at Sun.

The satisfaction Sun must feel at having chips available is probably comparable to its relief at the recent arrival of Java phones after years of promises from Sun. Sun has signed up all the major first-tier cell phone manufacturers and most of the second-tier ones, Sasaki said.

In a speech at the JavaOne conference Tuesday, Nokia President Pekka Ala-Pietila said his company will sell 50 million Java phones in 2002 and 100 million in 2003. Motorola is making all its phones Java capable. NTT DoCoMo in Japan, the leading Java phone adopter thus far, has already shipped 3 million.

Sasaki expects that all the phones NTT DoCoMo sells will be Java phones by the end of the year. That's a nice ally, considering NTT DoCoMo sells 60,000 phones a day.

"Java is clearly moving into the mainstream," said Stuart Creed, director of business development at Zucotto.

Creed, Levy, Sasaki and InSilicon Product Marketing Director Gervais Fong all believe phones with Java accelerator chips will arrive in 2002.

Java justified
Once Java is a staple, the next problem is figuring out what to do with it. Sasaki points to Java applications such as games for Japanese teenagers, gambling programs for Hong Kong high rollers and tools to help U.S. salespeople tap into corporate computers. But many analysts are skeptical.

"It still isn't clear that the applications are compelling enough. Whether somebody is going to buy a phone because of Java support, I don't think that's the case right now," Levy said.

Java accelerators could change that--especially in combination with high-speed third-generation (3G) cell phone networks that would enable data and Java programs to be downloaded more quickly.

Java chips would allow more powerful Java programs to run, such as MP3 players or better games. Or they could offload Java processing from the processor to let it run more slowly, thus consuming less power and providing longer battery life. In addition, cell phone makers could move some of the basic cell phone software--that used to remember phone numbers, log on to networks or synchronize with a PC--to Java versions that could be used with multiple models.

Once the Java foundation is laid, the money and the programmers likely will follow. "I also believe there are a lot of smart people who have a burning desire to make money. When that happens, Java is a big piece of it," Levy said.

Technicalities
Creating a Java accelerator chip isn't a trivial problem. And even when a chip is ready, there's the problem of updating the programming tools used by cell phone makers so software takes advantage of the new Java capabilities.

One difficulty is that Java programs use different types of instructions to perform tasks such as adding numbers together, so Java chips must be designed differently than conventional ones.

Chip size and capacity is also an issue. ARM Holdings engineer Howard Ho said his company's Jazelle chip executes 145 of the 228 total possible Java instructions, but that subset is what's used for about 80 percent of the Java processing.

Aurora VLSI takes a different approach. Its DeCaf chip runs 90 percent of Java instructions and thus is faster than competitors. But the added capacity means it's much bigger and therefore more expensive and power-hungry.

While Aurora and Zucotto plan standalone chips, most of the companies involved in putting Java on silicon sell only intellectual property in the form of chip designs meant to be grafted onto the main processor, Levy said.

It's not clear who's scoring with customers yet, but InSilicon's Fong said the company has signed at least one cell phone manufacturer. Zucotto's first chip, a combination Java accelerator and Bluetooth wireless communications chip called the x100, is due in the third quarter of 2001, and a Java-only one called the x120 is due in the fourth quarter.

Philips Semiconductor will use Zucotto's technology, Creed said. ARM has licensed its Jazelle design to LSI Logic and Sanyo, and its own models will be introduced in the fourth quarter, Ho said.

ARM, along with MIPS, has a major presence in the cell phone market CPU already. MIPS could be considering acquiring one of the smaller players, Levy said.

The merits of the different strategies and marketing efforts aren't certain yet, Levy said. Of the companies vying for customers now, "half will be gone by next year," he said. But the Java accelerator industry itself will be strong.