ANALYSIS

Biz environment that welcomed ARM is no more

By Ron Wilson

With the rising scent of carnage among CPU core intellectual-property vendors, some analysts are calling the entire IP business model into question. But it appears the problem is a combination of a bruising economy and a business model that proved more delicate than many investors had realized. Ironically, it is a highly successful CPU IP vendor, ARM, that has virtually defined the IP market. The company was early to license its RISC core as IP. To a great extent because of its platform’s fortuitous adoption by the cellular-handset industry ARM reached profitability, had a highly successful IPO and became the model for a gaggle of imitators. But in the wake of ARM’s success, the results have been decidedly more mixed. Some of the more recent IP vendors appear to have reached a level of stability. Some are doomed. Probably none are anywhere near sustained profitability.

A number of factors are at work here. One is that there just hasn’t been another explosive market as big—or as dominated by one IP vendor—as cell phones. Another is that in the highly conservative world of embedded CPUs, coming in second is very hard.

ARM had the fortune to be early into a market where there was no body of existing software that would have to be ported. For later CPU IP vendors, the cost to a prospective customer of converting software libraries, obtaining new tools and retraining software design teams is a huge barrier to entry. Nobody changes instruction sets without a compelling reason.

The need to offer a compelling reason has led later CPU IP vendors into some extreme marketing strategies. Some have scored design wins on systems-on-chip with huge numbers of parallel CPU cores—applications not well-served by nonconfigurable IP. But such complex architectures come with high attendant risks of failure in the real world.

Some recent CPU IP vendors have taken configurability to the instruction or even the compute engine level, raising serious issues about software compatibility, application portability and, frankly, the credibility of the vendor’s performance claims. On the one hand, huge performance claims are necessary if you are going to get a prospect's attention. On the other, as one IP marketer complained recently, “even when we can document the performance, people won’t believe us.”

Among the most frustrated by this dilemma are some vendors that offer huge performance claims as the result of run-time reconfigurability. That technology is proven in some applications. But it still tends to terrify most embedded-systems developers.

Beyond the relative merits of the various architectures and marketing models, all of the CPU IP companies—including ARM—have a potential vulnerability in common: their business model. Why ARM has succeeded while many of its followers will probably fail may come down to a simple matter of cash flows and timing.

Virtually all companies in the IP business use one of two business models, according to silicon analyst Jim Turley. The simpler, and less common, model bases revenue on two sources: a large license fee, paid at the inception of the contract, and a per-unit royalty, paid when the customer’s product is actually fabricated. The more common model adds a third important revenue stream: a monthly fee for maintenance and support.

Up-front fee

A number of issues make this model very brittle, Turley explained, even if the product is basically successful. The first is the up-front license fee. CPU IP providers ask licensees to fork over a large sum, often in the low seven figures, at the beginning of the contract. When a company is cash-rich, that may be an easy decision. But when cash is low and a new round of funding is unlikely — and when there is an engineering team just sitting there that could do a CPU itself — it can be hard to justify such a payment. License sales thus tend to decline rapidly when capital markets tighten.

Similarly, royalties, which can range from a dime to a few dollars per unit, are affected by timing. “A lot of people don’t realize it,” Turley said, “but the delay between the license fee and the first royalty check for a CPU is typically 18 to 24 months.” That fact separates the vendors that started two years before the recession from the latecomers.

Given those two factors, a company that grew its fixed expenses based on rapid early sales of licenses can find itself in a serious fix: high fixed expenses, slow or zero revenue from new licenses, and little or no revenue from royalties. That forces the company—which still must support its design wins and develop new products—to tough it out on little more than maintenance revenue and savings.

“On the average, an IP company has to assume at least six years from startup to profitability, even if they are successful,” Turley said. The recession has only lengthened that interval.

A further factor may doom some companies that might otherwise have survived the recession. Many CPU IP licensees have been startups clustered in a few networking- or communications-related markets. Most of those startups are expected to be weeded out. So if an IP vendor has sold 20 licenses into the same application segment, maybe only a few of its licensees will survive to ship product, and maybe one — or none — will reach high volume. Diversification is vital. But it’s at odds with the IP vendors’ tendency to push their experience in narrow technology segments.

There is a final irony. If you are selling to a technically hot startup, the central issue will be architecture: your technology. But the big, financially solid companies that could keep an IP vendor alive through a recession are a whole different ball game.

The big prospects are very aware of the business model risks. And they have no intention of basing a product line on an unsupported CPU from a defunct company. So the very fact that the IP business model is risky and the vendor is small may work against it in getting exactly the kinds of design wins it needs to escape that risk.

Nobody said it was a fair world.

Electrical Engineering Times January 28, 2002 www.eet.com

Cash-strapped processor core vendors struggle to stay afloat

Shakeout looms for IP startups

By Junko Yoshida and Anthony Cataldo

SAN MATEO, CALIF. — Once considered the shining stars of the venture capital community, many startup companies designing processor engines for next-generation networks and sophisticated system- on-chip devices are struggling to stay alive. As they contend with tenuous business plans and a lackluster economy, startups are cutting staff, seeking emergency cash infusions, courting acquisitions or radically changing their product plans. (Not all is gloom and doom: See www.eet.com/story/OEG2002 9125S0068.)

The young processor companies suffering the most are those that bet big on the networking infrastructure market, the hardest- hit segment of the electronics industry. But the market collapse has left few safe havens, casting doubt on a raft of processor startups that had hoped to sell their intellectual property in droves without having to dirty their hands with back-end chip design.

Among the startups that appear on the edge of collapse is Clearwater Networks, a once high- flying network processor vendor that observers said has struggled to get its product out in the face of funding problems. The company has seen the departure of founder and chief technical officer Mario Nemirovsky and two CEOs. The Los Gatos, Calif., company is now being headed by Dado Banatao, a managing partner at Tallwood Venture Capital (Palo Alto, Calif), which funded Clearwater. Banatao did not return phone calls from EE Times. Industry sources familiar with Clearwater said the company has solid technology intended for networking Layers 4 to 7 that is designed to hide memory latency while doing deep lookups in large memory tables. But the company underestimated the time it would take to design and deliver silicon, sources said. “The last time I checked, Clearwater was still struggling to get the product out,” said Linley Gwennap, founder and principal analyst of the Linley Group. Clearwater’s struggles are indicative of a wider shakeout among makers of network processors. At least two other companies—Acorn Networks and Entridia—have succumbed to the market slowdown and more casualties are expected. “The issue is that way too many companies, particularly in the network processor market, [were] funded under a good economy. But frankly, a lot of them were bad ideas,” Gwennap said.

One way out is merger. Bops Inc., a Mountain View, Calif., provider of programmable broadband DSP cores, is “looking at potentials of acquisition,” said CEO Carl Schlachte. As startups become more desperate for cash, management upheavals are common. TriMediaTechnologies Inc., a Philips spin-off that develops media processor intellectual property (IP), lost its CEO last summer when he couldn’t raise enough venture funding. He was replaced last month by Sunil Sanghavi, who is said to be bringing much-needed financial commitments to the company.

Bops recently backed off from its latest round of fund raising, said Schlachte, after being approached by an unnamed suitor for possible acquisition. Though Schlachte said Bops has “cash in the bank, and people are gainfully employed at our company,” he acknowledged it has been making cutbacks. “It has been a tough, tough year. We are watching nickels and dimes.”

A fundamental problem for IP companies is the tune it takes to generate revenue. Most can’t get royalties until customers ship products based on the licensed IP cores. And with every corporate move under increased scrutiny in a sluggish economy, customers are less inclined to enter multi-million-dollar licensing deals.

Bops claims to have three licensees, but the first product won't show up until the fourth quarter. “Every IP core company is facing exactly the same problems,” said Schlachte. To hasten its IP into real silicon, Bops is offering licensee more deliverable for use in building a system-on-chip, including sample hardware and software designs and engineering services, a strategy it calls “SoC in a box.”

“The sooner our licensees start shipping their products, the more quickly they pay me royalties,” said Schlachte.

Bops is also steering clear of the communications bloodbath in favor of consumers. Rather than pushing such applications as voice-over-Internet Protocol gateways, Bops is promoting its DSP cores for wireless LAN chip sets and for MPEG-4 encode/decode, MPEC-2, Dolby Digital and real-time streaming.

Other startups are trying to take control of when and how their technologies introduced by offering standard chips. Some are scrapping the pure-play IP model entirely and becoming fabless semiconductor vendors.

Lexra Inc. (San Jose, Calif.), a supplier of MIPS-compatible CPU cores, has taken the latter approach since settling a patent- infringement suit with MIPS Technologies Inc. last month. Lexra’s decision to settle with MIPS was part of a larger plan to jettison most of its RISC and DSP cores and focus on a single architecture for network cores. The shift occurred when Lexra found it couldn’t compete in some markets. “One of the areas that was at the heart of the battle was wireless LAN,” said president and CEO Charlie Cheng. “But we would have had to go head-to-head against [processor core heavyweight ARM, and it wasn’t clear that we could win.” As part of the settlement with MIPS, Lexra took a MIPS license—something it had tried to avoid since its founding. It also turned over its IP claims to MIPS. For these concessions, Lexra will continue to develop its LX8000 net processor for OC-768 core routers as a standard processor.

By using a foundry to produce standard devices, Lexra said it won’t have to rely on its customers to fund the $10 million to $20 million development effort. “In 2001, every single one of our licensees in [OC-768] canceled their project,” Cheng said. “We knew that unless we did something [dramatic] our technology won’t see the market.”

While Lexra dodged a bullet by settling with MIPS, picoTurbo Inc. (Milpitas, Calif.), did not fare as well in its battle to design and sell ARM clones. In a settlement reached with ARM last month, picoTurbo must cease licensing its ARM clones and work to migrate customers’ products to ARM Ltd.’s road map.

“It was costing them a bloody fortune to maintain lawyers and I know with the economic situation that the funding, especially for picoTurbo, was a little bit fuzzy,” said analyst Markus Levy at MicroDesign Resources.

Nazomi Communications, a provider of Java processor cores, also sees merit in standard devices. The company this week will announce its first standard chip, the JA108, designed to accelerate Java byte code (see story, page 26). The strategy is to augment royalty revenue from licensees with sales of standard products. “The main reason was to get the adoption going faster,’ said Mukesh Patel, CEO and president of the Santa Clara Calif., company. Nazomi still plans to support its Java co-processor IP, he added.

Java accelerators and co-processors are a market ripe for consolidation. There are some 10 providers of these specialty devices, most of which are still being evaluated by customers. Winners and losers should become clearer in six months, when a benchmark group called EEMBC releases results of its Java accelerator benchmarks, said analyst Levy, who serves as EEMBC’s president.

Now that the cell phone industry has endorsed Java, the market should be big enough to sustain at least four players, said Nazomi’s Patel. He said he is confident Nazomi has the right team and strategy to make the cut.

“We’re not profitable, but we do generate revenue and we have plenty of cash and have no concerns about funding,” Patella said.

Profits have also been elusive for TriMedia (Milpitas). Still, CEO Sanghavi said, “We believe that we are absolutely in the right market with the right business model.” The main target for the company’s very long instruction word core is consumer video applications, an unsettled market that can benefit from TriMedia’s programmable platform, he said.

But some industry experts are concerned that Philips Semiconductors’ successful TriMedia- based standard products maybe stealing the startup’s thunder. Philips is working on a stand- alone processor, custom products and a Digital Video Platform-based chip that combines MIPS and TriMedia cores. Given the economy, some industry watchers say customers are more likely to buy one of these standard products than to invest millions in taking a TriMedia license and building their own.

Sanghavi, however, said the custom option is attractive to some. National Semiconductor Corp., for example, is designing a reference platform for smart displays that ties together its X86-based Geode processor with Philips’ TriMedia processor, but plans to eventually integrate a licensed TriMedia core into a Geode processor. “It allows me to tell our customers ‘you can have it your way,’ “said Sanghavi.

National, for its part, said it expects to deliver a Geode processor with an integrated TriMedia core by early 2004 or sooner.

Electronic Engineer Times January 28, 2002 www.eet.com

By Chris Edwards

LONDON — Bucking the trend in the troubled intellectual-property sector, IP pioneer ARM Holdings saw pretax profit climb 42 percent in 2001, to nearly $71 million, on the growing strength of its licensing business, sales of development systems and royalties, the company reported last week.

Unit volumes of chips shipped with ARM cores grew 9 percent over 2000’s total, to 420 million. Sales climbed 45 percent, to $206.3 million, despite the chaos among ARM’s semiconductor customers.

For the fourth quarter of 2001, sales were $56.7 million, up 35 percent from fourth quarter 2000 and up 7 percent from the third quarter of 2001.

“While unit shipments from our semiconductor partners were slightly lower than in the previous quarter, royalty revenues were up 7 percent, due in part to better royalties on higher-value semiconductor products using ARM9-family core technology,” said chief executive Warren East. “ARM9 as a proportion is still pretty small, but we have seen a significant shipment in the wireless” market.

Royalties for cores used in wireless handsets remained the strongest, at 50 percent of royalty revenue.

East said ARM9 accounted for about half the company’s licensing revenue in 2001. “So two to three years out, it will be a significant chunk of our royalty revenue. But there is still a significant number of people doing ARM7 design starts.”

The company redesigned its ARM7 core last year to handle the sub-0.18-micron processes that some customers are now using to implement the core.

In the fourth quarter, ARM signed 15 licenses out of an annual total of 73. Of the total, more than one-third of the new licenses were from new customers. But only a third of those were for the traditional, company-wide licenses that ARM insisted on in the past. Of the 27 new licensees, 18 were for per-use deals obtained through the company’s foundry program.

That brought the total of per-use licensees to 25, with more than 40 separate designs purchased.

“We wanted the business from it to be purely incremental,” said East of the foundry business. “So, we developed a different product for a different sector. In a few years, if you look at our licensing, you will see a kink in 2001 [because of the addition of per-use licenses]. But it is really aimed at a different type of customer.”

ARM said its Gazelle core did well in 2001, attracting 16 existing customers for the ARM926EJ-S core. But the biggest growth came from its nonprocessor-core IP, such as peripherals and Java software. This segment “was 15 percent of revenue in 2001. It won’t be 30 percent in 2002, but it will grow steadily,” said East.

CHRIS EDWARDS IS THE EDITOR OF ELECTRONIC TIMES, E.E. TIMES’ SISTER PUBLICATION IN THE U.K.

Electronic Engineering Times February 4, 2002 www.eet.com

Coprocessor works with any baseband, startup says

BY TOM MURPHY

At some point in the future, a rollout of 2.5G and 3G cellular communication services will feature data- enabled phones and Java applications. But right now, Nazomi Communications Inc. is offering silicon, which it claims will enable smart phones to execute Java applications more efficiently.

The Santa Clara, Calif.-based startup introduces its JA108 Java Accelerator chip today after earning a patent in the technology Jan. 16. The company has also introduced intellectual property that system-on-a-chip (SOC) designers can incorporate into their baseband processors for wireless devices.

Jay Kamdar, COO and VP of marketing of Nazomi, estimates that the Java language and Java applications will be increasingly prevalent in wireless applications. As such, the RISC and CISC microprocessors now being used in wireless cell phones will require even more help in executing Java applications.

“That’s where Nazomi’s patented processor comes in,” Kamdar said. “We will offer superior Java performance while minimizing power consumption.”

Kamdar says Java applications in Java-enabled cell phones now in use in Japan are executing instructions on the baseband processor primarily with software. Many are impressed by how well NIT DoCoMo’s i- mode phones have gained acceptance in the Japanese market. One of the most popular applications for the digital, data-enabled phones is video games.

But as applications develop and swallow up more bandwidth and processing power of the baseband processor, there comes a point where software execution of Java applications reaches a point of diminishing returns. That is, they will swallow up too much processing power of the baseband and drain the batteries so quickly that the applications become less attractive to end-users.

Along comes the J108, which Kamdar and his fellow Sun Microsystems Inc. expatriates say provides the perfect balance. In a cell phone design, the JA108 will act as a coprocessor to the baseband, not unlike the way a graphics processor operates in a PC. The chip sits between the baseband and memory, and it allows the baseband to execute all of the typical communications and data instructions. The Java packets are then picked up by the JA108 and the applications are executed on the chip and sent back out through the baseband to the display.

The JA108 processor can operate universally with any processor, be it one geared for CDMA, OSM or another cellular communications protocol, Kamdar said. “A designer can just plug in the chip and enhance the current design,” he said. It requires no special tooling or software encoding, and the designers are still free to continue their code development in C or C++.

How does the company’s viability look in the marketplace? It may be too early to tell, according to some analysts.

“It certainly is an interesting product,” said Alan Brown, an analyst with Gartner Dataquest. “It looks to be an easy add-on to phone designs to accelerate Java applications. But the applications aren’t in the market yet. In my opinion, there is space for them (Nazomi Communications) to succeed, but there are other solutions out there to receive similar results.”

Markus Levyof Microprocessor Report says Nazomi faces the large task of going up against ARM Ltd. ARM processors have a stronghold on the cell phone market and recently the Cambridge, England-based company introduced enhancements to its ARM9 processor . Those improvements include higher performance, extensions to the instruction set for Java applications and DSP extensions. In addition, Nazomi faces competition from Zucotto Wireless and Digital Communications Technology.

Microprocessor Report is owned by Cahners Business Information, the parent company of Electronic News.

Electronic News Jnayary 28,2002 www.electronicnews.com