Perhaps the decision to name its flagship product after a fictional shipwrecked traveller who spent almost 30 years trapped on a small island was not the best of omens for Transmeta.
The company debuted in 2000 with Crusoe, a new chip with a low-power software-based approach to processing instructions. But four years later, Transmeta has racked up $591m (£330m) in losses and watched its Silicon Valley neighbour, Intel, claim all the spoils from an industry-wide move toward low-power mobile processors.
Transmeta warned investors in its most recent filing with the US Securities and Exchange Commission earlier this month that the company might be forced to scale back operations to concentrate on its fledgling licensing business if it needs to raise cash in future quarters and runs into problems securing credit.
In the five years before Transmeta was formally announced in January 2000, press and analysts were told little about what was going on inside the company. They knew its goal was to develop a next-generation mobile processor and its staff included technology luminaries such as Linux developer Linus Torvalds and Bell Laboratories processor designer David Ditzel. When Crusoe was finally unveiled, the PC industry had a chance to peek behind the curtain.
Crusoe's main benefit, and main problem, was its code-morphing software. Transmeta used a 128-bit VLIW (very long instruction word) architecture to build Crusoe, but that architecture was not compatible with the x86 architecture used by Intel and Advanced Micro Devices' processors and the vast majority of the world's PC software. So software was used to translate x86 instructions to Transmeta's hardware.
This approach allowed Transmeta to use fewer transistors on Crusoe than most x86 chips, cutting the chip's power consumption. The company recognised early that mobile processors needed to consume a lot less power if mobile computing devices were to take over the PC market.
An additional power-saving software technology called LongRun allowed Crusoe to adjust its clock speed thousands of times a second in sync with the processing requirements of an application. The company promised that users fed up with short battery life would have a better experience with a Crusoe system.
The first Crusoe processors consumed an average of one watt of power. Server suppliers such as RLX Technologies were excited by Crusoe's possibilities as the engine behind the new "blade server" concept. RLX pioneered the development of these thin devices, which held out the promise of helping IT managers reduce the space taken up by traditional servers.
Crusoe was ideal for blade servers used to host web pages because of its power characteristics and efficiency, said Chris Hipp, co-founder of RLX and now an independent technology consultant. These web servers did not require the most powerful processors on the market, he said.
However, PC suppliers wanted more general-purpose performance. Software simply cannot duplicate the raw performance of a well-designed collection of transistors. Early on, the industry knew that Crusoe was not the best-performing chip on the market, but Transmeta promised that the chip would have enough performance to satisfy the needs of most PC users.
Early reviews were not kind and suppliers and analysts did not get a chance to evaluate the processor until after it had been announced. These groups found Crusoe's performance to be much less than advertised.
Sony was the first PC company to release a Crusoe-based product and it continues to use the next-generation Efficeon processor, along with fellow Japanese PC suppliers Fujitsu and Sharp. But on IBM cancelled its Transmeta-based ThinkPad laptop in November 2000.
The former Compaq Computer was said to have flirted with the chip, but did not ship a product based on Crusoe until the tc1000 Tablet PC was released, after the company was acquired by Hewlett-Packard. Dell is famous for its arm-in-arm relationship with Intel, and never released a Crusoe notebook.
PC suppliers were further disappointed by manufacturing delays, as Transmeta tried to make the jump from 180-nanometer process technology to 130-nanometer process technology in 2001. Transmeta has no manufacturing facilities of its own, unlike Intel and AMD. It relies on third-party foundries to manufacture its designs.
IBM built the first Crusoe chips, but Transmeta chose Taiwan Semiconductor Manufacturing (TSMC) as its partner for the launch of the TM5800 Crusoe chip. Crusoe was the first chip that TSMC tried to build on its 130-nanometer process technology, and process technology jumps are often problematic for both chip designers and manufacturing processes, as has been seen this year with the jump to 90-nanometer process technology.
“It’s difficult when you’re a one-product company to execute 100% of the time,” said Kevin Krewell, editor in chief of the Microprocessor Report.
Whatever the reason, the delays irked PC suppliers. They could not plan on using Crusoe in upcoming product cycles because Transmeta could not guarantee sufficient quantities of the processors, Krewell said.
On the server side, the legions of dotcom companies that RLX had hoped would buy its Transmeta blade servers simply disappeared in late 2000 and 2001, along with the market for that technology, Hipp said. RLX was forced to turn to the general-purpose server market to survive, and Crusoe's performance was not up to the demands of that class of users, he said.
At about the same time, Intel started to gradually change its marketing strategy of emphasising raw clock speed as the most important indicator of performance, to focus more on the low-power market. In 2000, it quickly countered the Crusoe launch with low-power Pentium III chips of its own, although power consumption on those chips was not as attractive as Crusoe, according to analysts.
Intel also began to pay more attention to a design team in Israel that was developing a mobile processor core known as Banias. This contained some of the more power-friendly features of the Pentium III architecture, but also incorporated some of the performance features from the Pentium 4 family.
Details about Banias started to appear in 2001, and by its launch in 2003 as the Pentium M processor, Intel managed to steal the low-power spotlight from Transmeta.
Transmeta's technology is still well-regarded by power-conscious engineers and product designers, but the company has struggled to compete against Intel.
Because Intel enjoys a huge market share lead in desktop and notebook processors, it is often in a position to dictate pricing terms based on exclusive relationships with its customers. Dell's Intel-only processor strategy allows it to keep profits high while costs remain low, said Dell president and chief executive officer Kevin Rollins in a recent interview.
Transmeta's founders also overestimated demand for low-power mobile products, Krewell said. Low-power chips from Transmeta and Intel have helped notebook users who want increased mobility, but large desktop-replacement models designed for pure performance have been strong sellers over the past four years.
"The only sad part is that it took hundreds of millions of dollars to build these products, when the market segment itself really wasn't that big," Krewell said.
According to analysts, Transmeta's best hope for survival is the Efficeon processor, a revision to Crusoe that addressed many of the performance issues. Efficeon runs faster than Crusoe, uses a 256-bit VLIW architecture, and has an improved version of LongRun that will help boost performance and reduce power consumption.
"Transmeta's chances in the market depend on how the 90-nanometer [transition] plays out," said Dean McCarron, principal analyst with Mercury Research.
Transmeta is using Fujitsu's manufacturing facilities for its 90-nanometer Efficeons, and it made sure to pick a partner that had already built a 90-nanometer chip, said Arthur Swift, senior vice-president of marketing for Transmeta. It has produced samples of a 1.7GHz Efficeon processor, and expects to have limited production quantities available for customers this quarter, he said.
Crusoe has done well in Japan, where consumers are more interested in slim notebooks than their US or European counterparts. If that trend continues in other parts of the world, Transmeta's new chips could catch on, but Intel's Pentium M chip has had over a year to solidify its position as the leading chip in the market.
HP is using Efficeon in its new blade PCs, which is another relatively new concept that might gain momentum over the coming years.
Another hope is that Transmeta will make money from its research and development by licensing its technologies to other companies. NEC Electronics has licensed Transmeta's LongRun2 technology, and the company hopes to find other licensees that will pay for its intellectual property, Swift said.
The build-up surrounding Transmeta's launch and subsequent failure to make headway or profits is somewhat similar to the arcs of many companies launched in the late 1990s with a good idea and no profits. Several online retailers went out of business trying to create a market, but Transmeta's story is a little different. Armed with little more than a good idea and a team of talented engineers, it tried to compete headlong against a huge company with vast resources.
Robinson Crusoe eventually got off the island, but it took him 28 years. Transmeta won't have that long to right its ship as long as Intel sustains its advantage in mobile computing.
Tom Krazit writes for IDG News Service
This was first published in August 2004