Top secret chip less secret now

rohit@uci.edu
Fri, 13 Nov 1998 14:53:58 -0800 (PST)


This NEWS.COM (http://www.news.com/) story has been sent to you from rohit@uci.edu.

Message from sender:
Hmm. Improved performance within the current paradigm. Very worthy of respect, not necessarily going to turn the business upside down.

We do need to see PC-class power reduced to $1 chips -- but the real obstacle is that we don't have software architectures or networks for such widely embedded computing. More later, on Whitesides' vision of micron-scale electromechanical innovation as a predecssor to nanotech. His view: within ten years every material thing will have an electronic component of its behavior.

Rohit

PS. also testing news.com's email-to-a-friend feature
-------------------------------------------------------
Top secret chip less secret now
By Brooke Crothers and Stephen Shankland
November 13, 1998, 11:30 a.m. PT
http://www.news.com/News/Item/0%2C4%2C28737%2C00.html?sas.mail

[1PARA OPEN] Transmeta, the highly secretive, well-funded Silicon Valley chip start-up, may be offering the first glimpses of its well-guarded microprocessor design. [1PARA CLOSE]

The company recently received a patent for certain portions of the chip design and in the process may be revealing a broader picture of what the chip does. In short, it is a novel design that appears to be extremely fast and can potentially run all Windows software as well as other technologies including Java.

Speculation has been rife for the last couple of years about the nature of the company. It was formed in 1995 in Santa Clara, California, and is funded by Microsoft cofounder Paul Allen, among others. David Ditzel, an architect of early processors at AT&T's Bell Labs, and Linus Torvalds, the Finnish engineer who pioneered the Linux operating system, are among its principals.

[DISPLAY_RESOURCE_LINKS]

Because it has been so secretive, some reports have speculated in jest that the company is using alien technology. The patent reveals a slightly more down-to-earth design but still ground breaking because of its apparent ability to deftly handle Intel computer code and other technologies such as Java and RISC, according to Richard Belgard, who is an independent microprocessor patent consultant in Saratoga, California and has read the patent.

The upshot is this: Sections of the patent (No. 5,832,205) discuss a chip that can translate Intel chip instructions into a more advanced format referred to as VLIW (Very Long Instruction Word). VLIW is a catch-all term for a variety of technologies that essentially combine many simple computer instructions into a single long instruction which can then be executed more efficiently and quickly than current computer code.

Big implications The implications for the chip are tremendous. If the patent is truly a window on the finished product, the chip could tap into a huge market of Windows-Intel software and run it faster using a more advanced approach than Intel--or at least at a price-performance level that exceeds Intel. Though this has been done before by chip manufacturers such as Advanced Micro Devices and National Semiconductorís Cyrix arm, the Transmeta design appears to take this to a new level.

"This is not your motherís x86 [Intel-compatible] chip," said Belgard.

The technology could also be applied to other types of chips, according to the patent. For example, though the patent describes in detail how Transmeta's process would work to create a fast chip that's compatible with Intel silicon, the technique could work for "any family of ... computers," even Sun Microsystems' Java technology, the document says.

A peek at the patent • Transmeta's patent describes a combination of hardware and software that can translate instructions for processors such as those from Intel into the native language of Transmeta's chip.

"Code morphing" software, assisted by hardware on the chip itself, translates the instructions "on the fly," then stores the translated instructions so they can be re-used later without having to go through the relatively slow translation process again. The software also handles the task of tailoring the instructions so they're suited for the high-speed "very long instruction word" (VLIW) architecture of the chip. VLIW chips are small and fast because they don't have to devote precious silicon to keeping track of which results depend on other results.

Transmeta's patent also describes a technique to hold the processor's information in temporary registers where it can be quickly updated if the chip encounters an error.



The real power and beauty of the chip, on paper, is that it takes a hodgepodge of advanced technologies and integrates them into a potent mixture of software and hardware. The software is one of the keys for translating "on demand," said Belgard.

"It combines a lot of well known things but into a very elegant package," said Belgard. He also says that the design could avoid, to a large extent, patent wrangling with Intel because it deals with Intel technology in a novel way.

"Many, if not most, Intel patents are not used since much of the hardware on which Intel has patents is not employed. Rather, the Transmeta microprocessor uses a combination of software sequences and more elementary hardware. Therefore, Intel patents--at least the 'apparatus' patents--should not be a concern."

He cautioned though that Intel has other patents called "method patents" which could be a concern.

Colin Hunter, chief financial officer and a cofounder of Transmeta, declined comment on all questions about the technology and company.

The chip is potentially very fast though there is no mention of speeds in the patent. Itís a design "that is susceptible to very high speeds," Belgard said.

One-fourth the size It could also be cheap, according Belgard. He says it appears to be about one-fourth the size of an Intel chip. "Much smaller [and potentially] much less expensive."

The company behind all of this is still largely a mystery. In addition to being well-heeled, the company has some solid talent, according to Belgard.

David Ditzel is the other cofounder of Transmeta and is president and chief executive officer. Ditzel was the director of Sun Microsystems' Sparc Laboratories, and before that, an architect of early processors at AT&T's Bell Labs.

Transmeta has also managed to lure Linus Torvalds, the Finnish engineer who achieved near-mythic status as the initiator of a collective effort to write the Linux operating system.

In addition to the recent patent, one of the few clues to Transmeta's plan can be found on relatively obscure Web pages. One page says Paul Allen invested in Transmeta in 1997 and describes Transmeta as working on "engines for multimedia PCs."

It is not clear who would make the chip or when it might appear, though Belgard thinks this may happen in 1999. Companies rumored to be manufacturers of the chip include IBM and Texas Instruments.

"Code morphing" More detailed descriptions in the patent are quite esoteric and futuristic sounding.

Belgard says "code morphing" software is responsible for translating [Intel] instructions, and sequences of x86 (Intel) instructions, into the more powerful VLIW instructions.

"The code morphing software also then reorders and reschedules the instructions," he said.

The patent is enigmatically described as "Memory Controller for a Microprocessor for Detecting a Failure of Speculation on the Physical Nature of a Component Being Addressed." It was issued on November 3, 1998 and cites "inventors" Edmund J. Kelly, Robert Cmelik, and Malcom John Wing.

Paul Allen is an investor in CNET: The Computer Network, publisher of News.com.

-------------------------------------------------------