Sony obtains Transmeta's LongRun2 license

[one]

http://www.eet.com/semi/news/showArticle.jhtml?articleId=57703131

Sony takes license, backs Transmeta move back into IP

By Peter Clarke
Silicon Strategies
1 24, 2005 (7:31 HM EST)

LONDON — Global consumer electronics company Sony Corp. has licensed technologies for power management and transistor leakage control from Transmeta Corp. (Santa Clara, Calif.), joining a band of companies which have licensed technology from Transmeta.

Transmeta revealed on Friday (Jan. 21) that on the previous day the company had signed its third LongRun2 licensing agreement as it confirmed a move away from selling processors as a fabless chip company, and a move back towards an intellectual property (IP) business model.

That global consumer electronics company would appear to be Sony. Financial details of the deal were not revealed.

Sony has its own aspirations in the microprocessor market being on the verge launching its "Cell processor" in concert with Toshiba Corp. and IBM Corp., and may prefer to have Transmeta as a licensor partner rather than a potential competitor for some design slots.

LongRun2 technologies adjusts transistor threshold voltages to control transistor leakage in response to changes in runtime conditions, such as voltage and temperature, which are not predetermined when the chip is manufactured, the company said. "Transmeta's LongRun2 technologies represent a major practical innovation that addresses the problems of exponentially rising leakage and power consumption facing the semiconductor industry," said Bijan Moslehi, chief technology officer of The Noblemen Group, in a statement issued by Transmeta.

One of Transmeta's recent licensees is Fujitsu, which until recently was a foundry manufacturing partner.

Transmeta started out as a technology licensing company in the late 1990s striking deals with IBM Corp. and Toshiba Corp. which it revoked in 2000. At that time the company bought back its two technology licenses so that it could pursue a fabless chip company model.

Sony is the third company that was licensed for LongRun2 after NEC and Fujitsu. Hmm, PSP-2000 already in the mill?

 

[Farid]

It's not necessarily PSP related, actually i'm almost certain this deal is not PSP related, but the (very neat, BTW) technology might spawn on the PSP2 that's a sure bet, as long as the licensing deal is still up by the time the PSP2 hits the shelves
(Or PSPSP maybe, Ken Kutagari talked not so long ago about PSP battery life, in the future, lasting as long as a Tokyo-whatever* flight trip).


*In this season, whatever is rumored to have a beautiful weather...
Ok, I forget the name of the place KK cited.

 

[Guden Oden]

Clockrate-controlled voltage was already in use *before* Transmeta "invented" and patented it - what exactly is it they're licensing away here - their marketing name for it? :?

 

[Farid]

Clockrate-controlled voltage was already in use *before* Transmeta "invented" and patented it - what exactly is it they're licensing away here - their marketing name for it? :?

They "refined it" would be the term they would use, also check the infos about this technology, it's, logically, more advance than what we previously saw in the market. And it works quite well.

Also, keep in mind that Sony/NEC/Fujistu might do not want to develops theirs own solution (For various reasons), and/or others solutions available for licensing (If any) might not be interesting, so this might be the real reason behind this deal.

And about your second point, of course, it's marketing fluff powered.

 

[psurge]

Guden: according to the links, it reacts to temperature changes as well as voltage changes dynamically... Anyway, I am not at all familiar with process tech, so to anyone that is:

Is this at all similar or part of what Intel is doing with Montecito (dynamically adjusting clock rate and voltage to stay within a given power budget)?

Serge

 

[one]

The official press release came here.

http://investor.transmeta.com/ReleaseDetail.cfm?ReleaseID=153476

Though I picked up PSP as a possible candidate (Transmeta founder D. Ditzel was at the PSP session at Hot Chips 2004), LongRun2 has more usage than just to reduce power consumption in processors for handhelds. Simple clock & voltage control is what LongRun1 does.

http://investor.transmeta.com/ReleaseDetail.cfm?ReleaseID=119967

Transistor leakage is becoming an increasing problem as semiconductor technology scales to smaller dimensions. Industry experts have called transistor leakage one of the fundamental challenges to Moore's Law of technology scaling. Leakage problems are expected to get progressively worse as the industry scales to 90nm and 65nm transistors. Leakage power could easily dominate total chip power and prevent low power standby operation if not controlled.

Transmeta's first generation LongRun® power management technology was introduced in January 2000, and was the first technology in the industry to adjust MHz and voltage dynamically, hundreds of times per second, to reduce power consumption. Transmeta's second generation LongRun2 technology extends this approach further to include dynamic adjustments of transistor leakage under software control. Software control is important in order to adjust leakage due to changes in runtime conditions, such as voltage and temperature, that are not predetermined when the chip is manufactured.

Transmeta's new LongRun2 technology is able to control transistor leakage through software while a chip is running. Transmeta's LongRun2 software works to control leakage as an interdisciplinary solution in combination with special circuits in the Efficeon processor, and with a standard CMOS process. During the demonstration at the Microprocessor Forum conference, Transmeta showed the Efficeon processor adjusting leakage up to hundreds of times per second while playing a video game, playing a DVD movie and going into standby. In standby mode, Efficeon core leakage power was reduced by approximately 70 times by using LongRun2 technology.

With this leakage current control, higher clock speed is achievable with the same amount of power consumption under smaller process technologies (90nm, 65nm, 45nm...).

 

[aaaaa00]

During the demonstration at the Microprocessor Forum conference, Transmeta showed the Efficeon processor adjusting leakage up to hundreds of times per second while playing a video game, playing a DVD movie and going into standby. In standby mode, Efficeon core leakage power was reduced by approximately 70 times by using LongRun2 technology.

With this leakage current control, higher clock speed is achievable with the same amount of power consumption under smaller process technologies (90nm, 65nm, 45nm...).

All the Transmeta solutions have ever done is to control power usage when less than full CPU performance is demanded.

For example, at full clock, the 1ghz Transmeta chip in my TabletPC actually has WORSE power dissipation charactistics than the Intel Pentium M, simply because Intel has the better process technology.

At low clock rates, yes the Transmeta chip uses less power, but the performance of the chip is so horrible to begin with I'd rather have the Pentium M.

Hence I find it highly unlikely that Transmeta technology will help CELL improve performance. At most it what it means is that CELL will idle really well.

 

[one]

During the demonstration at the Microprocessor Forum conference, Transmeta showed the Efficeon processor adjusting leakage up to hundreds of times per second while playing a video game, playing a DVD movie and going into standby. In standby mode, Efficeon core leakage power was reduced by approximately 70 times by using LongRun2 technology.

With this leakage current control, higher clock speed is achievable with the same amount of power consumption under smaller process technologies (90nm, 65nm, 45nm...).

All the Transmeta solutions have ever done is to control power usage when less than full CPU performance is demanded.

For example, at full clock, the 1ghz Transmeta chip in my TabletPC actually has WORSE power dissipation charactistics than the Intel Pentium M, simply because Intel has the better process technology.

At low clock rates, yes the Transmeta chip uses less power, but the performance of the chip is so horrible to begin with I'd rather have the Pentium M.

Hence I find it highly unlikely that Transmeta technology will help CELL improve performance. At most it what it means is that CELL will idle really well.

Huh? You are talking about Crusoe, not Efficeon I'm afraid...

 

[aaaaa00]

Huh? You are talking about Crusoe, not Efficeon I'm afraid...

What does Efficeon do that's so different from what Crusoe did regarding dynamic clock and voltage management?

 

[one]

Huh? You are talking about Crusoe, not Efficeon I'm afraid...

What does Efficeon do that's so different from what Crusoe did regarding dynamic clock and voltage management?

Leakage control, as I quoted above... When a processor is under high load, for example gaming and video editing, LongRun2 lowers Vt (threshold voltage) and increases MHz. When a processor is idling it raises Vt, and decreases leakage current and power consumption.

In addition I'm not only talking about higher clock speed of a specific processor, but also about higher yield of higher-clockable chips in the manufacturer's fab side, as LongRun2 software can negate leakage dispersion in chips from the same production line.

http://informationweek.com/story/showArticle.jhtml?articleID=18402565

Transmeta Finds A Buyer And Investor In NEC March 25, 2004

NEC has licensed technology Transmeta says improves computer performance. NEC also bought another 2% of the chipmaker.
By Aaron Ricadela
InformationWeek

Microprocessor company Transmeta Corp. says NEC Electronics has become the first company to license Transmeta technology for reducing the "leakage" of electrical current in computer chips. In addition, the companies say Japanese chipmaker NEC took an approximately 2% stake in money-losing Transmeta in December.

Transmeta, which makes low-power chips and associated software for laptops, tablet PCs, and other computing devices, last fall introduced technology called LongRun2 to address an industrywide problem caused by the leakage of electric current in chips that happens as chip components shrink to miniscule sizes. The leakage can cause computers to overheat, degrade their performance, and reduce manufacturers' percentage of usable chips in a batch. NEC Electronics says it has licensed LongRun2 for use in chips made with 90-, 65-, and 45-nanometer processes. A nanometer is one-billionth of a meter.

Leakage is becoming a "colossal problem," says John Heinlein, director of strategic-partner initiatives at Transmeta. "As [NEC] moves into these smaller geometries, they--like everyone else in the industry--are up against transistor leakage."

In addition to the licensing agreement, NEC will pay Transmeta a royalty on chips sold using LongRun2. NEC bought about 10% of 25 million shares that Transmeta put up for sale in December, according to Heinlein. Shares of Transmeta closed Wednesday up 13 cents, at $3.59.

Earlier this month, Transmeta restated its 2003 results to reflect additional expenses related to bad-debt exposure, and posted an $87.6 million loss for the year. Computer makers including Hewlett-Packard and Sharp Corp. use Transmeta chips in some of their products.

LongRun2, which Transmeta uses in its own Efficeon chips, uses hardware and software techniques to change a processor's "threshold voltage" hundreds of times per second to control the leakage of electricity from transistors. Lowering a processor's performance in some situations to check the leakage of current can result in better-performing computers and higher manufacturing yields, says Heinlein.

Transmeta isn't alone in addressing leakage. In November, Intel said it had identified a new insulator that could be used in PCs, servers, and handheld computers to reduce current leakage. Intel says the insulation material, called High-K, could replace silicon dioxide in its products by 2007.

 

[Panajev2001a]

Lowering power consumption when the transistors are not switching, in idle parts of the circuit is an important thing: Intel did the same with the Pentium-M in a way turning off units on cache misses and branch mispredictions for example.

The problem with current leakage while the transistors are not switching is VERY important as we do not want to nullify the advantage of CMOS technology, that is very low passive/idle power consumption and peak power consumption only when the tranisstor is switching.

I cannot see how Sony getting more and more know-how in this area is suddenly BAD.

Sony/SCE+IBM+Toshiba are a bit better placed manufacturing processes-wise than Transmeta ever was.

Hence I find it highly unlikely that Transmeta technology will help CELL improve performance. At most it what it means is that CELL will idle really well.

See... would CELL be able to afford high clock speeds if, caused by bad leakage issues and other problems, at all times idle transistors and active/switching transistors consumed about the same amount of power and produced the same amount of heat ? The answer is no and you know it too... so quiet on the cheap puns please .

 

[aaaaa00]

Leakage control, as I quoted above... When a processor is under high load, for example gaming and video editing, LongRun2 lowers Vt (threshold voltage) and increases MHz. When a processor is idling it raises Vt, and decreases leakage current and power consumption.

It is my understanding that Efficon still implements normal Longrun. Longrun2 isn't in any Transmeta products yet.

When a processor is under high load, for example gaming and video editing, LongRun2 lowers Vt (threshold voltage) and increases MHz. When a processor is idling it raises Vt, and decreases leakage current and power consumption.

And that's my point exactly. All of Transmeta's Longrun technologies have basically delt with optimizing processor power usage when full performance is not demanded of the CPU.

You said:

With this leakage current control, higher clock speed is achievable with the same amount of power consumption under smaller process technologies (90nm, 65nm, 45nm...).

The point is there's nothing Transmeta can do to change the laws of physics. If you need those transistors running at full clock, then they're going to leak, end of story -- Transmeta can't change that.

In addition I'm not only talking about higher clock speed of a specific processor, but also about higher yield of higher-clockable chips in the manufacturer's fab side, as LongRun2 software can negate leakage dispersion in chips from the same production line.

http://informationweek.com/story/showArticle.jhtml?articleID=18402565

LongRun2, which Transmeta uses in its own Efficeon chips, uses hardware and software techniques to change a processor's "threshold voltage" hundreds of times per second to control the leakage of electricity from transistors. Lowering a processor's performance in some situations to check the leakage of current can result in better-performing computers and higher manufacturing yields, says Heinlein.

"Lowering a processor's performance".

Like I said, Transmeta technology will make CELL idle really well, but it can't really help if you actually need the performance for something.

Panajev is right that you don't generally need all parts of the CPU running at full clock at all times -- but that doesn't change the fundemental observation that Transmeta's techonologies revolve around optimizing power usage when you don't need the full performance -- not enabling higher maximum performance.

 

[aaaaa00]

See... would CELL be able to afford high clock speeds if, caused by bad leakage issues and other problems, at all times idle transistors and active/switching transistors consumed about the same amount of power and produced the same amount of heat ? The answer is no and you know it too... so quiet on the cheap puns please .

Transmeta is doing its work by reducing the performance of the pieces of the chip you aren't using at the present time. They don't increase the maximum possible clockrate of your CPU (that's determined by the design and the process), they just make it consume less power for the same thing (by twiddling with the voltage and the clockrate).

If that lets you in practice raise the clock of your machine because you have limited cooling/power for your solution, that's a second order effect -- the maximum possible clockrate is set by the process and your design.

 

[one]

"Lowering a processor's performance".

Like I said, Transmeta technology will make CELL idle really well, but it can't really help if you actually need the performance for something.

How many times do I need to repeat this?

Leakage control, as I quoted above... When a processor is under high load, for example gaming and video editing, LongRun2 lowers Vt (threshold voltage) and increases MHz. When a processor is idling it raises Vt, and decreases leakage current and power consumption.

LongRun and SpeedStep dynamically changes Vdd. LongRun2 changes Vt too. It lowers Vt and increases leak current when a chip needs best performance. Then check out your own first statement in this thread...

aaaaa00[/url]]All the Transmeta solutions have ever done is to control power usage when less than full CPU performance is demanded.

Isn't it wrong? I think it is.

 

[aaaaa00]

How many times do I need to repeat this?

Leakage control, as I quoted above... When a processor is under high load, for example gaming and video editing, LongRun2 lowers Vt (threshold voltage) and increases MHz. When a processor is idling it raises Vt, and decreases leakage current and power consumption.

LongRun and SpeedStep dynamically changes Vdd. LongRun2 changes Vt too. It lowers Vt and increases leak current when a chip needs best performance. Then check out your own statement...

All the Transmeta solutions have ever done is to control power usage when less than full CPU performance is demanded.

Isn't it wrong? I think it is.

In a processor with no Longrun and maximum performance with no power/thermal limitations (IE maximum possible clock desired), wouldn't you just set Vt to the lowest possible value in your design?

My memory is a bit foggy from my semiconductor classes, but I though Vt was largely defined by the physical parameters of your design and the materials you made your gate from. The only thing I can see Transmeta adjusting is Vsb (which is normally zero anyway).

 

[one]

How many times do I need to repeat this?

Leakage control, as I quoted above... When a processor is under high load, for example gaming and video editing, LongRun2 lowers Vt (threshold voltage) and increases MHz. When a processor is idling it raises Vt, and decreases leakage current and power consumption.

LongRun and SpeedStep dynamically changes Vdd. LongRun2 changes Vt too. It lowers Vt and increases leak current when a chip needs best performance. Then check out your own statement...

All the Transmeta solutions have ever done is to control power usage when less than full CPU performance is demanded.

Isn't it wrong? I think it is.

In a processor with no Longrun and maximum performance with no power/thermal limitations (IE maximum possible clock desired), wouldn't you just set Vt to the lowest possible value in your design?

My memory is a bit foggy from my semiconductor classes, but I though Vt was largely defined by the physical parameters of your design and the materials you made your gate from. The only thing I can see Transmeta adjusting is Vsb (which is normally zero anyway).

It's true that Intel desktop/server chips have fixed lowest Vt and typical embedded chips have relatively higher Vt, while Intel mobile chips have the Dual Vt configuration where some of transistors have lower Vt and others have higher Vt.

I should have been more specific when I wrote

With this leakage current control, higher clock speed is achievable with the same amount of power consumption under smaller process technologies (90nm, 65nm, 45nm...).

though I wrote in a subsequent post that

In addition I'm not only talking about higher clock speed of a specific processor, but also about higher yield of higher-clockable chips in the manufacturer's fab side

This improvement has nothing to do with idling of a processor.

Intel has a reseach in technologies called "body bias" and "adaptive body bias" which can be applied to this yield problem in smaller process technologies to the similar effect.

LongRun2 is useful to reduce the fluctuation in threshold voltage.

 

[aaaaa00]

I should have been more specific when I wrote

With this leakage current control, higher clock speed is achievable with the same amount of power consumption under smaller process technologies (90nm, 65nm, 45nm...).

Ah that's fine then.

though I wrote in a subsequent post that

In addition I'm not only talking about higher clock speed of a specific processor, but also about higher yield of higher-clockable chips in the manufacturer's fab side

I am saying Longrun2 can not directly increase the maximum possible clock rate of a design, since it can't really prevent leakage at maximum performance.

This improvement has nothing to do with idling of a processor.

Ok, I was being a little facetious with the idling remark.

Intel has a reseach in technologies called "body bias" and "adaptive body bias" which can be applied to this yield problem in smaller process technologies to the similar effect.

That's exactly what Vsb is in the equation up there Vsb = Voltage difference from source-body of the MOS.

I'm just saying Longrun2 is not a magical technology that will conquer leakage. It can be used to improve cases in which some of the time some transistors do not need to run at full voltage or clock (IE, dropping the performance dynamically), and (ok) it can also be used to compensate for fabrication inaccuracy, thermal and power supply drift and so forth to improve yields.

But when it comes down to it, your maximum clock rate is determined by your design parameters and the materials you select to build your gates -- and that is the upper limit of your performance. Longrun2 can't really do much to change that, and it is not some sort of universal solution for leakage.

Just look at your own digram:

You'll notice that the maximum clock rates converge as Vcc increases.

The point is playing with Vsb is clever and useful, but it won't change the laws of physics, nor can it significantly improve the maximum possible performance of a particular design.

 

[one]

You'll notice that the maximum clock rates converge as Vcc increases.

The point is playing with Vsb is clever and useful, but it won't change the laws of physics, nor can it significantly improve the maximum possible performance of a particular design.

To improve clockspeed it's not a magical solution. But as for improving the practical performance of a chip, it's a different story, for multicore design is the future. Intel will apply the body bias control to cores in their multicore desktop/server CPUs in the 65nm process. Likewise, LongRun2, which employs a software plus a body-bias controller hardware, can help Cell.

 

[one]

Bumping up with the new PR that specifically mentions CELL...

http://investor.transmeta.com/ReleaseDetail.cfm?ReleaseID=158938

Transmeta and Sony Group Reach Agreement on Multi-Year Strategic Alliance
Accelerates LongRun2 Adoption for Cell Derivatives; Builds Technology Collaboration in Other Areas

Santa Clara, CA - March 31, 2005 - Transmeta Corporation (NASDAQ:TMTA), the leader in efficient computing technologies, today announced that it has entered into a strategic alliance with Sony Corporation and Sony Computer Entertainment, Inc. (collectively referred to as "Sony Group"). Transmeta will provide engineering services to the Sony Group to accelerate and expand its adoption of Transmeta's LongRun2 technologies in products such as Cell derivatives, and engage strategic technology collaboration in other engineering areas.

The first aspect of this alliance enables Sony Group to strongly leverage the value of their recent license of Transmeta's LongRun2 technologies. By utilizing Transmeta's experience in advanced low power consumption semiconductor design, Sony Group can accelerate the introduction of LongRun2 and the benefits in their products. In particular, Sony Group expects to utilize Transmeta to help develop, and accelerate adoption of LongRun2 technologies into products such as Cell derivatives as well as other portable applications. This work is expected to encompass engineering on both current and future generation semiconductor process technologies.

Transmeta's LongRun2 technologies address the semiconductor industry's growing problems of excessive chip heat and transistor leakage in leading edge high performance semiconductors.

"We are very excited to be working with Sony Group on next generation technology products," said Dave Ditzel, Chief Technology Officer for Transmeta Corporation. "By working directly with Sony Group at the engineering level to integrate our LongRun2 technologies we will be able to provide the power and heat reduction benefits for Sony Group's advanced semiconductor projects in the shortest time possible."

Another aspect of this strategic alliance is the beginning of collaboration in other engineering areas, leveraging Transmeta's expertise in microprocessor design, software design and system design. This engagement centers on activities with Cell derivatives, and utilizes Transmeta's larger intellectual property foundation to enrich Sony's engineering projects.

"Transmeta's technology in building low-power consumption processors and their talented engineering team are highly respectable," said Ken Kutaragi, President and Group CEO of Sony Computer Entertainment. "We are excited to work with Transmeta as a way to further expand our capabilities in semiconductor chip design, software, power management and systems design for a broad range of products."

As part of the agreement, Transmeta will provide the services of over a hundred Transmeta engineers to work on advanced projects for the Sony Group.

 

[Titanio]

Yeah, Toms Hardware has a report too. It seems to be Cell related:

http://www.tomshardware.com/hardnews/20050331_183734.html

Seems Sony is providing some funding aswell to Transmeta.