Intel Gets Serious With Solar-powered CPU Tech

angry tapir writes "Intel's experimental solar-powered processor may have started off as a fun project, but the chip maker is now looking to extend the technology to hardware such as graphics processors, memory and floating point units. Intel last year showed the low-power processor — charged only by the light from a reading lamp — running Windows and Linux PCs. Intel is expected to share further details about the processor, which is code-named Claremont, at the International Solid-State Circuits Conference in San Francisco. The company is also expected to reveal information about efforts to integrate wireless capabilities into Atom chips for mobile devices."

Comments

Actually Solar is not the quest here folks...

[icebike]

Yes Intel did demo a solar cell powering a Pentium, but that was merely to make a point about the inefficiencies of near-threshold voltage (NTV) CPUs. They have no particular focus on Solar powered processors.

Near-threshold voltage (NTV) CPUs are the focus of Intel's research here.
NTV transistors can switch at voltages just the threshold for the device's powered state, and CPUs made of these can idle along at extremely low voltage doing real work (slower) or they can ramp up the power and work much faster.

The Register has a much better explanation of this technology than the linked article.

The idea is to have devices run at low voltages and power consumption rates that would be akin to a sleep mode in today's chips. And NTV techniques are not just limited to processors used in hand-held devices like smartphones and tablets, but to everything all the way up to exascale supercomputers, says Rattner. The important thing is that NTV techniques allow a chip's performance and power to scale as voltage scales up and down, and to do so across a wide dynamic range.

Also a good summary here:

Marketing spin aside, the "near-threshold voltage" chip is quite an achievement. Intel first revealed in March 2010 that it had a prototype chip running at such low voltages, but Claremont's creators took that technology and baked it into a full IA architecture processor. Based on a Pentium core, Claremont can not only be throttled down to "within a couple of hundred millivolts of the threshold voltage of the transistors," said Intel engineer Sriram Vangal, who demoed the chip during Rattner's turn, but – equally important – it also has a high dynamic range that allows it to be cranked up to deliver ten times the low-power performance by increasing the voltage.

Once again, the Register does a better job of reporting than Techworld.

Re:Actually Solar is not the quest here folks...

[icebike]

Essentially a throttle, but more likely a demand based system, such that non-busy processors can run at the lowest possible speed and voltage, and when work stacks up, it ramps up.

Great for the smart phone in your pocket which has nothing to do for hours at a time other than check the email and listen for calls.
Since its screen is off, you really don't care how fast it does those things as long as they are just barely fast enough.

There is a great deal of "stare time" that happens when people look at computers, and the processors are spinning away all the time while you are reading this. They could just as well drop to an extremely low power state, and wait for a mouse move, finger tap, or something else.

This much we've been doing all along, for the last 20 years. But power consumption still remained high, because even simple tasks like checking the clock to see if its time to increment that digital time read out took processing power, and historically any use of the processor kept it awake at something like full power for that task.

Now, those tasks can be performed at extremely low power, without ramping up the speed. Only when the processor can't meet the demand would the system increase the voltage and speed up the chip.

It's a Race

[GLMDesigns]

So many people are worried about how technological advances are ruining the environment. What many often forget is that technology is also the answer (unless you want to go back to a hunter-gather lifestyle and I hear that the drum/smoke-signal bandwidth really sucks, it's takes forever to download the latest movie.)

We're in a race - computational speed, new materials, new efficiencies versus the rate in which we're polluting the environment. Many things make me optimistic: photovoltaic paints for one - and now processing power so efficient that it can be solar powered. Wow. We may win this race after all. .

Re:It's a Race

[jamesjw]

In regards to your second statement regarding noise pollution, I believe AC/DC found an alternative to this issue, they formulated in the 1970's that Rock And Roll was not Noise Pollution and was sustainable indefinitely (Aint gonna die.) :)

Re:It's a Race

[ZankerH]

How many FLOPS does that get you at peak performance?

Saying the human brain is "more powerful" makes no sense by itself. It's better at certain tasks (like pattern recognition, jumping to conclusions and holding contradictory beliefs) because it's hard-wired to do them. When it has to use general-purpose computing (like when you try to do floating-point math), you'll find most computers a great deal faster and more efficient.

Re:It's a Race

[TheCouchPotatoFamine]

the basic rule is that neural networks can solve problems without knowing *how* precisely, and digital computers can do anything if you know exactly how. See the difference? You can't compare brains and computers. They are good at diametrically opposed things and always will be. Thats the law (of physics and computation).

Process Variation is anathema to NTV

[Theovon]

I'm not sure what transistor geometry Clairmont is manufactured at, but for really small transistors (e.g. 32nm), process variation is a serious problem, making it hard to scale voltage down that low. The results are unpredictable performance from die-to-die and within die and major reliability problems. Static RAMs are hit the hardest, because they use the smallest transistors. "http://www.cse.ohio-state.edu/~millerti/parichute-camera.pdf" is an example of a paper that explores the consequences of ultra-low voltage SRAMs and tries to solve it with forward error correction.

Re:Power consumption of processor

[rrohbeck]

Power is proportional to switching frequency and to the square of the supply voltage. Reducing the supply voltage is the main vehicle to reduce power consumption, but with standard CMOS you run into the problem that transistors leak a little current when they're run at or near the threshold voltage because they don't turn off completely (you need significantly more than the threshold voltage for that.)
So in a totem pole circuit (used in standard CMOS) current leaks straight from Vcc to ground - not good. They must have designed some tricky circuits that avoid this current path although the transistors are still conducting a little.
Of course the real reason behind this is that even standard CMOS designs suffer from leakage -- the smaller the more leakage -- so they can apply these techniques to standard designs as well. That will probably be a necessity at some point beyond 22nm.

Re:Intel late to the game for low power consumptio

[DigiShaman]

And here I was thinking that the first solar powered calculators were made in 1978. They have a CPU too, right?

Re:Interesting, but...

[Kell Bengal]

Imagine CPUs that run at 1.1V... you could power them with a potato!

Re:Interesting, but...

[mmmbeer]

At Wheatley Laboratories, CPU *is* potato!

Re:Power consumption of processor

[dkf]

Reducing the supply voltage is the main vehicle to reduce power consumption, but with standard CMOS you run into the problem that transistors leak a little current when they're run at or near the threshold voltage because they don't turn off completely (you need significantly more than the threshold voltage for that.)

Of course they do; CMOS transistors are analog circuit components. Yes, usually they're driven into a state where their non-linearity makes them behave almost like binary components, but they're very much not that. The closer you drive them to the limit, whether through raising the speed or through lowering the voltage, the more they behave like the analog devices they truly are.