The Not-So-Cool Future

markmcb writes "Researchers at Purdue University and several other universities are looking to start work on a major problem standing in the way of future chip design: heat. The team is proposing a new center to consolidate efforts in finding solutions for the problem that is expected to become a reality within the next 15 years as future chips are expected to produce around 10 times as much heat as today's chips. The new center would work to develop circuits that consume less electricity and couple them with micro cooling devices."

Comments

But think about the,,,

[Deltaspectre]

Think about the people up in northern Canada, who need that precious heat! Unless this is some evil conspiracy to kill them off?

Nothing new

[koreaman]

What this boils down to is "researches are looking at ways to make cooler chips." Well, duh, haven't they always?

Re:Nothing new

[lrichardson]

A few years back, I read a couple of articles about reversible chips ... run the op through one way, store the results, then run the exact mirror back through. Net heat result was (theoretically) zero. Reality was about 1-2% of regular heat build-up. But I haven't heard anything more on this. Sure, it effectively halves chip speed. And, even at the time, I thought it would be insane to engineer with the pre-emptive tasking coming into vogue. But something that drops heat production by two orders of magnitude seemed worthwhile pursuing. Anyone else heard where this research is at?

Re:Nothing new

[eliasen]

Why is the parent moderated funny?

Reversible computation is quite real, but it doesn't work in the way you explained. You don't need to actually run the computation backwards. To make a long story short, the only time that a reversible computer needs to expend energy as heat is when it's producing output, or setting/clearing variables to a known state. And then, it only requires energy proportional to the number of bits being output, and the temperature. So if you're testing whether a billion-digit number is prime, the entire calculation can take zero energy, except for the one bit of output.

Unfortunately, to get truly reversible computing, the computation has to be done arbitrarily slowly.

If you don't have it, Feynman Lectures on Computation has one of the clearest discussions of reversible computation. Very highly recommended, and fun. We're 35+ years past the time when Feynman made these lectures, and we're still nowhere close to the limits or the technology that he described. Techniques for varying the power supply on the chip alone would very greatly reduce energy usage.

Photonic chips?

[Mysticalfruit]

I thought the future of processors was going ot be photonic processors. I'm not sure if these will be producing any heat or not.

Re:Photonic chips?

[Rorschach1]

Thermodynamics makes sure of that.

Minimum Energy Requirements of Information Transfer and Computing

Re:Photonic chips?

[Have Blue]

Everything that performs work produces heat. This is what we mean by "nothing can be 100% efficient".

Re:Photonic chips?

[marcosdumay]

The tecnologies we have now for fotonics produce an incredible amount of hot (if you use milions of switches). Can't compete with CMOS. And there is no teoric limitation on either field that makes one more attractive than the other for low consumation.

Not Cooling

[LordoftheFrings]

I think that the solution to the heat problem will not come with better and more powerful cooling solutions, but rather radically changing how chips are designed and manufactured. The article doesn't contradict this, but I just want to emphasize that. Having some liquid nitrogen cooling unit is not the optimal, or even a good solution.

diamond cooling

[myukew]

they should look for ways to mass produce cheap diamonds.
Diamonds are about five times better at heat conducting as copper and could thus be used for passive cooling.

Re:diamond cooling

[Cheap Imitation]

The ultimate way to propose to that geek girl you love... a diamond engagement heatsink!

Re:diamond cooling

[LiquidCoooled]

Actually, diamond is looking better and better for use as a replacement for silicon.

see here for more info.
(This was reported extensively at the time)

Re:diamond cooling

[kebes]

Actually many researchers are in fact seriously pursuing using diamond as a future replacement for silicon. Both diamond and silicon are *very bad* conductors in their pure state. Both have to be doped (with phosphorous, boron, etc.) to become p-type or n-type semiconductors, which makes them useful as a substrate for microprocessors (note that when doped they are semiconductors, not conductors... your microchip would just short-out if the entire wafer was made of a metal/conductor).

Diamond's superior thermal, optical, and chemical-resistance properties make it attractive for future microprocessors... but unfortunately it is more difficult to make it work as a semiconductor, which is why silicon has always been the substrate of choice.

It's very interesting research, and we'll see where it goes. For more info, this C&E News article is good, or check here, or here and there's a bit here.

Re:diamond cooling

[N3Bruce]

Being able to conduct heat internally is a major asset. Conductivity of heat is based on the difference in temperature between the heated end and the unheated end of a material of a given shape and surface area. Think about this junior high school level experiment with a cigarette:

A 1 gram mass of loosely packed tobacco is wrapped into a paper sleeve .5 cm in diameter and 10 cm long and is a very poor conductor of heat. A match is applied to one end for a few seconds, causing the tobacco to smoulder red-hot, while the other end is cool enough to touch. The small area of combustion is kept warm enough to sustain combustion by the insulating properties of the tobacco and ash surrounding it.

If you repeat this experiment with an aluminum rod of the same size, such as an aluminum nail, the heat from the match would quickly conduct the entire length of the rod, making it hot to touch within a couple of seconds. While the rod would get hot enough to be uncomfortable to hold, the end which was heated by the match would definitely not be hot enough to light a cigarette, unless the whole rod was heated red-hot.

This simple experiment demonstrates the limits of heat sinks. While aluminum is a good conductor, it isn't perfect. The area closest to the heat source will always be hotter than the areas near the ends of the fins. The quicker heat can be conducted away from the area next to the heat source, the cooler that area stays. A heat sink made from a material that is a perfect heat conductor will have a uniform temperature throughout, and keep the temperature next to the heat source the same as the tips of the cooling fins.

Junction temperature of electrical components is the critical parameter in heat sink design. A heat sink today may have a temperature of say 50C at the ends of the cooling fins, but be 200 degrees at the chip/heatsink interface, which is a guesstimate of the maximum safe temperature of a junction. A perfect heatsink material might only need to be half the size or less to keep temperatures at safe levels. Heatsinks to dissapate larger amounts of heat could be scaled more easily than is currently possible.

1kW?!

[AaronLawrence]

("ten times as much heat as today's processors")
I don't think that 1kW processors will be practical. Nobody is going to want to pay to run that, and nobody will want a heater running in their room all the time either.

I'd say that they should be looking to limit it to not much more than current figures (100W) - maybe 200W if we are generous. After that it gets silly.

Re:1kW?!

[kebes]

FTA:
Current chips generate about 50-100 watts of heat per square centimeter.
"But in the future, say 15 to 20 years from now, the heat generation will likely be much more than that, especially in so-called hot spots, where several kilowatts of heat per square centimeter may be generated over very small regions of the chip..."

Let's not confuse power with power density. When the article says "10 times the heat" they mean kW/cm^2, not kW. Chips of the future will generate a few kW/cm^2 of heat in their hottest spots, but they will still be supplied from conventional 200W power supplies that run off of normal 120V power lines. It's the dissipation of so much heat in such a small area that is the issue, not the raw amount of energy being consumed.

So, again, it's not the the processor will draw 1 kW of power (it may draw considerably less), but rather that it's hottest spots will need to dissipate ~1 kW/cm^2 (i.e.: 1000 joules of heat per second per square centimeter).

Breeze

[MikeD83]

"Meanwhile, the cloud of electrons would be alternatively attracted to and repelled by adjacent electrodes. Alternating the voltages on the electrodes creates a cooling breeze because the moving cloud stirs the air."

Amazing, Purdue is developing the same technology used in such high tech devices as the Ionic Breeze air purifier.

Hot and bothered!

[3770]

Not that I claim to have a solution to the problem with overheating processors. But the power consumption of computers are starting to bother me.

I used to want the fastest computer around. But a few things have changed I guess.

First of all computers are starting to be fast enough for most needs.

Secondly, the way I use computers has changed with always on Internet. I never turn my computer off because I want to be able to quickly look something up on the web.

I also have a server that is running 24/7. Most of the time it is idling, but even when it is working I don't need it to be a speed demon.

So it is starting to be really important for me that a computer doesn't use a lot of power. I don't know if it affects my electric bill in a noticeable way, but it feels wrong.

Re:Hot and bothered!

[Hadlock]

So it is starting to be really important for me that a computer doesn't use a lot of power. I don't know if it affects my electric bill in a noticeable way, but it feels wrong.

well a quick google says it's about five cents per kWh... assume your server spins down the disk drives when idling, and your monitor turns off when not in use; you're probably averaging 200watts an hour. That comes out to be abour $6.72/month in electricity, or $80 per year.

If you're looking for power savings, an old laptop with an external hard drive only consumes about 15W at idle... or about $6 per year. In what you spend in two years running you "server" you could have a decent laptop + gianormous 120 gig external drive as your server, and look things up "instantly" from your bedside.

Screw this

[Timesprout]

We need to start working on the next generation of gerbil powered chips asap!!

Alliances

[Brainix]

The alliance proposed in the article, to me, seems similar to the AIM Alliance of the early 90s. Several companies united in a common goal. I've heard the AIM Alliance failed because competitors united in a common goal remain competitors, and as such tend not to fully disclose "trade secrets," even to further the common goal. If this proposed alliance takes off, I fear it will suffer the same fate as the AIM Alliance.

But can you make a cluster of them...?

[ites]

Not a joke.

The future is multi-core / multi-CPU boards where scaling comes from adding more pieces, not making them individually faster.

Yes, chips will always get faster and hopefully cooler, but it's no longer the key to performance.

hardware DRM

[GoatPigSheep]

When I think of future problems that will happen to hardware, Hardware DRM comes to mind.

heat has already been MOBO issue

[KarmaOverDogma]

Especially for those of us with newer motherboards who want a completely silent system with as few fans as possible

First it was CPUs with cooling and big/slow/no fans and big heatsinks, then PSUs GPUs and now MOBOs. My current custom box (now 14 months old) was built to be silent and I had a hard time settling on a motherboard that was state of the art, stable, and still used a passive heatsink to cool the board chipset fan-free. I finally settled on an Asus P4P800.

I can definately believe heat becoming even more of an issue. For those of us who want power/performance and quiet at the same time, this will become even more of a challenge as time goes on. I for one hope not to rely on expensive and/or complicated cooling devices, like peltier units, water pumps and the like. I hope the focus is on efficient chips that only clock up/power up as they need to, like the pentuim M.

my 2 cents.

10 times more heat?

[kennycoder]

Whoa that's cool, now it means no more petrol is needed.

If i take out my CPU cooler it reaches about 100'C. Now lets see, 100 x 10 = 1000'C in only 15 years of chip industry. If we manage out to put this heat into work, lets say we can have 'PC + hairdryer' packages or 'PC + free home-heating' winter offers or even 'PC - burn-a-pizza' boxes. Think about it, its only good news.
Funny, -1

Comment removed

[account_deleted]

Comment removed based on user account deletion

Why is heat reclamation not worth it?

[EbNo]

I'd like to hear from some engineering types about why we can't use the excess heat from CPUs to do useful work. I know virtually all large-scale methods of generating electricity involve generating large amounts of heat through some process (nuclear reactions, burning coal or oil, etc), using it to create a hot gas, which turns a turbine, generating electricity.

I also have some vague handwaving idea that there are processes for generating electricity that have to do with harnessing temperature differentials, but I really don't know what I'm talking about.

Anyway, why can't we have little gas turbine generators (or some other method) in our machines that reclaim some of this lost energy, instead of wasting it? Seems like the aggregate energy amounts would be pretty large.

Re:Why is heat reclamation not worth it?

[kebes]

In principle, yes, any temperature gradient can be harnessed to do some amount of useful work. Thermodynamics certainly allows this (without perfect 100% conversion, obviously).

AFAIK, it really is an engineering issue. Converting a temperature gradient to electricity works great when you have huge temperature gradients (like in nuclear reactors, coal plants, steam engine, etc.), but is not so useful in a computer tower. Firstly, the whole point of putting fins on a chip is to spread the heat out quickly, so that it doesn't build up and make the chip too hot (i.e. melt it and stuff). So for our chips to work, we can't run them any hotter than 60C (or maybe 100C or whatever). The gradient between 60C and room temperature, over a few centimeters, is not that great (imagine putting a paddle wheel above your CPU, and letting the current of up-flowing air turn it... now imagine how much useful work that puny paddle wheel is really going to do). If you actually built a device to extract that energy, it wouldn't be worth it. It would take a 1000 years (or whatever) of running it before the electricity savings would offset the cost of having built that little device.

So even though in principle you're right, in practice (from an engineering perspective) there's no economic advantage to doing this.

Another fun-fact is that it takes about ~7 years of using a solar-panel before the energy savings offset the production cost. So solar panels that burn out before this mark are actually *worse* for the environment that getting electricity from coal (or wherever)... (because producing a solar panel also pollutes the environment) Solar power is only going to be viable if they are either 1. cheaper or 2. longer-lasting or 3. more efficient than they are now (all of the above would be great).

Lastly, thermodynamics guarantees that in the winter, in a cold place, it's impossible to waste electricity (if you have a thermostated heating system). Basically any inefficiency in your home (be it from your vacuum cleaner or computer) ends up as heat, which makes the house warmer, and makes the thermostat's job a little easier. In the summer, however, it really is wasted energy.

Re:Why is heat reclamation not worth it?

[zippthorne]

The maximum amount of useful work you can extract from a heat engine with two temperature pools has been derived and is known as Carnot Efficiency:

eta = (Thot - Tcold)/Thot.

using absolute temperatures (Kelvin or Rankine)
So assuming the limit is Thot = 60C = 333 K and Tcold = 25C (average room temp) = 298 K, The maximum efficiency would be 10%. Assuming further that 100W is lost by the chip alone, only 10W would be potentially recoverable. Unfortunately it gets worse: The Carnot cycle is theoretical and no real carnot engine could ever be produced. There are some very efficient cycles available (stirling and rankine come to mind) however none can exceed the carnot efficiency.

It also gets worse as you make the engine smaller. Consider the tolerance of pistons or turbines. Suppose you must leave 1mm of gap between surfaces. For large engines this is no problem, but as the machines become smaller, the minimum gap becomes a greater percentage of the total area.

Machines to extract energy from such a small source at such a low temperature difference have significant theoretical inefficiencies before you even get to the practical ones. This does not mean that you can't recover any of the "wasted heat" but only that you've pretty much gotten all the useful work out of it that you can and recovering the rest would be very impractical.

Have you ever eaten a lobster? did you suck the meat from the legs?

Re:Why is heat reclamation not worth it?

[shadow_slicer]

"The gradient between 60C and room temperature, over a few centimeters, is not that great"
yeah, but it it's over a few nanometers it's pretty big. If we built a generator on that scale it might be worthwhile...

"Another fun-fact is that it takes about ~7 years of using a solar-panel before the energy savings offset the production cost."
Where do you get this from? I keep seeing that argument over and over again, but I can't seem to find any data to back it up.
A little googling, found this:
http://www.thecomma.co.uk/globalism/

"Lastly, thermodynamics guarantees that in the winter, in a cold place, it's impossible to waste electricity"
I call BS. Most home heating is not by resistive heating, but through heat pumps which are thermodynamically required to be more efficient than any resistive heat losses. Heat pumps operate like air conditioners in reverse, pumping heat from the outside into the inside. This means that the energy from a heat pump only goes to moving already existing heat, so they can enjoy effective thermodynamic "efficiencies" of greater than 100% (which aren't real efficiencies, because they don't take into account the heat drawn from the environment, and so are called Coefficients Of Performance).
A little googling provides this informative link:
http://energyoutlet.com/res/heatpump/efficiency.ht ml
In summary, that means that of the "wasted energy", you have a net energy waste of (COP_hp-1)*E_wasted in winter.

Re:A strange question, but...

[myukew]

it's the size.
compare the typical light bulb with the typical wire running through your house. the light bulb gets hot because of the thin wire.

w00t

[zionwillnotfall]

w00t, no more heaters! now we just need a new way to cool my house...

Missing an option?

[andreMA]

It sounds like (RTFA? who, me?) they're focussing on either reducing the amount of heat generated or finding ways to dispose of it more efficiently. Important, sure... but what about developing more heat-tolerant processors? If things ran reliably at 600C, you'd have an easier time moving x amount of waste heat away to the ambient (room-temp) environment, no? Proportional to the 4th power of the temperature difference, no?

Or perhaps I'm grossly physics-impaired.

Various solutions

[jd]

One "obvious" solution to the chip heating problem would be the following:

• Have a thin layer of some liquid like flourinert over the chip surface. It just has to conduct heat well, but not electricity.
  
• Put a Peltier device in contact with the top of the liquid. Peliters are metal, so that's why you want the electrically insulating layer.
  
• Have the top layer of the Peltier device double as a cold-plate.
  

This would let you get all the benefits of existing tried-and-tested cooling methods, but would eliminate the bugbears of the chip's casing being an insulator and the possibility of condensation screwing everything up.

A variant on this would be to have the chip stand upright, so that you could have a cooling system on both sides. The pins would need to be on the sides of the chip, then, not on the base.

A second option would be to look at where the heat is coming from. A lot of heat is going to be produced through resistance and the bulk of chips still use aluminum (which has a relatively high resistance) for the interconnects. Copper interconnects would run cooler, and (if anyone can figure out how to do it) silver would be best of all.

A third option is to look at the layout of the chips. I'm not sure exactly how memory chips are organized, but it would seem that the more interleaving you have, the lower the concentration of heat at any given point, so the cooler the chip will run. Similarly for processors, it would seem that the more spaced out a set of identical processing elements are, the better.

A fourth option is to double the width of the inputs to the chips (eg: you'd be looking at 128-bit procrssors) and to allow instructions to work on vectors or matrices. The idea here is that some of the problem is in the overheads of fetching and farming out the work. If you reduce the overheads, by transferring work in bulk, you should reduce the heat generated.

Re:Expect to see Asynchronous Processors instead

[dfghjk]

"And as a general rule, RISC processors are more efficient than CISC processors running at the same clock speed"

Where did that "general rule" come from? It's nonsense.