HP has found a revolutionary new use for its classic inkjet technology: cooling microprocessors.
A team from HP Labs attacking the growing problem of heat generation in powerful microprocessors determined that HP's inkjet heads could be used to target spray cooling for chips.
Working with engineers in HP's printing and imaging group and elsewhere in the company, the researchers re-configured the inkjet head to spray tiny droplets of dielectric liquid coolant instead of ink.
"For tomorrow's high heat densities, cooling is seen as a limiter to performance," says Chandrakant Patel, principal scientist, HP Labs. "Our patented spray cooling technology could help enable future high performance computing and comunication devices."
Patel and Cullen Bash, both of the Thermo-Mechanical Architecture Lab, developed the technology to direct the coolant onto specific areas of the chip according to its heat level.
"Because of its small size and construction, HP's spray cooling can be used in a wide range of applications," says Bash. "The technology is very adapaible and flexible."
Following is the text of a recent interview with Patel, discussing
problems with existing cooling methods and explaining how
researchers came up with their invention. To see a demonstration
of the technology, go here.
background on chip cooling
problems with heatsinks
inventing inkjet-based spray cooling
cooling data centers and more
more about research lead Chandrakant Patel
what is the significance of this invention?
For tomorrow's high heat density processing and communication devices, cooling is seen as a limiter to performance. Evaporative spray cooling is a solution to the high heat density. We've taken an existing technology, the inkjet printing cartridge, and reengineered it into an efficient, inexpensive cooling device for semiconductors.
why is it necessary to cool chips? what technologies existed before your invention?
A typical microchip is about the size of a postage stamp. A decade ago, a chip this size produced roughly 10 watts of heat. Subsequently, in the mid-'90s, the high-performance microprocessors reached 100 watts of heat dissipation - akin to a 100 W light bulb with a surface area equivalent to a postage stamp.
Heat sinks and heat pipes touched down on the chip and were mechanically held to the chip surface. The heat, produced uniformly over the postage stamp size surface area, diffused across the interface. The interface produced significant temperature difference between the chip and the heat sink. However, it was sufficiently within the cooling budget.
why will heatsinks no longer be adequate?
Whereas in the past, the CPU (central processing unit) made the entire surface area of the postage stamp size chip, the CPU of the next decade will share the postage stamp real estate with other functions. In other words, the CPU core will be relegated to roughly quadrant of the first class postage stamp. The rest of the chip will contain cache memory, possibly one or more additional CPU cores, memory controller and maybe even networking and graphics functions.
Now, even if the CPU portion (quadrant of a postage stamp) of the chip produced 75 W, the heat density would become too high for mechanical diffusion of heat across the chip wall to a heat sink. Thus, while we are at approximately 40 to 70 Watts per square centimeter today, we are moving towards 200 watts per square centimeter. At 200 Watts per square centimeter, we can't use mechanical means like a heat sink for cooling. We expect chips to reach the 200-Watt level in the next three or four years.
so if heatsinks based on heat diffusion are ineffective as power densities rise, what other technologies are available?
One well-known method is phase-change cooling by pooling a cooling fluid that undergoes a phase change around the heat source. An example of this is use of inert dielectric fluids such as 3M's Fluorinert that pools around the chip. One variant of this fluid, FC-72, boils at 56°C and a chip can be safely immersed in this fluid while operational. Such phase change, utilizing latent heat of vaporization of the liquid, removes significant heat flux. It does so at a constant temperature.
but aren't there limitations to phase-change cooling with pooling of fluid as power densities rise?
Yes, that's true. There are limitations with the phase-change mechanism where one "pools" the fluid around the heat source. The issues due to pooling are greatly exaggerated with uneven heat distribution and when high heat density of the order of 100 Watts per square centimeter is involved. The problem with pooling such liquid around a chip is the formation of insulating vapor bubbles on the chip surface.
As the pooled liquid changes phase, vapor bubbles form that adhere to the wall of the chip. And for the dielectric fluids, they form really quickly. At a certain point on a tiny chip, a bubble will form on the hottest spots. If a bubble sits on top, it becomes an insulator. At that point, heat transfer through the bubbles is greatly limited, and the chip wall temperature quickly exceeds specifications.
On a future chip like the one I described earlier with multiple functionalities on a single chip, there will be areas of high heat, low heat and no heat. So in areas of high heat a bubble would form if pool boiling were employed.
how did you come up with the idea for inkjet-based spray cooling?
Now, we're not the first to use spray cooling. In fact, there are examples of use of spray cooling in the market today. But the problem is, today's spray cooling is a "shower head" approach. A pump pressurizes the nozzle and tiny droplets strike the heat source and vaporize. Control based on heat load distribution is lacking.
The chips of tomorrow may have several areas where there are high heat loads and several areas with no load. So we need to control how much we spray and where we spray.
About four years ago, anticipating these higher power densities, we thought, wouldn't it be nice if we could put just enough fluid on the chip so that it goes 'psst' and vaporizes? Wouldn't it be nice if we're able to spray just enough fluid commensurate with the heat load, so that it goes from liquid to vapor right away without formation of any bubbles? For areas of low heat loads, less fluid; for areas of high heat load, more fluid.
And it occurred to us: What about Inkjet heads? Inkjet heads have nozzle-level control. A set of nozzles can be fired and another set commanded not to fire. Different patterns can be programmed, and so on. Well, Inkjet heads are made for water. And they are made for printing with ink. The question was, can we fire with nozzle-level control with dielectric fluid? Can we do it at all? We decided the only way to figure it out was to experiment.
what did it take to develop the technology?
We looked for an InkJet pen with highest flow rate, tested it independently at HP's Corvallis facility. Then, we built a test fixture in our Thermal Sciences Laboratory at HP Labs. We simulated a chip-based heat source, and we built a system to control firing and drop pattern, and drop volume.
To make it all work, we went and got the highest-capacity pen (or inkjet head). Where are you going to find the highest-capacity pen? In HP's DesignJet plotter.
In our test stand, fluid comes in from the reservoir into our high-capacity pen. It goes into the chamber. The Thermal Inkjet technology is used to fire out of these nozzles, firing a pre-defined pattern where cooling is needed. The liquid vaporizes from the chip wall and it goes up to a condenser, condenses to a liquid, then goes to the reservoir where it's stored and reused.
what problems are you still solving?
The InkJet head is self-priming. While other spray-cooling methods require high-pressure pumps to move the liquid, we need only a small pump to get liquid up to a reservoir. That means that less power is expended for pumping.
The dielectric liquid is pretty bad for priming. Its surface tension is poor compared to water. The other problem is that even the largest printer pen is not large enough in volume flow -- it is designed for printing, not spray cooling. We need more mass flow and thus more flow volume for spray cooling to remove enough heat. Essentially, we need to design a pen specifically for spray cooling.
how does the pen know where to spray?
Right now, spray configuration and control can be programmed into the nozzle. We have developed intellectual property that will allow us to map out the heat load and spray accordingly in a closed loop automated way.
do these chip-cooling technologies have other applications?
Oh yes, this is not for chips alone. Laser diodes, today used greatly in communications applications, are very high heat density sources. Moreover, the spray technique can be used for tomorrow's extremely high heat density "bladed" servers -- we can spray the entire board without contacting it.
what else is your group working on?
We take a holistic approach to cooling computer systems -- from the source of the heat, through interface, to the box, to the room and out to the environment. One has to take such a view to enable tomorrow's cooling. In this regard, we are quite unique.
An excellent example of our approach is our data-center cooling effort. We turned our attention to data-center cooling several years ago, even before the existence of large server farms. We looked at the data center's cooling requirements in the same way we look at a single computer's requirements -- in fact we even called our effort, the "Data Center is the Computer." The walls of the data center are akin to the walls of a computer system box. Today, we have a major effort underway with respect to the Smart Cooled data center of tomorrow.
At HP Labs, it is the job of the research team to anticipate the future and act accordingly. We are greatly aided in this effort by our partners -- The HP Cool Team, which is the thermal design team made up of product divisions.
what is your vision going forward?
It is really one of saving energy by Smart Cooling of planetary scale collection of computers in the context of HP Labs Planetary Scale Computing efforts. Our objective is twofold:
- To provide the appropriate amount of cooling where and when it is needed -- that is, dynamically provision cooling in a data center based on heat loads
- Actively control the heat load by redistributing computing within data centers and in the global network of data centers, based on the most efficient cooling available.
In my mind, a control volume drawn around a planetary scale collection of data centers, the confines of which maintain balanced cooling and compute loads, saves a world of energy.
how did you get interested in cooling technologies?
It's funny, because my background is not even in cooling. It's in structural dynamics. I worked in shock and vibration of systems when working on rigid disc drive mechanisms at Memorex/Unisys and subsequently at HP's Personal Computer Group.
I got into cooling and chip packaging is in the early 1990s. I wanted to work on HP Labs' VLIW (very long instruction word) project, and the only job available was in cooling and packaging of the high-performance microprocessor. So I said I'll learn the subject. The team had confidence that I would and hired me.
After we transferred the VLIW work to Intel (where it became Intel's Itanium processor architecture) we decided that we should take what we have learned from VLIW and turn it into a competency, and that became the cooling competency.
In 1996, we invited a bunch of people from around HP working in the area of cooling to come to HP Labs to talk shop. The people who showed up, approximately 90, became a distribution list on my computer. And I named that computer distribution list "Cool Team." And that name stuck.
Today, I don't believe any computer company has such a unique, cohesive virtual community of designers of cooling solutions. The "Cool Team" is a key partner in developing the HP Labs "portfolio of cooling solutions".
The HP Cool team meets annually, has vigorous conversations through e-mail throughout the year and through smaller core team on a quarterly basis, solves immediate problems that might arise for colleagues, and above all, advises HP Labs on development of a future portfolio of cooling solutions.
how did you get interested in technology?
I've always loved airplanes. I followed aircraft design for a long time. Ever since I was in kindergarten, I've been able to draw an airplane of any kind, anywhere, on any piece of paper. I follow most key commercial aircraft developments in detail.
I remember that when I was 11 years old, I read that Air India got delivery of a Boeing 747, and I had my father take me clear across the country (from my home near Calcutta) to Bombay Santa Cruz Airport (roughly 2,000 km or 1,500 miles) to see if there was a 747 there. They had taken delivery of four of these jumbo jets, but there were none at the airport. To this date, I remember there was a Swissair 707, but no 747. It was such a disappointment!
I really enjoy mechanical design, especially aircraft design, and so that got me started in engineering. I would have gone into aeronautics, but when I came to the United States, I was not a citizen, and it was very difficult for a non-citizen to get a job in aeronautics at the time.
do you fly?
No, I do not. It is really the design of aircraft that I marvel. I think aircraft exemplifies the peak of mechanical design.
I did get to fly a 737-400 flight simulator, though. A co-worker arranged it for me. The alarms were going on all the time that the plane was flying. I was sweating. And then, when I landed finally, the instructor said, "Let me show you how it's really done."
And he took off, and flew in simulated turbulence, storm, hail, and smoothly landed the aircraft. Upon landing, he turned around and asked my little son, who was sitting in the jump seat, "Who do you think did a better job?"
And my son said, "My Dad did. The plane kept going up and down. It was a lot more fun."