Steve Pawlowski, director of Intel's Microprocessor Technology Lab, knows Intel has a heat problem. As processors like Intel's Pentium 4 have increased in performance and power, they've also become generators of intense heat. Pentium 4 chips currently generate more heat than a kitchen hotplate and the company's projections show the heat generated by its processors will increase sharply in the coming years, perhaps rivalling the core of a nuclear power plant, unless solutions can be found to the problem.

The heat problem also has short-term implications. As well as dense server farms using blades, growing demand for notebooks and computers used as home entertainment centres means Intel has to find ways to reduce and dissipate heat more efficiently and more quietly than ever before. "As Intel keeps increasing the speed (of its processors) that generates more heat," said Dorothy Lai, a semiconductor analyst at Gartner. "The problem is that you can make a very big heatsink but you cannot use a very big heatsink as the product (PCs) is getting smaller."

Cooling alternatives
Faced with this challenge, Intel's engineers are looking at several different approaches to reduce the amount of heat generated by processors and to dissipate that heat more quickly, Pawlowski said. However, engineers have a limited set of options given current technologies and chip designs.

For example, they can adjust factors such as clock frequency, voltage and capacitance - the ability of a device to hold an electric charge - in a bid to reduce the amount of heat that a processor generates. "You can either drop the voltage and you get a quadratic improvement in terms of power, drop the frequency or don't increase the frequency as much and of course you can decrease the capacitance, if that's possible," Pawlowski said.

Engineers can also look to external devices that help dissipate the heat generated by a chip more quickly. In this area, Intel is investigating the use of improved cooling fans, impellers and other devices that can help keep processors cool. But these improvements are constrained by other factors, such as the packaging used to encase the silicon component of a chip.

"There's a certain transfer coefficient between the die and the package," Pawlowski said. "If you can't come up with a better package that will allow that heat transfer to occur from the die, no matter what you do on the outside, its going to be difficult. I think we're going to have to look at architectural solutions in order to try to minimise the heat as much as possible," he said.

New cooling research
Microarchitecture is important because heat is not generated uniformly across a die. A single die contains many different electrical components, such as logic and memory, which have different activity levels and generate different levels of heat. For example, memory generally has a lower activity level than logic and therefore generates less heat. By the same turn, some logic components generate less heat than others.

By spreading out the hottest components on the die, chip designers can help dissipate that heat more efficiently but this still doesn't eliminate the existence of hot spots on the die. "It looks more even but you'll still have the hot spots and those are the places you really need to worry about," Pawlowski said.

The importance of reducing and dissipating heat is changing the way that Intel's chip designers approach microarchitecture. "I think microarchitects are starting to look at heat when they do the design but that was never really a concern before; it was always performance, performance, performance," Pawlowski said.

Now that engineers are more focused on addressing the problem of heat, Pawlowski expects that cost-effective solutions will eventually emerge. "Confronted with a problem, engineers will become engineers and they'll come up with creative solutions," he said, noting that technologies such as MEMS (micro-electromechanical systems) and micro-fluidics may offer solutions to the heat problem.

Citing one example of a possible solution to the heat problem, Pawlowski said he was impressed by one young university researcher's idea for keeping chips cool by using a microthin layer of gas plasma to help dissipate heat. A gas becomes plasma when heat or other energy causes atoms in the gas to release some or all of their electrons. This leaves the remains of the atoms with a positive charge and creates freely flowing electrons, which are negatively charged. The researcher proposed accelerating plasma particles across the surface of a chip to create a micro-current to speed up the dissipation of heat.

His idea was submitted to Intel as part of a company program that sponsors research in leading-edge chip technologies. "If his test chip would have worked he probably would have won, but it didn't," Pawlowski said. "He had an arcing problem in the way he laid things out so it actually blew his chip up."

That setback doesn't necessarily mean the idea won't work. If the arcing problem can be resolved and the cooling technique is shown to work, Intel may get involved with the project at a future date, he said.