Intel Corp. will outline small but important changes next week to the way it expects to make transistors in the latter half of the decade, as it continues to battle the laws of physics to build ever-faster microprocessors.
The changes will affect both the structure of transistors and the materials used to make them, and will be described by Intel engineers at next week's International Electron Devices Meeting in Washington, D.C. Some of the changes could be introduced as soon as 2005, and attempt to address two of the biggest obstacles to building faster PC chips: heat and power consumption.
"We can make transistors that are very small and very fast, but that's not good enough any more," said Gerald Marcyk, director of Intel's components research lab. "The problem is that the power consumption is rising exponentially."
Transistors are basically miniature electrical switches that can be turned "on" or "off" to represent the ones and zeros of binary computer code. Increasing the number of transistors on a chip, while decreasing their size, has been a key factor in developing faster and faster chips over the decades.
Today's Pentium 4 contains about 42 million transistors. With the number of transistors in a microprocessor doubling roughly every two years, the count should hit a billion in the second half of the decade, resulting in microprocessors that have 25 times the computing power as today's chips, Marcyk said.
At the same time, transistors have been shrinking. Already they include structures only 70 nanometers wide, or 100 times thinner than a human hair. By 2007 those structures are expected to reach 20 nanometers or smaller.
As transistors shrink, however, they start to leak electricity, and can do so even when a computer is switched off. Electricity can escape through the thin layer known as the gate oxide, which separates the transistor gate from other parts of a chip. Leaky chips require more current to switch transistors on and off, which in turn creates more heat.
One of the papers to be presented next week proposes using a new class of material, called a High K gate dielectric, to replace the silicon dioxide used today. Intel doesn't know yet exactly which material it will use -- aluminium oxide and titanium oxide are among the contenders -- but thinks the new material can cut leakage by a factor of 10,000, Marcyk said.
Another paper will describe experiments mounting each transistor on a thin layer of silicon, which in turn sits on a layer of insulating material. The resulting "depleted-substrate" transistor should further reduce electrical leakage, Marcyk said.
The depleted-substrate transistor could also help reduce so-called "soft errors," another downside of microscopic transistors. Soft errors occur when alpha particles from contaminants in the chip packaging bombard charge-carrying electrons inside a transistor, and can force it to error momentarily, by skipping a one or a zero for example.
Intel admits it has its work cut out if it is to transfer these technologies from the laboratory to its production lines. It showed a 20-nanometer transistor at a technology conference in Kyoto earlier this year, but "those are research transistors," Marcyk noted.
"We can make a heroic effort to make a small number of transistors work, but it's a far cry from the scaling we need" to support Intel's high-volume manufacturing needs, he said.
That said, Intel engineers will announce some solid test results for their latest technologies at next week's conference in Washington, D.C., according to Marcyk. Conference rules prohibit the results being published ahead of time, he said.
Users aren't likely to shell out money on faster chips if there aren't any applications that make use of the extra power, but such software may be in the pipeline. The faster chips could be used to power new interface technologies for controlling computers using speech or physical gestures, to encrypt all the data sent back and forth between PCs, or power more realistic video games.