Intel red faced over breaking chip theory

Intel has attempted to retract a statement by one of its researchers who said that chips will become more unreliable as Moore's Law advances.

Intel researcher Padma Apparao made the comments in an interview with vnunet.com last week at an open day for Intel Research, and used similar wordings in a whitepaper that the company published last year.

"There isn't a Moore's Law tie-in on this," John Casey, a spokesman for Intel's Technology Leadership Group, told vnunet.com.

Apparao's research seeks to create an error analysis technology for server processors that is embedded in the firmware of a processor. The tool logs the errors that are caught by the error correction part of a chip and analyses these mistakes.

The analysis could create an early warning that a chip is starting to break down, giving the organisation a chance to swap out the component and increase server uptime.

The need for such a technology becomes greater as chips get smaller, Apparao said at last week's event.

Casey acknowledged that, as chip makers shrink the size of components on a chip and pack those parts more densely together, there is an increased risk that the components get hit by alpha particles or cosmic rays.

Such a collision inside the chip alters the outcome of calculations and could result in a system crash.

Smaller chips also run a greater risk of parts within the silicon becoming misaligned over time due to oxidation or minuscule shifts of components, causing such a high rate of faulty calculations that the chip would cease to function effectively, Apparao had said.

Even today's chips use error correction technology to prevent these problems. According to Apparao, however, the error rate could get out of hand as chips get smaller.

A chip being overwhelmed by a high error rate is unlikely, Casey contended, because the advances that Moore's Law creates also allow Intel to further bolster the error correction technologies.