Scaling Problem in Semiconductor Industry

Atoms don't scale, and this is the ultimate barrier to the continuation of Moore's Law, Bernie Meyerson, VP and chief technologist, IBM’s Systems and Technology Group, told the Globalpress Summit Conference in Monterey Tuesday.

"Whether that barrier happens in 2016, or 2020, or 2024 is under debate," said Meyerson, who was of the opinion that scaling would deliver 6nm processing in 2020.

While scaling is not the immediate problem, improving performance is. "I could make a 20GHz core so long as it was used to do absolutely nothing," joked Meyerson.

The constraint put on IC development by skyrocketing standby power meant that performance improvements can only be delivered by innovations such as strained silicon, design for manufacture, new dielectric materials, massively parallel architectures and many other factors.

The need for innovation has produced a huge increase in complexity in designing products. Meyerson gave an example of the problems faced when designing at very small geometries.
"Random fluctuations in the number of atoms in the dopant can affect the behavior of the device. You have to allow for that, or the circuits won't work," explained Meyerson.

Meyerson said that IBM would continue to develop the PowerPC core as well as the Cell processor because Cell is a chip architecture at the center of which is the PowerPC core. Cell surrounds that core with eight processing elements.

The next generation, Power6 will have a frequency between 4GHz and 5GHz and will be out in systems next year, said Meyerson.

(Thanks to Poorvaja kamalapuri for sending me the article)