网格计算被用来设计未来的晶片

A vast network of computers is being harnessed to design components¹ for the next generation of silicon² chips.

一组电脑搭建的巨大网路被用来设计下一代硅片的组件。

As components on chips shrink their properties become less dependable

Simulations假装，模拟 of transistors⁴晶体管 smaller than 30 nanometres (billionths of a metre) are being run on the UK e-science grid⁵, which links thousands of computers.

The results will help designers cope with the physical constraints系统规定参数 that occur when working at such tiny scales.

About 20 years worth of processing time has been used by simulating hundreds of thousands of tiny transistors.

Atomic impurity⁷

The researchers hope to get a sense of how such tiny components vary to work out the best way to produce future generations of chips with even smaller components.

"What we do in these simulations is try to predict the behaviour of these devices in the presence of atomic scale原子尺度 effects," said Professor Asen Asenov, head of the device modelling group at the University of Glasgow, which is leading the NanoCMOS simulation project.

The increasing power of silicon chips is largely dictated⁸ by the size of the components that chip makers⁹ can cram塞满，填满 on to each chunk¹¹组块 of silicon. The basic building block of a chip is the transistor³, tiny switches that can either be "on" or "off".

The current generation of chips use transistors with features around 32 nanometres in size, but many manufacturers will move to 22 nanometres soon.

"These problems started to appear a couple of generations ago but right now it's one of the most serious problems," said Prof Asenov.

"What's happening at such dimensions is that the atomic structure of the transistor cannot be precisely¹² controlled," he said. "In order to make them work we have to put in impurities杂质 to define different regions."

Prof Asenov and his team are not seeking the perfect design for a transistor, instead they are finding out how best to lay down materials so transistors perform consistently.

It used to be the case, said Prof Asenov, that silicon chips were identical and could be relied on to work in the same way. But as components shrink to 30 nanometres and beyond such certainty disappears.

No longer can designers expect to lay down crisp脆的，易碎的 ranks of perfectly¹⁴ formed transistors during manufacturing.

"Instead," he said, "designers have to introduce redundancy冗陈，备份, self-organisation and self testing."

The design and testing was done using a grid, essentially¹⁵ a piece of software that unites tens of thousands of PCs scattered¹⁶ across different sites.

Richard Sinnott, technical director at the National E-Science Centre in Glasgow, which brokers¹⁷ the grid resources for projects such as NanoCMO, said the team needed to use hundreds of thousands of hours of computer time.

"Prof Asenov wanted access to as much high performance computing¹⁸ as we could give them," he said.

His team gave them access to the number-crunching power they needed and helped them manage the huge amount of data being produced.

Over the course of a few weeks, he said, the project racked up获胜，累积 around 20 years worth of processing time as batches¹⁹ of hundreds of transistors were simulated.

"It's the biggest project we are involved with right now," said Mr Sinnott.