Intel announces 45nm breakthrough

Intel has created 153Mbit SRam memory chips using a 45nanometre process, which is the next stage of miniaturisation after the 65nm used in today's leading-edge processors.

SRam chips are always the first be made on a new scale of miniaturisation, allowing designs to be honed before progressing to processors. The 45nm refers to the size of the smallest feature within the transistors that make up the chip.

The first 45nm SRam modules do include logic circuits, proving that the process will work on processors, Intel senior fellow Mark Bohr told a press briefing.

Processors using the new 45nm process will ship towards the end of 2007 and will go mainstream over the following 18 months.

Each six-transistor memory cell on the 45nm SRam chips covers just 0.346 square micrometres, so that you could pack around 20,000 of them on a cross section of a human hair. The billion-transistor 153Mbit modules measure just 119 square millimetres.

Intel is still moving processor production from 90nm to 65nm, and expects shipments of the latter to outnumber their predecessors for the first time during the third quarter of this year. The company has already shipped more than a million 65nm dual-core chips.

One of the problems of miniaturisation is that insulation between the elements of a transistor tends to break down, leading to an increase in leakage current. This means transistors draw power even in an off state, reducing efficiency.

Yet the 45nm chips can decrease leakage current by a factor of five, according to Intel figures. Bohr declined to say how: 'We are not prepared to discuss specific techniques of this technology.'

He did say that switching times could be improved by 20 per cent by sacrificing the improved leakage and that the 45nm process reduced by 30 per cent the power taken to switch a transistor.

But he could not put a figure on how these improvements would increase the overall power efficiency, a factor becoming almost as important as processing speed. 'That would depend on the design and the number of transistors used,' Bohr said.

A 32nm process is scheduled to go into production in 2009, meaning Moore's Law (that transistor densities will double every two years) will hold good for another four years.