Electronics Weekly Magazine
Loading
You are in:  Research | Process R&D

Sign-up for newsletters:

Electronics Weekly newsletters - Sign up for Made By Monkeys, Mannerisms, Gadget Master and Daily and Monthly newsletters

'Nano-scale hell of physics' is upon us

Friday 20 May 2005 10:57

The days of happy scaling are in danger, and the nano-scale hell of physics is upon us, the MEDEA+ 2004 Forum was told by Professor Hugo De Man, co-founder of IMEC.

ElectronicsWeekly.com  
“Leakage is a threatening enemy," stated De Man, “Moore’s law for power goes the wrong way.”

The period from 1980 to 1995 was the era of happy scaling as Asics proliferated. Then, between 1995 and 2003, there came the ‘great SOC euphoria’ when Asics evolved into flexible platforms for domain specific embedded systems.

Now, asked De Man, ‘Is the era of happy scaling over?’ The conclusion he came to was an unequivocal ‘Yes’ - unless radical things dramatically change.

With microprocessor power up at 100W to 150W, and doubling every three years, the industry has to find new architectures.

“Von Neumann is a poor use of scaling – all the energy is going on the communication between the processor and the memory,” said De Man. “It’s much better to use 20 microprocessors running at 100MHz than one at 2GHz.”

The ‘threatening enemy’ of leakage has triggered a law of diminishing returns where the oxides will have to be kept larger than necessary.
 
"Leakage starts to dominate when we accept larger gate delays than performance scaling predicts," he stated, adding, “when we go to 45nm we will have to go to dual gate and fully depleted SOI”.

The ‘hell of nano-scale physics’ is hitting the industry at the same time as it is being hit by complexity issues. “We’re entering on nano-scale scaling, but if we look at design we’re into giga-complexity," said De Man.

The effects are already being seen in the commercial world. “The systems houses are stopping building Asics – their time is over – now it’s platforms, but the NRE for platforms is $50,” said De Man, “30 per cent of the NRE cost goes into software and this could go to two thirds.”  He complained: “Software moves up to the seventh heaven where I can’t understand it myself.”

“The semiconductor and fabless companies have to be specialists in a domain," added De Man, “they have to seek alliances with the systems guys.”

The problem for companies wanting to develop platform products is: “A lack of tool support at the platform level,” he said, adding “there’s no money in developing EDA tools any more – so who will do it?”

The answer to the process technology/design complexity crises, reckoned de Man, is a drastic one: “We need to lock up the hardware designers and energy-aware software designers with nano-scale aware architects and CMOS scalers under one roof and take time to talk to each other!”

The effect of this may be that fully integrated manufacturers – those semiconductor companies with their own fabs - may benefit at the expense of companies operating under the fabless/foundry model.

“IDMs have an advantage because they have all the people under one roof," concluded de Man, “even if they don’t talk to each other!”

www.imec.be
www.medea.org

 

Comments powered by Disqus

Share the content

Most Viewed

Products

Latest Jobs

Resources