The basic rules by which chips are being designed could be wrong, physicists have warned.
Researchers at the US National Institute of Science and Technology have warned that a flaw exists in transistor noise models which could fundamentally affect the efficiency of future chips.
The elastic tunnelling model predicts that as transistors get smaller, electronic noise within the transitors, which can cause erratic on-off states, should increase. However, a team of Nist scientists, who have been exploring nano-scale transistor behaviour, have found that transistor noise does not increase as transistors are scaled down.
"This implies that the theory explaining the effect must be wrong," said Jason Campbell, who lead the research, in a statement. "The model was a good working theory when transistors were large, but our observations clearly indicate that it's incorrect at the smaller nanoscale regimes where industry is headed."
The team also discovered that as less energy is pushed through nano-transistors, transistor noise increases. This could spell trouble for low-energy chips, which are being explored for use in devices including laptops and phones.
"This is a real bottleneck in our development of transistors for low-power applications," said Campbell. "We have to understand the problem before we can fix it — and troublingly, we don't know what's actually happening."
Campbell credits fellow Nist researcher KP Cheung with first identifying a possible problem with the elastic tunnelling model. Researchers from the University of Maryland College Park, and Rutgers University, also contributed to the team's work.
The researchers gave a presentation of some of their findings at an IEEE event last week. The team's initial results were published in a paper entitled The Origin of Random Telegraph Noise in Highly Scaled nMOSFETs in February.







Talkback
When I was doing serious electronic in the 70's I found that transistor noise was a pain. This was made worse when two or more transistors were in close proximity and I resorted to tuning out the noise with small inductors.
Translating this in to nano-tech could require designing ample spacing out of transistors and perhaps developing nano-inductors.
Interestingly I found that the best attenuated square wave peak was when there were only three connections: signal in; ground and signal out. You may well ask what about the positive power connection? The answer is that it is not necessary when clocking at high frequencies as the signal in connection provides the positive potential.
Splitting the signal in with phase shift could provide even greater stability as noise is cancelled out. This is is the same principal as noise cancelling headphones.
This is an oversimplification and would require greater explanation but is all about adapting known strategies.