What Moore Didn’t Tell Us about ICs
Gene Frantz
TI Principal Fellow and Business Development Manager, DSP
Now that we’ve painted some broad strokes about SoCs and the applications that use them, I’d like to take a step back and fill in some of the gaps. Three questions that keep us up at night: What are the physics that are driving the move to SoCs? How do we stay ahead of the limitations of those physics? And, just what are SoCs, anyway? My next few postings will attempt to answer these.
Let’s start with the basics. In 1965, Gordon Moore noted that the transistor density on integrated circuits had doubled every year since the IC was invented. Moore predicted that this trend would continue for the foreseeable future but would probably slow over the long term. In subsequent years, the pace did slow down a bit, but performance has doubled approximately every 18 months, and this is the current interpretation of Moore’s Law.
Feeling that power was being left out of the conversation (and noting that I didn’t want to have to carry a car battery around with me to use a portable device), I added my own codicil to this equation. Known as Gene’s Law (thank you very much), this stipulates that power dissipation will be reduced at roughly the same rate that performance increases.
In terms of cost, we have seen a 70 percent reduction as transistors double from process node to process node. And frankly, cost is probably more important than performance or even power.
This all seems very academic, and most experts see Moore’s Law continuing to follow the path it has held for nearly half a century. But there are things that Moore did not tell us about ICs that threaten the validity of his law.
In terms of performance, as measured in terms of clock speed, we effectively fell of the 18-month cycle of doubling nearly a decade ago. Then, how does performance continue to increase, I hear you ask. Well, we’ve managed to compensate for it with innovations in architecture such as deeper and deeper pipelines. And I’ll explain this more later.
The latest (and in my opinion the worst) casualty of Moore’s Law is power dissipation. The problem of leakage threatens. As IC technology has advanced, leakage power has increased significantly and the trend is to continue to increase. We can usually handle power leakage rates when they are insignificant compared to the active power dissipation, but we are beginning to see leakage power becoming equal to the active power dissipation. Unchecked, this trend will continue to plague us.
So, are we doomed? Are there answers here? Stay tuned.