[From the last episode: we looked at the concept of dynamic powerThe rate of energy consumption of a circuit while it's switching -- that is, while computing or otherwise doing work. – which applies to CMOSA way of connecting MOS (metal-oxide-semiconductor) transistors into a switch that conducts no current when sitting idle. even though static powerThe rate of energy consumption of a circuit when nothing is switching -- it's just sitting idle. is zero. (Or is it?…)]
So far, we’ve seen a pretty ideal world where, with the mere flip of a transistor switch, we can go from consuming no energy to consuming energy briefly while the switch makes a transition, back to consuming no energy once we’ve stabilized again. Pretty sweet. And we lived in that simple world until the last ten years or so.
When a Switch Isn’t Quite a Switch
To understand what’s going on here, let’s think of a real, physical switch – like what you have on the wall. It works by having two pieces of wire either contact each other or not. So, when the switch is closed, there’s a nice clean path along the wire; when the switch is open, the two contacts are far apart, and no electronsA fundamental particle found outside atoms. It carries a negative charge. It can move easily in a conducting material, which gives rise to electrical current. can jump across that gap.
There are two exceptions to this, just to be clear: if you raise the voltageVoltage is what gets electrons to flow. It's analogous to water pressure, which gets water to flow. Voltage is measured in units of "volts." high enough, then the electrons will have enough energy to jump the gap. You’d see that as a big ol’ spark. But it takes about 10,000 volts to create a spark in air that will go about a centimeter (that varies with humidity and such, so it’s not exact). That’s way more than the 110 V (or even 220 V) you may have in your house. So, no sparking. There is one other consideration… but let’s keep that one for next week. For the time being, this second consideration plays no part.
So, when your wall switch is off, no – zero – electrons flow.
Pinching Off
But that’s not how a transistor works. You can literally think of a transistor as shutting off by pinching shut the channel where they flow. It’s more like a balloon full of air. There’s pressure in the balloon to push the air out – just like the voltage trying to get the electrons to move. But if you twist the balloon opening shut enough, then you keep the air in there.
Or… do you? Ever notice that you think you have it shut and you hear a slight whine because air is leaking out? Or you tie it shut and let it float to the ceiling, and a little while later it’s a bit deflated and sagging?
That’s because, no matter how tight you tie those things, there always seems to be some way for air to leak out. You can reduce the leakage by using a better material or a tighter way of tying it, but it’s really hard to get it not to leak at all.
That’s how a transistor works too. The leakageThe very small amount of current that can flow through a transistor when it's "off." used to be small enough that no one really cared. But with the new, more aggressive processes, it seems like there are more and more ways for currentThe amount of electrical flow. Measured in amperes or amps (A). to leak. And there are millions of transistors in a circuit, so even a little bit of leakage from each one will add up quickly.
That means that circuits just sitting around actually do burn up energy – slowly, but enough to matter. If you make your transistors in a way that causes them to switch really quickly, then they leak more. So engineers need to work hard to balance the need for speed and the need to save energy.
Getting Sleepy
There are a lot of ways that they do this. The engineers that design the basic transistors and the processes that they’re made on work hard to keep leakage low. And circuit designers and architects find ways of shutting off entire blocks of circuits. And when I say “shut off” here, yes, it’s with a switch, but the switches we’ve been looking at were turning computing signals on and off. PowerThe rate of energy consumption. For electricity, it’s measured in watts (W). switching means shutting off the power supply to a block of circuits.
For instance, if the chipAn electronic device made on a piece of silicon. These days, it could also involve a mechanical chip, but, to the outside world, everything looks electronic. The chip is usually in some kind of package; that package might contain multiple chips. "Integrated circuit," and "IC" mean the same thing, but refer only to electronic chips, not mechanical chips. has a way of communicating with the internet, it might keep those circuits off until there’s something to communicate. Then it turns on, does its thing, and shuts back off.
That may sound simple, but it can get complicated.
- Designers have to figure out how to break the whole chip up into “power islands,” each with its own supply that they can shut off without shutting off or messing up the other islands. More smaller islands give better control at a finer level, but they also add complexity.
- It takes time to power these islands up and down. So that has to be figured into all of the timing calculations.
- They have to be careful at the boundaries between power islands. What happens when a transistor whose power is cut off is connected to the next transistor, which still has power? (Or vice versa?) Determining what “off” means comes with basic assumptions – like the fact that power will be on when a switch is turning on or off. If those assumptions aren’t met – like the power isn’t on – then there may be other leakage paths that open up (sometimes called “sneak paths”).
The New Mantra
Designing circuits used to be all about getting fast speed and low cost. Those are still important, but power is now right up there with them. Speed, in particular, comes at a cost in power, so sometimes “fast enough” is… well, fast enough. Going faster would burn more energy for no real benefit, so hold back where possible. There are three letters that engineers now have to optimize together: PPA. Performance, Power, and Area (where more area on the siliconAn element (number 14 in the periodic table) that can be a semiconductor, making it the material of preference for circuits and micro-mechanical devices. chip means more cost). These days, everything is about PPA.
Leave a Reply