A Dream for a Computing World Without Sleep Mode

The idea of computer sleep is as old as the idea of computer memory. The needs of many hundreds of millions of casual PC and device users were unimaginable in the 1950s, but the fundamental problem solved by sleep was then as much as now the fundamental problem of computer memory itself: volatility.

And, broadly speaking, the problem of computer memory—and, thus, of sleep modes—is the problem of on/off switches. Memory persists only when it's being provided with electricity. When power is lost to a computer system, its memory immediately evaporates. A planet's worth of batteries are at this very moment being drained because of a seemingly simple limitation.

Note that when I say "memory," I mean RAM. RAM is what keeps a computer supplied with information as it goes about its computing duties.

Probably most people think of memory has a hard-disc or cloud server, where memory functions something like a bank account for information. Not quite: In some senses, this sort of memory might as well not even exist for a computer—a popular analogy is that if we imagine that a processor's main (RAM) memory is about an hour's drive away from the processor itself, then the computer's hard-disc is roughly a distance equivalent to the time it takes to drive to Pluto. Hard-disc memory is non-volatile, but it's barely even in our (computer's) solar system.

So, we're talking about RAM. This is memory. It represents a computer's state at a certain time—running programs and their required data, mostly. This is what a sleep mode preserves.

The cost of accessing RAM relatively quickly (read: usefully) is energy. The whole mess is based on clever arrangements of transistors, which are circuit components that are often used as switches. Given some input power, a transistor will have some specific output power. Arrange a few transistors just so, and it becomes possible to preserve a certain state, like a bit. Put them all together and you might wind up with a much bigger state, like a computer system.

The catch is that transistors need some input power to do this.

This is the essence of a computer sleep mode. Shut down the computer, but keep its RAM powered. The system continues to draw just a trickle of power, but only the bare minimum.

This minimum has been progressively limited over the past couple decades thanks in part to the international IEA-led One Watt Initiative, which guides energy efficiency regulations in many countries. In 2001, for example, US President George W Bush issued an executive order mandating that all government agencies purchase computers in accordance with the initiative's benchmarks.

So, if sleep mode is as fundamental as volatile system memory, how fundamental is that? Non-volatile RAM is a looming, if slowly moving frontier. Flash memory is in this category, as is ROM (read-only memory, as in EPROM and EEPROM). ROM is fast to read, but prohibitively slow to modify. Flash memory, meanwhile, is limited in that it can only erase memory cells in large blocks at a time. Neither is a proper SRAM or DRAM replacement, and that may still be a ways off.

It won't be a for a lack of trying, however.

In 2017 or 2018, Intel's 3D XPoint memory will hit the market, though details are still scarce on what it actually is. Mostly we know that it will be a whole lot faster than Flash memory, but still slower than DRAM, making it another intermediate memory technology between actual system memory and hard-disc memory.

Other emerging memory technologies are based on changing the properties of certain materials, making them more or less electrically resistant or changing their magnetic properties. Crucially, these technologies eschew transistors entirely, and so, unlike RAM, they maintain their state when powered down.

It's thought that something like magnetoresistive RAM might someday be able to function as a "universal memory," compiling all of the benefits of RAM, Flash memory, and hard-disc memory into a single non-volatile technology. It turns out that this would solve a whole lot of computing problems at once—including the so-called memory bottleneck—but we can only really assume that one of those is the need for computers to sleep.

The sleep mode "problem" of concern to manufacturers remains energy efficiency. Keeping a memory grid full of switches powered doesn't take much, but across many hundreds of thousands of devices, machine downtime adds up quickly. Perfect sleep mode efficiency would mean no power consumption at all. Or the end of sleep mode itself.

You'll Sleep When You're Dead is Motherboard's exploration of the future of sleep. Read more stories.