|The Memory Revolution|
Issue #69 (August 1987)
The computer industry thrives on innovation. But even within this boiling-pot of ideas, there is a revolution in the works that I believe will turn much of the current technology on its head. There are one or two breakthroughs yet to be made; but [quoteright]when they are accomplished, probably
within ten years, the consequences will be enormous. Remember, you read about it here first.
...recall what desktop calculators looked like twenty years ago.
Let's start with some basic concepts, for those readers who are not active computer nerds. Every computer system — large or small — contains three kinds of equipment: input/output devices such as keyboards and viewing screens, one or more data processors, and a lot of memory storage equipment. When you look at any large mainframe installation, you won't see much input/output equipment; most of the terminals and printers are usually somewhere else, connected to the mainframe by cables or telephone wires. You also won't see a big processor, because this function requires only a few circuit boards. What you will see is mainly memory storage equipment — tape drives, disk drives, and the memory cards built into the processor. Moreover, a good bit of the computer's time is spent just copying information from one kind of memory to another, for reasons I'll explain below.
Memory storage is where the computer keeps its data — your credit card records, for example, or the instructions for navigating a rocket to the moon. A computer's "power" (as well as its cost) is measured largely by how it handles memory. Everyone who investigates buying a personal computer learns up front that its cost is primarily determined by how much RAM (random-access memory) and disk memory it has.
There are three desirable characteristics for any computer memory. The problem is that no single piece of equipment has all three — this is why information must constantly be copied from one memory to another. The first characteristic is that it hold a lot of data per dollar of equipment cost, which is satisfied by tape and hard disks. The second is that it allow the computer to find any single piece of data very quickly — ideally during the few billionths of a second between processing cycles. This is satisfied only by electronic random-access memory. The third is that it let the user take away a cheap permanent record of the data, for filing or transport to another computer; this is satisfied by tape and floppy disks.
If a memory storage device were developed that had all three characteristics, it would quickly sweep away all the others. Not only would it make computer systems cheaper and simpler, it would eliminate the wasted time they spend copying programs and data from one device to another. In fact, the development of such a device is the revolution that is about to happen.
The first part of the technology is already in place. CD/ROM is a technique for recording colossal amounts of data on cheap plastic disks (very similar to the audio "compact disks" presently on the market).
The computer version can hold every telephone listing in the US on two or three disks. A CD/ROM is read by bouncing a laser beam off its surface, using a relatively inexpensive device that resembles a record player. At present, however, it is a read-only memory — that is, a computer can only read off data that has been previously recorded by a complex manufacturing process.
The second part will consist of perfecting the so-called "WORM" (write once-and-read memory) version of CD/ROM, which will let the computer store its own data on the disks. This development appears to have nearly arrived.
The third and most crucial breakthrough will consist of constructing a device in which the laser beam that reads and writes on the CD/ROM can be deflected to any part of it nearly instantaneously. This will let it behave like a random-access memory. Because the laser beam is essentially weightless, moving it to specific parts of the disk (which of course will no longer move and which will probably no longer be circular) requires no unusual amount of energy. The problems will rest in the speed and accuracy of the deflection technology. The deflector will probably need to calibrate itself constantly by picking up prerecorded markers on the disk. But none of this is inherently impossible, or even very difficult. It just needs to be done.
The result will be a single kind of memory with all three of the desirable characteristics I listed earlier — lots of capacity per dollar of investment, access time on the order of a processor cycle, and a cheap permanent storage medium. Other forms of computer memory will cease to make economic or technological sense.
At this point I hear the computer mavens among us crying "Hold it — so far you've only described a non-erasable memory. How can that possibly supplant the traditional erasable RAM?" Contrary to many preconceptions, erasability is not essential in a computer memory if the medium is cheap enough. It's like paper — instead of trying to erase a piece we've used, we just reach for another sheet. To get technical for a moment, non-erasability will require some changes in software architecture. Pointers and tables will be used much more, and the traditional program heap will become stable. Some RAM will probably always be needed for stacks and master pointers. But none of this will degrade performance to any serious degree, and in many cases (such as large database sorts) pointer-based software will be more efficient.
Why is this a revolution? Imagine the classic computing center, with row after row of tape drives lining the walls. Now imagine white-coated movers trundling out all those expensive boxes. In the center of the large, empty room now stands a table, and on it is a small computer and half-a-dozen things that look like record players. From time to time the computer operator takes away a used record and replaces it with a fresh one. The whole operation now costs a tiny fraction as much to buy and maintain.
That's the revolution. And if you think it's impossible, just recall what desktop calculators looked like twenty years ago. Remember twelve pounds of gears and levers for $600 that could multiply two numbers in half a minute? They were the "state of the art" then, as mainframe computers are today.
Editor GEORGE TOWNER, unrestrained by cooler heads, dribbles his crystal ball across the floor in this issue. Of course, the nice thing about predicting the future is that by the time you can be proven wrong people have forgotten what you said. You can read about George's latest book here!
More Articles by George Towner
George Towner was born in Reno and grew up near Berkeley. As a teenager he began making gangster movies using an old 8mm camera, one of which featured a car being pushed over a cliff off State Highway 1. He has started and sold two successful technology firms, and currently works for Apple Computer, where he is the most senior in age. He lives with his wife in Sunnyvale. They have two daughters and a son.