NOT JUST A FLASH IN THE PAN

the Economist
Mar 9th 2006

Storage technology: Flash-memory chips are encroaching on markets previously dominated by hard disks, recordable DVDs and magnetic tape. Will flash displace all other storage technologies?

WHEN Apple launched the iPod nano, the slimmest incarnation to date of its prized music player, last September, it attracted plenty of attention. While consumers swooned over its sleek design, analysts were more likely to hail it as a turning point for the storage industry. Whereas the original iPod used a relatively bulky, power-hungry hard-disk drive to store music, the iPod nano instead relies on small, silent and sturdy "flash" memory chips, which retain data even when the power is switched off. It is a vivid illustration of the way in which flash memory is marching into markets previously dominated by other storage technologies: not just hard disks in music players, but photographic film in cameras and magnetic tape in camcorders.

Flash is everywhere. Digital cameras use plug-in flash cards to store photos. Flash-based memory sticks, or keychain drives, have dethroned floppy disks as the easiest way to carry data around. Mobile phones and handheld computers use flash memory to store software, documents, music tracks and photographs. Then there is all the flash you don't see, quietly storing settings and configuration data in cars, games consoles, printers, modems, satellite-positioning systems, and so on.

In 2005 global sales of flash chips exceeded $19 billion, says Alan Niebel, a semiconductor analyst at Web-Feet Research. That is enough to buy roughly 40 megabytes of flash for each person on earth. Such is the demand in consumer products that in November Intel and Micron Technology, two big chipmaking firms that are normally competitors, established a $5 billion joint-venture, called IM Flash Technologies.

Meanwhile, flash continues to expand its influence. Japanese camcorder-makers have been offering flash support (which allows video to be recorded directly to flash cards) for several years and, at the Consumer Electronics Show in Las Vegas in January, Sanyo previewed what it claims is the first camcorder able to record high-definition video to a flash-memory card. It goes on sale this month, just two years after the first hard-disk-based camcorder appeared and four years after the advent of the first camcorders able to generate recordable DVDs. Every storage technology, it seems, is threatened with replacement by flash.

Admittedly, the Sanyo camera records only about 20 minutes of high-definition video on to a one gigabyte flash card. But as such cards shoot up in capacity and tumble in price, the gap with other storage media will narrow. And, as the iPod nano demonstrates, capacity is not everything: an iPod nano with four gigabytes of flash storage sells for around $250, just $50 less than a "traditional" iPod based around a 30-gigabyte hard disk. Flash-based products may have less storage capacity at a given price point, but they are smaller, sleeker, more robust and require less battery power than devices based on mechanical storage technologies, which boosts their appeal.

FLASHBACK
The origins of flash memory go back to 1967, when Simon Sze and Dawon Kahng, two researchers working at Bell Labs in New Jersey, devised a new type of semiconductor memory device in which information could be stored and updated, and which was "non-volatile"--which means it retained its contents even when the power is switched off. This was a novelty, since most semiconductor memory was either random-access memory (RAM), the contents of which can be changed--but which forgets everything the moment the power goes; or read-only memory (ROM), which never forgets, but cannot be reprogrammed.

The beauty of the researchers' design lay in its simplicity. It required just a minor modification to a field-effect transistor, the basic building block of most microchips. Transistors use a small voltage, applied to the "gate" of the transistor, to control the flow of a large current. The zeroes and ones of digital information that whizz around inside a computer correspond to whether individual transistors are switched on or off.

The researchers' innovation was to add a second gate to the middle of the transistor, called a "floating" gate because it is insulated by a thin oxide layer. (The transistor's original gate is known as the "control" gate.) Initially, the floating gate does not interfere with the working of the transistor. But applying a suitable voltage across the oxide layer causes electrons to "tunnel" through it and become trapped on the floating gate, where they are stuck even if the power is disconnected. If there are enough electrons on the floating gate, it starts to interfere with the working of the transistor, so that a voltage applied to the control gate no longer controls the flow of current through the transistor. Whether current flows through the transistor or not can be detected using external circuitry.

The modified transistor can thus serve as a memory cell. When no electrons are on the floating gate, and the transistor works normally, it represents a one; and when electrons are trapped on the floating gate, so that the transistor stops working, the cell is deemed to represent a zero. Crucially, the process can also be reversed: a suitable voltage across the oxide layer applied in the opposite direction causes the electrons to "tunnel" back off the floating gate, restoring the transistor to its original state and resetting the stored value to a one. The cell is therefore both programmable and non-volatile.

Dr Sze, who now works at the National Chiao Tung University in Taiwan, recalls the initial reaction to this breakthrough. "My boss said 'Simon, tell me, what use can you think of for this device?' I could not think of anything." As a result, he was told to bury the result in an obscure journal, lest this useless invention draw the ridicule of his peers. Floating-gate memory did eventually find its way into production, but since it was expensive, it tended to be used in niche applications such as military equipment and the very earliest mobile phones, which were hardly mass-market items.

Then in 1980, Fujio Masuoka, a researcher at Toshiba, filed a patent for a novel variation on floating-gate memory. His new invention was dubbed "flash" memory, because it allowed entire sections of memory to be erased quickly and easily, by applying a voltage to a single wire connected to a group of cells. Dr Masuoka's design was a compromise between flexibility and cost. Being able to erase each memory cell individually made ordinary floating-gate memory complex, power-hungry and expensive. Dr Masuoka's design was less flexible, since it required entire groups of cells to be erased together, but it was far cheaper.

As with floating-gate memory, Dr Masuoka's idea was not initially appreciated by his superiors. "My daily work was to develop a one-megabit RAM," he says. "At home, after working hours, I worked to develop flash memory." In retrospect, however, his invention was a classic case of a well-prepared mind seeing an opportunity where others had not even realised there was a problem. Dr Masuoka happened to have a good understanding of the separate fields of computer memory and magnetic-storage systems. His invention offered a useful compromise: a new means of storing and accessing data files that was faster and more resilient than a hard disk, even if it could not compete with traditional memory (RAM) on speed or programming flexibility.

By 1986 Toshiba was producing the first batches of test chips, and two years later Intel licensed the technology and began its own production. In 1987 Dr Masuoka invented another type of flash memory that could be produced more cheaply and in denser arrays. That variety is now called NAND flash, to distinguish it from the earlier type, known as NOR flash. (NOR and NAND are the names of simple circuits that perform logical operations on binary information; in the context of flash memory, they describe the particular way in which the memory cells are interconnected.)

In the two decades since the first flash chips appeared, thousands of engineers have applied themselves to making the technology smaller, cheaper and better. Some of their advances are minor miracles of ingenuity, such as the development of "multi-level" flash-memory cells. These work by varying the quantity of electrons trapped on the floating gate, so that rather than preventing the transistor from working altogether, they disable its ability to control the flow of current to various degrees. Two binary digits (bits) of information can then be stored in a single cell, for example, depending on whether it is not working (0,0), 33% working (0,1), 66% working (1,0) or fully working (1,1).

Another feat of flash engineering has been to reduce the density of flash storage by stacking several chips on top of each other. This involves etching away the underside of the chips until they are almost as thin and as flexible as paper, bonding them together and then connecting them up via electrodes that project slightly out of the edge of the resulting sandwich (see picture). This trick is particularly favoured by makers of NOR flash, the kind used in mobile phones, where space is at a premium. Most stacked-chip flash products involve just two chips, though some producers make stacks up to five or six chips tall. As in the real world, when property gets expensive, engineers build upwards.

As such tricks improve the storage density and reduce the cost of flash memory, it will increasingly compete with hard disks to become the preferred storage technology in laptop computers; a showdown seems likely in the next few years. Laptops are portable devices, so any technology that can reduce volume and weight and increase battery life has obvious appeal. At the same time, however, as fully fledged computers running complex operating systems, laptops require tens of gigabytes of storage.

BATTLING FOR THE LAPTOP
That is why hard disks are the dominant form of storage in laptops. But last May, Samsung, a South Korean electronics giant, demonstrated a 16 gigabyte NAND flash drive. (Samsung is by far the largest producer of NAND flash, with more than 50% of the market.) The prototype drive contained 16 one-gigabyte flash chips packed into a standard hard-disk casing, which can slot into any computer. Being based on flash, however, it weighs half as much as an equivalent hard disk and uses much less power. Samsung has since announced a two-gigabyte flash chip, which will go into mass production later this year.

The main drawback with such flash drives, when they reach the market, will be their high cost. The price per gigabyte of flash storage is roughly 100 times that of hard-disk storage. The cost of a given amount of flash storage is falling all the time, of course, but the same is true of hard disks. Jim Handy, an analyst at Semico Research, a consultancy, estimates that it would cost around $4,000 to buy enough flash memory to replace an average laptop hard disk--a figure that has remained roughly constant for the past five years as laptops' storage requirements have grown. In other words, because both technologies continue to evolve, the price-performance gap between flash and hard disks is narrowing slowly, if at all (see chart).

However, says Mr Handy, the price per gigabyte does not give the complete picture. Hard disks have a fixed cost, independent of capacity, of around $100 for their internal mechanical parts. Flash has no such fixed cost. So at any given point there exists a range of storage capacities where flash, despite being much more expensive per gigabyte, can compete with hard disks on price. Today, flash is competitive for capacities of four gigabytes or less--hence its appearance in the iPod nano, which comes in two-gigabyte and four-gigabyte versions. Above about 20 gigabytes, hard disks win hands down.

In between is a battle zone where the two technologies compete. And as the price per gigabyte of both technologies falls, the fight shifts to ever higher capacities. In theory, it is just a matter of time before flash drives become competitive in the 50-100 gigabyte range of today's laptop hard-disk capacities. Yet each new operating system requires more storage capacity than the last. In January Microsoft signalled that laptops would need around 80 gigabytes of disk storage to run the next version of its Windows operating system, called Vista.

To further complicate matters, flash has the disadvantage that its memory cells eventually wear out--though the cells in modern flash chips can typically be reprogrammed around 1m times before this happens. This is not a problem in a devices that stores largely static files that are read but seldom modified (such as music tracks on an iPod). But using flash memory in place of a hard disk to store running software, such as an operating system, is a different matter, since information is constantly being modified and updated. To get around this drawback, flash-memory controllers keep track of how often particular groups of cells have been modified, and move data to different parts of the chip to maximise its lifespan. And if a fault is detected in a group of cells, that group can be tagged so that it is no longer used. It is worth remembering, however, that hard disks, particularly those in laptops, also have limited lifespans.

Although flash seems unlikely to replace hard disks just yet, it could soon start to infiltrate laptops in a more gradual way. Last year Microsoft and Intel demonstrated different ways to combine flash memory with hard-disk drives. In both cases the flash serves as a temporary storage for frequently used data, reducing the need to access the disk. This speeds up the rate at which data can be retrieved, enables laptops to wake up more quickly from sleep and extends battery life.

Not everyone is convinced, though. Mr Handy points out that neat demos usually take a couple of years at least to turn into real products. And in that time, hard disks will have improved further as well. Indeed, flash has traditionally done best against storage media that have more or less reached their technological limits, such as photographic film, floppy disks and CDs. Hard-disk firms, however, continue to innovate at great speed, and are by some measures managing to push down the cost per gigabyte of storage even faster than flash. Indeed, hard disks have lately been fighting back: both Samsung and Nokia have launched mobile phones that contain tiny hard disks to store music, for example. And the expected leap in demand for storage capacity, as phones double as TVs and video cameras, plays to hard disks' strengths.

THE QUEST FOR THE UNIVERSAL MEMORY
This is all a healthy reminder that no storage technology can afford to rest on its laurels. Indeed, the greatest threat to flash may come from its own success. Flash chips are, by most estimates, doubling in capacity roughly every year, outdoing even Moore's law, which predicts a doubling of the number of transistors on processor chips every 18 months. Sooner or later, this sort of exponential trend, which depends on shrinking each generation of transistors, is going to run into fundamental manufacturing limits. What new storage technology will step in then?

The field is wide open. Mr Niebel of Web-Feet Research says he keeps track of more than 40 pretenders to the flash throne. Many of them have yet to progress beyond laboratory prototypes. But one or two have made the leap to fabrication. Magnetic RAM (MRAM) and Ferroelectric RAM (FRAM) are, as their names indicate, variations of the RAM used in today's computers. These technologies combine the speed and programming flexibility of RAM with the low power consumption and non-volatility of flash. But to store data, they rely on special materials that respond to magnetic or electric fields, which makes their production trickier. As a result, prototype chips with more than a megabyte of capacity have begun to appear only in the past couple of years, and will be relegated to specialist markets until they can compete on cost and capacity with other technologies.

A radically different approach to storing data is the Ovonic Unified Memory (OUM), pioneered by Stan Ovshinsky, a highly successful and eclectic serial inventor. OUM is based on so-called chalcogenide materials, the atomic structure of which can be changed reversibly from a well ordered crystalline state to a disordered, amorphous state by applying a burst of electrical current. These two states can then be used to represent zeroes and ones. The technology is faster than flash, has a lifetime of trillions of cycles, and its performance increases as the size of the memory cells is reduced (since less energy is required to change the state of a smaller cell). Mr Ovshinsky's firm, Ovonyx, has already made prototype chips in conjunction with STMicroelectronics, a big chipmaking firm, and the results are said to be very promising.

OUM is just one of the candidates in the race to create a "universal" memory, the holy grail of the storage industry, which would combine the speed of RAM with the non-volatility of flash and the low cost and high storage-density of hard disks. Ed Doller, chief technology officer at Intel's flash-memory division, says the company is following the progress of OUM technology closely. Although it is not in a position to threaten flash in the next couple of years, he says, "beyond 2010, it has legs".

See this article with graphics and related items at https://www.economist.com/displaystory.cfm?story_id=5571544

Go to http://www.economist.com for more global news, views and analysis from the Economist Group. - ABOUT ECONOMIST.COM - Economist.com is the online version of The Economist newspaper, an independent weekly international news and business publication offering clear reporting, commentary and analysis on world politics, business, finance, science & technology, culture, society and the arts. COPYRIGHT - This e-mail message and Economist articles linked from it are copyright (c) 2006 The Economist Newspaper Group Limited. All rights reserved. http://www.economist.com/help/copy_general.cfm Economist.com privacy policy: http://www.economist.com/about/privacy.cfm