Billions of chips in today's computers, automobiles, cellphones, media cards, and those clever keychain memories are literally powerless when idle, yet they dispense data and instructions at the flick of the on-switch. They are almost all flash memory chips, a type of electrically erasable and programmable read-only memory.

Nonvolatility, flash's property of retaining data for years when unpowered, is crucial for most electronic systems any more complicated than a light bulb. A flash chip in a computer tells it how to boot up. In a cellphone, it holds the instructions and data needed to send and receive calls, and stores phone numbers. Electronic products of all types, from microwave ovens to industrial machinery, store their operating instructions in flash memory.

Almost as important as flash's nonvolatility is its programmability. It's the feature that lets users add addresses, calendar entries, and memos to personal digital assistants and erase and reuse the media cards that store pictures taken with digital cameras.

Engineer at Texas Instruments' Kilby fab in Dallas, Texas, holds up a 200-mm wafer containing developmental 64MB ferroelectric RAM chips.

Engineer at Texas Instruments' Kilby fab in Dallas, Texas, holds up a 200-mm wafer containing developmental 64MB ferroelectric RAM chips.

But flash is under assault by technologies bent on proving they can do better. These upstart random access memories (RAMs) have little in common. The ferroelectric memory picks up on the electric fields inside certain atoms and the directions in which they point. The magnetoresistive type stores data as the either-or directions of the alignment of small magnetic regions in a ferromagnetic material. A third, Ovonic Unified Memory, is based on a material that switches between crystalline and amorphous phases.

Different as these technologies are, they share two advantages over flash. First, they can write data in a few tens of nanoseconds, like the dynamic RAMs in a computer's main memory. Flash, on the other hand, takes at least a microsecond .

Second, the new memories can withstand constant rewriting for years, whereas flash cells begin to lose data after fewer than a million write cycles. Naturally, a million is fine for applications like media cards and cellphones, in which data may be written only a few times a day—as when a number is added to a cellphone's address book or a picture is taken with a digital camera. But if used to constantly write new data, as in a computer's main memory, flash cells would start breaking down in days or weeks. They also write much too slowly for computer-memory applications.

In contrast, the newcomers endure for a practically limitless number of write cycles. That endurance, together with their reasonable write speeds, takes them beyond flash to tackle the broader dynamic RAM market as well. DRAMs, after all, draw power continuously, even when data is not being read or written, and lose all data when power is switched off. Obviously, the electronics industry could be revolutionized by a memory as small as a DRAM and as fast, but able to retain data without drawing power.

A veritable pot of gold awaits such a successor to flash and DRAM. Taken together, flash and DRAM sales, which represent about 15 percent of the total semiconductor market, were projected last November by International Data Corp. (Framingham, Mass.) to total over US $22 billion in 2002 and exceed $25 billion in 2003. The big unanswered question is which one—or ones—will succeed?