New Memory Chips Don't Forget

A FIVE-YEAR-OLD company with a patented technology is trying to challenge the giant chipmaking Intel Corp. to set the next standard for the way computers store information. The battle ground is memory chips, little devices inside computers used to hold data and programs. Most of today's chips are ``volatile'' - they are erased when the power is turned off. Data for later reuse must be stored on magnetic disks.

But advances in memory chips might make magnetic storage obsolete within the next decade. Chips can be more than 10,000 times faster than disks and are more reliable, having no moving parts.

``It ultimately promises the perfect memory device,'' says Michael Gumport, a senior electronics analyst at Drexel Burnham Lambert Inc. ``I think that they are making good progress, but there is still a very long way to go,'' referring to the work of Ramtron International, Intel's Australian competitor.

Ross-Lyndon James, Ramtron's president, set out five years ago to create non-volatile memory chips using ferroelectrics, a technology that combines magnetism and electronics. With $60 million in start-up capital, Ramtron is now building a research and development factory in Colorado and has already produced its first chips. The Krysalis Corporation in California is another start-up company working with the same technology.

At the heart of Ramtron's integrated circuits is an extremely thin ceramic film made from crystals of lead-zirconate-titanate (PZT). Each PZT crystal holds a single titanium atom. A tiny electric field can flip the atom between two positions, like a microscopic toggle switch. This allows writing on the chip in the binary code that computers understand. The thousands of atomic-switches on the chip remain in position even after the power is turned off.

Competing technology

Ferroelectrics face steep competition from the flash memory, a non-volatile system that has been developed by Intel, Seeq Technology of San Jose, Calif., and Toshiba in Japan. Flash memory stores information using a device called a floating gate, which is like a tiny box that can hold an electrical charge. It also has two positions: empty or full. The electric charge remains inside the box even when the power is off.

Both flash and ferroelectric technologies have drawbacks. Ramtron's chip works at the normal high speeds of volatile memory chips, but some engineers question whether or not its flipping atoms will be susceptible to fatigue. Flash memories can be read at normal chip speed but are a thousand times slower at remembering new information, slower even than a hard disk (though faster than a floppy).

Intel leads the race to market. Currently, Ramtron's chips can only store 4,096 bits, or a quarter-page of text, making them practical only for a few applications. One of the first will be an electronic odometer for automobiles, Mr. James says. Ramtron doesn't expect to be in production of a 1 megabit ferroelectric memory chip until 1990, says Richard Horton, president of Ramtron's Colorado facility.

Intel, by contrast, will have the satisfaction of seeing its flash memories for sale in the Psion hand-held computers this fall. ``We are in production with 1 megabit today,'' says Bruce McCormick, the flash-memory marketing manager at Intel. ``That is a big difference between flash and ferroelectric.''

(This fall, two other computer manufacturers introduced hand-held computers that use wallet-sized cards containing memory chips to store documents and programs. But the chips aren't truly volatile, because they are powered by tiny batteries. When the batteries fail, the data is lost.)

Flash technology leads

Ferroelectrics will be chasing a moving target in seeking to catch up to flash technology, says Stefan Lai, who is managing Intel's 16-megabit flash memory effort. ``Our current technology will allow us to do a 4-megabit right now,'' says Dr. Lai. He expects Intel to complete its 16-megabit design in 1992.

James is confident that Ramtron will find it easier to push ferroelectrics to higher memory densities than has been the experience of other companies working with conventional designs. ``The smaller we make it, the better it works,'' he says. Other chipmakers, by contrast, have often needed to totally redesign their circuits to achieve higher storage capacity.

Intel manufactures and sells its flash memories directly. Ramtron hopes to license its ferroelectric technology to other chip manufactures. The company's 13 patents, James contends, means that ``anybody who wants to go apply [ferroelectrics] to memory will have to come into an arrangement with Ramtron.''

To make hard disks obsolete, however, chipmakers will have to make their chips much cheaper and dramatically increase the storage capacity.

A typical 3 1/2-inch hard disk that can store more than 200 megabytes of information (about 100,000 typed pages) costs $1,000-$2,000. To store that much information on today's state-of-the-art chips would cost more than five times as much and require 1,600 chips.

``I would probably say that we are looking at a good decade before we see hard disk technology easily eclipsed by a solid-state solution,'' says Liz Baird, a spokeswoman for Quantum, a maker of disk drives.

Other technologies such as optical storage will complicate the picture, making it likely that no single solution will be adopted, Ms. Baird contends.

``What we really think is going to happen is that there will emerge a lot of complementary ways for these technologies to work together.... We think that the hard disk business has a long life ahead of it.''