MirrorBit stores two bits per cell by creating two distinct storage nodes per cell in a layer of nitride deposited below the cell's gate. By contrast, Intel's StrataFlash approach uses four distinct levels of charge to create two bits per cell.

"Instead of a unidirectional conduction that you see in floating-gate flash, we've created a bidirectional gate," said Kevin Plouse, vice president of technical marketing and business development at AMD's memory group. "The entire material under the gate is nitride, and we trap charge there on either side of the multibit cell."

Promising weapon

The approach reads a bit in the opposite direction from the one in which it is programmed. A read voltage of about 2 V is applied to the drain, the source is grounded and 4 to 5 V are applied to the gate. That should pull electrons from the channel, but the presence of charge blocks the path of the electron flow and no channel is induced. With no current flowing, the bit is read as data zero, as in normal flash memory, said Plouse.

If AMD delivers on all that it promises with MirrorBit, it will finally have a weapon to compete with Intel's StrataFlash products, analysts said.

"Intel pretty much owns the market for flash used in [cellular phone] handsets," said Mike Nguyen, wireless technology analyst at International Data Corp. "The latest and greatest 2.5G handsets shipping in Europe now have two 128-Mbit flash parts and one 64-Mbit flash, so the handset market is a great market for AMD to target. On paper, their cost and pricing claims for MirrorBit are great. We'll now have to see if they can do it."

The 64-Mbit product being launched into the market this week is a 3-V part aimed at networking, telecom and automotive customers. The cellular handset market requires 1.8-V operation, with a burst mode, for the data-enabled phones that are becoming popular in Korea, Japan and, to a lesser extent, Western markets.

A 1.8-V MirrorBit part will require AMD to upgrade to a 0.13-micron process at its flash fabs in Japan, where the company jointly operates two manufacturing plants with Fujitsu Ltd., and later at AMD's main fab here, which is being converted from microprocessor production to flash.

The 0.13-micron process will be ready next year at one fab in Japan, and later for the other two flash facilities, the company said. By 2004, that process will support a gigabit-density MirrorBit part, said Plouse. At the 0.09-micron process node expected in 2005 or so, all of AMD's flash products will be based on MirrorBit. That means that nearly all of AMD's flash design resources will be dedicated to MirrorBit products by the end of this year, Plouse said.

AMD is claiming big things for MirrorBit, including a 90-nanosecond random-access time, standby power of 1 microamp and data retention of 20 years, specs that either match or improve on floating-gate flash. MirrorBit products will support a page mode read, with four 16-bit words read together, and a 16-word page write mode. Those functions will be a superset that can be enabled or not; otherwise MirrorBit products are pin-compatible with conventional floating-gate flash parts.

MirrorBit requires about 30 percent fewer process steps than floating-gate flash, which will translate into fewer mask layers, Plouse said. With a buried bit line, contacts to each cell are simpler, reducing the number of etching steps.

Intel has shipped StrataFlash products since 1997, and about one-quarter of all flash bits shipped by Intel are in multilevel-cell technology, said Troy Winslow, flash marketing manager at Intel. Intel will move to 0.13-micron process technology for flash later this year, and Winslow said that "all the new and advanced flash products from Intel will use multilevel-cell technology."

Competitive setup

That sets up a head-to-head competition in the high-density flash market. "If you look at AMD's product line, they really don't have high-density parts now," said Rich Wawrzyniak, flash analyst at Semico Research. "With MirrorBit, AMD will be potentially equal to what Intel can do. AMD should benefit just by having high-density parts on the market when demand materializes in the handset market."

Alan Nogee, who tracks wireless markets for In-Stat/MDR, said many wireless handset and PDA users in Asia are likely to want to store images and MP3 audio files in flash. Voice-recognition technology also will demand high-density flash parts. Handset displays with 65,000 colors are taking off in Korea and Japan, requiring more flash.

"The U.S. and Europe may be a few years behind Asia. But since you can't put a hard drive in a cell phone, flash becomes the equivalent of the HDD in a PC," Nogee said. "And that makes these multibit flash parts important to AMD. Intel has a pretty good lock on the high-density market now, but from the handsets we've peeled open and looked at, AMD already is making some inroads in the handset market."