BY HANK HOGAN
AUSTIN, Texas – Freescale Semiconductor Inc. (www.freescale.com) has embarked on a memory technology-magnetic tunnel-junction random access memory, or MRAM-that is proving to have implications for manufacturing in Freescale’s own cleanrooms. Another memory alternative, ferroelectric RAM, or FRAM, is already in production, and both could impact semiconductor cleanrooms throughout the industry.
While Freescale doesn’t make stand-alone memory chips, it inserts considerable storage into its automotive and consumer-embedded products. Embedded memories can run the gamut of technologies-from DRAM and SRAM to flash-which can add to product development complexity.
“We believe that MRAM can displace that multiple memory and provide a single unique solution to those markets,” says Saied Tehrani, director of MRAM technology at Freescale. The company thinks MRAM can provide a performance advantage at relatively the same or perhaps even a better cost.
Freescale expects to begin volume production of a 4 Mbit MRAM next year in a 0.18 μm process running on 200 mm wafers. MRAM stores information in a magnetic bit. As a result, storage is as permanent as a disk drive, but MRAM’s magnetic materials, which are contaminants to standard silicon transistors, have to be introduced to the fab in the process.
Tehrani says Freescale has controlled cross-contamination by employing some of the same segregation techniques used when copper was introduced to the fab. On the tool side of contamination control, the company is working with manufacturers to drive down tool-derived particulates and contaminants.
FRAM’s future
Meanwhile, Ramtron International Corp. (www.ramtron.com) has been researching FRAMs for years and works with foundry partners to develop the manufacturing process. In FRAM, data is stored in the position of atoms within a ferroelectric crystal. This storage is nonvolatile. Megabit FRAM devices are shipping today, many of them as replacements for battery backed-up SRAMs.
Tom Davenport, Ramtron’s vice president of quality and technology, says the company’s FRAM manufacturing process is compatible with standard CMOS processing and testing. The FRAM module, he says, is inserted after transistor formation but before metallization. “The low mobility and good glass-forming characteristics of the materials make for ease of integration and minimize contamination concerns,” says Davenport. “Copper is much tougher.”
Davenport acknowledges, however, that the FRAM process will require changes to the manufacturing toolset to ensure contamination control. In fact, a new tool will be needed to lay down the FRAM module, and Davenport says it will be similar to what is used on high-K DRAM lines.
But in their quest to be a universal solution, both memory alternatives face a similar problem. MRAM and FRAM are competing against established memory technologies that benefit from high volume-and hence, low cost-manufacturing. Richard Gordon, research vice president of Gartner’s semiconductor research organization, says other manufacturers have managed to overcome this chicken-and-egg problem in the past. Gordon points, for instance, to the NAND-flash memory that’s now found in USB drives and elsewhere.
While a new memory technology’s acceptance into the market isn’t impossible, it will take time. Gordon notes the need for a universal memory technology to replace the current array of different memory types. “There’s a good reason for the use or design of a universal memory in something like a cell phone,” he says.