PORTLAND, Ore. For over a decade, materials researchers have sought a ferroelectric material that could work in flash-sized bit cells that retain information for as long as a decade--the base requirement for nonvolatile memories.
Researchers at Yale University and the Semiconductor Research Corp. (SRC) claim that ferroelectrics are more appropriate for replacing DRAM than flash. Current DRAM technology has to be refreshed every few milliseconds; ferroelectric materials could last minutes without freshing.
Yale and SRC researchers recently demonstrated an experimental ferroelectric transistor for FeDRAMs that retained information 1,000 times longer than DRAMs, consumed 20 times less power and can, they claim, be scaled to even the most advanced nodes on the International Technology Roadmap for Semiconductors.
"Our memories are as fast a DRAM, if not faster, but are as small as flash, and more scalable," claimed Yale engineering professor Tso-Ping Ma. "Flash runs into a brick wall at 25-nanometer node, but FeDRAMs can scale as small as CMOS, which could extend below the 10-nanometer node."
Ferroelectric memories use oxides like lead zirconate titanate that spontaneously form into nanoscale dipoles that can be electrically switched. To make memories nonvolatile, chip makers must shield bit cells from the depolarization fields normally created by the circuitry of silicon chips. The resulting package size makes them non-competitive with flash.
On the other hand, FeDRAMs combine the virtues of both DRAM and flash, and potentially could scale to smaller sizes since a bit cell is essentially a CMOS transistor using a ferroelectric material for its gate oxide.
"Several chip makers have nonvolatilie FRAMs, but they shield the bit cell from depolarization fields by sandwiching the ferroelectric material between two metals, making their cell sizes too large to be competitive with flash," said Ma. "But our cell sizes are very competitive [with] flash."