Freescale claims reliability as the motivation when describing its switchover to nanocrystals from both the split-gate and single-transistor (1-T) flash architectures it uses today.

"With traditional flash you have to create islands—slicing the floating gate layer apart for individual bit cells—but with nanocrystals you don't have to slice them, because they are already in discrete units," said Gowrishankar Chindalore, flash memory engineering manager at Freescale.

"Even tiny defects will eventually leak away the charge on a traditional floating gate, but nanocrystals are far enough apart that any defects will only drain individual crystals."

For over 15 years Freescale has been using polysilicon floating gates. About five years ago the company realized that polysilicon gates were going to be subject to process uniformity woes beyond the 90 nanometer node.

Freescale built prototype nanocrystal memory devices in 2005, and over the past four years has also experimented with using nitride charge traps as an alternative means of isolating charge and thus tolerating defects at advanced nodes.

But nitrides could not stand the high temperatures required by the automotive applications for which many of Freescale's microcontrollers are used. Nanocrystalline floating gates, on the other hand, are naturally immune to high temperatures just as they are naturally immune to defects.

"You can bake a nanocrystal part at 150 degrees C [302 Fahrenheit] for 20 years, and you will still be able to read the data inside it," said Chindalore.

Modern automobile engines depend on their flash programs to run properly, so longevity and zero-faults are tantamount to meeting stringent safety requirements demanded by car makers.

And because the nanocrystalline film does not have to be cut into gates, it also reduces the number of mask layers needed thus lowering costs. A standard CMOS line can be used too, simplifying the processing, and costs, further, according to Freescale.

Today Freescale uses two different floating-gate architectures for its flash memories, depending on the application, namely, either split-gate or 1-T. But in the future, the company sees more and more applications for silicon nanocrystal-based memories.

"We used split-gates today when we wanted less logic overhead, and we used 1-T when we want the highest density arrays, but silicon nanocrytals give us the best of both worlds—its a split-gate architecture so it has less logic overhead, but nanocrystals are even higher density than our 1-T solution, so in the future Freescale will only need a single flash architecture," said Chindalore.

Freescale is advancing future developments to overcome the challenges that Angstrom-scale defects impose on the efficiency of its current split-gate and 1-T architectures. But nanocrystalline films divide floating gates into more than 100 islands each only 100-to-150 Angstroms in diameter—far enough apart not to affect each other.

The company is detailing the developent at the IEEE International Memory Workshop (IMW, May 10-14, 2009, Monterey, Calif.).

Freescale is demonstrating its new silicon nanocrystal process with a 32 Mbit test chip at the workshop.