by Andy Patrizio
Buried in a report out of Japan earlier this year, Toshiba disclosed plans to release ReRAM products in 2013, making it either the first or one of the first to bring what promises to be a revolutionary form of non-volatile storage memory to market, if it delivers on those promises.
ReRAM, or RRAM, is short for Resistive Random-Access Memory, a non-volatile memory type under development by a large number of memory and semiconductor makers, including Toshiba, Micron, HP, IBM and Rambus.
ReRAM promises the speed of DRAM and the storability of NAND flash. The potential, then, is eliminating the line between storage and execution memory. Right now, you have to load an app or data off of an HDD or SSD into memory, but since ReRAM is as fast as DRAM and holds its data, there is no need to separate execution memory from storage.
The question is when it could get to market, shaken out, proven and made affordable. SSD has actually been around more than two decades; the drives for laptops first appeared in 2006, and now enterprise servers are just beginning to warm to SSD. The tech industry loves a revolutionary product but it’s often slow to adopt it.
Toshiba will begin sampling 64Gbit ReRAM test chips in 2013, with engineering samples in 2014 and volume production in 2015. They will be stacked in up to 16 layers. Unlike standard memory, ReRAM will stack its memory up, rather than spreading it out in two dimensions, to achieve greater density in a smaller space. The devices will use a three-dimensional structure with 16 layers of stacking to provide maximum density.
At the same time, the company plans to ship BiCS NAND 128Gbit and 256Gbit Flash memories on the same time frame. ReRAM has faster writes than BiCS NAND, so ReRAM will be for high speed storage.
BiCS NAND is a new process design that Jim Handy, principal analyst with Objective Analysis, said would be Toshiba’s successor to the existing NAND flash design. BiCS stands for Bit Cost Scaling and its Toshiba’s method of manufacturing smaller NAND memory, which has proven really hard to get below 20 nanometers.
“I would expect Toshiba to run with NAND, then move to BiCS for a few generations, then maybe after that ReRAM has a chance of displacing NAND. That means four generations of NAND, and if you figure on two years per generation, that’s eight years before ReRAM has a chance at being a financial success,” he said.
So don’t get your hopes up for a ReRAM device any time soon, because it is not close to market. NAND still has a lot of life and keeps finding ways to live on. A few years back, Handy said engineers thought they would never get NAND flash below 60nm. Then they thought 35nm was the bottom. Now it’s at 20nm and creeping down into the teens.
“These different technologies are all contenders to take the place of NAND when NAND stops scaling, but those geniuses who find ways to move NAND to the next process geometry keep pulling a rabbit out of the hat. So the new technologies can never get to the economies of scale against an established market like NAND,” he said.
At this point, ReRAM looks to cost several times as much as NAND flash. To make ReRAM memory, you have to take a process as complex as NAND manufacturing and then put another layer on top of that, he said.
When NAND stops scaling down and ReRAM gets one to two process nodes smaller then NAND, then it will be feasible, said Handy, and not before. ReRAM is thought to be able to shrink down as low as 6nm, while companies like Toshiba, Micron and Intel are wrestling with 19nm for NAND.
“Everyone knows NAND will stop scaling. The only tricky thing is they can’t say when that will happen. If ReRAM is well poised when that happens, then it could take over the market very quickly,” said Handy.