By Mark LaPedus, SemiMD senior editor

Several years ago, an executive from Intel Corp. declared that flash memory would run out of gas and would stop scaling at about 90nm, prompting the need for a next-generation — or universal — memory type.

FRAM, MRAM, phase-change, ReRAM and others fall into the so-called universal memory category. Developers of these technologies claim their respective technologies can replace DRAM, NAND, NOR — or all three.

Seeking to jump on the “next big thing” in memory technology, Elpida, Hynix, Micron, Toshiba, Samsung and a number of startups are scrambling to develop various next-generation memory types — and for good reason. Memory scaling could soon hit the wall and there could be a huge market for a new category of storage-class memories and other products.

But in general, most next-generation memory types have failed to live up to their promises. Most technologies are still not in production despite years of R&D. They are expensive to make and difficult to scale. Vendors also have no clue when they will ship their product in volumes.

And despite the early predictions from Intel, flash and DRAM have become cheaper — and scaled far beyond what was originally thought — thereby pushing out the need for the next-generation memory types. In NAND, for example, the Toshiba/SanDisk duo are shipping the world’s most advanced chips at 19nm. In NOR flash, Micron Technology Inc. is the world’s first vendor to move to 45nm. And several DRAM makers are sampling sub-30nm parts.

Is the end near for memory scaling? Louis Parrillo, chief technology officer of Unity Semiconductor Inc., said the floating gate structure for today’s flash devices is expected to hit the wall and end around 18nm to 15nm.

At that node or before, both flash and DRAM makers also face what Parrillo called the “Achilles Heel” in scaling:  lithography. “Planar CMOS requires fine lithography like EUV,” but the trouble is that there are uncertainties and throughput issues with EUV, he said during a panel at Applied Materials Inc. on Wednesday (Feb. 1). The panel was sponsored by MIT Club of Northern California.

Instead of scaling, future next-generation memory devices will go vertical, he said. Vertical memories will not require bleeding-edge processes, nor EUV. “I think there is an opportunity for us,” he said.

Seeking to replace NAND, Unity is developing a so-called CMOx Cross-point Memory, a vertical, four-layer memory array and peripheral control logic device. The memory effect of CMOx is created by moving oxygen ions between two metal oxides under an electric field.Last year, Unity entered into a joint development agreement with Micron to develop CMOx-based chips within Micron’s fabs. Micron is an investor in Unity. The company reportedly hopes to roll out is first devices in 2014.

There are several pressing questions for Unity and other developers of next-generation memories. First, what are the driving applications for newfangled devices? Second, when will vendors ship? Third, can vendors drive down the cost but boost the performance? Fourth, will the vendors with deep pockets ultimately survive? And finally, will risk-adverse OEMs adopt these products?

Ishai Naveh, co-founder and vice president of marketing and business development for Adesto Technologies Inc., said the market is ripe for a new memory type. “We believe the market is changing,” he said during the panel. “People are looking for new types of memory.”

There are many challenges for next-generation memory startups, however. For a mainstream DRAM or NAND vendor, the cost is $500 million for process R&D alone to move from one node to another, Naveh  said.

Obviously, a startup does not have that type of funding. Another challenge is to get qualified by major OEMs, many of which are risk adverse and may not take a chance on a new technology. OEMs also want multiple, pin-compatible suppliers in each segment, and, of course, cheap parts.

So, next-generation memory startups must look for more specialized or niche markets, he said. “Unless you go with someone like Samsung that buys you, you need to create a new market,” he said.

In fact, Samsung Electronics Co. Ltd. recently acquired one next-generation memory startup – Grandis, a developer of spin-torque MRAM technology. Some speculate that the big memory houses will end up gobbling up the smaller players.

In any case, Adesto, a developer of conductive bridging RAM (CBRAM) memories, appears to have found a niche – at least initially. It has been sampling its first product – a 1-megabit part geared to replace EEPROM.
CBRAM was originally developed at Arizona State University and is also known as Programmable Metallization Cell (PMC).  The technology has been shown to be scalable down to 12nm. Applied, Altis and others are the investors in Adesto.

Others are already shipping. For example, Everspin Technologies Inc. – the MRAM spinoff from Freescale Semiconductor Inc. – is the only MRAM vendor to be shipping product in volumes for the commercial market.  Everspin is shipping first-generation MRAMs based on a toggle-write technology, mainly for the SRAM replacement market.

For some time, the company has been shipping MRAMs for automotive, energy and industrial applications. “For us, we are selling into all kinds of applications,” said Jon Slaughter, vice president of technology, research and development at Everspin, during the panel. “The biggest applications for us are enterprise-class data storage.”

Everspin’s MRAM products use a one transistor, one magnetic tunnel junction (MTJ) memory cell for the storage element. The next trick for Everspin is to develop a next-generation spin-torque MRAM (ST-MRAM), which is due out in 2012. ST-MRAM uses an alternate method for programming an MTJ element that has the potential to simplify the MRAM cell and reduce write power for storage and other applications.

Another ST-MRAM developer, Avalanche Technology Inc., outlined the recipe for success in the next-generation memory race. “Performance and cost, cost, cost,” said Rajiv Ranjan, chief technology officer for the startup.

Ranjan sees a new and emerging application for Avalanche’s so-called Spin Programmable Memory (SPMEM), which has yet to be introduced. It is ideal as a replacement for the embedded DRAM space, he said during the panel.

Rob Aitken, a fellow for ARM Holdings plc, believes there is a pressing need for a new technology in the embedded memory space. “There is no viable alternative to embedded SRAM,” he said during the panel. “We see it lasting a couple of more nodes,” possibly to 14nm.

The big opportunity for next-generation memory types includes mobile devices, cloud computing and solid-state drives, said Steve Hudgens, the former chief technology officer at Ovonyx Inc., who is now a consultant to the provider of IP for phase-change memories.

Some call this segment storage-class memories. In fact, there is a pressing need for a new and faster memory that sits between the processor and DRAM in a system to boost I/O performance. In theory, the so-called storage-class memory would offload many of the functions in a power-hungry DRAM.

Others believe that a storage-class memory could one day replace the DRAM, especially in datacenter servers. DRAM remains a power hog despite the shift to smaller geometries and so-called green architectures.

It’s unlikely a next-generation memory would replace NAND – and for good reason. “NAND is almost free,”  Hudgens said.

Tags: Adesto, Elpida, Everspin, Hynix, Micron, phase-change, Samsung

This entry was posted on Thursday, February 2nd, 2012 at 5:04 pm and is filed under News Stories. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.