News and New Products

When will emerging memory arrive?

Driven by several factors, new approaches to memory are starting to materialize as alternative memory start-ups try different routes to a market they have high hopes for.

By Tam Harbert, Contributing Editor -- Electronic Business, 10/6/2009

The problem with emerging memory is just that – it always seems to be emerging but never quite arrives.

Over the last several years, a slew of start-ups as well as some major semiconductor companies – including Samsung and IBM – have been trying to figure out what comes next in the memory business. That’s because current memory technology – particularly NAND flash – is reaching its scaling limit. The common wisdom is that NAND can’t scale below 20 nanometers, according to Jim Handy, analyst at Objective Analysis. “So there are a lot of people focusing a lot of effort on whatever the next technology will be.”

On the other hand, such common wisdom in the past – for example warnings in the 1980s that DRAM had reach its limit – has been wrong. “What usually happens is that somebody comes up with a novel way of using existing technology, which adds several generations to the technology’s life,” he noted.

The jury is still out this time around. Companies are examining a variety of approaches, including phase-change memory, resistive RAM, and magnetic RAM. Some start-ups, such as MagSil, EverSpin, and Crocus Technologies, have been around for at least a couple of years. But a handful of new companies that emerged from stealth mode just in the last few months have renewed attention to emerging memory technologies. These include Adesto Technologies, Four Dimensional Storage, QS Semiconductor, and Unity Semiconductor.

In addition to the idea that the end is near for NAND, several other factors are behind the new approaches to memory.

First, memory is becoming the place in a system where an OEM can differentiate and add value, said Bob Merritt, founding partner of Convergent Semiconductors, a consultancy that focuses on semiconductor memory. Single-core microprocessors, such as Intel’s Atom, are becoming commodities in the mobile market, particularly in netbooks, he said. With these products, it’s not so much about the processing power as it is about the characteristics of the memory, including its power consumption and volatility.

Also, there is a lack of innovation among most of the established memory companies, according to Darrell Rinerson, founder, chairman, and CEO of Unity Semiconductor. Business has been so bad for most memory manufacturers that they have not generated enough cash to invest in R&D. “To innovate, you have to have the money. Some memory companies just have not had the ability to do that.”

And yet, those very memory companies are the ones with the big fabs – capacity that has been idle for awhile. One key aspect for any of these emerging technologies is to figure out how to tap into that manufacturing capacity. “The challenge for these emerging memories is not only to develop a process, but then to find a way to manufacture it reliably and cost effectively,” said Barry Hoberman, a marketing consultant for Crocus Technology. “That manufacturing source is a huge hurdle for start-ups to get over.”

Most emerging memory vendors have adopted a two-stage manufacturing scheme while in the early phase of development, said Handy. They source standard CMOS wafers from foundries, and then build their own proprietary layer on top of that at another facility.

For example, EverSpin, an MRAM company spun out of Freescale Semiconductor that has been manufacturing product for more than three years, uses a two-stage process. The company sources standard CMOS logic wafers from foundries, bringing them into its leased space at Freescale's Chandler, Ariz, fab and using a machine from the hard disk drive industry to deposit magnetic material and connect the transistors, said Saied Tehrani, co-founder and chief operating officer of EverSpin.

MRAM company Crocus Technology has been doing something similar during its development phase, but has not shipped any product. In June, however, it announced that Tower Semiconductor had agreed to port Crocus' MRAM technology into Tower's fab. Tower also took a $1.25 million equity position in Crocus. Although the companies did not say when they would begin manufacturing products, “we are the only MRAM company to announce even a path to a single-site wafer flow,” said Hoberman.

Unity Semiconductor, which has developed a memory technology called CMOx, isn’t looking for a single site. In fact, it is the only new memory company that intends to keep a two-stage process for volume manufacturing, said Handy.

Using two sites is actually an important part of the company’s business plan, said Rinerson. Unity intends to source standard CMOS wafers from a foundry, and then build its own proprietary technology on top of those wafers at the fab of a joint venture partner, which the company plans to announce by the end of this year, he said. The joint venture partner – which Rinerson said will be a large IDM (integrated device manufacturer) – presumably has empty fab space available. “We can start with a shell and then put in enough equipment to do a certain number of these back-end wafers,” he said. That will cost a lot less than building a fab. “We can get into the manufacturing game for maybe $500,000,” he estimated. The joint venture partner will get an exclusive license to manufacture Unity’s devices. Unity plans to limit licensing to only its joint venture partner, so that it can avoid the commodity pricing that has plagued the traditional memory market, said Rinerson.

Vijay Rakesh, senior analyst at ThinkEquity, said Unity’s approach makes sense. “There’s plenty of manufacturing capacity out there,” he said. “This is attractive to a lot of these DRAM guys who are in trouble” because it would enable them to supply the market, yet at the same time avoid spending the $3 billion it typically costs to build a new fab.

This assumes, of course, that current memory technologies are indeed losing steam. As long as they stay cheap and plentiful, emerging memory will have a tough time finding a broad market, said Merritt. Rather, they will have to compete on strong performance attributes and savvy business plans.

“How do you convince people to buy this new stuff when DRAM and SRAM vendors are already losing money?" Merritt asked. "There is no way you're going to make money by competing with something that's losing money. Nobody will buy [emerging memory] unless it's cheaper than the others.”