Everspin Targets Niches for MRAM

　　Consistent challenge with emerging memories: the volume must go up for the price to come down, and the price has to come down for the volume to go up. Everspin Technologies is hoping, however, that specific industries will favor value over price-per-bit as it ramps up production of its MRAM.

　　The MRAM pioneer just released an application white paper for its nvNITRO solution being jointly delivered with SMART Modular Technologies. The two companies collaborated on the development of the nvNITRO NVMe storage accelerator and are focusing on jointly developing several vertical markets, the first being the financial technology (FinTech) sector, where nvNITRO can increase data throughput and reduce critical storage bottlenecks, according to Pat Patla, senior vice president of marketing at Everspin.

　　In a telephone interview with EE Times, Patla said the nvNITRO NVMe storage accelerator enables a 90 percent reduction in latency for financial transactions, which helps fintech companies meet compliance rules that often need synchronous logging. This adds latency and creates transaction processing bottlenecks. As outlined in the whitepaper, nvNITRO's ability to reduce latency so significantly enables fintech companies to log transactions faster so they can increase their overall transaction volume, and ultimately, their revenue. Meanwhile, the persistence of STT-MRAM also contributes to necessary compliance as all data logs are protected.

　　The vertical focus comes as Everspin has started mass-production of its STT-MRAM. The company hit a notable milestone of recording revenue for its first 40nm 256Mb STT-MRAM products in the fourth quarter of 2017 and is ramping its volume production in 2018. There are handful of companies working an MRAM, but Everspin is the only company that's commercially delivering both discrete and embedded MRAM products, thanks in part to its collaboration with GlobalFoundries.

　　Everspin is touting its 256Mb STT-MRAM as the first ever perpendicular MTJ STT-MRAM entering mass production. Patla said its ST-DDR3 / ST-DDR4 interface was designed to be as close to standards as possible so that customers can easily integrate it into their systems with a couple of tiny changes, primarily to the memory controller. Everspin's STT-MRAM is also NVMe-compliant.

　　Patla acknowledges that until there's a standard in place, there will be concerns from customers about sole-sourcing the technology. “DDDR5 is where everyone is convening," he said.

　　The company is focusing on areas where there is a need for fast, persistent memory by offering near-DRAM performance combined with non-volatility. “Flash offers great performance next to rotating media, but against DRAM or MRAM, it significantly slower," Patla said. He describes spin torque MRAM as a three-legged stool comprised of performance, endurance and bit air rate, with non-volatility has the seat.

　　Aside from specific industry verticals such as fintech that see value in adopting MRAM, Patla said Everspin is targeting three buckets: artificial intelligence (AI), edge computing and digital transformation, which includes data center applications. Because MRAM only users power when reading and writing, it's ideal for edge computing environments as demand for processing outside the data center increases and Internet of Things deployments grow.

　　Price is not the key driver of MRAM adoption, said Patla. Instead, specific applications that need fast, persistent memories are guiding customers' decisions. “People are buying based on the value proposition rather than dollars per gigabyte," he said.

　　Right now, the price of MRAM is still rather high, said Thomas Coughlin, founder of Coughlin Associates, but it is the most interesting emerging memory technology because its performance is close to SRAM and DRAM, and its endurance is very high. Coughlin said MRAM makes sense for cache buffering, and for specific applications, such as the nvNITRO NVMe storage accelerator for financial applications.

　　“Doing a transaction quickly is important, but having a record is just as important," Coughlin said. "It's an interesting case study. Time is money, and they're making more money than they're spending."

　　Coughlin said specialized niche markets are willing to pay for the combination of speed and non-volatility of MRAM. He noted that Everspin has already shipped more than 70 million units, and its partnership with GlobalFoundries for wafers and embedded products opens up possibilities for increased volume. Satisfying niche markets that pulls in more volume will help Everspin's MRAM become more mainstream. “That's where GlobalFoundries' products come in," he said.

　　The interesting possibilities for MRAM are bolstered by the fact that Samsung and TSMC will be producing MRAM products this year, too, said Coughlin. “That leads me to the think there's a general perception MRAM has a place," he said.

　　As other MRAM makers mature, such as Spin Transfer Technologies, competition could lead to a squeeze on prices. Having more than one supplier creates a more vibrant market, Coughlin said, and 2018 will be an important year in terms developing a viable supply chain.

　　MRAM is not going replace SRAM and DRAM right in any applications right away, Coughlin, but there is a gap between those memories and NAND flash to filled. “3D Xpoint is trying to fill this gap, but is still far away from DRAM," he said. It's also important to remember that flash took decades to become competitive with hard drives, he added.

　　In the meantime, ReRAM and FRAM are both worth watching as emerging memories to fill the gap between DRAM and flash, said Coughlin. “It's a veritable zoo of technologies and we'll have to wait and see which animals survive the evolutionary process," he added.