Bringing NVMe/TCP Up to Speed NVMe Technology in the Real World
Guest Post

MRAM: Latest High-Performance Applications

This is a discussion around MRAM and how the technology can benefit your organization.

Download the presentation: MRAM: Latest High-Performance Applications

00:03 Troy Winslow: Hi, my name is Troy Winslow, with Everspin Technologies. Thank you for joining our virtual class at FMS. Sorry, we can't join you in person. Joining me today is Brent Yardley, he is the chief architect at IBM storage systems. He'll walk you through in detail a high-performance application that is utilizing the MRAM. I'm going to start off though, about 15 minutes, a few slides, telling you a story about MRAM and how it could be relevant to your situation today, regardless if we're in this COVID environment for the next year or so. It's a great time to consider the possibilities of a new memory technology that you may have never considered before.

00:57 TW: So MRAM, if you're not familiar, is categorized as an emerging memory technology, but it's actually been around for a couple of decades in development, and over 10 years in production. There's new discrete suppliers and product offerings coming to market throughout 2020 and will in throughout the foreseeable future. And there's discrete memory MRAM and then there's embedded MRAM for a system on chip and microcontroller type memory, and every major memory, OEM or foundry has announced plans for developing embedded MRAM or is actually in production. GlobalFoundries, Intel, TSMC, Samsung and others.

01:57 TW: The industry currently is supplying 128 kbit to 1 gigabit densities, and I know that may not sound like huge densities when it comes to comparing to DRAM or NAND, but there is a huge market, SRAM, FRAM still plays in the kilobit and low megabit densities. MRAM has been able to scale up to one gig and it's a scalable technology that is using more mature process lithographies, all the way from 180 down to 28 and 22, but it has the capability of going down all the way to 12 nanometer. Everspin and GlobalFoundries have announced a joint development agreement to go down to 12 nanometer. So it's scalable unlike some other maturing memory technologies that are no longer scalable, and that gives the promise that MRAM can continue to grow in its interface offerings, but also its density points. And most importantly, for many of you, its cost effectiveness. And so again, MRAM while categorized as emerging technology is definitely out there, it's proven, and we're one of several vendors now offering discrete memory in the marketplace. Competition is good for the industry, as well as providing second sources.

03:39 TW: But going back to a story I wanted to tell you is, we all remember SSDs versus hard drives, and what we saw there is the promise of a much higher performance and durable drive provided by NAND and SSDs over the incumbent hard drives. Now, did one totally displace the other? No, hard drives are still around today, and they have a role in our storage architecture. But SSDs have created a new hierarchy and innovation continues pretty significantly and has drove the SSD business into the tens of billions of dollars. Yeah, it was a much higher price technology back when it was originally launched, less than 20 years ago from SATA and SAS, now we have PCIe and NVMe interfaces, and that is driving innovation and improvements in the architecture and the performance of your systems.

04:51 TW: And I just ask that you consider MRAM as just as viable a technology with just that potential, because it really does have all the makings of what the industry has been looking for as a universal memory. So the four building blocks of MRAM are its persistence, its performance, its endurance and reliability. And persistence, the non-volatility means that if the power goes out, your data is still protected. That also has to do with how long that data is protected and some MRAM suppliers, Everspin including can provide up to 20 years of data protection, and... I'm sorry, retention. Pretty incredible performance metrics, and I'll show you on the next slide, but MRAM has the ability to deliver SRAM or DRAM like performance, and we're not limited from a write performance like NOR, but equally can deliver reads and writes with low latency and high bandwidth. So that's a key building block. But the endurance is a key point of MRAM, that it's a non-destructive cell, and so we can deliver essentially unlimited performance in your applications, very similar to SRAM and DRAM. One E 10 to the 15th, endurance capabilities.

06:35 TW: Now, some applications, like the data center, don't require that. And we have products that deliver less because there are trade-offs to make when it comes to design, but the capabilities of the MRAM are unlimited endurance. So, that provides for a potential, all the working class persistent non-volatile memory, and then reliability. So, this is one of the key points that we've been able to establish after 10 years of being in production with our MRAM, is very few. Less than 0.5 defects per million and field failures, and most of those were not even dyer related, they were packaging related. So, unbelievable reliability. It's also inherently radiation tolerant, so that gives it broad applications in space and yes, our data center is going out into space.

07:37 TW: So again, for code execution or data storage or a combination of the two, MRAM is uniquely positioned for the next universal memory or the first universal memory. And here you can see where you have other non-volatile memories in the lower right-hand quadrant from a slower latency, or a higher latency and lower endurance. And many of those are scaling the wrong way, especially when it comes to an endurance. But MRAM has the ability and is constantly increasing its endurance, as well as lowering its latency and improving its performance. So, you can see our toggle memory is already at that virtually unlimited cycling capability in endurance. As well as near DRAM-type speeds from a serial, and we offer a toggle and a serial and parallel interface.

08:42 TW: Our STT-MRAM, which is characterized in our one gigabit device, is currently designed for the data center, so endurance wasn't a key priority, but inherently in the technology, and we've even announced that we will have an STT product next year that will deliver one eighth to the 14th endurance with an STT, as well as improved performance. So, exciting stuff, exciting scalability and we believe MRAM is a superior technology on many aspects. Again, there's uses for the other memory types, especially from a storage class in high capacity storage. Just like there's plenty of uses for SSDs and hard drives, the examples are multiple throughout our industry.

09:46 TW: But the available market continues to grow, and you can see here, I've just highlighted in yellow, the data center, the other 50% of the market is in industrial and IOT, which also drives information and data to the data center. But that has a 28% CAGR, depends on the analysts and your judgment of that, but it's a significant available market. And even our favorite analysts at Flash Memory Summit, Tom Coughlin and Jim Handy, they're forecasting that stand-alone MRAM by itself will not reach a TAM, but actually a revenue of $4 billion by the end of this decade. Now, it's because more and more people are adopting it. Everspin, we have over 1,300 customers already in production on our devices. So, I encourage you to consider MRAM when you're thinking about and architecting your next product. But, for today, we're focused on the enterprise, and whether it be storage, server storage, all-flash arrays, hardware accelerators or in traditional RAID, because RAID requires lower density. So, let me talk about the RAID here in a second.

11:14 TW: But whether you're utilizing a one megabit serial device in your RAID application or you're looking at higher density up to the one gigabit, you're looking for a low latency, non-vol memory that can generally provide power loss protection, yet deliver a non-chemical, solid-state persistent memory, simplifies the design. You're able to improve your performance, whether it be latency, quality of service or a higher number of streams. You're simplifying it because you no longer require the same number of capacitors and backing up your RAM, and that of course, results in smaller board space and a simpler design and longer life because you don't have the maintenance associated with that. So, plenty of examples of the benefits of using MRAM. In RAID, we're talking, again, the one to four megabits for journaling and you can see from the block diagram that for error logs and system event logging, RAID devices are using MRAM today.

12:39 TW: And, that's a key component of the data center applications but in essence, because of its range of applications, its range of performance, this is an example where you have a particular MRAM that is specifically suited for your requirements. Again, whether it's low density serial or parallel, or if it's high density from a DDR aspect. So, I want to save time and turn this over to Brent and how IBM has used MRAM in their approach in a variety of different value-added propositions and deliverables. But if you're interested in MRAM, I encourage you to visit our website or contact me directly. But right now, one, I want to thank you for joining this broadcast and now I'm going to turn it over to Brent. Brent, take it away.

Dig Deeper on Flash memory and storage

Disaster Recovery
Data Backup
Data Center