Inside The MRAM

Today, the industry is shipping various next-generation nonvolatile memory types, such as 3D NAND, MRAM and ReRAM. In fact, MRAM has been shipping for some time. To get a handle on MRAM, Semiconductor Engineering sat down to discuss the technology with Phillip LoPresti, president and chief executive of Everspin, a supplier of MRAMs. What follows are excerpts of that conversation.

SE: Where are we with MRAM today?

LoPresti: We are still the only company shipping any kind of MRAM product. We have several families of products. Our first family is field switch technology. We have shipped over 50 million units right now, both standalone and embedded products. Those fit all kinds of applications, including enterprise, server, storage and industrial control. We also have it qualified for automotive.

SE: Everspin has been shipping those parts for some time, right?

LoPresti: That’s been in production since 2007 and 2008.

SE: What are the applications for field switch?

LoPresti: That part is targeted for either battery-backed SRAM or high-endurance and high write-speed EEPROM or NOR flash applications. In addition, we work with all of the big storage companies for RAID.

SE: What’s new with STT-RAM?

LoPresti: The second family of products from us is another variation of MRAM. Instead of field switch, it’s spin-torque switching, or what we call ST-MRAM. We’ve been sampling that since 2012 or early 2013. That technology fundamentally changes computing and storage. But initially, a lot of people were not sure what to do with it. So getting out products early was really our premise. We wanted people to get out there and start looking at it. We wanted to get an ecosystem established, so people can use it when it’s ready to go into production.

SE: Have you made any progress with ST-MRAM?

LoPresti: We are starting to see that pay off. There are lots of applications that are making the transition for the need for persistent memory. Our first part in the spin-torque family is a 64-megabit DDR-3. Last year, we announced a design-in with a company called Mangstor, which does high-performance SSDs. Now, we are in-house testing our 256-megabit chip, which is also DDR-3. That also triggers the change from 200mm to 300mm production.

SE: Everspin has been shipping parts on 200mm wafers. The change you are talking about relates to Everspin’s announcement last year, when the company formed a 300mm foundry alliance with GlobalFoundries, right?

LoPresti: That’s a big strategic partnership we’ve done with GlobalFoundries to enable our product development and production, which is really key for us.

SE: Many other companies are still working on MRAM, right?

LoPresti: Other companies are tinkering with MRAM. They are coming out with prototypes, samples or test chips. But they are not really targeted to go to production. We’re more focused on a part that can be used in a system. That’s why the partnership with GlobalFoundries is really pivotal in taking us to the next densities like 256-meg. Our team is already in design with our 1-gigabit DDR-4 part. Again, that’s tied to the relationship with GlobalFoundries to have that capability.

SE: What is the difference between field switch and ST-MRAM?

LoPresti: The primary difference is the way the bit switches. We use magnetic materials to create resistive behavior, either high impedance or low impedance. We detect it with a single transistor. The only thing that changes is how we switch the bit from high state to low state. In field switch, it’s exactly the way it sounds. It’s switched with a field. You generate a certain level of current on the chip. You direct them to the bit. You implement some clocking mechanism. And then you can switch the bit into one state to another by generating a magnetic field. In spin torque switching, we use a polarizing spin current that either goes up or down the bit. And depending on the direction of travel, it switches the bit from one state to the other. That technology requires a lot less current to switch bits. As you shrink the size of the bit, there’s a proportion in relationship to the size of the current needed. That allows scalability.

SE: ST-MRAM delivers the speed of SRAM and the non-volatility of flash with unlimited endurance. In the past, you’ve talked about several applications. One is write buffers for SSDs, right?

LoPresti: We are trying to buffer or cache the writes, depending on the densities needed. By adding MRAM to your system, it improves the overall performance and certainly the reliability.

SE: Will STT-MRAM ever replace DRAM?

LoPresti: DRAM has a fundamental problem. You have to refresh it. People want persistence. DRAM isn’t persistent. DRAM will reach a limit and will only start servicing areas where persistence doesn’t matter. That will probably go on forever. Besides its scalable limits, DRAM will be used less. It will get replaced by other technologies in different areas.

SE: Does MRAM compete with NAND, ReRAM or other technologies?

LoPresti: A lot of people ask us how does MRAM compete with the other technologies. We don’t compete with NAND. In a lot of cases, we make NAND better. Resistive memory is coming out. They can’t do our function. We aren’t trying to do their function. They can all work with each other.

SE: What’s the status with the foundry deal with GlobalFoundries?

LoPresti: It’s going well. We’ve received our first silicon, which is based on 300mm CMOS technology at 40nm. This is the 256-megabit chip I mentioned. Since November, they’ve been getting the line up and running. We’ve transferred our process from 200mm to 300mm. So we now have a fully-running 300mm process. We can now target this to various higher-density solutions.

SE: Micron, Samsung, Toshiba and other big companies are working on MRAM. Does that worry you?

LoPresti: We would actually like other people in the market. If there are more people shipping the same product, then customers are more prone to engage in high-volume applications. This is because they don’t have to worry about being sole-sourced. Big companies don’t like it when one of the most important parts in their systems comes from only one company or one fab. So long term, we would like to proliferate MRAM to the biggest memory companies and so on. We don’t see that as a bad thing.

SE: So what are the challenges for the MRAM hopefuls?

LoPresti: This isn’t easy. Any new memory technology is very hard to do. It takes time. You’ve got to stick with it.

SE: Any other challenges?

LoPresti: One thing is being able to get customers comfortable with the technology. That’s why we have a systems group. All of these big companies used to have huge R&D groups. They used to have three groups building the same thing. Then, they would have a face off and decide which one to use. Now, they are not doing any of that. So when a new technology comes and you go to them, they say: ‘This is really great, but I have no time to figure out how to use it.’ So now, even as a memory company, you have to deliver the solution. Hence, you have to build the actual SSD or build the evaluation board to convince people it works. And not only does it work, but it must add value to your system. Those things are challenging. We’re overcoming them and moving in the right direction.