Information demands are increasing dramatically as digital transformation and other business trends are creating the need for more real-time decision making. Collecting, transmitting and storing the data that helps drive business insights is putting strains on businesses as they grapple with optimizing the increasing flow of data. Nowhere is this taking its toll on traditional systems more than in storage, where both the volume and critical nature of the information is driving rapid changes in how data is handled and tiered. Storage is both the bottleneck of most environments while simultaneously being the most critical component of any application.
Everspin created its nvNITRO™ technology to help address the growing needs for faster and more persistent storage. Built on magnetoresistive random access memory (MRAM) that is fabricated by GLOBALFOUNDRIES, nvNITRO brings both high performance and persistence to data storage, enabling a new generation of application performance.
We recently showed the power of nvNITRO at Supercomputing 17, the worldwide event for high performance computing.
In a demonstration with SMART Modular Technologies, SMART’s NVMe accelerator card was able to drive high performance with ultra-low latency. The demo showed an NVMe accelerator acting as a front-end buffer for an enterprise SSD. While SSDs are transforming businesses today and all flash arrays are gaining popularity because of their performance advantages over rotating media, NAND memory still can’t match the high speed and low latency of MRAM.
Transaction processing is just one of the areas where we see opportunity for MRAM. In these environments, to guarantee the integrity and compliance of transactions, many systems require logging or journaling of each transaction before beginning of the next new transaction. These applications – such as banking, payment processing, stock trading, e-commerce, supply chain, or ERP/CRM – can all benefit from nvNITRO technology. As message traffic increases, that additional logging can become a bottleneck if not handled quickly and efficiently.
With an MRAM storage accelerator as a front end to an SSD, transaction logging can be achieved in a fraction of the time required with just an SSD. The lower latency of MRAM means that these logs can be written faster, freeing the system up to begin the next transaction without delay. nvNITRO’s 9X reduction in latency, through the use of MRAM, means more transactions can be recorded per second, bringing the potential for greater overall application throughput.
The other key benefit that MRAM delivers is its ability to maintain the state of the data without requiring batteries or supercapacitors. For these businesses, writing huge volumes of transactions also presents a second challenge beyond speed – maintaining the data regardless of the state of the underlying system. Typically, when a system loses power or has a power interruption, transactions that are “in flight”, either being written, or being journaled, can be lost because standard DRAM memory is not persistent and the NAND memory in SSDs just can’t write fast enough to capture all of the in-flight data before power is lost. With the persistence of MRAM, this data can be written out faster, reducing the data stored in the buffer. If the system does need to restart, that data would still be persistent in the MRAM upon initialization. In a world where regulatory entities scrutinize every transaction and may need a financial company to “replay” its transactions, ensuring everything was logged properly the first time is invaluable. This sort of protection goes beyond just protecting the data; at that point, it is actually protecting the company.
The traffic at Supercomputing was brisk and we were happy to see the level of excitement that our demo was producing. Technology like MRAM can become a great foundation for many future platforms. The ability to integrate nvNITRO technology into storage solutions through a variety of interfaces – directly as a PCIe or U.2 device, integrated into the chassis or integrated directly into the system boards – means there is a wide variety of implementations to match specific needs. Discussions about nvNITRO always start with the specific use case being shown, but eventually becomes “hey, could you…?” And that is where it gets interesting.
Along with the STT-MRAM that we were displaying in the nvNITRO demo, STT-MRAM is also available as embedded MRAM (eMRAM) through GF for those applications that demand the persistence, durability and write performance that embedded flash (eFlash) cannot deliver. As we see growth in areas like drones, IoT and autonomous vehicles, the value of embedding MRAM directly into designs will grow. Today’s nvNITRO solutions are built on Everspin 40nm STT-MRAM technology that is produced by our partner GF. Additionally, GF is now offering process design kits for 22FDX eMRAM. GF expects customers to start prototyping MRAM on multi-project wafers (MPWs) to start in Q1 2018.
We see the immediate opportunity today in accelerating the storage of massive data streams. These large amounts of telemetry need to be efficiently handled in a manner that ensures both fast capture and long-term retention. But as MRAM and eMRAM continue to gain market momentum (moving traditional memory and storage products aside) and the form factors shrink, we will see even greater opportunity present itself. Today we are accelerating the back-end storage and processing piece of the equation but it is not a stretch to see MRAM and eMRAM potentially integrating into the front-end and edge devices that are creating this data – and that is where things begin to get even more interesting.
If Supercomputing 17 was any indication, the future is bright for MRAM.