Eversping has a bunch of MRAM announcements today starting with the first Gigabit device. On top of that there are a few PCIe based accelerator boards to talk about.
SemiAccurate has always thought Everspin technology was interesting and saw a lot of potential for the products. Until today you could only get their MRAM in 256Mb capacities, enough for some applications but a bit of a psychological downer to be in the Mb class in a Gb world. Sure the tech was radically different but those numbers do make some skeptical. As we mentioned that changed with the introduction of today’s 1Gb class MRAMs. They are built on a Globalfoundries 28nm process, something SemiAccurate previously told you about, and will probably port down to the 22/20nm node soon enough.
Everspin is showing off a 1GB and 2GB DDR4 form factor MRAMs based on the new 1Gb die. The former is 8x 1Gb dies while the latter is a 8x dual-die packages. Both should be able to plug into any DDR4 capable device that has the correct timings, nothing radical, just a few tweaks here and there. If you have a Xilinx device, you can make it 1Gb Everspin MRAM compatible with a simple script.
This 1Gb device is sampling now and there is supposedly an NVMe over fabric demo with Mellanox at Flash Memory Summit now. Better yet the MRAMs support both block and byte modes, can be memory mapped, and in general do what you want. There is SPDK support if you want to play around on your own.
On the end-user product side, Everspin is announcing new nvNITRO accelerators in 1GB and 2GB sizes. These are based on the older 40nm 256Mb MRAMs in DDR3 form factor. The nvNITRO devices are PCIe3 8x devices in half-height or M.2 form factors. These devices deliver a claimed 6μS latency and unlike some others, can do it consistently with consistent power. Deliveries will start in Q4 at $2200 or so.S|A
Latest posts by Charlie Demerjian (see all)
- Mediatek Sensio brings six biometric measurements to devices - Dec 14, 2017
- AMD Adrenaline drivers bring new GPU features - Dec 12, 2017
- Intel doesn’t want you to know about Gemini Lake Atoms - Dec 12, 2017
- What is going wrong on Intel’s 10nm process? - Dec 11, 2017
- More on the Qualcomm Snapdragon 845 architecture - Dec 11, 2017