Intel Confirms Its 22nm FinFET MRAM Is Production-Ready
MRAM and the associated grab-bag of alternative memory technologies have long seemed to occupy an unfortunate no man’s land between “Never going to happen,” and “Next Big Thing.” When Intel debuted its own 3D XPoint technology 3.5 years ago, it seemed as if the already narrow space for these alternative memory technologies might be squeezed further. But that hasn’t happened. Instead, Intel has moved ahead with its own research into MRAM and has designed a version of the technology compatible with its own process nodes.
At the International Solid State Circuits Conference (ISSCC) this week, Intel presented a paper on its own spin-transfer torque MRAM (STT-MRAM). MRAM is a promising technology in ultra-low power circuits, IoT devices, and next-generation embedded devices because it can lower the total power consumption required for memory accesses while offering reasonably high performance and a smaller physical die. According to EETimes, Intel has created 7Mb perpendicular STT-MRAM arrays on its 22FL FinFET process.
22FL isn’t a process node we’ve written much about, but WikiChip has additional details. Wikichip writes that 22FL is designed for ultra-low power, RF, and low design costs. Despite carrying the 22FL moniker, the process node is apparently based on a relaxed variant of Intel’s 14nm node, with very low leakage (lower than what Intel specifies for its own transistors on typical process nodes) and a low Vmin. Yields on the new MRAM are supposed to be quite high, with a bit yield rate of 99.9 percent.
The idea that Intel developed this technology for its foundry customers is interesting, in part because the firm has been so quiet about who its foundry customers actually are. GlobalFoundries has offered MRAM technology on its 22FDX node since 2017, but it isn’t known if anyone is actually using the technology. Intel has been quiet about its own design wins as well, but EETimes reports that the technology is believed to be shipping to customers. Chipzilla is apparently also working with ReRAM, with a goal of developing low-cost embedded arrays for IoT and automotive applications as well.
The commercialization challenges for technologies like ReRAM and MRAM are still real — there’s a reason the CPUs inside our collective machines use SRAM and DRAM rather than MRAM today — but the growth of new computing markets and the need for power-saving technologies that don’t rely on process node shrinks to deliver improvement means research into these alternative approaches will continue. And Intel, apparently, has foundry customers sufficient to drive investment into both technologies, despite how quiet the firm has been about its partnerships.