An emerging class of persistent memories (HP Memristors) aims to break a bottleneck in storage access time that is one of the obstacles in delivering higher computing performance. However both the chips and a programming model for them are still works-in-progress.
Some databases already support in-memory operations with non-volatile flash.
To accelerate efforts, HPE released Atlas, a programming environment for servers using non-volatile memory. It uses today’s flash-based non-volatile DIMMs to start work on a new programming model for systems with persistent memory.
Atlas can speed up operations by an order of magnitude compared to systems relying on hard-disk storage, assuming read/write speeds in hundreds of nanoseconds, Chakrabarti said. “We are far from an end state, but we are at a point where we can try it out,” he said.
The environment runs on servers using NVDIMMs that typically employ 8-16 Gbytes of flash, but with next-generation memories such as Intel’s 3D XPoint they could pack 500 Gbytes to a TByte running at much lower latencies. Unfortunately it could be two years or more before such chips are widely available, observers said.
He expects the hardware to be available for use with databases and storage file systems in about 2020. It could take 5-10 years before it is enabled in programming languages for a wider array of server applications, he estimated.
“The hope is people will start to write those apps even of they do it through libraries to convince standards committees to put persistent memory support in programming languages so it gets supported in compilers and tool chains,” he said.
MRAM and Other Memory
Everspin has released 256 Mbit devices and is promising a Gbit chip next year for its MRAM technology which has faster write speeds and longer endurance than XPoint. For its part, the XPoint chips will sport faster reads and a hundred-fold greater density.
Startup Spin Transfer Technologies (Fremont, Calif.) is expected to bolster the case for MRAM with news next month about its plans. “It adds more weight when you see a second or third company using the technology,” Niebel said.
Looking further out, Western Digital promised to deliver before 2020 an alternative based om resistive RAM. Sony is working on its own variant aiming at machine learning which could be paired with its imagers for computer vision apps. And startup Nantero claims by 2o19 it will have Gbit chips using its carbon-nanotube approach.
“The acid test is getting volume products to market at Gbit density,” said Niebel.
SOURCE – EETimes