Rambus and IBM develop hybrid memory for future data centres

Combining new memory systems with DRAM should lead to higher performance at a lower cost

Rambus has announced a partnership with IBM to research hybrid memory systems, through its Rambus Labs department. These systems will be used in ‘future' data centres, it says.

Together, Rambus Labs and IBM plan to optimise DRAM and emerging memory technology use to create ‘a high-capacity memory subsystem that delivers comparable performance to DRAM alone'.

Hybrid systems, with high-capacity and high-density memory near the processor, show improved system performance and an increased ability to utilise CPU resources, IBM said. The new hybrid systems will combine standard DRAM with technology like Flash, Phase Change Memory (PCM) and Spin Torque Transfer Magnetic RAM (STT-MRAM). The result is similar performance to DRAM at a lower cost-per-bit.

Rambus Labs will create a flexible prototype hybrid memory platform, using the OpenCAPI (Coherent Accelerator Processor Interface) interface, to show how multiple memory types perform in real-world server applications.

Laura Stark, senior VP and general manager of Rambus' Emerging Solutions Division, said that the work was being driven by "The exploding volume of data and rapidly evolving workloads for big data applications".

IBM will supply its new POWER9 processor and OpenCAPI interface to build the system prototype. As part of the work, Rambus has joined two open standards organisations: the OpenCAPI Consortium and OpenPOWER Foundation, based on CAPI and the IBM Power Architecture, respectively.

Steve Fields, chief engineer of Power Systems at IBM, said:

"IBM believes in transforming the architecture of server memory to allow open innovation and to fully exploit the diversity of memory technologies that will emerge in the coming years. This project leverages the new architecture to combine the best attributes of multiple types of media to achieve new levels of system cost/performance for memory-intensive cloud deployments and AI applications."