Flash Memory Summit: Netlist Demos HybriDIMM Storage Class Memory

Netlist, Inc. demonstrated its HybriDIMM storage class memory product at the Flash Memory Summit.

This demonstration showcased two widely used open-source databases running concurrently on a Storage Class Memory (SCM) solution: MongoDB exercises HybriDIMM’s in-memory channel block storage capabilities while the Redis in-memory database utilizes memory semantics to achieve a true persistent memory solution.

Click to enlarge

C.K. Hong, CEO, Netlist, said: “With the flexibility to operate as main memory, block storage and persistent memory in a standard server – a first in the industry, HybriDIMM can serve as both a memory or storage solution for the rapidly growing markets in big data analytics, in-memory computing and artificial intelligence. This demonstration shows the versatility of HybriDIMM’s revolutionary architecture and the practical benefits of data access through multiple operating modes in the memory channel SCM. We are pushing forward with our plans for commercializing this ground-breaking product, in concert with our development partners.“

HybriDIMM combines existing memory semiconductors, DRAM and NAND flash, with intelligent ‘on-DIMM’ co-processing to manage the data transfer between on-DIMM storage and the system. It is recognized by any x86 server as a standard LRDIMM with no BIOS modifications or system changes. HybriDIMM’s three operating modes include main memory access via on board DRAM, and Block Storage and Persistent Memory accessed via on-board co-processors with NAND flash. HybriDIMM’s cost-effective fast storage and memory extension allows datacenter operators to lower the cost of running database applications, online transaction processing, big data analytics, and in-memory computing, and capitalize on the promise of storage class memory.

Technical summary of HybriDIMM:

Unifies memory and storage into one module (DIMM), breaking the decades-long limitations of the Von Neumann computing architecture
Applies NVM Libraries to scale memory to storage capacities
Accelerates storage access time in the memory channel
Allows multi-terabytes of persistent memory to run real-time in-memory applications on existing Intel-based servers at speeds an order of magnitude faster than NAND
Generation I HybriDIMM: configurable 256-512GB NAND+8-16GB DRAM
Generation II HybriDIMM: configurable 1TB NAND+32GB DRAM