Non-volatile DIMMs: immortal DRAM?

DRAM's speed. Flash's data preservation. The NVDIMM has both. Too good to be true? It isn't.
Written by Robin Harris, Contributor on

DRAM's speed. Flash's data preservation. The NVDIMM has both. Too good to be true? It isn't.

Lots of cool flash apps at the Flash Memory Summit last week. One of the best was the NVDIMM from San Diego-area Agiga Tech, a Cypress Semiconductor company.

Nothing beats synchronous DRAM's combination of price, performance and durability for main memory. But lose power and DRAM's data is gone. Critical apps have long used battery-back up (BBU) to guard against wonky power.

But BBUs have problems. Heavy. Slow charging. 2-3 year life. Costly. Compared to solid state devices, a pain.

The NVDIMM backs up data without a battery. Their NVDIMM integrates standard DDR3 SDRAM with NAND flash, a system controller and an ultracapacitor power source.

Ultracapacitors are cool. In a demo at the conference, Agiga Tech showed how the UCs would fully charge in less than 10 seconds and run a power-hungry toy over a minute.

Orchestrating the DRAM off-load onto the flash when there is a power outage is the problem. The system has to:

  1. Sense power failure.
  2. Initiate DRAM-to-flash offload.
  3. Restore DRAM contents while the ultracapacitors recharge.
  4. Inform system when process is complete.

The NVDIMM includes a simple host-controlled protocol that let's the system call the shots. Only the system knows if a power-down is accidental or not, while only the NVDIMM knows if the ultracapacitors are charged or not.

The Storage Bits take Will NVDIMMs replace standard DIMMs on PCs and notebooks? No. Consumers won't pay the extra cost for the occasional benefit.

But for embedded systems that already use batteries to ensure data protection - RAID controllers for example - NVDIMMs should prove popular. Even if they cost more, they are lighter, faster to recharge, smaller and have longer data retention than battery backed DRAM.

What NVDIMMs don't do is condition power like many whole system BBUs. But data centers already condition their power even if few consumers do.

The bigger issue is that system designers need prodding to take advantage of non-volatile DIMMs. Not only do they need to code to Agiga's protocol, they need to think about how, for example, an NVDIMM RAMdisk could benefit their application.

Despite these issues, ultracapacitor-based flash backup promises to extend non-volatility to smaller and lower-cost devices. That is a welcome improvement.

Comments welcome, of course.

Editorial standards