Micron Exits 3D Xpoint, Moves to CXL

Article By : George Leopold

Micron exits the 3D Xpoint memory market. How long will Intel hang on?

Micron Technologies is exiting the once-promising 3D Xpoint non-volatile memory market, instead focusing its data center efforts on the emerging Compute Express Link interface as bandwidth requirements soar.

Sanjay Mehrotra, Micron’s president and CEO, said this week the memory maker would cease 3D Xpoint development and “reprioritize our R&D investments” toward new memory products based on Compute Express Link (CXL), the emerging CPU-memory industry standard interface.

“Our decision was driven by our assessment of the 3D XPoint market opportunity in light of the expected impact of CXL and our new emerging memory products, on the future data center,” Mehrotra said.

Micron is currently in discussions with “several potential buyers” of its dedicated 3D Xpoint fab in Lehi, Utah. It hopes to complete a sale later this year, Mehrotra added. The Lehi fab led Micron’s entry into the NAND market the early 2000s, later bringing Micron’s version of 3D Xpoint memory technology into mass production.

Micron’s departure marks a turning point for 3D Xpoint technology. While Intel has doubled down on its version of the memory technology with its Optane family of persistent memory, Micron is instead hitching its wagon to the emerging CXL standard designed to ease data center bottlenecks.

Intel announced last fall it would use the proceeds of the sale of its NAND flash memory business to SK Hynix to invest in further development of Optane as the memory technology gains traction among data center operators. Observers noted that Intel has effectively been able to subsidize Optane by integrating it with server systems.

Since 3D Xpoint non-volatile memory technology emerged in 2015, Micron said its strategy was providing data center customers with persistent memory technology for cloud-native infrastructure such as applications containers. The initial value proposition was cheaper DRAM and faster NAND.

Evolving data center workloads changed those requirements.

Sumit Sadana, Micron’s chief business officer, noted that CPU-DRAM bandwidth has since emerged as the limiting factor as data-intensive AI workloads dominate enterprise data centers. “More DRAM is needed to ensure adequate memory bandwidth per CPU core,” Sadana said this week in announcing the memory maker’s embrace of the CXL spec.

Micron said it would seek to leverage 3D Xpoint process and product technologies as it transitions to CXL. “We are going to be building products that are going to work with the CXL interface, and have a different optimization point as it relates to the performance and cost and power tradeoffs that exist in the hierarchy from DRAM to flash storage,” said Sadana.

Only the timing of Micron’s announcement surprised observers. “They don’t have a successful 3D Xpoint product,” noted chip industry analyst Tom Coughlin, who also noted that Micron could adapt its phase-change memory process technology developed for Xpoint to CXL.

Coughlin and others note that data center demand for persistent memory such as Optane remains, and CXL is designed to handle heterogenous memory architectures.

That also leaves an opening for Intel’s Optane technology, which the chip maker is essentially subsidizing as it seeks to boost volume and find new memory-intensive big data applications in data centers.

“Intel is in an interesting situation” having spun off its NAND technology to SK Hynix, Coughlin said, estimating it is selling Optane products at half the price of DRAMs. As DRAM prices rise, Coughlin asked: “Will Intel have the intestinal fortitude” to stick with its Optane strategy as it seeks to ramp volume?

Leave a comment