As the non-volatile memory express (NVMe) protocol moves to an annual release cycle, the latest updates reflect the growing maturity of the ecosystem and the rapid pace of innovation driving its advancement, including AI.
NVMe 2.4 incorporates updates across security, sustainability, and management as well as specification clarifications. In a briefing with EE Times, Mike Allison, chair of the NVMe Errata Task Group, said the update bundles together a year of work across existing specifications rather than introducing entirely new ones. “We’ve been very busy,” Allison said.
Since the previous release, the non-profit NVM Express consortium has authorized 17 new technical proposals and completed 15 proposals, along with multiple engineering change notices. Allison said most of them were modified because the organization now has an errata group, which clarifies text when needed or when specification conflicts arise.
Due to the age of the storage protocol, some features are simply no longer being used, he added.
View All When NVMe 2.0 was released in 2021, it was “refactored” to break away the base specification from others that provide specific functionalities to allow for faster and simpler development in what has become an increasingly diverse device environment.
Securing NVMe for the post-quantum era
A major focus of this release is strengthening security, particularly for virtualized and cloud-based environments, as well as addressing the threat to encryption posed by quantum computing.
The update adds an information package for Trusted Execution Environment Device Interface Security Protocol (TDISP) that enables controllers to report detailed state information. Allison said the capability can be used to validate controller status when systems or workloads are migrated between servers. “You can validate what was the state when you left the controller in the source to what is the state in the destination when you’re continuing,” he said.
The specification also introduces improvements to lockdown functionality. Previously, lockdown settings were lost after a power cycle. The update now allows those settings to persist across reboots and introduces separate lockdown controls for virtualized environments, Allison said.
NVMe 2.4 also addresses emerging cybersecurity requirements through support for post-quantum cryptography (PQC). “We’ve updated our asymmetric cryptographic key exchange algorithms to be prepared for the PQC changes,” Allison said.
Sustainability features improve autonomous power management
The NVMe update also builds on earlier power-monitoring capabilities with the addition of voltage monitoring, which enables SSDs to identify discrepancies between the voltage supplied by the platform and the voltage actually received by the device. Allison said this feature can track measurements over time and across power cycles, helping organizations diagnose potential infrastructure issues.
While power efficiency remains important across all markets, Allison noted that client devices continue to drive much of the innovation. “Most of the power work coming in is for client devices and laptops,” he said, though enterprise and data center deployments ultimately benefit from the same advancements. The release also introduces idle I/O exit latency limits that give hosts greater control over autonomous power management. Rather than allowing the SSD to optimize solely for power savings, hosts can specify latency thresholds that must be maintained.
Another addition enables restoration of a device to go back to the way it was when it was shipped so it can be reprogrammed from scratch. Allison said this simplifies deployment for OEMs that configure devices with multiple operational personalities.
Enhanced management subhead
The most substantial architectural developments in NVMe 2.4 include PCIe exported NVM subsystem migration and rate-limiting capabilities designed for shared PCIe environments.
The exported subsystem work enables the creation of virtual SSDs within physical SSDs, allowing virtual machines to access only designated controllers and namespaces, Allison said. “We’re moving the hardware abstraction from the hypervisor into the SSD itself.”
He said this allows virtual machines to communicate more directly with storage resources while maintaining migration flexibility.
The rate limiting feature adds new controls for allocating PCIe bandwidth and IOPS among multiple controllers sharing a transport. Allison said that NVMe previously had no ability to decide which controller got which portion of the PCIe bandwidth and IOPS.
Additional management improvements include host-addressable subsystem local memory, reclaim-unit scope reporting, NVMe-MI status reporting, and a redesigned NVMe-MI architecture that reduces redundant data structures and memory consumption. The release also adds support for the new SFF E2 form factor and improves management behavior in inter-integrated circuit (I2C) environments.
Looking ahead
NVMe technology continues to expand far beyond its origins as a PCIe SSD interface, becoming the common storage architecture across client devices, enterprise systems, cloud infrastructure, AI platforms, and even automotive applications. NVMe now spans nearly every major computing market. “It’s used everywhere,” Allison said. “Automotive is doing NVMe. Phones are, tablets are, laptops are. It’s really growing.”
But despite NVMe’s widespread adoption, Allison believes the technology is still accelerating rather than slowing down. “If I took the specs and printed them all out, we grew by 10%,” he said. “That 10% is bigger than the initial NVMe release. When I look at the last two releases, the rate of change is greater in the last two releases than in the history of NVMe.”
Beyond the current release, Allison said the NVMe community is already exploring future capabilities aimed at AI workloads, I/O prioritization, distributed namespaces, and new transport technologies. While AI is influencing development priorities, he emphasized that the focus remains on improving efficiency across diverse workloads and deployment models.
Read also:
[Synopsys Updates CXL IP Portfolio for AI-Era Infrastructure](https://www.eetimes.com/synopsys-updates-cxl-ip-portfolio-for-ai-era-infrastructure/)
[Marvell Targets AI Bottlenecks with Memory-Disaggregation Portfolio](https://www.eetimes.com/marvell-targets-ai-bottlenecks-with-memory-disaggregation-portfolio/)
[Samsung Lays Out AI Memory Roadmap](https://www.eetimes.com/samsung-lays-out-ai-memory-roadmap/)