Building on years of HP research
The technical advances in this new class of memory – and its potential for improving compute performance – caught the attention of HP researchers over a decade ago, when HP Labs first demonstrated the existence of memristors, a fourth basic passive circuit element that acts as a logic gate that can “remember” whether it is open or closed even when powered off.
“That helped set off a boom in fundamental memory research at HP Labs and led us to open an initial research collaboration with HP’s Personal Systems group and the Office of the CTO focusing on the different ways in which persistent memory could be developed for different kinds of client devices, asking what the potential benefits and challenges of each kind of memory might be,” Haas recalls (when Hewlett-Packard split into two companies, the effort was brought fully within HP Labs). Both research strands also generated a large portfolio of patents that Haas and his colleagues continue to build upon.
As part of their work, the persistent memory client research team created an emulator to better understand how both operating systems and applications might need to be reconfigured in systems deploying different kinds of persistent memory architectures. It also offered important insights into how client devices like workstations or laptops would need to be redesigned in order to take full advantage of the new technology.
By 2015, it was clear that one of these methodologies, resistive random-access memory (also known as ReRAM), was likely to be the first to be commercially developed. That year Intel and Micron, for example, announced their 3D XPoint technology, a phase change-based version of ReRAM which today powers Intel’s DCPMM Optane™ Persistent Memory DIMMs. Industry interest in ReRAM attracted the attention of HP’s Workstations group, which approached the HP Labs team to undertake a second collaborative program that would support a deeper investigation into the possibilities of ReRAM-based persistent memory computing.
Re-engineering the workstation ecosystem
Within a year, the new cross-company team had demonstrated how persistent memory can significantly improve compute performance in several workstation use cases, advanced the state of the art, and outlined several innovative paths to take maximum advantage of the new technology.
Given that persistent memory would also require both operating systems and applications to change, the team began working with industry partners to establish new standards for operating systems and applications that are optimized for persistent memory-based computing.
In addition, the workstation they were developing necessitated a new approach to issues like device security, since persistent memory introduces a large new data repository into workstation architecture. While Intel DCPMM persistent memory has its own security protections, they are targeted for data centers, and thus needed to be enhanced to make them compatible with client usage models.
That led the team to develop a new security solution, leveraging TPM encryption and key management, and integrate it within HP’s existing computer management framework. At the same time, they made it possible to migrate encrypted persistent memory modules from one workstation to another, without losing data.
Dramatic results
Testing applications optimized for persistent memory in the new environment produced startling results, leading the HP Labs and Workstation teams to develop persistent memory support for a number of open-source and commercial applications across a range of markets including energy and media/entertainment.
“Depending on the application, we’re seeing tremendous performance gains from these devices,” Haas reports. “Large data sets, for example, that might have taken tens of minutes to load before are now quasi-instantly and permanently available.”
In addition, HP Labs’ persistent-memory optimized prototypes have demonstrated 5x less energy consumed by memory in idle mode, and require no energy at all when in sleep and other reduced power modes.
The cross-company team has also demonstrated new functionalities such as persistent undo/redo functions across application closures or system power-downs, and replacing auto-save functions with real-time user action and data update recording, allowing users to resume or recover virtually any work last performed. These set the stage for even more dramatic future improvements in the user experience and the development of as-yet unimagined applications.
Many of these innovations are now incorporated in HP’s most powerful commercial client systems, or accessible to software developers in partnership with HP. HP’s Z6 G4 and Z8 G4 workstations support Intel Optane™ DC Persistent Memory Modules running Linux, and will support Microsoft’s persistent memory-compatible version of Windows 10 for Workstations, which is expected to debut early in 2020.
“We’re thrilled to be offering best-in-class options for some of our most demanding customers and partners, thanks to the incredible work that researchers in the cross-company team put in,” says Raphael Gay, Distinguished Technologist in HP’s Workstations group. “Our partners are even more emphatic. One of them, a large 5000+ employee company shared that they are “excited about the project’s progress to date and all the different collaborative things that we could do … our CEO is requesting a study on the use cases for persistent memory across our portfolio of applications.””
Looking ahead, the HP Labs/Workstations team expects HP’s persistent memory-ready workstations to impact activities as diverse as gaming, product development, artificial intelligence, data science, and high-end animation. They plan to keep optimizing persistent memory-enabled workstation performance and are working with industry partners to grow a Persistent Memory Ecosystem for client applications, including partners and independent software vendors.
Haas and his HP Labs colleagues are also exploring the use of persistent memory in 3D-Print workflows, and in mobile and IoT devices.
“We’re tracking other memory technologies that are capable of being integrated into our platforms in the future, such as other kinds of resistive RAM, M-RAM, and carbon nanotubes,” Haas says. “We want to understand the tremendous potential that these new memories will have for further improving a wide variety of compute-intense applications.”
