SNIA PM & CS Summit Advances And ISI SOT MRAM Tester
The two-day 2021 SNIA Persistent Memory and Computational Storage Summit (PM & CS Summit) is the continuation of the previous one-day Persistent Memory Summits. This year, with the change in name of the Solid State Storage Initiative (SSSI) to the Compute Memory and Storage Initiative (CMSI), the organizing committee decided to add various talks around computational storage to the prior PM focus.
There were speakers from SMART Modular, Lenovo, Eideticom, Samsung, HPEG2M, Silinnov Consulting, NGD Systems, MemVerge, Los Alamos National Laboratory, Intel, NVIDIA, Futurewei and Dell gave talks on PM and CS, accented by two panel discussions each day. Here are a few highlights from the meeting.
Arthur Sainio from SMART Modular gave a talk that included the chart below showing the move from near memory attached via DDR or OMI to also include far memory organized with a switched fabric such as CXL, GenZ or CCIX by 2022. These changes are needed to provide memory with faster performance and larger amounts of data in memory.
Jonathan Hinkle from Lenovo put the traditional storage/memory hierarchy on its head in him presentation. He spoke about the advantages of including PM with DRAM and spoke about a three step program going from PM NVMe SSDs and DIMMs to NVDIMM-P to CXL or CCIX. Far memory can be attached via CXL to greatly expand memory sizes without causing difficulties with capacitive loading via DDR and without excessive latencies. He also reported test results showing advantages using CCIX hardware and Formulus Black software.
Steve Bates from Eideticom told us that the future of storage is offload. He also spoke about 3-C’s to make customers happy; compatibility, consumability and composibility. He then described how the company’s compute storage processor (CSP) can be usd in storage devices or in a smart NIC.
David Wang from Samsung Electronics gave a breakdown of CXL 2.0, the CXL consortium and uses for CXL, as shown below. He also spoke about computational storage and security with CXL.
David McIntyre, also from Samsung, spoke about security with computational storage drives. He discussed several activities to deal with security issues that might otherwise arise with computational storage devices. These include establishing roots of trust in components, secure enclaves for boog drive firmware and unencrypted keys, future keys per I/O and key management.
Eduardo Berrocal of Intel Corporation and Keith Orsak of HPE talked about the challenges of measuring persistent memory performance, referring to a SNIA white paper from 2020 and a test specification working draft. Their presentation gave several test examples and they also called out for companies to submit their real-world workloads captures to the working group.
Jerome Gaysse from Silinnov Consulting gave its outlook for computational storage in the cloud.
Scott Shadley from NGD Systems talked about productization of computational storage as shown in the image below for current computational storage products ranging from storage devices to network-based products.
Charles Fan from MemVerge told attendees about four top use cases for computational storage for big memory computing applications. The image below shows their Memory Machine software which they discussed using in cloud infrastructure, databases, genomics as well as animation and VFX (video special effects).
Bradley Settlemyer from Los Alamos National Laboratory talked about the performance, value and limitations for computational storage in HPC applications.
Andy Rudoff and other people from Intel spoke about the use of persistent memory in CXL. They discussed the current use of Intel’s Optane memory in the memory bus and then discussed efforts underway to implement CXL with various command sets and features. They also said that the momentum around CXL is huge and this will drive software support, with some software already changed to support CXL features.
Michael Kagen, CTO of NVIDIA helped us undersatand why distributed AI in data centers needs computational storage to bring compute and data closer together. Chun Liu from Futurewei Technologies talked about Zoned Name Spaces and their use with applications logs on persistent memory.
Shyam Iyer, from Dell and chair of SNIAS SDXI Technical working Group spoke about SNIAs smart data accelerator interface (SDXI). This is a memory to memory data mover that can support faster data movement by offloading data movement from CPU operations. IN particular he discussed offload direct memory access (DMA) engines and the need for industry standards to enable this.
There was some other recent news about advances in MRAM device testing capabilities from Integral Solutions International (ISI). Integral Solutions International’s (ISI) Gen3 Pulser is a cost-effective, fast, and versatile STT-MRAM measurement solution. To meet the industry demands for even narrower pulse width and higher pulse amplitude, and in support of emerging MRAM technologies such as SOT-MRAM, ISI is preparing the release of its Gen4 Pulser platform.
ISI’s Gen4 Pulser will allow a minimum pulse width of 500pS, an order of magnitude improvement over their Gen3 solution. Further, pulse amplitude will be increased up to +/-3V, 150% higher than available on most pulsing equipment today.
ISI’s Gen4 Pulser will continue to support FAST cycle timing in support of high-repetition measurement cycles such as Write Error Rate. Each Gen4 Pulser will be equipped with 2 parallel channels, which can be configured for either single or dual STT-MRAM measurements, or single SOT-MRAM measurements where the SOT and MTJ structures must be pulsed simultaneously.
For even higher-throughput applications, ISI’s Gen4 Pulser is modular, where up to 4 of these Pulser channels can be installed at once, providing up to 8x STT-MRAM or 4x SOT-MRAM device testing in parallel. Adding to ISI’s large family of user-configurable Magnet solutions, in addition to the recent release of the 1T/1.5T Perpendicular Magnets, ISI is also releasing a 3D Magnet ideally suited for SOT device testing applications.
The SNIA Persistent Memory and Computational Storage Summit demonstrated the increasing importance of persistent memory and computational storage in data center and other applications. ISI’s Gen4 Pulser will enable Spin Orbit Torque (SOT) as well as Spin Transfer Torque (STT) MRAM testing.