High Bandwidth Memory (HBM) is the advanced memory technology underpinning the recent explosion of AI-based applications.  HBM uniquely satisfies the key drivers for all AI workload demands: high bandwidth and low latency.  As AI workloads grow more complex, increased HBM densities and performance become a necessity.

• High-Performance: HBM is a unique kind of high-performance memory made of a set of vertically stacked dies. Its stacked profile improves the memory’s density for a given footprint.
• High Bandwidth: The distinguishing feature of HBM is its significantly higher bandwidth compared to traditional DIMM-based memory.  Its higher bandwidth allows for a wider array of inputs from CPUs, FPGAs, and GPUs, and enables much larger throughputs for parallel processing: the key to AI’s power.

Netlist Innovation: In 2010, Netlist invented proprietary designs underpinning the successful creation and proliferation of stacked HBM today. Netlist stands at the forefront of HBM innovation allowing for the recent explosive growth of AI. Netlist’s innovations help provide those performance gains, thus allowing AI to tackle more challenging tasks and execute operations that were previously unimaginable.

DDR5, or Double Data Rate 5, is a type of memory technology used in computer systems for primary memory. DDR5 memory is offered in different module form factors offering higher data transfer rates compared to its predecessor, DDR4. This means it can move data between the memory and the processor at a faster rate, leading to improved system performance.

• High-Performance: DDR5 modules have two independent subchannels, each with 32 data I/Os that increase concurrency and bandwidth.
• High-Reliability: DDR5 chips have error detection and correction mechanisms within the DRAM die which improve reliability and enables high density chips.
• Power Management: DDR5 modules have power management integrated circuits (PMICs) that provide local regulation and reduce the complexity of the motherboard design.

Netlist Innovation: As with many other important memory technologies forming the backbone of our digital world, Netlist’s innovations power the adoption and growing success of the industry’s latest DDR5 Dual In-line Memory Modules (DDR5 DIMMs). Netlist’s localized module-based power management solutions allow for high-efficiency power delivery to every DIMM component, high-precision voltage regulation, and dynamic power adjustments tailored to each DDR5 DIMMs’ unique demands. The result is improved overall system stability, higher speed, and increased energy efficiency.

DDR5 Multiplexer Combined Ranks (MCR) Dual In-line Memory Modules (MCRDIMMs) are designed to manage large amounts of data quickly and efficiently.  MCRDIMMs are ideal for use in powerful servers in enterprise and data center applications, especially when dealing with complex tasks like AI and big data processing.

• Higher Capacity: DDR5 MCRDIMM modules can support up to 1024 GB of memory per module, which is substantially higher than standard DDR5 DIMMs.
• High Performance: Simultaneous operation of two ranks with a special buffer to make both ranks work at the same time, effectively double the data rate and bandwidth.
• Low Power with Power Management: DDR5 MCRDIMM modules have on-board power management circuitry that provide local voltage regulation and reduce the power consumption of the module.

Netlist Innovation: Beyond their use of localized DIMM management technologies pioneered by Netlist, MCRDIMMs will also incorporate Netlist’s proprietary on-module intelligence features, including innovations in load reduction and rank multiplexing. These crucial improvements allow the use of early Netlist innovations like distributed data buffers, a design first created by Netlist in the early 2000s. MCRDIMMs are expected to become the next-generation of server memory modules used in the bulk of AI-workload and data center servers — Netlist’s innovations help make this possibility a reality.

Compute Express Link (CXL™) is a significant advancement in computer technology, representing a departure from the long- standing Peripheral Component Interconnect (PCI) standard that has been in use since 1992. CXL brings a range of features that cater to the evolving needs of high-performance data center computing and artificial intelligence (AI).

• High Speed Connections: CXL provides cache-coherent, high-speed connections between various components, including Central Processing Units (CPUs), storage, and memory. This ensures that data remains consistent across all components and allows for efficient data sharing.
• Superior Capacity: CXL offers high-capacity solutions, which are essential for data-intensive applications and the increasing demand for larger memory capacities.
• Expanded Memory: CXL opens the door to a new era of memory expansion and pooling. This means that data centers can significantly increase their memory capacities and allocate them more efficiently, addressing the requirements of modern computing workloads.
• Cost-Efficiency: CXL ushers in a new era by providing the industry the ability to economically expand capacities through the adoption of CXL-based DRAM cards, DRAM and NAND combined cards, and NAND storage products.

Netlist Innovation: Netlist continues to drive R&D development  that will provide lower-cost high-capacity CXL solutions that have DRAM-like performance.  Netlist’s new CXL solutions will:

• Build upon its 20+ year legacy of R&D-based innovations, including localized module intelligence and on-module power management.
• Empower data centers to adopt and deploy a wider array of applications, while overcoming existing cost and performance barriers.