News Posts matching #HBM

Return to Keyword Browsing

Silicon Motion Showcases Enterprise SSD Controllers and PCIe NVMe Boot Drive Solutions at GTC 2026

Silicon Motion Technology Corporation, a global leader in designing and marketing NAND flash controllers for solid-state storage devices, today announced that it will showcase a rich portfolio of differentiated enterprise SSD controllers and PCIe NVMe BGA boot SSD at NVIDIA GTC 2026, Booth #3015, purpose-built to meet the evolving requirements of the NVIDIA AI ecosystem. As AI models scale, inference architectures are extending beyond HBM and System DRAM into high-performance NAND storage tiers, as reflected in NVIDIA's ICMS initiative. In this new architecture, NAND-based Storage becomes a performance-critical layer that requires deterministic latency and quality-of-service differentiation.

Silicon Motion delivers a vertically integrated storage solution encompassing advanced SSD controller design, full firmware development, and compact Reference Design Kits (RDKs) aligned with leading enterprise form factors such as E1.S, E1.L, E3.S, E3.L, and U.2 for AI server deployments. The company also provides enterprise PCIe NVMe BGA boot SSD with strong endurance and long-term operational stability, deployed in AI systems.

MSI Launches XpertStation WS300 on NVIDIA DGX Station Architecture

MSI today announced the launch of XpertStation WS300 on NVIDIA DGX Station Architecture, a next-generation deskside AI supercomputer built to support the accelerating demands of large language models (LLMs), generative AI, and advanced data science workflows. Powered by NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, supporting up to 748 GB of large coherent memory and dual 400GbE networking, the platform extends advanced AI infrastructure capabilities into a compact deskside deployment model and is available for order starting today.

"MSI has a strategic vision to advance AI-first computing," said Danny Hsu, General Manager of MSI's Enterprise Platform Solutions. "With NVIDIA, we are defining the next era of AI infrastructure, bridging centralized performance and distributed innovation, and enabling organizations to move from experimentation to production with greater speed, scale, and confidence."

Samsung Unveils HBM4E Solutions at NVIDIA GTC 2026

Samsung Electronics, a global leader in advanced semiconductor technology, today announced the comprehensive AI computing technologies it will showcase at NVIDIA GTC 2026 in San Jose, California, scheduled for March 16-19. As the industry's only semiconductor company offering a total AI solution spanning memory, logic, foundry and advanced packaging, Samsung will exhibit its full suite of products and solutions that enable customers to design and build groundbreaking AI systems. To learn more about Samsung's AI solutions, please visit the company's GTC 2026 booth (#1207).

The centerpiece of Samsung's showcase at NVIDIA GTC 2026 will be the new sixth-generation HBM4, which is now in mass production and is designed for the NVIDIA Vera Rubin platform. Samsung's HBM4 is expected to help accelerate the development of future AI applications, delivering consistent processing speeds of 11.7 gigabits-per-second (Gbps), which exceeds the industry standard of 8 Gbps, and can be enhanced to 13 Gbps.

KIOXIA Announces New Super High IOPS SSD Series

Kioxia America, Inc. today announced the development of its Super High IOPS SSD, a new type of SSD enabling the GPU to directly access high-speed flash memory as an expansion to High Bandwidth Memory (HBM) in AI systems. The new Super High IOPS SSD, the KIOXIA GP Series, is purpose-built to meet the growing performance demands of AI and high-performance computing, providing larger GPU-accessible memory capacity for faster data access to AI workloads. Evaluation samples of KIOXIA GP Series will be available to select customers by the end of 2026.

The NVIDIA Storage-Next initiative addresses the anticipated shift from compute-intensive to data-intensive workloads and the expanded need for GPU-accessible memory space, currently limited by HBM size. Expanding the GPU's usable memory space allows access to larger data sets and improves GPU utilization by moving more data closer to compute resources.

Micron Completes Acquisition of PSMC's Tongluo P5 Site in Taiwan

Micron Technology, Inc. (Nasdaq: MU) today announced it has completed the acquisition and assumed ownership of Powerchip Semiconductor Manufacturing Corporation's (PSMC) P5 site in Tongluo, Miaoli County, Taiwan, under the acquisition agreement previously announced on January 17, 2026.

The new site will complement Micron's existing operations in Taiwan as an extension of the company's vertically integrated mega campus in Taichung, located approximately 15 miles away. The site includes approximately 300,000 square feet of existing 300 mm cleanroom space and will support Micron's efforts to expand supply of leading-edge DRAM products, including HBM, to meet growing AI-driven demand.

Memory Makers Expect Shortages to End in Late 2028, Could Pause Expansion Plans

Memory manufacturers are forecasting that memory shortages will persist until the end of 2028, when the supply chain balance is expected to be restored and memory will once again reach its commodity status. However, according to the South Korean Chosun Biz newspaper, memory manufacturers are reconsidering their expansion plans to ensure they don't overinvest in building new capacity after demand cools down. Reportedly, Samsung's internal projections indicate that demand for DRAM in forms like HBM and regular DDR will peak in the next few quarters, up until the end of 2028. This is the point where Samsung expects industry-wide demand to cool down and balance to be restored.

When the AI expansion began, memory manufacturers received massive orders from hyperscalers around the world, ordering DRAM supply months in advance. With no end in sight, SK Hynix, Micron, and Samsung started ordering more lithography tools from ASML to expand their wafer manufacturing capacity to meet their customers' needs for more HBM, DDR, and GDDR memory. According to ASML, the company estimates that in 2027, based on current order information and demand, it will deliver 56 Low-NA EUV scanners. This includes seven units for Samsung and as many as 20 units for SK Hynix, specifically for memory and storage. According to South Korean media, SK Hynix plans to install 20 Low-NA EUV units in the next two years, all designed for HBM memory and advanced storage solutions. However, as the demand is expected to stay in line with previous expansion plans, this is stopping memory makers from considering taking the fab capacity expansion further than initially planned.

Samsung Reportedly to Use 2 nm Process on HBM4E Base Die

Samsung is reportedly planning to bring its 2 nm process to the base die of HBM4E, its 7th-generation high-bandwidth memory. This comes just a month after the company shipped the industry's first commercial HBM4, and alongside a separate effort to redesign the HBM4E power delivery network to handle the increase in power bumps from 13,682 to 14,457 within the same footprint. Until HBM3, the base die had a somewhat passive role, sitting at the bottom of an HBM stack and handling power and signal control. Starting with HBM4, it took on a more active role by handling some compute tasks directly, making the underlying process node increasingly important. For HBM4, Samsung already had an edge here, using 4 nm logic base dies from its own foundry paired with its latest 1c DRAM, well ahead of the 12 nm (N12) process SK hynix sources from TSMC. Moving to 2 nm for HBM4E would extend that lead further, improving power efficiency, thermal management, and area utilization.

The other key players in the industry are also focusing on custom HBM4E as the next battleground, with TSMC stating it plans to use its 3 nm process for custom HBM4E, and SK hynix also working on its own variant. With its latest 2 nm node push, Samsung appears to be making a deliberate move to stay a step ahead on process technology. Standard HBM4E is expected mid-year, with custom product tape-in following in the second half of the year. The Business Korea report also points to the foundry angle. Producing HBM base dies internally helps Samsung Foundry keep fab utilization high, and the 2 nm node is expected to play a major role in ramping production at the Taylor fab in Texas. There, equipment installation is already underway, with the first wafer tape-in targeted before year-end.

Applied Materials and Micron Partner to Accelerate HBM, NAND and DRAM Development

Applied Materials, Inc. today announced it is working with Micron Technology to develop next-generation DRAM, high-bandwidth memory (HBM) and NAND solutions that increase the energy-efficient performance of AI systems, bringing together advanced R&D capabilities from Applied's EPIC Center in Silicon Valley and Micron's state-of-the-art innovation center in Boise, Idaho to strengthen the semiconductor innovation pipeline in the United States.

"Applied Materials and Micron have a long-standing partnership focused on driving higher performance and more energy-efficient advanced memory chips by pushing the boundaries of materials engineering and manufacturing innovation," said Gary Dickerson, President and CEO of Applied Materials. "We are excited to deepen our collaboration with Micron as a founding partner at the EPIC Center as next-generation memory technologies play an increasingly vital role in the future of AI systems."

Fujitsu Showcases "MONAKA" CPU Sample with 3.5D XDSiP Packaging

During MWC, Fujitsu partnered with networking equipment maker 1FINITY to unveil the first silicon wafers and an engineering sample of its "MONAKA" CPU. Scheduled for release in 2027, the initial Fujitsu MONAKA CPU utilizes the Armv9-A architecture and a 3D chiplet layout that combines a core die with separate SRAM and I/O dies. A single chip features 144 cores, and two-socket configurations can scale up to 288 cores per node. The platform supports 12-channel DDR5, PCIe 6.0 with CXL 3.0, and Arm SVE2 for AI and HPC workloads. Fujitsu has chosen TSMC to manufacture this chip using the 2 nm node, paired with Broadcom's 3.5D eXtreme Dimension System-in-Package (XDSiP) packaging architecture. This packaging allows MONAKA to become a 144-core design featuring four 36-core chiplets. These chiplets are stacked face-to-face with SRAM tiles through hybrid copper bonding, utilizing TSMC's N5 process for the cache layer.

In the pictures below, we can see the silicon complex in its early sample packaging, which shows a large central I/O die, HBM memory around the CPU, and the new packaging technology. Reportedly, this CPU has already reached a working version, with Broadcom shipping the CPU to Fujitsu in late February this year. After initial testing and early performance validation, Fujitsu plans to ship these processors to customers around summer, with mass shipping to commence in 2027. The company envisions this SoC as a powerhouse for AI inference, simulation, and large-scale data processing. It will sell these systems to external customers, who showed great interest in Fujitsu's A64FX when the Fugaku computer came online. Fugaku was the most powerful supercomputer back in 2020, achieving 415.53 PetaFLOPS of FP64 and an impressive HPL-AI score of 1.421 ExaFLOPS using lower FP16 precision. Hence, we expect the new MONAKA CPU to enable much greater speeds and some efficiency improvements as well.

Rambus Announces Its New HBM4E Memory Controller IP, Up to 16 Gbps per Pin

Rambus Inc. (NASDAQ: RMBS), a premier chip and silicon IP provider making data faster and safer, today announced the industry's leading HBM4E Memory Controller IP, extending its market leadership in HBM IP. This new solution delivers breakthrough performance with advanced reliability features enabling designers to address the demanding memory bandwidth requirements of next-generation AI accelerators and graphics processing units (GPUs).

"Given the insatiable bandwidth demands of AI, it's imperative for the memory ecosystem to continue aggressively advancing memory performance," said Simon Blake-Wilson, SVP and general manager of Silicon IP, at Rambus. "As a leading silicon IP provider for AI applications, we are bringing the industry's leading HBM4E Controller IP solution to the market as a key enabler for breakthrough performance in next-generation AI processors and accelerators."

NVIDIA Lowers HBM4 Specs for "Vera Rubin" VR200 as Memory Suppliers Miss 22 TB/s Target

NVIDIA has reportedly lowered its performance requirements for the HBM4 memory used in "Rubin" GPUs, as SK hynix and Samsung are reportedly struggling to meet the ambitious performance targets set by NVIDIA. According to a new note from SemiAnalysis, NVIDIA is reducing its specification requirements for the upcoming GPU generation. Originally, NVIDIA targeted a total bandwidth of 22 TB/s for the Rubin chip, but memory suppliers seem to be having difficulty meeting these requirements. Initial shipments are expected to achieve closer to 20 TB/s, which translates to approximately 10 Gbps per pin for HBM4. This indicates that NVIDIA's aggressive upgrade plan for "Vera Rubin" is facing a setback, and the final performance will differ slightly.

Interestingly, NVIDIA's initial target for the "Vera Rubin" VR200 NVL72 system was 13 TB/s in March 2025, which was later upgraded to 20.5 TB/s by September. At CES 2026, NVIDIA confirmed that the VR200 NVL72 system is now operating at 22 TB/s of bandwidth. Compared to AMD's Instinct MI455X accelerator, which has 19.6 TB/s, NVIDIA initially had lower system bandwidth. They addressed this by using faster DRAM and improving interconnects between CPUs, GPUs, and the entire system. However, as memory makers like SK hynix and Samsung struggle to meet NVIDIA's performance requirements, we will see HBM4 speeds of about 20 TB/s for the entire "Vera Rubin" system.

Samsung Redesigns HBM4E Power Delivery to Cut Defects and Explore HBM-GPU Separation

Samsung is overhauling the power delivery network in its HBM4E memory to tackle what's becoming one of the bigger engineering headaches in next-gen AI chips. This comes just two weeks after the company shipped its first commercial HBM4, already pushing 11.7 Gbps consistently with headroom up to 13 Gbps. As designs move from HBM4 to HBM4E, the number of power bumps grows from 13,682 to 14,457, packed into the same space with thinner, denser wiring. That drives up current density and resistance, causing IR drop (voltage weakens as it travels through the circuits) while the heat generated in the process makes things worse, creating a feedback loop that can affect performance or even cause circuit failure.

To fix this, Samsung segmented the power network. The large centralized MET4 power block on the base die, previously laid out in big honeycomb-like sections near the interposer, has been split into four smaller sections, with upper layers further broken up to reduce congestion and shorten routing paths. According to Samsung, the results were significant, metal circuit defects dropped 97% compared to HBM4, and IR drop improved by 41%, giving the chip more voltage headroom for higher speeds and better reliability.

DRAM Market Grows 29% in 4Q25, Samsung Regains Top Spot

TrendForce's latest findings reveal that the expansion of AI applications from LLM training to inference has prompted CSPs to broaden data center build-outs beyond AI servers to include general-purpose servers. This shift has extended memory procurement beyond HBM3e, LPDDR5X, and high-capacity RDIMMs to RDIMMs across multiple densities. Aggressive additional orders have driven a sharp increase in conventional DRAM contract prices, lifting total DRAM industry revenue to $53.58 billion in 4Q25, up 29.4% QoQ.

Across segments, buyers have struggled to secure sufficient supply amid a widening supply-demand gap. This has significantly strengthened suppliers' pricing power. Conventional DRAM contract prices rose 45-50% QoQ, while blended contract prices for conventional DRAM and HBM increased 50-55%, marking an accelerated upswing across all product categories.

NVIDIA Ships First "Vera Rubin" VR200 Samples to Customers

NVIDIA reported its full-year 2025 results with massive revenue of $215.9 billion, with $68.1 billion coming in the fourth quarter. The company's earnings call after the results were published contained some interesting information and confirmed that the first "Vera Rubin" VR200 racks are shipping to customers as samples, with volume shipping to commence in the second half of 2026. NVIDIA confirmed that the upcoming platform, which includes the Vera CPU, Rubin GPU, NVLink 6 switch, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-6 Ethernet switch, will be powering the next-generation trillion-parameter models with only one-fourth of the GPUs compared to the previous-generation "Blackwell" and will reduce inference costs by up to 10 times.
NVIDIA CFO Colette KressWe shipped our first Vera Rubin samples to customers earlier this week, and we remain on track to commence production shipments in the second half of the year. Based on its modular, cable-free tray design, Rubin will deliver improved resiliency and serviceability relative to Blackwell. We expect every cloud model builder to deploy Vera Rubin.

Samsung Foundry Utilization Jumps to 80% This Quarter

Samsung Foundry has struck gold as its utilization rate has jumped to an impressive 80% this quarter. This comes after many years of struggling to achieve a good utilization rate, while competitors like TSMC have consistently led the adoption of newer nodes and gained a massive customer base. However, the situation has now changed, and Samsung's foundries are now operating at about 80% of total capacity, consistently producing silicon, with the Pyeongtaek Campus P2 and P3 leading the charge. Previously, these lines were only achieving a 50% booking rate last year and even struggled with insufficient production volume in the second half of 2024. Samsung uses these sites to manufacture 4 nm, 5 nm, and 7 nm nodes, which are now considered "mature," as the leading edge shifts to sub-3 nm production.

Part of the foundry revival is attributed to the strong demand for its 6th-generation HBM, coming in the form of HBM4 with a custom base die manufactured on the 4 nm node. While competitors are making HBM4 base dies on older nodes, Samsung is using a 4 nm custom base die that provides higher design density for any logic that ASIC makers want to implement. This can include some data processing that will aid the main accelerator in AI workloads. This has translated into high demand for Samsung Foundry products from external customers, who are now keeping production at high capacity.

SoftBank and AMD Validate GPU Partitioning for AI Workloads Using Instinct GPUs

SoftBank and AMD have started joint validation of AMD Instinct GPUs for next-generation AI infrastructure by focusing on a GPU partitioning mechanism that lets a single GPU handle multiple AI workloads simultaneously. SoftBank developed an Orchestrator system that divides AMD Instinct GPU resources based on workload requirements: model size, number of concurrent executions, and memory needs. The system splits compute workloads across multiple GPU instances running on individual Accelerator Complex Dies (XCDs). These can range from single instance mode (SPX mode) up to eight instances (CPX mode). The HBM memory pools are also divided into individual regions for each GPU instance to prevent latency spikes. The goal is to avoid the inefficiency of uniform GPU resource allocation, which can lead to GPU resource shortages or waste depending on workload demands.

SoftBank says the enhanced Orchestrator runs multiple AI applications on a single GPU with minimal resource strain. No performance figures have been shared yet, though SoftBank mentions improved resource allocation for small and mid-size language model workloads. The company also plans to explore similar orchestration for other AI accelerators beyond AMD. A live demonstration is planned at the AMD booth during MWC Barcelona 2026 in March 2-5. SoftBank also published technical details on architecture and Orchestrator management methods on its Research Institute of Advanced Technology blog. In a recent report, we noted that AMD's next-gen Instinct MI455X accelerators, set to compete with NVIDIA's Vera Rubin, are running into serious manufacturing problems that are pushing back AMD's roadmap. Only limited production is expected this year, with mass production pushed to Q2 2027.

Samsung Ships Industry-First Commercial HBM4 With Ultimate Performance for AI Computing

Samsung Electronics Co., Ltd., a global leader in advanced memory technology, today announced that it has begun mass production of its industry-leading HBM4 and has shipped commercial products to customers. This achievement marks a first in the industry, securing an early leadership position in the HBM4 market. By proactively leveraging its most advanced 6th-generation 10 nanometer (nm)-class DRAM process (1c), the company achieved stable yields and industry-leading performance from the outset of mass production—all accomplished seamlessly and without any additional redesigns.

"Instead of taking the conventional path of utilizing existing proven designs, Samsung took the leap and adopted the most advanced nodes like the 1c DRAM and 4 nm logic process for HBM4," said Sang Joon Hwang, Executive Vice President and Head of Memory Development at Samsung Electronics. "By leveraging our process competitiveness and design optimization, we are able to secure substantial performance headroom, enabling us to satisfy our customers' escalating demands for higher performance, when they need them."

Applied Materials Unveils Transistor and Wiring Innovations for Faster Chips

Applied Materials, Inc., the leader in materials engineering for the semiconductor industry, today introduced new deposition, etch and materials modification systems that boost the performance of leading-edge logic chips at 2 nm and beyond. The technologies supercharge AI compute through atomic-scale improvements to the most fundamental electronic building block - the transistor.

The transition to Gate-All-Around (GAA) transistors is a major industry inflection and a critical enabler of the energy-efficient computing needed to deliver more powerful AI chips. As 2 nm-class GAA chips ramp to volume production this year, Applied is introducing new material innovations to enhance next-generation GAA transistors for angstrom nodes. The combined impact of the new chipmaking systems contributes a significant portion of the total energy-efficient performance gains of GAA process node transitions.

CXMT Reportedly Plans to Dedicate 20% of Mass Production Capacity to HBM3 Line in 2026

According to fresh news reports ChangXin Memory Technologies (CXMT) has just announced the initiation of HBM3 module mass manufacturing activities. The Chinese computer memory maker is slowly catching up with more mainstream international rivals—most notably, South Korean companies started full production of fourth-generation high-bandwidth memory (HBM3) products back in 2023. Over the past year, CXMT has demonstrated improved capabilities across a promising range of (mostly) commercial lines—for example, their proprietary DDR5 and LPDDR5X designs were showcased in preview form, last November. By early February (2026), the likes of ASUS, Acer, Dell, and HP were reportedly weighing up the sourcing of CXMT consumer-grade memory products.

The first signs of CXMT's diversification into the development of "homegrown" HBM3 modules were divulged by industry insiders, last May. By early autumn, a joint venture—involving YMTC—seemed to pave the way for an acceleration of the Chinese industry's overall HBM manufacturing footprint. Huawei's newest/futuristic Ascend AI accelerator relies on "in-house" HBM tech, due to global sanctions preventing the supply of cutting-edge Micron, Samsung, SK Hynix components to Chinese firms. An MK South Korea reports cites recent information disclosed by an anonymous source, regarding a secret partnership: "Huawei, which is leading the development of AI semiconductors in China, is developing HBM together with CXMT. It is expected to enter mass production despite (talk of) low yield(s)." Further inside track info posits that CXMT is planning a per month output of 60,000 wafers for "HBM3 production." This impressive guesstimate allegedly accounts for about 20% of the memory maker's total manufacturing capacity of 300,000 wafers per month, at least for 2026.

SEMI Reports 2025 Annual Worldwide Silicon Wafer Shipments and Revenue Results

Worldwide silicon wafer shipments in 2025 increased 5.8% to 12,973 million square inches (MSI) while wafer revenue slipped 1.2% to $11.4 billion over the same period, the SEMI Silicon Manufacturers Group (SMG) reported in its year-end analysis of the silicon wafer industry.

2025 marks an inflection year for wafer shipments, with silicon MSI resuming growth supported by strong demand for advanced epitaxial wafers in logic and polished wafers for high-bandwidth memory (HBM), driven by AI applications. Softness in wafer revenue is mostly attributed to the slow momentum in traditional semiconductor applications where the demand and pricing environment are yet to improve.

NVIDIA to Use SK hynix and Samsung HBM4 for "Vera Rubin" Without Micron

NVIDIA's upcoming "Vera Rubin" AI systems are scheduled for late summer shipping in the form of VR200 NVL72 rack-scale solutions that will power the next generation of AI models. However, not every memory maker of HBM4 qualified for a design win, as Micron has reportedly fallen out of the equation, with only Samsung and SK hynix left to supply the precious HBM4 memory. According to leaked institutional notes from SemiAnalysis, which tracks the supply chain in great detail, SK Hynix will represent about 70% of the HBM4 supply for VR200 NVL72 systems, with Samsung getting the remaining 30% of the supply. For a major memory maker like Micron, there is reportedly zero commitment for the supply of HBM4 memory.

Interestingly, this is not the end of Micron's share of memory in NVIDIA VR200 NVL72 systems. Instead of HBM4, the company will supply LPDDR5X memory for "Vera" CPUs, which can be equipped with up to 1.5 TB of LPDDR5X, making up for the lost share with the HBM4. It is possible that Micron didn't qualify for the significant system upgrade that NVIDIA performed for VR200 NVL72, which went from the initial system target of 13 TB/s in March 2025 to 20.5 TB/s in September. However, at CES 2026, NVIDIA confirmed that the VR200 NVL72 system is now operating at 22 TB/s of bandwidth, marking a nearly 70% increase in system bandwidth, all derived from aggressive memory specification scaling that the company demanded from the memory makers.

Forge Nano Transforms Advanced Semiconductor Chip Manufacturing with High-Speed 1000:1 Aspect Ratio Atomic Layer Deposition Coatings

Forge Nano, Inc., a technology company pioneering domestic battery and semiconductor innovations, today announced a breakthrough that fundamentally redefines the economics and architecture of advanced semiconductor manufacturing. The company has demonstrated high-speed, defect-free atomic layer deposition (ALD) coatings in semiconductor features at a 1000:1 aspect ratio. Conformality is maintained at production-scale while providing coverage on features 2 orders of magnitude greater than line-of-sight techniques.

This breakthrough removes the primary constraint that has limited 3D semiconductor scaling and enables architectures previously considered uneconomic or impossible. The demonstration was performed on production-representative wafers supplied by C2MI, a leading semiconductor process development and manufacturing innovation center in Canada, with conformality, defectivity and electrical performance independently validated through the partner's internal metrology and reliability testing.

Intel and SoftBank Subsidiary Saimemory Partner on Next-Gen Z-Angle Memory

Intel and SoftBank subsidiary Saimemory have signed an agreement to work together on developing and selling Z-Angle Memory (ZAM) technology aimed at AI and high-performance computing tasks. ZAM is described as a stacked DRAM architecture designed to go beyond current high-bandwidth memory (HBM) solutions. The goal is higher memory capacity and bandwidth at lower power. The name refers to vertical (Z-axis) stacking, according to reports from EE Times Japan. The project builds on Intel Next Generation DRAM Bonding (NGDB) work, developed under the U.S. government-backed Advanced Memory Technology (AMT) program in collaboration with Sandia, Lawrence Livermore, and Los Alamos national laboratories. Recent NGDB prototypes featuring eight vertically stacked DRAM layers have demonstrated full functionality using this new vertical stacking and bonding approach, as disclosed by Sandia in January.

According to reports from Nikkei and Wallstreet.cn, Saimemory is targeting two to three times the capacity of today's HBM, with 40-50% lower power consumption, while keeping costs competitive. Saimemory plans to have prototypes ready by early 2028, with commercial products following in 2029. SoftBank is said to be investing around ¥3 billion through the prototype phase. For Intel, the partnership marks a return to advanced memory technologies after exiting the DRAM market decades ago.

Qualcomm and AMD to Adopt SOCAMM2 Memory in AI Products

Qualcomm and AMD are reportedly considering the integration of SOCAMM2 memory into their AI product lines to enhance memory capacity and speed. Following NVIDIA's "Vera" CPU, which features LPDDR5X memory on the SOCAMM factor, offering 1.2 TB/s of memory bandwidth and supporting up to 1.5 TB of LPDDR5X memory, Qualcomm and AMD aim to pursue a similar strategy to achieve high-speed, high-capacity AI systems. These systems require fast memory close to AI accelerators, and SOCAMM can complement the extensive HBM memory on the accelerators. It acts as a fast memory pool with sufficient capacity to store entire models in memory, reducing the need for data transfer to SSD flash storage.

NVIDIA's SOCAMM approach involves surrounding the CPU with multiple SOCAMM modules, a strategy Qualcomm and AMD are expected to adopt in their AI solutions. AMD's Instinct MI accelerators, paired with EPYC CPUs, might be the next candidates for this integration, or AMD may develop a completely new system. Qualcomm already offers AI solutions like the AI200 and AI250 inference accelerators, which include up to 768 GB of LPDDR5 per card. These could soon be complemented by SOCAMM, allowing for standardized memory expansion based on the SOCAMM standard rather than soldering individual memory modules. This approach would enable manufacturers to offer various system versions and solutions by simply adding or removing SOCAMM modules, without the need for extensive PCB soldering work.

Samsung Electronics Announces Fourth Quarter and FY 2025 Results

Samsung Electronics today reported financial results for the fourth quarter and the fiscal year 2025.

In the quarter ended December 31, 2025, the Company posted its highest-ever quarterly consolidated revenue at KRW 93.8 trillion, representing a quarter-on-quarter (QoQ) increase of 9%. Operating profit was also an all-time high, at KRW 20.1 trillion. For the full year, the Company reported KRW 333.6 trillion in annual revenue and KRW 43.6 trillion in operating profit.
Return to Keyword Browsing
Nov 17th, 2026 13:49 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts