News Posts matching #HBM

Return to Keyword Browsing

DRAM Prices to Continue Rising in 4Q25, Server Demand Surges Ahead While Legacy Process Products See Steeper Increases

TrendForce's latest investigations reveal that the three major DRAM suppliers continue to allocate advanced process capacity primarily to high-end server DRAM and HBM, crowding out capacity for PC, mobile, and consumer applications. Combined with diverging demand across end products, this dynamic is keeping legacy-process DRAM price hikes pronounced, while newer-generation products are seeing more modest gains. Overall, conventional DRAM prices are expected to rise 8-13% QoQ in 4Q25, and when HBM is included, the increase could widen to 13-18%.

PC DRAM prices edge up slightly, U.S. CSPs consider early server DRAM procurement
Shipments of finished PCs are projected to decline with PC promotions and restocking momentum slowing in 4Q25, leading major OEMs to cut DRAM bit procurement. Meanwhile, as the three suppliers aggressively shift capacity toward server DDR5, the supply of both PC DDR5 and DDR4 remains constrained. As a result, PC DRAM prices are forecast to continue edging upward through the quarter.

Micron's 12-Hi HBM4 Delivers 2.8 TB/s Bandwidth, 11 Gb/s Per-Pin Transfer Rate

Yesterday, Micron reported its fourth-quarter and full-year fiscal 2025 results, beating the consensus and providing interesting comments on its upcoming HBM technology. During the earnings call, Sanjay Mehrotra, CEO of Micron, confirmed that HBM4—the company's next-generation memory—will be available next year with several enhancements over the base HBM4 JEDEC specification. "Micron Technology's HBM4 12-high remains on track to support customer platform ramps, even as the performance requirements for HBM4 bandwidth and pin speeds have increased. We have recently shipped customer samples of our HBM4 with industry-leading bandwidth exceeding 2.8 TB/s and pin speeds over 11 Gbps." The JEDEC specification for HBM4 is 2 TB/s bandwidth and 8 Gb/s pin transfer across a 2048-bit interface. Micron is pushing this to 11 Gb/s, resulting in 40% higher bandwidth at 2.8 TB/s.

There is a simple reason why this is the case. Customers like NVIDIA are asking HBM makers to create a more performant HBM memory to keep up with the exponential growth that their compute dies are experiencing. To satisfy these requirements, Micron is moving beyond JEDEC specifications. "We believe Micron Technology's HBM4 outperforms all competing HBM4 products, delivering industry-leading performance as well as best-in-class power efficiency. Our proven 1-gamma DRAM, innovative and power-efficient HBM4 design, in-house advanced CMOS base die, and advanced packaging innovations are key differentiators enabling this best-in-class product," said Sanjay.

InnoSilicon's Fenghua No.3 GPU Launches with 112GB+ HBM Memory and Claims to Support CUDA

InnoSilicon has unveiled the Fenghua No.3 GPU marking a significant shift in how the Chinese company develops GPUs. While for previous versions, they used Imagination Technologies' PowerVR IP, now the Fenghua No.3 uses an open-source RISC-V design based on OpenCore Institute's Nanhu V3 project. The Fenghua No.3 comes with over 112 GB of HBM memory putting it right in the AI accelerator market. InnoSilicon says one card can run 32B and 72B parameter language models. They also claim that if you use eight cards together, you can handle huge 671B and 685B parameter models.

The GPU support DirectX 12 Vulkan 1.2, and OpenGL 4.6 APIs for games, and it can handle even ray tracing. InnoSilicon showed the card running Tomb Raider, Delta Force, and Valorant, however it didn't share exact performance numbers or the settings in use. The card can power up to six 8 K screens at 30 Hz and supports YUV444 format to improve color accuracy. This aims at medical, CAD, and scientific computing markets. What's even more interesting is that InnoSilicon says that Fenghua No.3 it's CUDA compatible—a bold claim that, if true, could expand the GPU's software options. However, since CUDA is proprietary, this claim requires verification.

SK Hynix Begins 1b DRAM Pilot Production While Negotiating Customer Price Adjustments

SK Hynix is installing a new pilot production line at its M15X facility in Cheongju, bringing online a new 1b DRAM manufacturing line designed to address surging HBM demand, The Bell reports. Equipment installation is scheduled to be completed by November with the pilot line targeting an initial monthly output of 10,000 wafers. The move highlights different approaches among memory leaders, while Samsung pursues 1c DRAM technology for HBM4 applications, both SK Hynix and Micron have committed to 1b DRAM architecture, according to Aju News. This difference reflects varying technical roadmaps as companies race to capture as much as they can from the expanding AI memory markets.

On the other hand it seems that memory price adjustments have become a widespread practice. TrendForce reports citing SeDaily and Business Korea notes that SK Hynix has entered price negotiations with customers, however, no formal announcement has been made yet. The company is tweaking its prices in line with current market trends. This hints at a unified industry shift towards pricier memory across all big-name suppliers. Micron kicked things off in September first letting its channel partners know about 20-30% price hikes for DRAM. Samsung followed by informing major customers that Q4 contract prices for DRAM products, including LPDDR4X and LPDDR5/5X variants, would rise 15-30%. NAND pricing isn't immune either—eMMC and UFS contract prices are expected to climb 5-10%. On the same note, TrendForce indicates suppliers are ramping QLC SSD production, with capacity utilization projected to continue its growth through 2026. Inference AI workloads are expected to sustain demand into 2027, potentially creating enterprise SSD supply constraints by 2026.

Samsung Reportedly Raises DRAM and NAND Contract Prices up to 30%

According to Korean media outlet News Daily, Samsung Electronics has confirmed that it has raised contract prices for DRAM by as much as 15-30% and mobile NAND flash by roughly 5-10%, in response to a noticeably tighter supply/demand balance. Samsung reportedly notified major customers ahead of fourth-quarter contracting, following a market shift, where older-generation parts are being phased down while demand from hyperscale cloud operators and device makers racing to ship AI-enabled PCs and flagship phones has surged. Additionally, Samsung's rivals such as SK Hynix and Micron have signaled similar adjustments, with several suppliers notifying clients of double-digit increases and temporary limits on new quotes. For buyers, the result is immediate price pressure across server and mobile storage purchases, while for vendors, it is a tactical reprioritization of wafer capacity toward higher-margin, next-generation product lines.

Samsung's DRAM LPDDR4X, LPDDR5, and LPDDR5X families, as well as embedded NAND formats such as eMMC and UFS, are among those experiencing contract uplifts. Meanwhile, shortages in commodity DDR4 briefly pushed spot quotes sharply higher. Memory makers have redirected capacity to DDR5 and HBM, prioritizing HBM orders used by AI accelerators, tightening the supply of consumer-focused DRAM. Samsung, which retains roughly a third of global DRAM and NAND market share, is accelerating development of next-generation LPDDR6 even as it tweaks production schedules for legacy nodes. The pricing reset is expected to materially improve semiconductor division margins in the near term, cushioning against earlier cyclical weakness. For customers, it will force revisions to inventory planning and, likely, a renewed focus on design choices for price/performance.

Huawei Unveils Homegrown HBM and Ascend 950, Bets on Massive SuperClusters

Huawei used its Connect 2025 event to unveil an ambitious hardware roadmap that pairs the company's first self-developed HBM memory with a new generation of Ascend AI accelerators, and the message was clear: Huawei will compete on scale rather than single-chip supremacy. While still lagging behind NVIDIA on per-chip performance, the system-scale solutions can be competitive. The company confirmed the Ascend 950 family for early 2026 in two variants. One model, labeled 950PR, is described with about 128 GB of in-house HBM and roughly 1.6 TB/s of bandwidth, while the 950DT increases memory to 144 GB and pushes bandwidth toward 4 TB/s. Huawei also sketched longer term plans for Ascend 960 and Ascend 970 devices arriving in 2027 and 2028, promising larger aggregate memory capacities, higher interconnect bandwidth, and wider support for FP8 precision formats.

AMD Launches ROCm 7.0, Up to 3.8x Performance Uplift Over ROCm 6.0

AMD today unveiled ROCm 7.0, a massive update to its open GPU software platform for AI workloads across datacenter racks and even client devices. The release focuses on support for the new Instinct MI350 accelerator series, built on CDNA 4. It pairs hardware-level refinements, including advanced HBM utilization and expanded datatype handling, with software improvements designed to enhance throughput for both training and inference. AMD classifies this release as a generational step that widens deployment choices for enterprises, from single-node servers to multi-rack clusters. Company benchmarks shared with partners indicate multi-fold gains over previous ROCm 6.0 versions on select large-model tasks, and AMD shows tighter integration with open-source serving stacks to help teams transition from prototype to production more quickly.

Under the hood, ROCm 7.0 delivers algorithmic and systems advances intended to reduce cost and improve utilization: enhanced attention and reasoning primitives, sparse mixture-of-experts techniques, and support for FP4, FP6, and FP8 precision formats that trade a bit of accuracy for massive memory savings. ROCm 7.0 also focuses on scale and manageability by enabling GPU-direct communications, distributed inference across nodes with vLLM-d and DeepEP compatibility, and cluster tools such as AMD Resource Manager and a new AI Workbench for orchestration and monitoring. To smooth adoption, AMD provides prebuilt Docker images with quantized model examples and turnkey stacks for common frameworks. AMD is also expanding endpoint support to Ryzen AI processors and Radeon RX series GPUs, so developers can deploy consistent pipelines from edge devices to cloud racks. ROCm 7 is available now through AMD's developer resources, with documentation, examples, and enterprise support to help teams get started.

TCS Unveils Chiplet-Based System Engineering Services to Accelerate Semiconductor Innovation

Tata Consultancy Services a global leader in IT services, consulting, and business solutions, announced the launch of its Chiplet-based System Engineering Services, designed to help semiconductor companies push the boundaries of traditional chip design. By using chiplets (which are small integrated circuits that serve as building blocks of larger chips), TCS is enabling chipmakers to deliver faster, efficient, and more powerful processors at a time when global demand for advanced semiconductors is soaring.

The timing of this launch is significant for India. The country's semiconductor market valued at $45-50 billion in 2024-2025, is projected to more than double to $100-110 billion by 2030. With government support through the ₹76,000 crore India Semiconductor Mission, the country is positioning itself as a major global hub for chip design and manufacturing. Already, India is home to 20% of the world's chip design engineers, and leading global players are investing in manufacturing and assembly facilities. TCS' new services are expected to strengthen this momentum by giving both Indian and international companies access to world-class expertise in chip-to-system engineering.

China's AI Chip Ambitions Limited by HBM Memory Supply, Notes Report

According to SemiAnalysis, Chinese AI semiconductor buildout faces an unexpected bottleneck that's more critical than manufacturing constraints: high-bandwidth memory shortages. While domestic foundries like SMIC can produce sufficient AI processors, the lack of HBM supply chains has emerged as the primary limitation throttling China's semiconductor ambitions. Huawei's Ascend 910C illustrates this mismatch. The company has foundry capacity to manufacture 805,000 units annually through TSMC, which is a result of stockpiling before the cutoff, as well as SMIC. However, HBM shortages prevent realizing this production potential. Chinese firms accumulated roughly 11.4 million HBM stacks from Samsung before export controls tightened, but these stockpiles won't sustain long-term AI scaling. The memory crunch effectively neutralizes Chinese manufacturing capabilities, giving Western competitors like NVIDIA and AMD a sustained advantage despite Beijing's substantial semiconductor investments.

The path toward HBM independence runs through domestic memory manufacturers like CXMT and the recent collaboration with YMTC. Converting standard DRAM production to HBM manufacturing requires specialized equipment, which is predominantly supplied by Western suppliers, creating another dependency layer. China could achieve competitive HBM3E production by 2026 if current investment trajectories continue and equipment restrictions remain stable. However, this timeline assumes successful technology transfer and uninterrupted access to critical manufacturing tools. If export controls expand to encompass additional HBM-related equipment, the development window could become considerably longer. YMTC is reportedly ready to enter DRAM production and could begin purchasing equipment by late 2025, utilizing its secret weapon, Xtacking technology, one of the most advanced hybrid bonding techniques available. This matters because HBM requires stacking at least 16 memory chips with incredible precision, something YMTC has been doing for years with their NAND products.

YMTC and CXMT to Jointly Accelerate China's HBM Production

Two of the biggest names in the Chinese memory/storage industry, YMTC and CXMT, are reportedly teaming up to accelerate the domestic high-bandwidth memory development. YMTC, best known for making NAND flash storage, is apparently ready to jump into DRAM production and could start buying equipment by late 2025. What makes this interesting is YMTC's secret weapon: their Xtacking technology, which TechInsights calls one of the best hybrid bonding techniques out there. This matters because HBM requires stacking at least 16 memory chips with incredible precision, something YMTC has been doing for years with their NAND products. Meanwhile, CXMT has been ramping up fast, pushing out 70% more DRAM wafers than last year and aiming to get HBM3 certified soon, with HBM3E coming within two years.

CXMT already produces HBM2 and has the DRAM manufacturing infrastructure, while YMTC's bonding expertise could help solve the technical headaches that come with stacking memory chips higher and higher without them overheating. Additionally, companies like Wuhan Xinxin and Tongfu Microelectronics are developing the packaging and assembly capabilities necessary to transform these chips into finished products. CXMT expects to produce 2.73 million wafers in 2025, up from 1.62 million this year, and they could soon match the volume of Micron's shipments. However, all we need to watch now is the feasibility of this HBM development, as Chinese companies require cutting-edge DRAM for their accelerators, like Huawei's Ascend. If global demand persists, we may even see some Western customers use HBM originated in China.

2Q25 DRAM Revenue Jumps 17.1%, SK hynix Expands Market Share as Taiwanese Vendors Deliver Strong Growth

TrendForce's latest investigations demonstrate that global DRAM industry revenue reached US$31.63 billion in the second quarter of 2025, up 17.1% QoQ. This growth was fueled by rising contract prices for conventional DRAM, robust shipments growth, and expanding HBM volumes. Inventory digestion at DRAM suppliers accelerated with stronger procurement momentum from PC OEMs, smartphone makers, and CSPs, driving contract prices for most products back into positive territory.

SK hynix posted bit shipments above its targets in 2Q25. However, the higher share of lower-priced DDR4 shipments limited ASP growth. Even so, revenue surged 25.8% QoQ to $12.23 billion, boosting market share to 38.7% and securing the top spot once again.

Marvell Unveils Industry's First 64 Gbps/wire Bi-Directional Die-to-Die Interface IP in 2nm to Power Next Generation XPUs

Marvell Technology, Inc. (NASDAQ: MRVL), a leader in data infrastructure semiconductor solutions, today announced the industry's first 2 nm 64 Gbps bi-directional die-to-die (D2D) interconnect, enabling chip designers to significantly boost the bandwidth and performance of next-generation XPUs while reducing power and silicon area. Delivering 32 Gbps of simultaneous two-way connectivity over a single wire, the interface IP—also available in 3 nm—sets a new standard for performance, power efficiency, and resiliency to meet the scaling demands of next-generation data centers.

The Marvell 64 Gbps bi-directional D2D interface offers bandwidth density over 30 Tbps/mm, more than three times the bandwidth density of UCIe at equivalent speeds, and a minimal depth configuration that reduces compute die area requirements to 15% compared to conventional implementations. The interface IP is also the industry's first in its class to feature advanced adaptive power management that automatically adjusts device activity to bursty data center traffic. This innovation reduces interface power consumption by up to 75% with normal workloads and up to 42% during peak traffic periods.

NVIDIA Prepares Cut-Down B30A "Blackwell" AI Accelerator for Chinese Market

NVIDIA's China-specific accelerators are not unknown by now, and the last export-modified accelerator the company prepared was H20, which recently got an export license. However, Chinese AI labs are on the lookout for a more performant design, and NVIDIA is here to deliver. According to Reuters, NVIDIA has designed a new SKU called B30A, based on the B300 "Blackwell Ultra." Designed as a single-die design, it takes the same base die from the dual-die B300, and just embeds a single one onto the package for half the performance to comply with export regulations.

As the Blackwell Ultra, in its B300 dual-die configuration, features 15 PetaFLOPS of FP4, 7.5 PetaFLOPS at FP6/FP8, 3.75 PetaFLOPS at FP16/BF16, and 1.88 PetaFLOPS at TF32 precision, we can expect the single-die B30A to halve these numbers. This yields roughly 7.5 PetaFLOPS FP4, 3.75 PetaFLOPS FP6/FP8, 1.875 PetaFLOPS FP16/BF16, and 0.94 PetaFLOPS at TF32. Additionally, NVIDIA will not skimp on the mission-critical features like HBM and NVLink, which enable AI training and inference in a much easier way. These chips may be sufficient to counter any current Chinese-designed AI accelerator with a superior software stack and higher performance. As the U.S. administration decided to grant NVIDIA access to the export license, NVIDIA is again serving the Chinese market with a high-performance accelerator for their domestic needs and AI labs.

NVIDIA Could Target 2027 for 3nm HBM Base Die Production

According to market reports, NVIDIA is planning to develop its own High Bandwidth Memory (HBM) Base Die design, with production targeting 3 nm process technology and trial manufacturing expected to begin in late 2027. TrendForce, citing Commercial Times, sees this as a strategic move that would allow NVIDIA to control the underlying logic components of HBM modules regardless of which memory manufacturer supplies the stacked memory layers. The news disrupted the HBM industry, where SK Hynix currently dominates. While SK Hynix uses primarily in-house Base Die solutions, it relies on advanced manufacturing processes from companies like TSMC to achieve speeds above 10 Gbps. Memory manufacturers typically lack the complex design capabilities needed for sophisticated Base Die and ASIC development.

NVIDIA's plan seems to aim at tightening its grip on its ecosystem through the NVLink Fusion platform giving customers more mix-and-match choices. However, cloud service providers might push back against NVIDIA's Base Die solutions, as many first went for ASIC development to cut their reliance on NVIDIA. At the same time, SK Hynix keeps pushing its HBM4 technology forward. The company just recently shipped 12-layer samples with 36 GB capacity and speed over 2 TB per second (60% faster than HBM3E). HBM4 is expected to deliver higher speeds and stacking with more complex packaging integration. As we previously reported, NVIDIA and AMD will deploy customized HBM4 memory in their upcoming AI accelerators. Combined with NVIDIA's plans to manufacture its own HBM Base Die and SK Hynix's accelerated HBM4 production timeline, the HBM market is about to enter a new phase of competition and change.

SK hynix Surpasses Samsung as Leading DRAM Manufacturer

SK hynix has taken the lead from Samsung Electronics as the top DRAM maker worldwide signaling the first change in leadership in over three decades. The main reason for this market shift is high bandwidth memory (HBM) where SK hynix has an exclusive deal to supply NVIDIA while Samsung hasn't yet secured this partnership (it does however supply HBM3E to AMD and Broadcom). Samsung's slice of the global DRAM market dropped significantly from 41.5% at the end of 2024 to 32.7% in the first half of 2025. This 8.8% drop is Samsung's biggest market share loss since they started keeping track in 1999. The company's market position has gone back to levels not seen since 2014.

SK hynix grabbed a 36.3% market share in the first half of 2025 beating Samsung for the first time since 1992. SK hynix DRAM market share has been growing from 27.7% in 2022 to its current top spot. A key component for this growth was also SK hynix's U.S. subsidiary which reported sales of $17.79 billion in the first half of 2025, representing a 103 percent increase from the previous year. Industry analysts expect SK hynix to maintain its competitive advantage in the HBM market despite anticipated increased competition from other manufacturers.

NVIDIA and AMD Customize HBM4 Memory for Next-Generation AI Accelerators

NVIDIA's and AMD's next-generation accelerators will definitely utilize a next-generation HBM4 memory, too. However, interesting news, confirmed by SemiAnalysis, is that these accelerators will utilize a customized HBM memory for the first time. In an HBM stack, consisting of stacked DRAM dies up to a 12-high, connected with TSVs, there is the possibility of embedding an optional base die with customized logic/accelerator circuitry tailored to a specific need. It appears that NVIDIA and AMD are not only exploring this option to gain more performance, but are also actively working on an implementation with a custom base that will push these accelerators ahead of any third-party ASIC solution by a third. Since everyone is trying to grab a bit of market share in the AI ASIC world, NVIDIA and AMD have to innovate their designs to stay ahead constantly.

"We expect NVIDIA and AMD to come out with their custom HBM implementations with HBM4, with other accelerator designers likely to only have a custom base die implementation by HBM4E in the 2027 timeframe," notes the SemiAnalysis post. This means that while ASIC designers, such as Broadcom and MediaTek, will have a working solution in 2027 chips, NVIDIA and AMD will have a working custom HBM4 in 2026 products. For NVIDIA, this is the "Rubin" architecture, supposed to outperform the current "Blackwell" implementation by a wide margin. For AMD, this is the Instinct MI400 series of accelerators, which is coming for the first time in a rack-scale flavor.

Sandisk to Collaborate with SK hynix to Drive Standardization of High-Bandwidth Flash Memory Technology

Sandisk Corporation (NASDAQ: SNDK) today announced it has signed a landmark Memorandum of Understanding (MOU) with SK hynix to work together to establish the specification for High Bandwidth Flash - a new technology designed to deliver breakthrough memory capacity and performance for the next generation of AI inferences. Through this collaboration, the companies expect to standardize the specification, define technology requirements and explore the creation of a technology ecosystem for High Bandwidth Flash.

"By collaborating with SK hynix to define the High Bandwidth Flash specification, we are addressing the critical need for scalable memory in the AI industry," said Alper Ilkbahar, Executive Vice President and Chief Technology Officer, and HBF Technical Advisory Board member at Sandisk. "This collaboration accelerates innovation and will offer the industry new tools to handle the exponential data demands of tomorrow's applications. Our work will help provide an effective solution to meet the world's technology needs and exceeds expectations of our respective customers."

NEO Semiconductor Introduces World's First Extreme High Bandwidth Memory (X-HBM) Architecture for AI Chips

NEO Semiconductor, a leading developer of breakthrough memory technologies, today introduced the world's first Extreme High Bandwidth Memory (X-HBM) architecture for AI chips. Built to meet the growing demands of generative AI and high-performance computing, X-HBM delivers unmatched performance with a 32K-bit data bus and potentially 512 Gbit per die, dramatically surpassing the limitations of traditional HBM with 16X greater bandwidth or 10X higher density.

"X-HBM is not an incremental upgrade, it's a fundamental breakthrough," said Andy Hsu, Founder & CEO of NEO Semiconductor. "With 16X the bandwidth or 10X the density of current memory technologies, X-HBM gives AI chipmakers a clear path to deliver next-generation performance years ahead of the existing roadmap. It's a game-changer for accelerating AI infrastructure, reducing energy consumption, and scaling AI capabilities across industries."

Samsung Electronics Announces Second Quarter 2025 Results

Samsung Electronics today reported financial results for the second quarter ended June 30, 2025. The Company posted KRW 74.6 trillion in consolidated revenue, a decrease of 5.8% compared to the previous quarter. Operating profit decreased to KRW 4.7 trillion.

The Device Solutions (DS) Division reported an increase in revenue on the back of expanded sales of high density, high-performance memory products, but inventory value adjustments in memory and one-off costs related to the impacts of export restrictions related to China in non-memory had an adverse effect on profit. In the Device eXperience (DX) Division, operating profit declined quarter-on-quarter due to a sequential decline in sales volume following the launch of new smartphone models in the first quarter. Looking ahead to H2, the DS Division plans to proactively meet the growing demand for high-value-added and AI-driven products and continue to strengthen competitiveness in advanced semiconductors. The DX Division will seek to minimize the impact of uncertainties stemming from tariff policies that are likely to persist.

Sandisk Forms HBF Advisory Board for High-Bandwidth Flash Memory Development

Sandisk Corporation today announced the formation of a Technical Advisory Board to guide the development and strategy of its groundbreaking High Bandwidth Flash (HBF) memory technology. The board includes industry experts and senior technical leaders from both within and outside the company. Appointed today, Professor David Patterson and Raja Koduri will provide strategic guidance, technical insight, market perspective, and shape open standards as Sandisk prepares to launch HBF.

"We're honored to have two distinguished computer architecture experts join our Technical Advisory Board," said Alper Ilkbahar, Executive Vice President, Chief Technology Officer, and HBF Technical Advisory Board member at Sandisk. "Their collective experience and strategic counsel will be instrumental in shaping HBF as the future memory standard for the AI industry, and affirming we not only meet but exceed the expectations of our customers and partners."

SK hynix Announces 2Q25 Financial Results

SK hynix Inc. announced today that it has recorded 22.232 trillion won in revenues, 9.2129 trillion won in operating profit (with an operating margin of 41%), and 6.9962 trillion won in net profit (with a net margin of 31%) in the second quarter. Both revenues and operating profits stood at all-time highs, beating the previous best results in the fourth quarter of last year. The company said that an aggressive investment by global big tech companies into AI led to a steady increase in demand for AI memory. Shipments of both DRAM and NAND flash were higher than expected, helping the company log the best quarterly results.

SK hynix expanded sales of 12-high HBM3E in the DRAM space, while registering a growth in sales of NAND for all applications. With the industry-leading competitiveness in AI memory and profitability-first management discipline, the company has maintained a positive earnings trend. With the latest financial results, the company's cash and cash equivalents increased to 17 trillion won at the end of June, up by 2.7 trillion won a quarter earlier. Its debt ratio and net debt ratio stood at 25% and 6%, respectively, as net debt fell by 4.1 trillion won, compared with the previous quarter.

Nikon Introduces 600x600 mm Substrates for Advanced AI Silicon

Nikon announced that it will begin accepting orders today for its Digital Lithography System DSP-100, a back-end tool purpose-built for the next-gen of advanced semiconductor packaging. The company will deliver the first units during fiscal year 2026. The DSP-100 is specifically designed for panel-level packaging (PLP), a technology that industry leaders such as TSMC, Intel, and Samsung are adopting to overcome the cost and area limitations of 300 mm wafers. Standardized rectangular substrates, already fixed at 510x515 mm, allow more compute chiplets, HBM4 stacks, and I/O dies to be integrated into a single package. Nikon's engineers pushed the standard further by enabling exposure on glass or resin panels up to 600×600 mm. Key specifications released today show a 1.0 µm line-and-space resolution, overlay accuracy within ± 0.3 µm and throughput of 50 panels per hour at the 510x515 mm size.

By eliminating photomasks and instead steering an i-line equivalent light source through a spatial light modulator (SLM), the DSP-100 sidesteps both the optical distortion inherent in large single-lens systems and the expense of mask sets. A proprietary multi-lens array borrowed from Nikon's flat-panel display division stitches the pattern seamlessly across the full substrate. With a 600×600 mm panel accommodating 36 of the 100 mm square packages, the system delivers nine times higher throughput productivity compared to 300 mm wafer processes, dramatically reducing the cost per die. The maskless architecture also shortens design iteration cycles, a critical advantage as generative AI workloads drive rapid evaluation of data center accelerators. Nikon states that volume production of the DSP-100 will commence in 2026, with pricing disclosed only to qualified buyers.

LG Electronics to Enter Semiconductor Equipment Market with Hybrid Bonding

LG Electronics has quietly launched a plan to become a semiconductor equipment maker. Its Production Technology Research Institute has begun developing a hybrid bonding machine tailored for next-generation high bandwidth memory (HBM), with an internal goal of shipping production units by 2028. Hybrid bonding is a wafer-level joining technique that eliminates the need for solder bumps. Instead, copper pads on each die are planarized to nanometer-scale smoothness and pressed together at room temperature, forming a permanent direct electrical contact. This method yields thinner memory stacks, operates at lower temperatures, and delivers faster electrical performance than those assembled with conventional thermal compression bonding. HBM consists of tightly stacked DRAM layers, where current packaging supports up to eight layers with thermal compression bonding. Stacking beyond twelve layers causes the solder bump pitch to collapse, making hybrid bonding essential for taller, more efficient stacks.

LG's push into hybrid bonding development comes as industry leaders SK hynix, Micron, and Samsung race to supply HBM4 and HBM4E memory for NVIDIA, AMD, Intel, Google, and Amazon's AI systems slated for 2026. To accelerate its program, LG is recruiting dozens of PhDs in semiconductor packaging and collaborating with Seoul National University. The company views the hybrid bonding as a natural extension of its existing B2B offerings in HVAC and robotics. Currently, only the Dutch firm BESI and the US supplier Applied Materials offer commercial hybrid bonding, and neither has operations in Korea. Local competitors, including Samsung's Semes, Hanwha Semitec, and Hanmi Semiconductor, remain in prototype or pilot phases. If LG meets its 2028 target, its first customer shipments could align with SK Hynix's HBM4E volume ramp and Samsung's HBM4 risk line launch.

Transition between DRAM Generations Drives Diverging 3Q25 Price Trends; Consumer DDR4 Prices Expected to Surge Over 40%

TrendForce's latest findings show that the three major DRAM suppliers are shifting capacity toward high-end products and have begun announcing end-of-life (EOL) plans for PC and server-grade DDR4 and mobile LPDDR4X. This has triggered aggressive restocking of legacy-generation products, further fueled by traditional peak-season demand. Consequently, average contact prices for conventional DRAM are projected to rise by 10% to 15% in 3Q25. Including HBM, overall DRAM prices are expected to increase by 15% to 20%.

TrendForce notes that DDR4 demand remains strong in 3Q25, and suppliers are aiming to raise prices more aggressively. As Taiwanese manufacturers currently lack the capacity and product specifications to fully meet market needs, DDR4 supply is expected to remain tight in the short term. Moreover, DDR4 output is being prioritized for server applications, leaving limited availability for consumer-grade user.

Intel "Jaguar Shores" Uses HBM4, "Diamond Rapids" Pairs with MRDIMM Gen 2 Memory

During the Intel AI Summit in Seoul, South Korea, Intel teased its upcoming product portfolio, featuring next-generation memory technologies. Being in Seoul, memory makers like SK Hynix are Intel's main partners for these products. Teased at the summit is Intel's upcoming AI accelerator, called "Jaguar Shores," which utilizes the next-generation HBM4 memory, offering 2.0 TB/s of bandwidth per module across 2,048 IO pins. SK Hynix plans to support this accelerator with its memory, ensuring that Intel's big data center-grade AI accelerator is equipped with the fastest memory on the market. Since the "Falcon Shores" accelerator is only intended for testing with external customers, we don't have an exact baseline to compare to, and Jaguar Shores specifications are scarce.

Next up, Intel confirmed that its upcoming seventh-generation "Diamond Rapids" Xeon processors will use the second generation of MRDIMMs (Multiplexer Rank Dual Inline Memory Modules), an upgrade from the first-generation MRDIMMs used in the Xeon 6 family. The upgrade to MRDIMMs Gen 2 will allow Intel to push transfer rates to 12,800 MT/s, up from 8,800 MT/s in Xeon 6 with MRDIMMs Gen 1. Alongside this 45% speed bump in raw transfer rates, the memory channels are jumping to 16, up from 12 in the current generation, yielding an additional bandwidth boost. Given that MRDIMMs operate by connecting more memory ranks using a multiplexer, and that these modules buffer data and commands, the increased data transfer rate comes without any additional signal degradation. As Intel is expected to pack more cores, this will be an essential piece in the toolbox to feed them and keep those cores busy on the Oak Stream platform, based on the LGA9324 socket.
Return to Keyword Browsing
Nov 17th, 2026 13:49 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts