Samsung in Production of Second-gen 10nm DRAM

发布时间:2017-12-21 00:00
作者:Ameya360
来源:Dylan McGrath
阅读量:2090

  South Korea's Samsung Electronics said it has commenced production of the second generation of its 10nm-class 8-Gb DDR4 DRAM. The chips achieve speeds of up to 3,600 megabits per second (Mbps) and are produced without the need for extreme ultraviolet (EUV) lithography.

  Gyoyoung Jin, president of Samsung's memory business, said through a press statement that new technologies in DRAM circuit design and process technology enabled Samsung to break through "a major barrier for DRAM scalability." To meet booming market demand, he said Samsung would rapidly ramp up its second-generation 10nm-class DRAM production and also aggressively expand production of the first generation of the chips.

  Devices labeled 10nm-class have feature sizes as small as 10 to 19 nanometers.

  The market for DRAM chips has been robust all year, as demand growth has outstripped supply and led to continual price increases. According to market research firm IC Insights, the DRAM market will grow by 74 percent this year, its largest rate of expansion since 1994.

  Samsung is planning to begin transitioning to EUV for logic chips next year at the 7nm node, although it is unclear when the technology will be put into production for DRAM. Because of the relative structural simplicity of DRAM cells, it is generally accepted that it will require EUV later than other types of devices.

  Samsung said it is able to produce second-generation 10nm-class DRAMs that achieve speed, performance and efficiency advantages over first-generation devices by employing advanced techniques, including a high-sensitivity cell data sensing system and a progressive “air spacer” scheme.

  The data sensing system enables a more accurate determination of the data stored in each cell, increasing the level of circuit integration and manufacturing productivity, Samsung said. Meanwhile, placing an air spacer around bit lines decreases parasitic capacitance, enabling a higher level of scaling and rapid cell operation, according to the company.

  Samsung claims its second-generation 10nm-class 8Gb DRAM boasts about 30 percent better productivity than the first generation of the chips, while also offering performance and energy efficiency advantages of 15 and 10 percent, respectively. The new devices can operate at 3,600 Mbps per pin, up from about 3,200 Mbps for the first-generation chips, which have been in production since 2016.

  The innovations used in the second-generation 10nm-class DRAMs will enable Samsung to accelerate its plans for faster introductions of future DRAM chips, including DDR5, HBM3, LPDDR5 and GDDR6, the company said.

(备注:文章来源于网络,信息仅供参考,不代表本网站观点,如有侵权请联系删除!)

在线留言询价

相关阅读
Samsung’s 4nm GAIA Could Mark First PIM Commercialization in AI PCs, Mass Production as Early as 2027
Samsung is exploring a new approach to boost AI performance and efficiency on PCs. According to ZDNet, Samsung showcased its LPDDR5X-PIM paired with an edge AI accelerator SoC at Hot Chips. The company’s AI PC chipset under development, known as “GAIA,” is expected to incorporate LPDDR5X-PIM, which integrates computing capabilities directly into memory. This would mark the first integration of Processing-in-Memory (PIM) into an actual AI PC chipset. If GAIA reaches mass production, it could also become the first commercial application of PIM beyond the research and validation stage, the report notes.  GAIA is an AI accelerator for AI PCs being developed by Samsung Electronics’ System LSI division. Built on Samsung Foundry’s 4nm process, the chip is optimized for AI workloads. ZDNet adds that Samsung has reportedly provided GAIA prototypes to major PC makers, including Lenovo and HP, for performance validation, with mass production potentially beginning as early as next year.  LPDDR5X-PIM Delivers Up to 3.01x Higher Output in Testing  The key to the approach is reducing data movement. As the report explains, LPDDR5X-PIM adds computing capabilities directly to low-power LPDDR5X memory, allowing some data to be processed where it is stored rather than repeatedly transferred between memory and CPUs or GPUs. This can ease AI computing bottlenecks while reducing power consumption.  Samsung validated the SoC with both LPDDR5X and LPDDR5X-PIM, with PIM showing clear performance gains. LPDDR5X-PIM completed the test workload in 5.4 seconds versus 12.3 seconds for conventional LPDDR5X, a 2.28x improvement, while output jumped 3.01x from 27.0 to 81.3 per second, the report notes. ZDNet adds that Samsung previously said simulations showed up to 6.2x higher AI matrix-computation performance than conventional memory. These characteristics make LPDDR5X-PIM particularly suited to on-device AI, where laptops and smartphones face tight power and thermal limits.  GAIA Could Put PIM Commercialization to the Test  Samsung has been extending PIM technology across its memory portfolio, according to ZDNet. The company has introduced HBM-PIM for servers and integrated PIM with LPDDR for mobile applications. However, technical and ecosystem challenges have continued to limit its adoption in commercial products, the report adds. GAIA could therefore serve as a key test of PIM’s commercial potential. Sources cited by ZDNet note that if GAIA moves into mass production after customer validation in the AI PC market, PIM could evolve from a next-generation memory technology into a key solution for on-device AI.  Samsung is also lowering barriers to PIM commercialization. As noted by Seoul Economic Daily, its Address Alignment Mode (AAM) allows LPDDR5X-PIM to work with existing DRAM memory controllers, reducing the need for major system changes. The report adds that the first JEDEC specification for next-generation LPDDR6-PIM is nearing finalization, as Samsung looks to expand PIM from a proprietary solution into a standardized memory technology.  Meanwhile, SK hynix is also advancing PIM-based AI solutions. At CES 2026, the company showcased AiMX, an AI accelerator prototype built around its GDDR6-AiM PIM chip and designed for LLM workloads. The company also demonstrated other computational-memory technologies, including Compute-using-DRAM (CuD) and CMM-Ax.
2026-08-27 13:46 阅读量:465
Samsung Foundry Eyes 2H26 Full Utilization as HBM4 Base Dies and U.S. Orders Fuel Turnaround Hopes
  At last week’s earnings call, Samsung surprised the market with a strong performance, driven almost entirely by its memory business, which contributed 99.7% of total operating profit. While the foundry unit remained in the red, a turnaround could come sooner than expected, with Chosun Biz reporting that Samsung Foundry aims to reach 100% capacity utilization by the second half of this year, up from the current 70%–80% range.  According to Newsis, Samsung’s non-memory businesses, including Foundry and System LSI, are estimated to have posted a loss of around KRW 600 billion in 2Q. The improvement has fueled expectations of an earlier-than-expected turnaround, with some industry observers speculating that profitability could return within this year, ahead of Han Jin-man’s previously cited 2028 timeline, the report adds.  The most bullish outlook may come from HuffPost Korea, which cited an analyst report projecting that Samsung Foundry could return to profitability as early as the third quarter. The forecast was based on the ramp-up of HBM4 base die production and increased output of Samsung’s in-house Exynos application processors, according to the report.  Chosun Biz attributes Samsung Foundry’s utilization recovery to stronger demand for advanced products, particularly HBM base dies as well as growing orders from major U.S. tech customers. The improvement marks a sharp reversal from last year, when utilization at key nodes such as 4nm and 5nm fell below 50%, according to the report.  Samsung’s 2nm Gains Traction Despite Yield Hurdles  Still, the recovery of advanced nodes and further yield improvements will determine whether Samsung Foundry can sustain its turnaround momentum. According to Chosun Biz, Samsung needs to raise 2nm yields to around 70% to secure large-scale mass production contracts, with current yields estimated at roughly 60%.  Despite the remaining 2nm yield challenges, Samsung is gaining traction among global tech giants. According to News1, Samsung is expanding advanced-node engagements with major customers such as Broadcom, with 2nm order volume expected to more than double this year from last year’s level.  The Information also reported earlier that while Google plans to have TSMC manufacture the main computing die for its next-generation TPU, codenamed “Icefish,” Samsung could be tapped to produce an I/O-related component that connects the chip to memory using its 2nm technology.  Meanwhile, Samsung is already reshaping its business mix. As noted by Chosun Biz, advanced-node processes are expected to account for more than 50% of the company’s foundry revenue this year, while AI and HPC-related sales are projected to rise from the high-teens percentage range last year to over 30%.
2026-08-04 14:54 阅读量:768
Samsung Reportedly Builds Hybrid Bonding Mass Production Line; Full-Scale Deployment Seen in 2029–2030
  With next-generation HBM advancing, attention is increasingly turning to the commercialization of hybrid bonding. According to The Elec, citing industry sources, Samsung Electronics is establishing a mass production line equipped with around 50 die-to-wafer (D2W) hybrid bonders at its Pyeongtaek campus to support next-generation HBM and logic semiconductor production. Equipment deliveries and installation are expected to begin toward the end of 2026.  However, The Elec notes that Samsung internally expects large-scale hybrid bonding production to begin around 2030. Weekly Post similarly reports that Samsung Electronics aims to begin full-scale hybrid bonding mass production in 2029. The era of hybrid bonding-based HBM is expected to begin with the deployments of NVIDIA’s next-generation AI GPU, Feynman. While Rubin Ultra is expected to use HBM4E, Feynman is highly likely to adopt HBM5 or custom HBM.  BESI Remains the Front-Runner as Samsung Explores Alternatives  Notably, Samsung Electronics has reportedly selected Dutch semiconductor back-end equipment supplier BESI as its preferred partner for D2W hybrid bonding equipment, The Elec notes. However, the order has yet to be finalized as the companies continue negotiating Samsung’s requested equipment modifications. The report also cites the price of BESI’s hybrid bonder—around KRW 6 billion (approximately US$4.3 million) per unit, roughly twice that of competing products—as a factor contributing to the prolonged negotiations. BESI, which also supplies hybrid bonders to TSMC and Micron, is reportedly reluctant to modify the tool architecture exclusively for Samsung, The Elec adds.  While BESI remains Samsung’s preferred supplier, the company is also evaluating domestic equipment makers as alternatives. SEMES, Samsung Electronics’ semiconductor equipment subsidiary, has also reportedly been asked to supply D2W hybrid bonders for the planned mass production line. Industry sources cited by The Elec said the equipment is believed to have passed Samsung’s qualification process. Samsung is also said to be considering Hanwha Semitech as a potential hybrid bonder supplier. Hanwha Semitech completed development of its second-generation D2W hybrid bonder, SHB2 Nano, in February and supplied an evaluation system to SK hynix in April, the report notes.  Hybrid Bonding for cHBM and 3D Packaging  Hybrid bonding is expected to play a particularly important role in customized HBM (cHBM), The Elec notes. Samsung plans to replace the cHBM base die with a logic die incorporating customer computing IP, enabling it to support external CPUs and GPUs while optimizing data transfer. The company also aims to implement a 3D system-in-package (SiP) structure that stacks HBM DRAM layers directly on the logic die.  Separately, Samsung Foundry introduced its next-generation hybrid bonding-based 3D vertical stacking solution, 3D Cube-H, last month, according to The Elec. Designed for next-generation AI chips and high-performance computing (HPC) systems, the heterogeneous integration technology is expected to deliver higher performance, improved power efficiency, and greater bandwidth than existing packaging solutions. Samsung Foundry is currently promoting the technology to customers.
2026-07-23 10:14 阅读量:950
Samsung Reportedly Completes Tesla AI5 Tape-out, Paving the Way for 2nm Ramp at Taylor Fab
  Samsung’s 2nm ramp at its Taylor fab may be moving closer to reality. According to Yonhap News, a Samsung Foundry official recently shared on LinkedIn that Tesla’s AI5 chip has completed tape-out and is expected to enter production afterwards on Samsung’s 2nm process at the Texas facility.  As highlighted by Yonhap News, the disclosure marks the first time Samsung Foundry’s production timeline for the Tesla AI5 chip has been revealed in concrete terms.  Notably, Samsung is not expected to be the sole foundry partner for Tesla’s next-generation AI chips. Yonhap News adds that Tesla will split AI5 production between Samsung and TSMC, while Samsung will reportedly take full responsibility for AI6 and TSMC will handle AI6.5.  Meanwhile, Samsung is already mass-producing Tesla’s existing AI4 chip on its 7nm process at the Pyeongtaek foundry line, with the upgraded AI4 version also expected to be produced at the same campus, according to Yonhap News.  Taylor Fab Becomes Key to Samsung’s Foundry Recovery  The positive development, as noted by The Guru, also signals that preparations for Samsung’s Taylor fab are accelerating. Samsung began moving key front-end and back-end semiconductor equipment into the facility in late April and has deployed core engineers from its South Korean headquarters to secure initial yields, the report adds.  As previously reported by The Elec, Samsung has reaffirmed that the Texas fab will begin ramp production in 2027. Beyond standard 2nm, industry observers cited by the report also expect Samsung to introduce its SF2P+ (second-generation enhanced 2nm process) at the facility, further strengthening its advanced-node capabilities.  A successful 2nm ramp at Taylor could provide fresh momentum for Samsung Foundry’s long-awaited recovery after years of losses, ET News reports. With its mid- to long-term order backlog nearing KRW 50 trillion, Samsung is also expanding its foundry customer base beyond Tesla, securing partnerships with major tech companies including Meta and Anthropic, the report adds.
2026-07-14 15:24 阅读量:990
  • 一周热料
  • 紧缺物料秒杀
型号 品牌 询价
RB751G-40T2R ROHM Semiconductor
TL431ACLPR Texas Instruments
MC33074DR2G onsemi
CDZVT2R20B ROHM Semiconductor
BD71847AMWV-E2 ROHM Semiconductor
型号 品牌 抢购
IPZ40N04S5L4R8ATMA1 Infineon Technologies
BP3621 ROHM Semiconductor
TPS63050YFFR Texas Instruments
ESR03EZPJ151 ROHM Semiconductor
BU33JA2MNVX-CTL ROHM Semiconductor
STM32F429IGT6 STMicroelectronics
热门标签
ROHM
Aavid
Averlogic
开发板
SUSUMU
NXP
PCB
传感器
半导体
相关百科
关于我们
AMEYA360微信服务号 AMEYA360微信服务号
AMEYA360商城(www.ameya360.com)上线于2011年,现 有超过3500家优质供应商,收录600万种产品型号数据,100 多万种元器件库存可供选购,产品覆盖MCU+存储器+电源芯 片+IGBT+MOS管+运放+射频蓝牙+传感器+电阻电容电感+ 连接器等多个领域,平台主营业务涵盖电子元器件现货销售、 BOM配单及提供产品配套资料等,为广大客户提供一站式购 销服务。

请输入下方图片中的验证码:

验证码