Recent #DRAM news in the semiconductor industry

3 months ago

➀ The impact of generative AI will lead to a sharp increase in DRAM wafer demand.

➁ DeepSeek, a Chinese AI startup, has released a large-scale language model (LLM) called DeepSeek-R1, which is comparable to OpenAI's GPT-4.

➂ DeepSeek's R1 was developed in just two months at a cost of 5.6 million US dollars, a fraction of what other companies spend.

➃ The rise of smartphones is an example of disruptive innovation, replacing PCs in the market.

➄ The growth of AI servers, especially those with AI semiconductors, is expected to drive the demand for advanced node logic and DRAM wafers.

➅ The demand for DRAM wafers is expected to soar as generative AI becomes more prevalent.

➆ The global wafer demand is projected to increase significantly due to various factors, including government subsidies and the rise of AI.

AIDRAMGenerative AI
4 months ago

➀ The SIA reports that the semiconductor sales in 2024 reached $627 billion, a 19.1% increase from the $526.8 billion in 2023.

➁ The market topped $600 billion for the first time last year and is expected to see double-digit revenue growth this year.

➂ Logic led the market in 2024 with $212.6 billion in sales, followed by memories, which grew 78.9% year-on-year to reach $165.1 billion.

DRAMSIAmarket growthmemorysalessemiconductor
5 months ago
➀ Kioxia Corporation and Nanya Technology have developed a low-power DRAM technology called OCTRAM, which uses oxide-semiconductor transistors for ultra-low leakage and high efficiency. The technology aims to significantly reduce power consumption in AI, post-5G communication, and IoT devices. ➁ The OCTRAM features a cylindrical-shaped InGaZnO vertical transistor as its cell transistor, offering enhanced memory density. ➂ The technology achieves a high 'on' current exceeding 15μA per cell and an 'off' current as low as 1aA per cell, reducing energy leakage to negligible levels.
DRAMenergy efficiency
5 months ago
➀ Compute Express Link (CXL) technology is expected to move from a niche to mainstream use in 2025; ➁ CXL's support for memory expansion is a significant driver, with various server and memory solutions now available; ➂ CXL 2.0 and future generations of PCIe/CXL will enable more advanced use cases like switching and dynamic memory allocation.
2nm3D IC3nmAIAI ChipAI PCAMDArmAsusCXLChipletCoolingDDR4DRAMDellEDAEMIBEUVGDDRGPUGaNHBMHPCInfineonIntelLaptopLinuxMicrochipNPUNVIDIAPCIePrivacyRaspberry PiSSDSoftwareSwitchTIXeonautomotivecpucybersecurityddr5iosmemorymonitorsemiconductorserver
6 months ago
➀ China experienced a decline in chip investment, with new investment deals dropping by 35.9% year-on-year in the first 11 months of 2024; ➁ The total funding decreased by 32.4%; ➂ ChangXin Memory and Unisoc were the largest and second-largest investors, respectively. JW Insights notes that the US semiconductor market surpassed China's in Q3, driven by massive AI infrastructure investments in the US.
AIChinaDRAMUNISOCinvestmentsemiconductor
6 months ago
➀ SEMI forecasts semiconductor manufacturing equipment sales to grow 6.5% year-on-year to $113 billion in 2024, with projections of $121 billion for 2025 and $139 billion for 2026. ➁ The wafer fab equipment (WFE) segment is expected to grow 5.4% to $101 billion in 2024, driven by DRAM, HBM, and China. ➂ Back-end equipment sales are projected to surge, with test equipment sales increasing 14.7% in 2025 and 18.6% in 2026.
ChinaDRAMEquipment SalesGrowthHBMKoreaNANDSEMITaiwanfoundryinvestmentmemorysemiconductor
6 months ago
➀ Hynix has started mass production of the world's first 321-layer NAND with 1Tb capacity; ➁ The 321-layer device offers a 12% improvement in write time and a 13% improvement in read time compared to the 238-layer device; ➂ Hynix has adopted the '3 plugs' process technology and a low-stress material for stacking more than 300 layers.
3D ICDRAMNANDmemory
7 months ago
➀ NVIDIA has announced a new Grace Blackwell platform with the NVIDIA GB200 NVL4, which combines two Arm-based Grace CPUs and four Blackwell GPUs. The platform is designed for lower power deployments with a power consumption of just over 6kW. The NVL4 module is targeted to be a large node with up to 1.3TB of combined coherent memory. ➁ The GB200 NVL4 is set to ship in the second half of 2025. ➂ The NVL4 is designed as an alternative to the higher power and higher GPU count platforms offered by NVIDIA, targeting specific deployment needs.
AIAI PCArmChipletDRAMEDAEMIBEUVGPUHBMHPCNPUNVIDIASSDcpu
7 months ago
➀ Hynix plans to start sampling HBM3e 16hi memory with 48 GB capacity per cube in H1 2025; ➁ HBM3e 16hi offers a high-capacity alternative before the production of HBM4 16hi; ➂ HBM3e 16hi can deliver a maximum capacity of 384 GB per system, surpassing NVIDIA’s Rubin at 288 GB; ➃ Transitioning from HBM3e to HBM4 will double the I/O count and increase die size, but the per-die capacity remains at 24 GB; ➄ HBM3e 16hi will adopt the Advanced MR-MUF stacking process for higher stack counts and computational bandwidth.
HynixChipletDRAMHBMMicrochipTSMC
7 months ago

➀ TechInsights has obtained and analyzed Samsung Electronics' D1b DRAM, revealing its cell size and bit density.

➁ Samsung's D1b DRAM has a cell size significantly smaller than SK Hynix and Micron's D1b versions, with a wordline and bitline pitch of 32.6nm and 37.6nm respectively.

➂ The cell design rule or feature size for the 7.8F^2 DRAM cell is 12.54nm, indicating that Samsung's D1b DRAM is in the 12nm mid-range class.

DRAMSamsung Electronicstechnology
7 months ago
➀ A new 3D imaging technique reveals key details of the magnetic skyrmion, a nanoscale object with potential for new microelectronic devices with high data storage capacity and low power consumption. ➁ Researchers at Lawrence Berkeley National Laboratory have created 3D X-ray images to characterize the spin orientations within skyrmions, which could lead to more efficient quantum computing and data storage devices. ➂ The study demonstrates the importance of understanding the 3D spin texture of skyrmions for the development of advanced spintronic devices.
3D ICDRAMData StorageNVIDIAQuantum Computing
7 months ago
➀ Researchers from Queen Mary University of London have developed new nanocomposite films from starch, offering a sustainable alternative to petroleum-based materials; ➁ The films are made from abundant natural polymer starch and highly conductive MXene, with potential applications in electronics and sensing; ➂ The nanocomposites can be tailored for various uses, including wearable technology and healthcare, and are biodegradable.
2nm3D IC3nmAIAI ChipAI PCAMDArmAsusChipletCoolingDRAMDellEDAEMIBEUVGDDRGPUGaNHBMHPCInfineonLaptopLinuxMicrochipNPUNVIDIAPCIePrivacyRaspberry PiSSDSoftwareSwitchTIautomotivecpucybersecuritygamingiosmemorymonitorsemiconductor