EUV The Focal Point

EUV The Focal Point - Team

EUV The Focal Point is your Extreme Ultraviolet (EUV) lithography podcast. Industry Briefings: Cover leading-edge nodes, DRAM, HBM and strategy moves from ASML & co. and from the end customers Apple & co. Focus Deep Dives: Unpack physics, plasma, optics and how EUV scanners really work. Hosted by EUV experts Samantha and Jack, built entirely with AI (a technology that itself relies on EUV-made chips ;-) using company newsrooms, Wikipedia and news sites. AI can make mistakes: always verify the information independently before using it as a basis for business decisions.

  1. 2日前

    [060] Industry briefing - EUV The Focal Point

    This post was created using AI. Please check the information if you want to use it as a basis for decision-making. This week is light on new EUV scanner announcements but rich in semiconductor economics. NVIDIA’s demand, SK hynix’s U.S. HBM investment, Hot Chips memory disclosures, and Intel’s 18A-P versus possible TSMC sourcing choices show how utilization, wafer intensity, packaging, and capital allocation are becoming the decisive variables around advanced lithography. Key takeaways - NVIDIA reported Q2 FY2027 revenue of $96.2 billion and Data Center revenue of $89.0 billion, with Q3 revenue guided to $108.0 billion ±2%. - SK hynix broke ground on a more-than-$4-billion HBM packaging facility in Indiana; cleanroom opening is targeted for October 2028 and mass production for H2 2029. - The Indiana site will package and test advanced wafers made in South Korea, regionalizing the back end without duplicating the full EUV-intensive front-end flow in the U.S. - Micron said at Hot Chips that HBM requires roughly three times the wafer area of DDR5 for the same memory capacity, and that the ratio is not improving. - SK hynix indicated at Hot Chips that hybrid bonding is unlikely for HBM4E and may arrive with HBM5 at the earliest, extending the economic life of its current packaging approach. - Intel said Diamond Rapids uses Intel 18A-P, giving its leading-edge manufacturing platform additional internal volume. - TrendForce reported that Intel may consider TSMC N2X for some Razor Lake compute dies in 2027 depending on 18A-P performance; this remains unconfirmed. - Intel’s recent $20 billion equity offering and its Q2 Intel Foundry operating loss of about $2.1 billion underline why utilization and cost per good die matter as much as process milestones. - A useful follow-up to last week’s outlook arrived from SK hynix: a concrete capacity date is now available for Indiana, although not yet for the front-end M15X question. Glossary Extreme Ultraviolet (EUV) — Leading-edge lithography using 13.5-nanometer light for advanced logic and memory patterning. High Numerical Aperture (High-NA) EUV — ASML’s higher-resolution EUV platform used selectively by Intel in production learning. High Bandwidth Memory (HBM) — Stacked memory optimized for very high bandwidth in AI and high-performance computing systems. Dynamic Random-Access Memory (DRAM) — Volatile semiconductor memory; HBM is built from specialized DRAM dies. 18A-P — Intel’s performance-enhanced version of the Intel 18A process family. N2X — TSMC’s performance-oriented member of its 2-nanometer process family, targeted for 2027. Process Design Kit (PDK) — Foundry-provided design rules, models, and collateral used to build chips for a process. Hybrid bonding — Fine-pitch direct bonding approach intended to improve density and interconnect performance in advanced packages. Cost per good die — Manufacturing cost divided by the number of functional dies produced, combining yield, cycle time, and capital utilization.

  2. 6日前

    [059] Industry briefing - EUV The Focal Point

    This post was created using AI. Please check the information if you want to use it as a basis for decision-making. This week is less about a new EUV scanner milestone and more about the economics that determine when lithography capacity becomes valuable. Intel connects 18A-P to Diamond Rapids, Samsung and SMIC show renewed foundry pricing power, and Chinese memory investment continues to scale. Key takeaways: - Intel presented Diamond Rapids at Hot Chips 2026 and tied the product to Intel 18A-P and Foveros Direct 3D packaging. - The Hot Chips update partially answers an earlier podcast outlook asking whether Intel 18A-P risk production would become concrete product activity. - TrendForce reported that Intel may use TSMC N2X for some Razor Lake compute dies in 2027, but the sourcing plan remains unconfirmed. - Reuters reported Samsung raised prices for some new foundry orders by up to 15%, with its SF4 line at Pyeongtaek running at full capacity. - Samsung’s price increases are an indirect EUV signal because high utilization improves the economics of installed advanced-node lithography capacity. - SMIC reported second-quarter revenue above $3 billion, wafer shipments up 14% sequentially, average selling prices up 5.7%, and utilization at 93.7%. - YMTC’s parent plans a roughly RMB 33 billion Shanghai IPO to fund production upgrades and advanced-storage R&D. - The week’s central theme is manufacturing optionality: Intel can push 18A-P while preserving TSMC options, and customers are paying more to secure scarce capacity. - No major new official ASML EUV shipment or High-NA production milestone was identified for this week. Glossary: Extreme Ultraviolet (EUV) — 13.5-nanometer lithography used for the most demanding layers in advanced logic and memory manufacturing. Intel 18A-P — Intel’s performance-enhanced extension of the Intel 18A process family. N2X — TSMC’s high-performance extension of its 2-nanometer process family, targeted at performance-sensitive applications. Foundry — A semiconductor manufacturer that fabricates chips designed by external customers. Utilization — The share of available manufacturing capacity actively being used. Average selling price (ASP) — Average revenue received per unit sold, such as a processed wafer. Foveros Direct 3D — Intel’s advanced 3D packaging technology for direct die-to-die integration. High Bandwidth Memory (HBM) — Stacked DRAM designed for very high bandwidth in AI and high-performance-computing systems. Deep Ultraviolet (DUV) — Older lithography technology still used across many semiconductor layers and in advanced production where EUV is unavailable or uneconomic.

  3. 8月18日

    [058] Industry briefing - EUV The Focal Point

    This post was created using AI. Please check the information if you want to use it as a basis for decision-making. Samsung’s revised foundry roadmap puts High-NA EUV around the 1-nanometer-class generation in 2030 while extending the life of its Low-NA EUV platform. At the same time, record equipment spending, new equity financing, and packaging investments show that the economic race is increasingly about sequencing capital across the whole manufacturing system rather than adopting one tool first. Key takeaways - Samsung reportedly moved SF1.4 from a 2027 target to 2029 and plans High-NA EUV insertion with its 1-nanometer-class process around 2030. - Samsung’s roadmap implies continued Low-NA EUV use for SF1.4+ while the company concentrates near-term resources on the SF2 family. - Applied Materials reported record fiscal Q3 2026 revenue of $9.12 billion, up 25% year over year, and guided fiscal Q4 revenue to about $10.25 billion, plus or minus $500 million. - Applied Materials raised its calendar-2026 advanced-packaging growth expectation to more than 70%, reinforcing the broader process-equipment signal highlighted in the August 4 outlook. - Intel upsized its common-stock offering from $15 billion to $20 billion and expects approximately $19.7 billion of net proceeds; the company says proceeds may include capital expenditures and working capital. - TSMC approved about $29.44 billion of capital appropriations covering advanced technology, advanced packaging, specialty capacity, fabs, and facility systems. - TSMC and Sony signed a definitive agreement for a next-generation image-sensor joint venture in Kumamoto, with volume production targeted for 2029. - SPIL reportedly broke ground on a nearly NT$100 billion Douliu plant that will add CoWoS capacity, with first-phase operations planned for 2028. - Nanya Technology says it is evaluating potential expansion sites but has not made a decision on reported new fabs in Yunlin and Pingtung. - U.S. policy pressure is increasingly reaching foundries themselves, with lawmakers pushing stricter scrutiny of advanced-chip orders that could indirectly serve sanctioned Chinese firms. Glossary Extreme Ultraviolet (EUV) — Lithography using 13.5-nanometer light for critical patterning at advanced semiconductor nodes. High Numerical Aperture (High-NA) EUV — The newer EUV platform intended for tighter critical-layer patterning and selective future-node insertion. Low Numerical Aperture (Low-NA) EUV — The established 0.33-NA EUV platform used broadly in current advanced logic and memory production. SF2 — Samsung Foundry’s 2-nanometer-class process family. SF1.4 — Samsung Foundry’s 1.4-nanometer-class process generation, now reported for 2029. Intel 14A — Intel Foundry’s next-generation process platform planned after Intel 18A. CoWoS — Chip on Wafer on Substrate, an advanced packaging approach used for large AI accelerators and HBM integration. HBM — High Bandwidth Memory, stacked memory used with advanced AI and high-performance computing processors. DRAM — Dynamic Random-Access Memory, the volatile memory technology underlying mainstream memory products and HBM.

  4. 8月14日

    [057] Deep Dive Topic - New Ways to Store Bits

    This post was created using AI. Please check the information if you want to use it as a basis for decision-making. EPISODE TEASER Why is the memory industry exploring RRAM, MRAM, and new DRAM concepts when SRAM, DRAM, and NAND already work so well? This episode explains the physical storage mechanisms, the engineering trade-offs, and the most realistic application paths. It also clarifies why “DRAM+” is an umbrella term rather than one standardized memory technology. KEY TAKEAWAYS - Every memory tier trades speed, density, energy, endurance, retention, cost, and integration complexity. - Conventional DRAM stores charge in a capacitor and must refresh it because the charge leaks away. - “DRAM+” can describe enhanced DRAM approaches such as HBM, gain-cell or capacitorless DRAM, three-dimensional integration, embedded DRAM, and processing near the memory. - RRAM stores information as resistance by moving ions or defects in a switching layer; variability is one of its central challenges. - STT-MRAM writes through a magnetic tunnel junction, while SOT-MRAM separates the read and write paths. - Nonvolatile memories reduce standby and refresh energy, but active write energy and peripheral circuits still matter. - RRAM and MRAM can support compute-in-memory, but array parasitics, data conversion, write control, and device variation limit idealized gains. - Advanced DRAM is the strongest near-term route for main memory; MRAM and RRAM are more likely to expand first in embedded and specialized roles. - The likely future is a more heterogeneous memory hierarchy, not one universal replacement. GLOSSARY DRAM — Dynamic random-access memory. Volatile memory that usually stores each bit as charge in a capacitor and requires periodic refresh. DRAM+ — A nonstandard umbrella term used here for enhanced DRAM cells, architectures, packaging, stacking, and processing-near-memory concepts. SRAM — Static random-access memory. Fast volatile memory commonly used for processor caches; a typical cell uses several transistors. NAND flash — Dense nonvolatile storage that retains data without power and is programmed and erased in larger groups than RAM. RRAM or ReRAM — Resistive random-access memory. Nonvolatile memory that stores data in different resistance states. MRAM — Magnetoresistive random-access memory. Nonvolatile memory that stores data in magnetic orientation and reads it through a resistance difference. STT-MRAM — Spin-transfer torque MRAM. A form of MRAM in which write current passes through the magnetic tunnel junction. SOT-MRAM — Spin-orbit torque MRAM. A form of MRAM with a separate lateral write path that acts on the magnetic free layer. Crossbar array — A grid of crossing conductors with a memory element at each intersection; useful for dense storage and parallel analog operations. Compute-in-memory — An architecture that performs selected operations in or very near memory arrays to reduce data movement.

    [057] Deep Dive Topic - New Ways to Store Bits
  5. 8月10日

    [056] Industry briefing - EUV The Focal Point

    This post was created using AI. Please check the information if you want to use it as a basis for decision-making. Nanya Technology has turned EUV memory expansion into a funded fab schedule, while TeraFab has gained a confirmed Texas site and an initial investment figure. AMD and TSMC provide fresh demand-side evidence, and reported Chinese-tool evaluations by Samsung and SK hynix show how export controls can reshape equipment qualification beyond lithography. Key takeaways - Nanya approved up to NT$346.6 billion of Fab 5A capital expenditure for 2026-2029, including EUV equipment. - Fab 5A is designed for 45,000 wafer starts per month, with first-phase capacity targeted at 35,900 WSPM. - Nanya plans wafer input in the second half of 2027, with 30,000 WSPM targeted for 2028 and 35,900 WSPM for 2029. - SpaceX and Tesla identified Grimes County, Texas, as the TeraFab site and announced an initial investment of $16.8 billion. - TeraFab still has no disclosed EUV scanner count, mask strategy, detailed process flow, or public production qualification schedule. - AMD reported Q2 2026 revenue of $11.5 billion; Data Center revenue was $6.7 billion, up 107% year over year. - TSMC reported July 2026 revenue of NT$467.58 billion, up 44.7% year over year. - Reuters reported evaluations of AMEC etch tools by Samsung and SK hynix, while both companies denied testing AMEC equipment for use in China and no broad deployment decision was reported. Glossary Extreme Ultraviolet (EUV) — Advanced lithography using 13.5-nanometer light for critical semiconductor patterning. Dynamic Random-Access Memory (DRAM) — Volatile memory used widely in servers, PCs, mobile devices, and AI systems. Wafer starts per month (WSPM) — A fab-capacity measure describing how many wafers enter production each month. 1C / 1D / 1E DRAM — Successive 10-nanometer-class DRAM process generations in Nanya’s roadmap. Capital expenditure (capex) — Spending on long-lived manufacturing assets such as fabs and production equipment. Etch — A semiconductor process that selectively removes material to transfer patterns into underlying layers. Mask strategy — The plan for photomasks and patterning steps used to manufacture a chip design. Process flow — The ordered sequence of fabrication steps used to build semiconductor devices.

  6. 8月4日

    [054] Industry briefing - EUV The Focal Point

    This post was created using AI. Please check the information if you want to use it as a basis for decision-making. This week’s episode looks beyond the EUV scanner to the design and manufacturing tools that turn advanced patterns into profitable silicon. SK hynix’s HBM4 ramp, strong results from Lam Research and KLA, and certified Synopsys flows for Intel 14A all point to the same conclusion: process integration and yield control are becoming as strategic as exposure capacity. Key takeaways - SK hynix reported second-quarter revenue of 79.3 trillion won and operating profit of 60.5 trillion won. - SK hynix began mass shipments of HBM4 in the second quarter and plans to ramp production in the second half of 2026. - SK hynix has finalized long-term agreements with around 10 customers and is accelerating the M15X mass-production schedule. - Lam Research reported record quarterly revenue of 6.72 billion dollars and guided the September quarter to about 8.1 billion dollars, plus or minus 400 million dollars. - KLA reported quarterly revenue of 3.66 billion dollars and guided the September quarter to 4.0 billion dollars, plus or minus 200 million dollars. - Synopsys certified AI-powered EDA and multiphysics flows for Intel 14A, including support across silicon, packaging, thermal, power-integrity, and electromagnetic analysis. - High-NA EUV economics depend on the full manufacturing flow, including masks, resists, etch transfer, metrology, design rules, and packaging. - Reuters reported that China has begun producing domestically developed immersion DUV lithography systems, although high-volume reliability and performance remain unproven. - No new official EUV scanner shipment count was identified this week; the strongest signals came from the surrounding equipment and design ecosystem. Glossary Extreme Ultraviolet (EUV) — Lithography using 13.5-nanometer light to pattern the most advanced semiconductor layers. High Numerical Aperture (High-NA) EUV — A higher-resolution EUV platform intended for selected critical layers at future advanced nodes. Deep Ultraviolet (DUV) — Lithography using longer-wavelength ultraviolet light; immersion DUV remains important across mature and advanced process flows. High Bandwidth Memory (HBM) — Vertically stacked memory designed to provide very high data bandwidth for AI and high-performance computing. Dynamic Random-Access Memory (DRAM) — Volatile working memory used in servers, computers, mobile devices, and HBM stacks. Electronic Design Automation (EDA) — Software used to design, verify, optimize, and prepare chips for manufacturing. Design Technology Co-Optimization (DTCO) — Joint optimization of chip design rules and process technology. Process control — Inspection, metrology, analytics, and feedback used to maintain yield and manufacturing stability. Multiphysics analysis — Combined modeling of electrical, thermal, mechanical, and electromagnetic effects in a chip and package.

  7. 7月30日

    [053] Deep Dive Topic - Heterogeneous Integration: The Evolution of System Architecture

    This post was created using AI. Please check the information if you want to use it as a basis for decision-making. Heterogeneous integration is changing how advanced semiconductor systems gain performance. This episode explains why shrinking transistors is no longer enough on its own, how chiplets and advanced packaging connect specialized logic, memory, input-output, photonics, and other functions, and why artificial intelligence is accelerating the transition. It also examines the technical, economic, and supply-chain trade-offs behind the trend. KEY TAKEAWAYS - Transistor scaling remains essential, but modern performance gains increasingly come from system-level integration. - Heterogeneous integration combines dies made for different functions, process nodes, and sometimes different materials. - Chiplets can improve yield, reuse, product flexibility, and development speed. - Die-to-die links introduce latency, power, protocol, packaging, and test overhead. - Interposers, silicon bridges, through-silicon vias, and hybrid bonding enable dense communication inside advanced packages. - High-bandwidth memory is a central driver because artificial intelligence and high-performance computing are often limited by data movement. - Thermal management, mechanical stress, package warpage, and final assembly yield are major constraints. - UCIe aims to create a more interoperable chiplet ecosystem, but physical and manufacturing compatibility remain difficult. - Advanced packaging increases the strategic importance of foundries, assembly and test providers, materials, equipment, metrology, and design software. - Lithography remains critical both for leading-edge chiplets and for fine-patterned package structures. GLOSSARY Chiplet: A die designed to operate as one building block inside a larger multi-die system. Heterogeneous integration: Combining different dies, process technologies, functions, or materials in one package or system. Interposer: An intermediate wiring layer that provides dense connections between dies and the package substrate. High-bandwidth memory: Vertically stacked dynamic random-access memory designed for very wide, energy-efficient links to nearby processors. Through-silicon via: A vertical electrical connection passing through a silicon die or wafer. Hybrid bonding: Fine-pitch die or wafer bonding that joins copper connections and surrounding dielectric surfaces without conventional solder bumps. Two-and-a-half-dimensional integration: Side-by-side dies connected through a dense interposer or bridge. Three-dimensional integration: Active dies stacked vertically and connected with dense vertical links. Known good die: A die tested before assembly to reduce the risk of placing a defective component into an expensive package. UCIe: Universal Chiplet Interconnect Express, an industry specification for die-to-die communication.

    [053] Deep Dive Topic - Heterogeneous Integration: The Evolution of System Architecture

番組について

EUV The Focal Point is your Extreme Ultraviolet (EUV) lithography podcast. Industry Briefings: Cover leading-edge nodes, DRAM, HBM and strategy moves from ASML & co. and from the end customers Apple & co. Focus Deep Dives: Unpack physics, plasma, optics and how EUV scanners really work. Hosted by EUV experts Samantha and Jack, built entirely with AI (a technology that itself relies on EUV-made chips ;-) using company newsrooms, Wikipedia and news sites. AI can make mistakes: always verify the information independently before using it as a basis for business decisions.