Semiwiki on LinkedIn Semiwiki on Facebook Semiwiki on Twitter Register / Log in 
Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/search/354859/?c[users]=Fred+Chen&o=date
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2030970
            [XFI] => 1060170
        )

    [wordpress] => /var/www/html
)

Search results

Search the forum Search Semiwiki WordPress
  1. F

    Inside NVIDIA Groq 3 LPX: The Low-Latency Inference Accelerator for the NVIDIA Vera Rubin Platform

    784 billion transistors over 8 chips makes more sense.
  2. F

    Inside NVIDIA Groq 3 LPX: The Low-Latency Inference Accelerator for the NVIDIA Vera Rubin Platform

    4GB is a lot of SRAM (is it supposed to be 4 Gb)? From: https://www.cna.com.tw/news/ait/202603170011.aspx
  3. F

    Samsung and SK hynix are scaling back their DRAM expansion plans: Is there concern about potential overcapacity?

    16. March 2026 06:01 Samir Bashir The memory industry is currently raking in huge profits thanks to the AI boom, but behind the scenes, nervousness is growing. While Samsung and SK hynix are benefiting massively from the current demand for DRAM and HBM, that is precisely the problem: anyone who...
  4. F

    Samsung Foundry nabs Nvidia

    Groq already lost confidence in themselves last year: https://www.investing.com/news/company-news/groq-slashes-2025-revenue-projections-to-500-million--the-information-93CH-4158309 Investing.com -- AI chipmaker Groq has significantly reduced its 2025 revenue projections from more than $2...
  5. F

    Samsung Foundry nabs Nvidia

    Must be a big chip, i. e., hard to yield.
  6. F

    China’s No. 2 chipmaker Hua Hong readies 7 nm production as Beijing ramps up self-sufficiency drive

    So this is a fully domestically supplied fab?
  7. F

    BEOL M0 32nm with EUV Low NA0.33 Single Exposure, Nvidia cuLitho helps but not easy for HVM

    This cuLitho P32 print looks risky:
  8. F

    BEOL M0 32nm with EUV Low NA0.33 Single Exposure, Nvidia cuLitho helps but not easy for HVM

    I don't think these are realistic patterns, with such large gaps between line ends. Are they leaving room for SRAFs? Aside from the long lines which look like they didn't clear, these could be stochastic occurrences. In that case, the effectiveness itself will be stochastic, since defect...
  9. F

    Build China’s ASML

    Well, this may not age so well, as there is plenty of blur that does not come from ASML's machine, namely, the resist (from the electrons, and in a lot of cases, acids).
  10. F

    Build China’s ASML

    Also, the components of ASML's EUV systems are individually expensive and many would only be used just for those kinds of systems. Even something that may sound trivial like precision maglev vacuum stages. Hydrogen-proofing makes it sound less trivial.
  11. F

    Elon Musk kills résumés for chip hires and wants these 3 bullet points

    Well, "a résumé is still required to apply for most other jobs at Tesla in the U.S.—with some positions even calling for an “evidence of excellence” statement." https://fortune.com/2026/02/20/how-does-elon-musk-hire-dojo3-chip-team-no-resumes-3-bullet-points/
  12. F

    Build China’s ASML

    The article briefly mentioned that Shanghai Microelectronics Equipment (Group) Co., Ltd.'s 28nm lithography machine (ArF, Immersion) "has entered the process testing stage." So how well they will do there is still a question.
  13. F

    Build China’s ASML

    Originally, I had posted their development goals directly from the auto-translation, but the translation contained misleading errors. Focusing on the process development, their suggested goals for the 15th Five-Year Plan (which ends 2030) are: 1. A controllable and self-reliant 28nm ecosystem...
  14. F

    Intel’s EMIB Challenges TSMC’s CoWoS as America’s Answer to the AI Packaging Bottleneck

    Muhammad Zuhair •Mar 6, 2026 at 01:15pm EST Intel's packaging services are being considered a viable alternative to TSMC's CoWoS, as supply constraints are forcing US fabless customers to seek other options. Intel's EMIB Packaging Orders Could Reach 'Billions in Revenue' Moving Into H2 2026; a...
  15. F

    Build China’s ASML

    For them, Moore's Law ended at 28nm: This news is interesting: "Shanghai Microelectronics Equipment (Group) Co., Ltd.'s 28nm lithography machine (ArF, Immersion) has entered the process testing stage." In contrast, the EUV status report seems to highlight a national-level challenge: "Reports...
  16. F

    Build China’s ASML

    NIL is said to target < 10 nm features. The biggest issue is the mask or template, it has tighter defect, placement, roughness, and CD specs than scanner masks, and it also has limited lifetime. The authors did not understand that the SADP is for the template not the wafer. Throughput is also...
  17. F

    Build China’s ASML

    China has been doing EUV research and following ongoing EUV developments, although I don't know if they have been attending the SPIE Advanced Lithography Conferences in person, or if they're banned. But they should know about stochastics. And that TSMC uses multipatterning. (They should...
  18. F

    Build China’s ASML

  19. F

    Build China’s ASML

    Well, taking SMEE as a reference, they're not learning or deploying fast enough.
  20. F

    Intel has manufacturing capacity issues. They may take years to fix.

    The company redirected some of its chips production to meet surging server demand, but analysts say it’s missing out on emerging AI-driven trends. Published March 3, 2026 Nathan Owens Technology companies are clamoring for processors, especially hyperscalers looking to power their data...
Back
Top