DIGITIMES Insight: AI chip upgrades shift substrate supply bottleneck from upstream materials to equipment
<em>DIGITIMES Intelligence</em> has observed that as the performance of AI accelerators and server CPUs continues to advance, chips are increasing not only in compute core counts and I/O channels but also in integration complexity. Higher integration of high-bandwidth memory (HBM), high-speed SerDes, and chiplet designs makes signal transmission, power delivery, and component interconnects within the package far more complex, driving high-end IC substrates toward larger surface areas and higher layer counts.
Read original article ↗
Related Articles
ARTPRICE-NACHRICHTEN: EIN WELTBUCH WIRD ZUM SPIEGEL DER KÜNSTLICHEN INTELLIGENZ
OpenAI/Astra, Perplexity, DeepSeek, Google Gemini und xAI/Grok stehen vor derselben Herausforderung: Ein Meta-Lese-Exper
Intel’s bug bounty program on Intigriti now shows as suspended, and a new Intel disclosure program there pays no bountie
An Intel Celeron 1200 (Tualatin) was revived from the dead after a ripped-off pin was successfully replaced.
CXL adds memory capacity, but HBM retains its bandwidth edge
Growing memory requirements are increasing interest in Compute Express Link (CXL) as a way to add capacity beyond locall