Rapidus Eyes Eight-Reticle AI Packaging by 2030
Rapidus’s plan to manufacture eight-reticle interposers on 600 mm glass panels is a high-risk attempt to make the largest AI-class chiplet packages more economical by 2030.
Tag
52 published articles
Semiconductors are materials whose electrical conductivity can be precisely controlled, making them essential to modern electronic devices and systems. Silicon is the most widely used, while compounds such as gallium nitride and silicon carbide serve specialised roles. Semiconductor chips provide computing, memory, sensing, power control and communications functions across consumer, industrial and infrastructure technologies.
Rapidus’s plan to manufacture eight-reticle interposers on 600 mm glass panels is a high-risk attempt to make the largest AI-class chiplet packages more economical by 2030.
Micron’s planned $10 billion Research Labs network in Boise signals a long-term bet that memory architecture, packaging and manufacturing science will be central to the next phase of AI computing.
Retail prices for some DDR5 memory kits have risen by almost 500% year on year, reflecting an exceptional squeeze in the wider DRAM and NAND markets rather than a uniform increase across every type of RAM.
A report linking Intel’s unannounced Razor Lake laptop processors to TSMC’s N2X process and a larger last-level cache points to a potentially more ambitious high-performance mobile strategy, but the specifications remain unconfirmed.
Sony has not publicly set the timing or price of its next-generation PlayStation, with elevated memory costs making the economics of a new console less predictable.
Rising memory and component costs are beginning to affect graphics-card and monitor pricing, with budget buyers likely to have the least room to absorb further increases in late 2026.
A reported rise in high-bandwidth-memory shipments to Malaysia is consistent with Intel preparing to expand advanced packaging in Penang, but the available trade signals do not establish that Project Pelican is the destination.
Reports that Nvidia has reserved TSMC’s A16 process for its 2028 Feynman generation underline how power delivery, packaging capacity and supply commitments are becoming central to AI-chip competition.
Intel has confirmed that its Diamond Rapids Xeon generation will arrive in 2027, but the reported 256-core flagship remains unannounced and should be treated as a roadmap claim rather than a product specification.
Lam Research’s plan to invest more than $3 billion in its global laboratory network signals a push to place development capacity closer to the customers and technologies shaping the next semiconductor cycle.
TSMC’s reported yield above 98% for its 5.5-reticle CoWoS-L packaging signals that exceptionally large AI packages are becoming manufacturable at scale, although the company’s longer-term roadmap still carries substantial execution risk.
A report that CXMT has pushed DDR5 manufacturing yield above 90% would mark a notable operational milestone, but it does not by itself prove parity with the established DRAM leaders.