TSMC prepares CoPoS shift with 750 × 620 mm panels

TSMC plans a move from CoWoS to panelized CoPoS packaging to accommodate larger reticles and more HBM for Artificial Intelligence accelerators, with pilots in 2026 and mass production by 2029.

TSMC is reportedly preparing a major shift in its advanced packaging strategy, moving from chip-on-wafer-on-substrate CoWoS to a panelized approach called CoPoS. pilots could start as soon as 2026, with the company targeting mass production between late 2028 and the first half of 2029. the transition involves replacing roughly 300 mm circular silicon interposers with large glass or sapphire panels in both square and rectangular formats.

early CoPoS formats are said to be around 310 × 310 mm, with later options extending to about 515 × 510 mm and an even larger 750 × 620 mm panel. the larger, near-square panel is intended to reduce wasted edge area and to allow bigger reticles and masks, a benefit aimed directly at packaging for Artificial Intelligence accelerators. that space also makes it easier to place multiple dies and multiple stacks of high bandwidth memory on a single package, improving integration density and potentially boosting performance for workloads that depend on tight die-to-die connectivity.

technically, CoPoS merges concepts from CoWoS with fan-out panel-level packaging, building redistribution layers on glass or sapphire rather than on round silicon wafers to improve area utilization. TSMC is lining up development and production capacity at its AP7 chiayi campus, with phases 4 and 5 expected to host panel work, and plans to add related capacity in future arizona facilities. equipment and tooling orders are already being placed; suppliers reported to be involved include kla and applied materials, alongside taiwanese firms such as innolux. the panel route promises higher area utilization, lower cost per packaged accelerator and fewer warpage issues as package sizes increase, while TSMC will continue to offer its existing CoWoS lineup during the transition.

77

Impact Score

Analog computing from waste heat

MIT researchers developed an analog computing approach that uses waste heat in electronic devices to process data without electricity. The technique performs matrix vector multiplication with strong accuracy and could also help monitor heat in chips without extra energy use.

How Artificial Intelligence is reshaping financial services oversight

Financial services regulators are largely treating Artificial Intelligence as another technology governed by existing rules rather than building new securities-specific frameworks. History suggests that clearer expectations will emerge through examinations, enforcement, and supervisory guidance.

Nvidia faces gamer backlash over Artificial Intelligence shift

Nvidia is facing growing frustration from gamers as memory supply is steered toward data center chips and DLSS 5 becomes more central to game performance. The dispute highlights how far the company’s priorities have shifted toward enterprise Artificial Intelligence.

Executives see limited Artificial Intelligence productivity gains so far

Corporate enthusiasm around Artificial Intelligence has yet to translate into broad gains in employment or productivity, reviving comparisons to the long lag between early computing breakthroughs and measurable economic impact. Recent surveys and studies show mixed results, with strong expectations for future benefits but little consensus on present gains.

Nvidia skips a new GeForce generation as Artificial Intelligence chips dominate

Nvidia is set to go a year without a new GeForce GPU generation for the first time since the 1990s as memory shortages and higher margins in Artificial Intelligence hardware reshape the market. AMD and Intel are also struggling to capitalize because the same supply constraints are hitting gaming products across the industry.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.