Ming-Chi Kuo: TSMC's CoPoS Expected to Enter Mass Production in H2 2028
0xBroomberg
TF International's Ming-Chi Kuo says TSMC's next-gen advanced packaging technology CoPoS is set for mass production in the second half of 2028, with Nvidia's Feynman chip named as a likely first adopter — giving the industry a concrete timeline for solving the economics of packaging oversized AI chips.
What problem is CoPoS designed to solve?
AI chips keep getting bigger. Packaging sizes now exceed the limit of a single reticle — the maximum area one lithography exposure can cover — by more than 9.5×, crushing yield and inflating cost.
In plain terms = the chip is too large to "print" in one shot; CoPoS is built to make "stitching it together" cheap and reliable at scale.
Nvidia's next-gen AI chip Feynman has a production window that lines up with CoPoS timing. Kuo names it as a probable first adopter.
What does the three-layer architecture look like?
The core structure is a sandwich: glass as the central layer, with ABF build-up layers on both sides (ABF-GCP) — three layers in total.
Glass processing involves TGV (through-glass vias — tiny channels drilled through glass to carry signals), copper filling, and metallization, all technically demanding steps.
Interconnect is handled jointly by the RDL on the chip side (a thin-film rewiring layer that reroutes chip pins) and the TGV + ABF build-up on the substrate side — no single layer does it all.
Which three misconceptions did Kuo correct?
Misconception 1: Glass acts as an interposer in CoPoS. In fact, it does not perform unified interconnect — it is a structural core layer, not a signal "hub."
Misconception 2: Glass replaces ABF. In fact, glass and ABF coexist inside the substrate stack — the three-layer architecture itself is the proof.
Misconception 3: Chips sit directly on glass. In fact, chips attach to the ABF build-up surface on the outside of the glass core substrate, separated by a layer.
What does this mean for TSMC and the industry?
All three clarifications point to one judgment: CoPoS does not use glass to overthrow the existing packaging stack — it inserts glass as a structural backbone while preserving ABF process continuity.
This means → TSMC chose a "gradual embed" path, not a "tear it down" one. The ABF supply chain is not replaced — it is upgraded.
Kuo believes CoPoS will sustain TSMC's advanced-packaging lead through roughly 2032. This reflects a packaging moat that may stretch longer than the market currently prices in.
Content is for reference only, not financial advice.