先进封装推动代工厂与 OSAT 业务边界日渐重叠

Key Takeaways:

• Foundries have been doing backend work for advanced packages. OSATs are now taking that on.
• This particularly affects 2.5D and panel assembly with new equipment and materials.
• Hybrid bonding is mostly a foundry activity now but will likely move to OSATs as well.

Foundries and outsourced assembly and test (OSAT) houses have traditionally had distinct lanes, but the lines are blurring as OSATs acquire select foundry capabilities and foundries return some of the jobs they have taken from OSATs.

“We’re at this bridging point where the IDMs and the foundries are doing more of what OSATs can do, and the OSATs are also trying to get much closer to what the foundry does at the back end,” said Raghu Nayak, head of sales at Zeiss.

OSATs will never compete with foundries on advanced nodes and front-end wafer fabrication. But for backend activities, where dimensions aren’t as demanding, OSATs are acquiring capabilities that replicate what foundries can do. The biggest areas of overlap are 2.5D integration and panel assembly, with hybrid bonding likely to move there as well.

The change affects equipment and materials. “We are seeing the migration of lithography-related materials and processes from traditional wafer fabs into advanced packaging environments,” said Poupak Khodabandeh, chief business officer at Brewer Science. “As advanced packaging increasingly relies on fine-pitch redistribution layers (RDLs), OSATs and packaging organizations are using materials and process flows that historically would have been associated with wafer fabrication.”

“The industry has traditionally operated in silos across equipment, front-end fabrication, design, packaging, and test,” said Jack Berg, vice president of business development at Silvaco. “We need to connect those disciplines so teams can use shared, physics-grounded insight to make faster, better-informed design decisions, while also providing greater manufacturing flexibility — especially at the backend.”

It used to be “to each his own”
Foundries and OSATs have taken on complementary duties for many years. Foundries fabricate the silicon chip, while OSATs put the chip into a housing and test the final product. A vast array of EDA tools and fab equipment makes what the foundry does possible, while OSATs have worked more with tools that originated in the printed circuit board (PCB) world.

Two industry trends are challenging this status quo. The first is the development of advanced packaging. It’s no longer enough to put a chip on a substrate, wire it up, and pour molding compound over it. This doesn’t minimize the effort required to do those basics well, but they’re very different from what advanced packaging requires.

“Advanced packaging is becoming more lithography-intensive,” said Khodabandeh. “The industry’s push toward higher interconnect density for AI, HBM, chiplet architectures, and heterogeneous integration is driving the adoption of wafer-fab-style patterning processes within packaging facilities. RDL fabrication is a good example, where increasingly sophisticated lithography is required to create finer features.”

Intel Foundry works both sides of the aisle, with foundry and packaging capabilities. “When you do fine pitch bumps, the bump count goes up quite a bit compared to coarse pitch,” said Ravi Mahajan, fellow at Intel Foundry. “So you have to figure out things like embedding [bridge dies such as EMIB]. You have to figure out alignment and equipment precision.” For them, they can attempt these things on either side of the divide.

Foundries have been heavily involved in 2.5D packaging, but a second industry trend is making that more difficult — a lack of capacity. Foundries can’t make all the chips their clients want, so they’re prioritizing and allocating. That may mean turning away some business they would otherwise entertain, likely in backend and legacy processes, since foundries will focus on newer-node silicon processes they are uniquely capable of.

The OSATs are ready to pick up those processes. These are lower-resolution processes that foundries have historically employed but for which OSATs share capabilities. Legacy-node builds can spill over to assembly houses, especially since advanced packaging requires tooling consistent with those older technologies.

Although this overlap largely affects older processes, as constraints continue, even more aggressive technology, which foundries have often packaged (or done the first packaging steps) themselves, may be offloaded. “With further volume restrictions, both legacy-node and leading-edge products are shifting toward OSAT packaging,” said Thomas Uhrmann, director of business development at EV Group.

Three primary tasks
These market dynamics have driven three primary assembly-related processes from foundry to OSAT: 2.5D packaging, panel packaging, and eventually, hybrid bonding.

2.5D packaging in a manner resembling TSMC’s CoWoS (chip on wafer on substrate) involves mounting one die atop another. When the base is an interposer, this is 2.5D. When it’s another active die, it becomes 3D. Typical processes place the top die, already singulated, on a bottom die on a wafer that hasn’t been diced up yet. Once the top die is in place, the bottom wafer can be diced, with the stack-up headed for a package.

One fast-growing area that needs this technology is photonics. “Customers are coming to us asking if we can do electrical die to photonics die or wafer attach,” said Gerard John, senior director in Amkor’s Chiplets/FCBGA Business Unit. “We take this module, dice it out, and then attach it to a substrate. That’s a new trend we are seeing, and we can do that.”

It may, however, bring some tough new requirements. “This is a big challenge because now you have the optic side of the wafer or die,” added John, and this needs active alignment to attach fibers and test them.

Panels, meanwhile, have grown to the 500 mm range per side, but moving to a smaller size actually helped. “Several OSATs had early development lines for panel-level packaging, mainly targeting large panels with 500 mm dimensions,” explained Uhrmann. “TSMC’s decision to move to panel-level packaging with 310 × 310 mm², which is a format closer to classical 300 mm round process tools, has enabled traditional wafer-based equipment manufacturers to scale up for high-performance materials and resolution, rather than PCB equipment manufacturers to scale down.”

Meanwhile, some assembly houses are doing hybrid bonding. But for high volume, it largely remains a foundry task, owing to the challenges the process presents. As hybrid bonding loses its novelty and becomes standard, it represents another task that foundries could hand off to OSATs.

How are you paying for that?
Whereas foundries typically have the equipment for these processes, OSATs have to invest to match those capabilities. They’re not buying leading-edge stuff (you’re not going to find EUV in an OSAT), but they are equipping themselves with tools they haven’t traditionally used.

The return-on-investment (ROI) outlook for these capital purchases appears to be very good. The first thing one needs for a good ROI is a sizeable, growing market. Advanced packaging is now common enough (while supporting high-revenue markets such as AI in the data center) to promise a reasonable payoff. Fortune Business Insights says advanced packaging grew to over $45B in 2025 and expects it to reach $100B by 2034, with a CAGR (compound annual growth rate) just over 9%.

“If you look at ASE alone, the company has already passed its $3.5 billion target for leading-edge packaging this year, and it is expecting to double that in 2027,” noted Uhrmann.

Backend processes are becoming more important to foundries. Much of the performance advantage being sought comes from innovative packaging technologies rather than smaller transistors.

“This performance race is driven primarily by AI and high-performance computing, where margins are exceptionally high, partly due to constrained supply,” observed Uhrmann. “The margin is shifting from front-end chip manufacturing toward packaging and system integration, because a growing share of system performance now comes from advanced packaging, heterogeneous integration and 3D integration rather than from node scaling alone.”

So fabs would typically want to hang on to that business. Not only is there an ROI, but it is likely to come sooner than the ROI from advanced front-end tools.

OSATs have different economics. “Their business model is based on cost competitiveness and volume, with structurally lower margins than foundries, which makes ROI highly dependent on utilization and long-term customer commitments,” said Uhrmann.

Because foundries are oversubscribed, OSATs are also fully utilized, which favors ROI. “The sector is constrained by equipment lead times and cleanroom space rather than by orders,” said Uhrmann. “In this environment, investments in 2.5D, panel-level and test equipment pay back quickly.”

But if OSATs largely fulfill a spillover role when markets get tight, then foundries are in a position to win under most circumstances, whereas OSATs could be more exposed to economic cycles as the need for overflow comes and goes.

Waiting in the wings
Hybrid bonding is yet another technology that could spill over, but which hasn’t yet in high volume. OSATs are getting into the game, but foundries still dominate, and that could continue as long as advanced packages are managed vertically, with one company sourcing all the chiplets.

“Hybrid bonding is a foundry and IDM-dominated technology,” explained Uhrmann. “As long as the demand can be covered by the incumbent suppliers, there is no logic in investing by the OSATs.”

But when heterogeneous integration starts bringing different dies from different technologies and sources — think memory and photonics — it may make more sense for OSATs to participate in that business as well. That will require more equipment purchases.

“We expect hybrid bonding to follow the same path as CoWoS: adoption begins at the foundries, and once demand exceeds foundry capacity and processes mature, it will expand to OSATs as a second source,” said Uhrmann.

All in all, OSATs are buying equipment at a healthy clip, not just to take on new processes, but to simply to add capacity. Amkor, for instance, is investing in a new Arizona facility. And purchases are increasing. ASE spent $3.4B on equipment and raised its 2026 budget by 27%. And even after that, mid-year it raised its capital spend by another $1.5B, of which $600M is for equipment, according to Seeking Alpha.

Ultimately, having each side work on what the other side used to do can result in new business. “It creates more of an opportunity to understand these new architectures,” said Sanjiv Bhatt, senior director of global marketing and business development for the semiconductor business at Mitsubishi Chemical. “But it requires us to work with all stakeholders, including the tool makers, on how they are going to incorporate these new materials and tools.”

Related Articles
Advanced Packaging Limits Come Into Focus
Mechanical and process control limits are now shaping what can be manufactured at scale.

Panel-Level Packaging’s Second Wave Meets Engineering Reality
The cost case is strengthening, but glass, warpage, and bonding yield stand in the way.

Foundry Capacity Is Limiting Who Competes At Leading Edge Nodes
But the inability to utilize leading-edge process nodes has created opportunities for small and midsize chip developers in multi-die design, along with some sophisticated architectural design tradeoffs.