AI News

The boundary between foundries and OSATs is blurring: 2.5D and panel-level packaging are the first to spill over

Jiwei.comUpdated Nov 16 19:48

AI summary

The boundary between foundries and OSATs is blurring due to advanced packaging and tight capacity, with 2.5D integration, panel-level packaging, and future hybrid bonding becoming the areas of greatest overlap between the two sides. The advanced packaging market already exceeded US$45 billion in 2025 and is projected to reach US$100 billion by 2034; ASE has this year surpassed its leading-edge packaging target of US$3.5 billion and will raise its 2026 equipment budget by 27%. OSATs are investing in equipment they traditionally did not use to take on spillover processes, but large-scale mass production of hybrid bonding remains dominated by foundries.

Body · AI translation

Foundries and outsourced semiconductor assembly and test providers (OSATs) have traditionally had a clear division of labor: foundries handle front-end wafer manufacturing, while OSATs handle packaging and testing. But as OSATs acquire some foundry capabilities and foundries take back some business previously given to OSATs, the boundary between the two is blurring.

A recent Semiconductor Engineering report analyzed this industry shift in depth. Raghu Nayak, head of sales at Zeiss, said: "We are at a bridging point: IDMs and foundries are doing more of what OSATs can do, and OSATs are trying to get closer to what foundries do in the back end." However, OSATs will not compete with foundries on advanced nodes and front-end wafer manufacturing. In the back end, dimensional requirements are relatively lower, and OSATs are acquiring processes that can replicate foundry capabilities. The largest overlap between the two sides is 2.5D integration and panel-level assembly, and hybrid bonding may join this list in the future.

This change is affecting equipment and materials. Poupak Khodabandeh, chief commercial officer at Brewer Science, said lithography-related materials and processes are migrating from traditional fabs to advanced packaging environments. "As advanced packaging increasingly relies on fine-pitch redistribution layers (RDL), OSATs and packaging organizations are using materials and flows historically associated with wafer manufacturing." Jack Berg, vice president of business development at Silvaco, noted that the industry has traditionally operated in silos across equipment, front-end manufacturing, design, packaging and test, and now needs to connect these links so teams can make design decisions faster based on shared, physics-compliant insights while providing greater manufacturing flexibility in the back end.

The past division of labor: foundries make chips, OSATs do packaging and test. For years, foundries and OSATs have had complementary roles. Foundries manufacture silicon chips, and OSATs put chips into packages and test the final product. A large number of EDA tools and fab equipment support foundry processes, while OSATs rely more on tools derived from the printed circuit board (PCB) sector. Today two major trends are challenging this status quo.

The first trend is the development of advanced packaging. Placing a chip on a substrate, wire bonding, and then encapsulating it in molding compound is no longer enough to define advanced packaging. Doing these basics well is still important, but the requirements of advanced packaging are very different. Khodabandeh said advanced packaging is becoming increasingly "lithography-intensive." AI, HBM, chiplet architectures and heterogeneous integration are driving higher interconnect density, prompting packaging facilities to adopt fab-style patterning processes. RDL manufacturing is a typical example: forming finer features requires increasingly complex lithography.

Intel Foundry has both foundry and packaging capabilities. Ravi Mahajan, an Intel Foundry fellow, said that when making fine-pitch bumps, the number of bumps increases substantially compared with coarse pitch, so issues such as embedded bridge dies (such as EMIB), alignment and equipment precision must be addressed. For vendors with both foundry and packaging capabilities, they can try these processes on both sides.

The second trend is tight capacity. Foundries cannot produce all the chips customers need, so prioritization and allocation are required. This may mean turning down some business they would otherwise take, especially back-end and mature-process business, because foundries will focus on new-node silicon processes where their unique capabilities lie. OSATs are ready to take on these processes. These lower-resolution processes were historically used by foundries, but OSATs also have the corresponding capabilities. Mature-node products may spill over to assembly and test houses, especially as advanced packaging requires equipment consistent with these older technologies. Thomas Uhrmann, vice president of sales at EV Group, said: "As capacity constraints intensify further, both mature-node and leading-edge products are shifting to OSAT packaging."

Three main spillover tasks: 2.5D, panel-level packaging and hybrid bonding. Market dynamics are driving three major assembly-related processes from foundries to OSATs: 2.5D packaging, panel-level packaging, and eventually hybrid bonding.

2.5D packaging is similar to TSMC's CoWoS (chip-on-wafer-on-substrate) and involves mounting one die on top of another. When the bottom is an interposer, it is 2.5D; when the bottom is another active die, it becomes 3D. A typical flow is to place a already-singulated top die onto a bottom wafer die that has not yet been diced, then dice the bottom wafer after the top die is in place, and the stacked structure enters the packaging step. Photonics is a fast-growing field that needs this technology. Gerard John, senior director of the Chiplets/FCBGA business unit at Amkor, said customers come to ask whether electrical die-to-photonic die or wafer attachment can be done. "We dice this module out and then attach it to the substrate. This is a new trend we are seeing, and we can do it." But this brings new challenges: the optical side of the wafer or die requires active alignment to connect optical fibers and test.

In panel-level packaging, panel sizes once grew to around 500 mm per side, but shrinking the size actually helps. Uhrmann explained that several OSATs developed panel-level packaging lines early on, mainly targeting large 500 mm panels; TSMC decided to adopt 310×310 mm panel-level packaging, a format closer to traditional 300 mm round process equipment, enabling traditional wafer equipment makers to scale up for high-performance materials and resolution rather than forcing PCB equipment makers to scale down.

In hybrid bonding, some OSATs are already working on it, but large-scale mass production remains mainly a foundry task because the process challenges are greater. As hybrid bonding ceases to be novel and becomes a standard process, it may also become another task that foundries hand over to OSATs.

Investment and returns: OSATs must fill in the equipment. Foundries typically own the equipment needed for these processes, while OSATs must invest to match capabilities. They will not buy the most leading-edge equipment—there will be no EUV at OSATs—but they are equipping themselves with tools they traditionally did not use. The return-on-investment outlook for this capital spending looks good. Good ROI first requires a sizable and growing market. Advanced packaging is now common enough and underpins high-revenue markets such as data center AI, promising reasonable returns. Fortune Business Insights says advanced packaging grew to more than $45 billion in 2025 and is expected to reach $100 billion by 2034, with a compound annual growth rate of slightly above 9%. Uhrmann noted that ASE alone has already exceeded its $3.5 billion leading-edge packaging target this year and expects to double it by 2027.

Back-end processes are becoming increasingly important to foundries. Many performance advantages come from innovative packaging technologies rather than smaller transistors. Uhrmann observed that this performance race is driven mainly by AI and high-performance computing, with extremely high margins, partly because supply is constrained. "Margins are shifting from front-end chip manufacturing to packaging and system integration, because a growing share of system performance comes from advanced packaging, heterogeneous integration and 3D integration, not just from node scaling." As a result, fabs generally want to keep this business. There is not only ROI, but the returns may also come earlier than with advanced front-end equipment.

OSATs have a different economic model. Uhrmann said their business model is based on cost competitiveness and scale, with structurally lower margins than foundries, so ROI depends heavily on utilization and long-term customer commitments. Because foundry capacity is oversubscribed, OSATs are also running at full capacity, which is good for ROI. "The industry is constrained by equipment lead times and cleanroom space, not by orders. In this environment, investments in 2.5D, panel-level and test equipment can pay back quickly." But if OSATs mainly play a spillover role when the market is tight, then foundries remain in a favorable position in most cases, while OSATs may be more vulnerable to economic cycles as spillover demand comes and goes.

Hybrid bonding is still waiting. Hybrid bonding is another technology that could spill over but has not yet shifted on a large scale. OSATs are entering, but foundries still dominate. This pattern may persist as long as advanced packaging is managed vertically and all chiplets are purchased by one company. Uhrmann explained: "Hybrid bonding is a foundry- and IDM-dominated technology. As long as existing suppliers can meet demand, there is no logic for OSATs to invest." But when heterogeneous integration begins to bring in different dies from different technologies and sources—such as memory and photonics—it may make more sense for OSATs to participate in this business, which will require more equipment purchases. He expects hybrid bonding to follow a path similar to CoWoS: first adopted at foundries, then expanding to OSATs as a second source once demand exceeds foundry capacity and the process matures.

Overall, OSATs are buying equipment at a healthy pace, not only to take on new processes but also to add capacity. For example, Amkor is investing in a new plant in Arizona. Purchasing is also increasing. According to Seeking Alpha, ASE spent $3.4 billion on equipment and raised its 2026 budget by 27%; even so, it added another $1.5 billion in capital expenditure mid-year, of which $600 million was for equipment. Sanjiv Bhatt, senior director of global marketing and business development for Mitsubishi Chemical's semiconductor business, said letting both sides do what the other used to do may ultimately lead to new business. "This creates more opportunities to understand these new architectures, but it requires us to work with all stakeholders, including equipment makers, to study how to integrate these new materials and tools."

Source

Jiwei.com