The packaging stage, long the least glamorous part of making a chip, has become one of the industry’s tightest bottlenecks. As AI accelerators grow, the way chiplets are connected determines how fast a product can ship and how much it costs. Samsung Electro-Mechanics and Qualcomm have been working on one answer for more than a year.
The two companies are developing a technology known as an organic bridge for advanced packaging of AI chips, according to people familiar with the work. The project has been in research and validation for over a year, the people said. Neither company has announced the effort publicly.
The idea is to replace the silicon bridges used in some high-end packages with structures built from organic materials. A bridge sits between chiplets and carries the dense wiring that lets them communicate. Organic versions are cheaper to make and can be built over larger areas than silicon, which is why the approach has drawn interest.
The shift toward chiplets began years ago, when AMD used the approach to compete with Intel in server processors, and it has since spread through the industry. Breaking a chip into smaller dies improves yields and cuts cost, but it creates a new demand. The connections between the dies must carry enormous amounts of data at high speed, and that demand is what bridge technologies exist to serve.
Samsung Electro-Mechanics is also developing 2.1D packaging substrates based on the organic bridge with other customers, the people said. The company’s plan calls for embedding an organic bridge with five to seven redistribution layers into a flip-chip ball grid array substrate, a format used widely in high-performance chips.
Redistribution layers are the fine wiring that spreads signals from a chip’s dense contacts to the wider spacing a package can handle. In a flip-chip ball grid array substrate, the chip sits face down on a substrate studded with solder balls that connect to a board. Adding an organic bridge with multiple redistribution layers into that substrate would let several dies share one high-density link.
The technical target is aggressive. The company aims to shrink line width and spacing to between 1.5 and 2 microns, a density that rivals what silicon bridges achieve today, according to people familiar with the plan. Meeting that specification on an organic substrate is the central engineering challenge.
The clearest comparison is Intel’s EMIB, the embedded multi-die interconnect bridge Intel has used for years to tie chips together. EMIB relies on a small silicon bridge. The organic approach Samsung Electro-Mechanics and Qualcomm are pursuing is widely seen in the industry as a potential alternative that costs less.
Why it matters now is straightforward. AI chips are hitting the limits of how large a single die can be built economically. The alternative, stitching several smaller dies together in one package, depends entirely on the quality of the connections between them. Packaging has become a performance problem, not a cost detail.
Qualcomm’s interest fits its push into AI and data-center silicon. The company is known for smartphone processors but has been expanding into infrastructure, where advanced packaging is a requirement. A cheaper bridging technology would lower the cost of the multi-die designs it wants to sell.
The work also reflects pressure on the supply chain. The most advanced packaging capacity, dominated by TSMC’s CoWoS and related techniques, has been in short supply for years. Chip designers and component makers are searching for alternatives that do not depend on a single supplier or on silicon interposers.
Samsung Electro-Mechanics is a natural partner. The company, part of the Samsung group, is a major supplier of substrates for high-end chips and has invested heavily in advanced packaging. Its interest in organic bridges extends its reach from the substrate into the interconnect itself.
The Korean supply chain has a stake of its own. Samsung competes with TSMC across foundry and advanced packaging, and the country’s component makers want a domestic path that does not route every advanced package through a rival’s facilities. An organic bridge that works at scale would widen that path.
The market for advanced packaging has grown quickly alongside AI. Analysts estimate the segment is now worth tens of billions of dollars a year, and it is growing faster than the broader semiconductor market. Suppliers that can offer a credible alternative to silicon-based approaches are positioned to capture a share of that expansion.
Analysts said the project is a signal rather than a product announcement. Nothing has shipped, and validation of a new bridge technology can take years. But the fact that a major substrate maker and a major chip designer are jointly pursuing an alternative to silicon bridges shows how much the economics of packaging have shifted.
The uncertainty is in timing. Even if the technology validates, volume production would require new tooling and qualification across multiple chip designs. People familiar with the work cautioned that commercial use is not imminent, and no timeline for deployment has been set.
What the effort establishes is that the industry no longer treats packaging as a commodity. Whoever can connect chiplets more cheaply and densely than silicon bridges allow will hold an advantage over the cost of every AI chip that follows, and Samsung Electro-Mechanics has put itself in that contest.
