New high-density fan-out module, ASE Technology pushes advanced packaging for AI chips

ASE Technology is highlighting its FOCoS fan-out module as a high-density packaging option for AI and data center processors, combining multiple dies and high-bandwidth memory in a single advanced substrate stack for hyperscale and HPC customers.

Edited by ad hoc news New Releases & Launches Desk. Reviewed before publication on 06/15/2026 at 9:35 PM ET. Details in the imprint.

With artificial intelligence workloads pushing chip power and bandwidth limits, ASE Technology is putting its FOCoS fan-out module in the spotlight as one of its most advanced high-density packaging options for AI accelerators and data center processors. The FOCoS (Fan-out Chip-on-Substrate) family is designed to integrate logic dies and high-bandwidth memory on a single advanced substrate, giving semiconductor customers a path to wider I/O, shorter interconnects and better thermal performance than traditional multi-chip modules. According to ASE, the latest FOCoS variants support chiplet-style partitioning, fine-line redistribution layers and large package sizes that target GPUs, custom AI ASICs and networking silicon for hyperscale data centers. The official FOCoS product page describes configurations that combine multiple logic dies and stacks of high-bandwidth memory on an organic substrate to approach system-in-package integration.

How ASE’s FOCoS fan-out module addresses AI and HPC packaging limits

FOCoS sits within ASE’s advanced packaging portfolio alongside 2.5D interposers, fan-out wafer-level packaging and system-in-package designs, but it is specifically optimized to put multiple large dies onto a single high-density substrate without the cost of a full silicon interposer. In FOCoS, bare dies are first attached to a thin core or carrier, then encapsulated and connected through redistribution layers that fan out the I/O to an organic laminate, enabling much finer wiring pitches between dies than conventional flip-chip assembly while keeping the package mechanically robust for data center deployment. This structure allows designers to split a monolithic GPU or accelerator into several chiplets placed side by side, while still achieving high-bandwidth die-to-die communication and reducing the risk and cost associated with manufacturing very large single dies.

For AI accelerators that rely on high-bandwidth memory, FOCoS supports mounting multiple HBM stacks close to the logic dies to minimize trace length and latency and to maximize data throughput. The tight physical proximity between logic and memory in a fan-out configuration is especially important for transformer-based models and other bandwidth-hungry workloads that need to move large parameter sets across the memory interface with minimal bottlenecks, and the FOCoS substrate can be laid out to provide parallel signal paths from each memory stack to the compute chiplets. In practice, this means an AI accelerator vendor can design a package in which several HBM stacks flank a central compute complex, all tied together through fine-pitch redistribution layers that reduce parasitics and help maintain signal integrity at high data rates.

From a manufacturing standpoint, ASE has been scaling FOCoS production capacity at its advanced packaging facilities to meet demand for heterogeneous integration in networking, high-performance computing and data center markets. The company emphasizes process control for warpage and yield when dealing with large package sizes and multiple thick HBM stacks, using tailored mold compounds, underfill materials and warpage-control techniques so that FOCoS substrates remain flat enough for surface-mount assembly and reflow onto customer boards. At the same time, the FOCoS line is designed to be modular, so that different customers can choose configurations with varying numbers of chiplets, memory stacks and power-delivery layers without redesigning the entire process flow.

Power delivery and thermal management are two of the more challenging aspects of AI accelerator packaging, and FOCoS tackles them through dense power planes and heat-spreading options in the substrate and encapsulation. Because FOCoS packages can be physically large, ASE can embed wide power and ground planes and thick copper structures in the laminate to bring current to high-power dies while keeping voltage droop within tight margins at high load. On the thermal side, the flat encapsulated surface over the chiplets and memory stacks is compatible with large vapor chambers or cold plates, and the company can expose critical die surfaces or add thermal vias in the mold to shorten the path to the heat spreader, which is important for GPUs and custom accelerators that often operate at several hundred watts in data center racks.

ASE positions FOCoS as a bridge technology between traditional multi-chip modules and full 2.5D/3D integration, aiming to offer high interconnect density with lower cost and greater design flexibility than silicon interposers. For chipmakers experimenting with chiplet-based architectures, FOCoS provides an intermediate option where logic dies and memory can be partitioned and laid out on a single organic substrate with fan-out routing instead of insisting on a monolithic die or a costly interposer. Industry coverage of advanced packaging trends has pointed out that fan-out and chiplet approaches like FOCoS are becoming key enablers for AI and networking processors that need more memory bandwidth and I/O than legacy organic packages can deliver without massive increases in layer count, cost and complexity. A recent analysis in Semiconductor Digest notes that fan-out and embedded bridge technologies are increasingly used to connect accelerators with memory and networking interfaces as cloud providers roll out larger-scale AI clusters.

FOCoS does not target end consumers directly; instead, its customers are major semiconductor vendors designing chips for hyperscale cloud operators, enterprise data centers and high-performance computing sites. Those customers can tailor FOCoS-based packages to their own floorplan and thermal requirements, choosing the number and arrangement of chiplets and memory stacks, which makes FOCoS part of a broader ecosystem shift toward co-design of silicon, package and system. For US investors and industry watchers, FOCoS matters because it is one of the advanced packaging technologies that can capture a growing share of value in the AI hardware stack as chip architects push beyond the limits of traditional monolithic designs and look to outsource complex back-end integration to experienced OSAT providers.

Within ASE Technology’s portfolio, FOCoS sits alongside chiplet, 2.5D interposer and fan-out wafer-level packaging offerings that together position the company to serve a broad set of AI and high-bandwidth customers, and the firm has highlighted advanced packaging as a core growth driver in its investor communications. In its most recent investor presentation, ASE pointed to rising demand for heterogeneous integration, high-bandwidth memory attachment and system-in-package solutions as a multi-year opportunity, with AI accelerators and data center networking devices among the most important application segments for its advanced packaging business. The latest ASE investor presentation underlines that advanced packaging and system-level solutions are expected to outpace more mature wire-bonding services as cloud and AI customers migrate to chiplet architectures.

Advanced packaging solutions like FOCoS are strategically important for ASE Technology because they allow the company to move up the value chain beyond commoditized assembly, and they tie ASE closely to customers building differentiated AI and data center products that require long-term collaboration on package design, thermal management and reliability. For investors, FOCoS is one of several advanced substrate and fan-out technologies that could support higher-margin growth if demand for AI accelerators, high-bandwidth memory and chiplet-based networking gear remains robust across cloud and enterprise markets. Shares of ASE Technology Holding (TW0003711008) closed on the New York Stock Exchange at $38.49 on 06/14/2026, reflecting how the market is currently valuing its mix of traditional packaging services and higher-value advanced offerings.

ASE FOCoS fan-out module in brief

• Product: FOCoS fan-out module
• Manufacturer: ASE Technology Holding Co.
• Category: New Release, Advanced Packaging for AI and HPC
• Launch date: Initial FOCoS introduction mid-2010s; current AI-focused variants highlighted in recent ASE advanced packaging roadmaps
• MSRP / Price: Not publicly disclosed; priced per customer design and volume
• Availability: Offered as a packaging service through ASE’s advanced packaging facilities in Asia for global semiconductor customers
• Target audience: Semiconductor companies designing AI accelerators, GPUs, networking ASICs and high-performance computing processors
• Key differentiator / USP: Combines multiple logic dies and high-bandwidth memory stacks on a single high-density fan-out substrate as an alternative to full silicon interposers

This article was a.i.-assisted and editorially reviewed. Product information without warranty; prices and availability may change at short notice. Not investment advice and not a buy or sell recommendation. Trading involves risk up to and including the total loss of invested capital.