Chiplet Architecture: Companies, Startups & Research

Intel Corporation, TSMC, Advanced Micro Devices Inc., Qualcomm, Samsung and Siemens AG are among the companies active in chiplet architecture. Their work spans die-to-die links, advanced packaging, processor designs and multi-die design tools. Cadence and Synopsys Inc. also develop design and verification tools, while Arm Ltd. is working on its Chiplet System Architecture.

Lightmatter Inc., Rebellions Inc and Celestial Ai are notable startups pursuing distinct chiplet approaches. Lightmatter and Celestial Ai focus on photonic interconnects, while Rebellions develops AI accelerators and has patented chiplet communication and security methods. Their focus areas point to opportunities in high-bandwidth data movement and multi-chiplet systems, though their approaches address different technical needs.

IMEC, Massachusetts Institute of Technology and Georgia Institute of Technology are prominent research organizations in this field. IMEC works on interconnect testing and repair for 2.5D and 3D systems. MIT has researched photonic chiplet integration for quantum processors, while Georgia Tech develops interconnect simulation, thermal prediction and packaging-aware design methods.

FACTS records $1,188,461,269 in grants across 82 awards since 2020. This is the clearest dedicated grant indicator provided, but it should not be read as total investment in chiplet technology. The funding-round records aggregate $1,154,369,657,666 across 69 rounds and include broad AI and infrastructure financing, making that total a noisy measure of chiplet-specific funding.

Intel Corporation leads the listed patent holders with 95 patents, followed by Qualcomm with 60 and Unifabrix Ltd. with 30. Advanced Micro Devices Inc. has 25, and Robert Bosch GmbH has 23. These figures identify organizations with substantial patent activity in the dataset; they do not by themselves establish the quality, scope or commercial use of those patents.

Patent activity declined 44.0% in the latest 12-month period, with 755 patents in the 24-month window. Publication activity moved in the opposite direction: 585 publications appeared last year, up 28.3%. The split suggests research output is rising even as patenting has slowed, but neither measure alone establishes whether a technology is ready for broad commercial adoption.

October 2026 work from Tsinghua University reported CMP-free all-copper bonding using porous copper, while Unimicron Technology Corporation examined glass packaging and Cu-Cu hybrid bonding. Recent research also addresses signal integrity, thermal analysis and package routing. In September 2026, GigaRDL proposed GPU-resident routing for redistribution layers, reflecting efforts to make complex package design more efficient.

The main approaches include 2.5D interposers, 3D die stacking, bridge-based integration, fan-out integration, die-to-die interfaces and composable die partitioning. Interposers and bridges connect side-by-side dies; stacking places dies vertically. Fan-out uses redistribution layers, while interface standards define how dies communicate. Each approach trades off bandwidth, package complexity, thermal behavior, routing flexibility and testability.

The United States leads with 9,973 documents, followed by Germany with 2,369 and Taiwan with 1,984. The United Kingdom, China and South Korea are also active, with 1,914, 1,669 and 1,607 documents respectively. These figures indicate where activity is concentrated in the dataset, not necessarily where all chiplet manufacturing or supply-chain capacity is located.

AI and high-performance computing lead the listed applications with 3,983 documents, followed by automotive computing with 3,512. Other areas include consumer and mobile electronics, quantum computing, and networking and communications. Chiplets can combine compute, memory and specialized functions in one package, but each application brings different requirements for performance, power, reliability, cost and system validation.