IMEC has developed a chip platform that operates at frequencies of up to 325GHz, potentially making 6G hardware affordable enough for practical deployment.

      TL;DRIMEC’s 300mm RF silicon interposer platform achieves record-low signal loss at 325GHz, marking a significant progression towards scalable, cost-effective 6G chip production. IMEC, the Belgian semiconductor research institute collaborating with over 600 players in the chip industry, has enhanced its 300mm RF silicon interposer platform by adding three new manufacturing capabilities, bringing 6G chip production closer to commercial feasibility. The platform reaches record-low signal loss at frequencies up to 325GHz, covering the millimeter-wave and sub-terahertz bands essential for 6G networks. This research was showcased at the IEEE International Microwave Symposium in Boston this month.

      IMEC aims to address challenges related to materials and economics. 6G radios must operate at frequencies much higher than traditional silicon can support, necessitating the use of compound semiconductors like indium phosphide, gallium arsenide, and silicon germanium. While these materials provide enhanced RF performance, they are produced on small, costly wafers that do not scale like standard 300mm silicon production lines.

      The approach taken by IMEC involves utilizing a silicon interposer as a carrier substrate, enabling the integration of small chiplets made from compound semiconductors onto a standard 300mm silicon wafer. The interposer manages the digital interconnections and passive components, allowing the III-V chiplets to focus on RF signal processing. This results in a versatile platform where different materials can be combined without needing each to scale individually.

      The three new manufacturing capabilities introduced in June target specific production challenges. High-density embedded capacitors, or MIMCAPs, enable passive components to be transferred from the expensive III-V chiplets to the more affordable silicon interposer, which minimizes both chiplet size and cost. A scalable modeling framework for passive components provides designers with simulation tools to predict performance before fabrication. Laser-assisted bonding allows for the precise placement of III-V chiplets onto the silicon carrier.

      “With this work, we demonstrate a uniquely integrated platform that brings together performance, scalability, and manufacturability,” stated Xiao Sun, principal member of technical staff at IMEC. “Our next focus is to further enhance the technology readiness of the platform and facilitate low-volume manufacturing support, enabling our partners to easily develop and scale next-generation RF systems.”

      The timing is significant as Nvidia aims to make telecom one of its primary growth areas. Last October, the company invested $1 billion in Nokia for a 2.9 percent stake, and at Mobile World Congress in March, it formed a coalition that includes Ericsson, Deutsche Telekom, T-Mobile, SK Telecom, and SoftBank to develop 6G networks based on what it calls AI-native platforms. Nvidia has been broadening its AI partnerships across various hardware sectors, including robotics, data centers, and automotive, with telecom infrastructure being the next logical step in that strategy.

      Jensen Huang has suggested that every radio access network in a 6G landscape will effectively function as an AI computer, blurring the line between communication hardware and AI inference. If this vision proves accurate, the challenge will shift from software to silicon, specifically regarding the cost-effectiveness and reliability of manufacturing the underlying RF chips at scale. This is precisely where IMEC’s platform is relevant.

      IMEC holds a unique position within the semiconductor ecosystem. Based in Leuven, Belgium, it is a non-profit organization with over 5,000 researchers from 96 countries. Its business approach involves developing pre-competitive chip technology in collaboration with the industry and then transferring the results for commercial use. TSMC, which is expanding its chip manufacturing presence in Europe, is one of IMEC’s longstanding partners, alongside Samsung, Intel, and many major foundries globally.

      The importance of the 325GHz insertion loss benchmark lies in its relevance not just to the frequencies initially expected for 6G but also to the sub-terahertz range being investigated for ultra-high-bandwidth short-range connections. Achieving low signal loss at these frequencies using a standard silicon manufacturing platform, rather than specialized substrates, enhances the work’s relevance to deployment economics rather than purely laboratory performance.

      However, this does not imply that 6G hardware is production-ready. IMEC’s roadmap acknowledges that further development is needed for the platform to achieve low-volume manufacturing readiness, much less the high-volume production required for global telecom deployment. The semiconductor industry typically sees a five to seven-year gap between research breakthroughs and commercial chips, and 6G networks are not anticipated to begin standardization until at least 2028. IMEC’s platform is on a timeline that could coincide with the telecom industry's needs.