The San Jose startup is developing a modular chiplet platform with TSMC sample plans for 2027 as demand rises for custom AI semiconductors beyond proprietary designs from Broadcom and Marvell Technology.
TYLSemi said it has raised an oversubscribed $43 million early-stage round to help companies build custom AI chips using open-standard chiplets, or modular chip components, rather than proprietary custom-chip technology. Matter Venture Partners led the financing, with Viola Ventures, GHOVC, Egis Technology and other strategic investors focused on semiconductor and AI infrastructure participating. The San Jose startup, co-founded by Mohit Gupta, chief executive, and Sunil Bhardwaj, is developing a full-stack chiplet platform built around reusable components that customers can combine instead of designing every part of an AI processor from scratch. Its product lineup includes TYL.IO for chiplet connectivity, TYL.Power for intelligent power delivery and TYL.Forge as a full-stack custom silicon production platform. TYLSemi says its chiplets are built on UCIe and other industry standards so they can work with existing chip architectures and advanced packaging technologies. The company says that approach can cut the time and cost of custom AI silicon development by as much as 50%. The company is entering a fast-growing market for AI semiconductors, where large groups such as Meta Platforms are already working with Broadcom and others on custom designs. Broadcom and rival Marvell Technology offer proprietary systems that let chips communicate at high speed, while TYLSemi is positioning itself around standards-based interoperability. TYLSemi has also lined up a collaboration with TSMC to produce samples of its TYL.IO and TYL.Power chiplets, with first samples expected in 2027.