Leading foundries’ constrained capacity and their insistence on a vertical business model could delay the arrival of chiplets in a form originally promised as a “Lego block-like,” mix-and-match design approach.
What’s at stake:
Technologies to build 2D and 2.5D multi-die chips have existed for almost a decade. Yet, demand for chiplets languished until the dawn of the generative AI era. Nvidia’s AI processors have swept the market, triggering a change so abrupt that foundries and Outsourced Semiconductor Assembly and Test (OSAT) companies are left flat-footed. At stake now is how long aspiring chiplet vendors can wait, and whether going to China for chiplet production is an option.
As global chiplet demand soars, the shortage in production capacity has focused sharply for developers of AI processors, high-performance computing chips and automotive OEMs/Tier Ones looking for scalable, automotive semiconductor designs.