UK firm Infinitesima targets 3D chip metrology gap as AI reshapes fabs

Demand for AI accelerators and high-bandwidth memory (HBM) is pushing semiconductor makers to scale on two fronts, and UK metrology developer Infinitesima says its technology is aimed at one of the resulting bottlenecks: fast, reliable measurement.

At transistor level, 3D gate-all-around (GAA) architectures and High-NA extreme ultraviolet (EUV) lithography are extending scaling through finer patterning and better imaging resolution. At system level, advanced 3D integration combines multiple compute and memory dies in a single package to raise performance and bandwidth.

That shift is changing packaging. HBM roadmaps are heading towards taller stacks, including 16-high configurations, which add thermal, interconnect and assembly-yield challenges. Hybrid bonding allows denser integration but sets strict surface requirements:

AI is also being used to optimise chip manufacturing itself. As fabs adopt AI-driven process control, trustworthy measurement data becomes essential. Manufacturers need non-destructive readings of surface height, feature depth, sidewall profiles and nanoscale roughness at speeds suited to high-volume production.

Dense 3D data of this kind can feed run-to-run and feed-forward control, support virtual metrology and anchor hybrid schemes where optical or electron-beam measurements are calibrated against a 3D reference. In High-NA EUV, 3D resist profile and roughness data can help calibrate and validate stochastic resist models, linking line-edge and line-width roughness to patterning defectivity.

Atomic force microscopy (AFM) gives detailed 3D surface data but has historically been held back by throughput. Infinitesima's Metron3D is a hybrid-optical AFM using 3-axis interferometry and photothermal probe actuation, with measurement speeds 10 to 100 times faster than conventional AFM. The fully automated 300mm system is designed to deliver true 3D information non-destructively at high-volume manufacturing speeds.

The UK is building its semiconductor position through its National Semiconductor Strategy and 2026 AI Hardware Plan, concentrating on chip design, compound semiconductors, photonics, advanced materials and research. Initiatives include funding for AI hardware innovation, the UK Semiconductor Centre and support for commercialising research.

The article argues that opportunity extends beyond design into enabling manufacturing technologies. Investment in metrology, process control, advanced packaging, pilot manufacturing and equipment innovation could help bridge the gap between research and high-volume production. Closer collaboration between government, researchers, equipment suppliers and chipmakers could also speed adoption of UK-developed technologies.

With High-NA lithography, GAA devices, hybrid bonding and stacked memory all needing measurement that keeps pace with the process, techniques that capture atomic-scale 3D detail quickly and economically are expected to ramp faster and help fab yield.