NTT uses AI and robotics to triple semiconductor film experiment speed

This is a nice example of AI not just analysing experimental data, but closing the loop with robotics to run the lab itself and then explain what it learned. NTT’s system is notable because it explicitly aims for interpretability: the AI not only finds optimal deposition conditions but distils them into rules that human researchers can transfer to other materials and tools. That is exactly the kind of capability needed to speed up the hardware and materials pipeline that underpins the AI boom, from power devices to advanced packaging. ([news.mynavi.jp](https://news.mynavi.jp/techplus/article/20260827-4872315/))

For AGI timelines, faster and more systematic materials discovery for power semiconductors and other components reduces one of the physical bottlenecks on compute growth. If similar “self-driving labs” proliferate across cooling, energy storage and new transistor structures, the capital being poured into AI data centres can translate into usable infrastructure more quickly. At the same time, interpretability in robotics-heavy scientific workflows is a kind of alignment practice: humans retain a conceptual model of what the system is doing rather than treating it as an opaque optimiser.