Computer chips have scaled dramatically over the past five decades, with billions of transistors now integrated onto a single processor. This scaling is enabled by advanced manufacturing technologies, most notably Extreme Ultraviolet (EUV) photolithography. EUV systems enable the patterning of extremely fine features required for modern semiconductor devices. These systems are manufactured by ASML and used by leading chip producers such as TSMC, which supply companies like NVIDIA and Apple.

A key challenge is that EUV systems are highly complex, and optimal operating parameters vary across machines and drift over time. Identifying and maintaining optimal input settings is therefore difficult in practice. We developed an adaptive optimization algorithm that iteratively probes system outputs and adjusts control inputs to approach optimal operating conditions. This algorithm improves manufacturing yield and continuously tracks optimal settings as system conditions drift over time.