Sparse data, vast spaces: navigating materials discovery with AI and atomistic simulations

The discovery of new materials for nanoelectronics has been a cornerstone of innovation and device scaling, sustaining transistor progress over many years. Prominent examples include the introduction of high-k dielectrics, new metals for gate control and interconnect, and high-mobility channel materials. As device dimensions shrink, integration complexity grows, and new deposition techniques expand the accessible chemical design space, identifying high‑performance materials becomes increasingly challenging. Moreover, finding materials with the right intrinsic properties is only the first step; they must also retain these at the nanometer scale, allow conformal deposition, remain stable through processing, integrate seamlessly, and align with sustainability objectives.

This talk discusses how AI and atomistic simulations are transforming the materials discovery process by addressing these challenges. By leveraging sparse modeling datasets and exploring expansive compositional spaces, these tools enable virtual screening and predictive modeling of material properties without prior measurements. The approach will be illustrated through the identification of new dielectrics and the design of amorphous semiconducting materials, where machine-learned models built on atomistic simulations accelerate candidate selection and guide experimental validations. To ensure adequateness and efficiency, this approach needs to be tightly iterated with experimental demonstrations, accounting for processing, patterning and device-relevant dimensions constraints. Ultimately, this hybrid methodology provides a scalable and sustainable pathway to identify new material candidates for next-generation nanoelectronic devices.