As the semiconductor industry advances toward smaller nodes, heterogeneous integration, and increasingly complex device architectures, traditional process development methods are hitting scalability and efficiency limits. High cost, long iteration cycles, and the growing combinatorial complexity of materials and process conditions are now significant barriers to innovation. To overcome these challenges, new paradigms are needed—ones that combine human domain expertise with intelligent, data-driven automation.
At Cosmos Innovation, we have built an industry-first solution for accelerating Design of Experiments (DoE), which has conventionally relied on extremely slow trial-and-error approaches. Our AI tools are capable of learning from sparse, noisy, or incomplete data in the small data regime; optimizing toward defined engineering goals; and adapting strategies based on real-time feedback from experiments or production data. These systems are designed not only to accelerate discovery and optimization, but to evolve alongside the manufacturing environment—laying the foundation for agentic AI systems that can operate with increasing autonomy in industrial settings.
Our AI platform combines closed-loop self-optimization workflows, hybrid AI-physics engines, and generative models to accelerate semiconductor innovation from lab to fab. These systems enable intelligent, adaptive process development that complements human expertise—allowing engineers to explore broader, high-dimensional process and design spaces, converge on optimal solutions faster, and reduce costly trial-and-error cycles.
In this presentation, we will share a series of real-world case studies in collaboration with partners and customers across the semiconductor industry. The case studies will demonstrate: • Up to 10x acceleration in recipe development and process iteration cycles • Automated discovery of high-performance process windows with minimal data • Disruptive improvements in device performance and cost-efficiency in production-relevant settings
We will also cover an industry-first end-to-end demonstration of building a new semiconductor technology on our AI platform: perovskite-silicon tandem technology. We have used our AI platform to dramatically accelerate the identification and tuning of complex material stacks and process recipes, making major progress on longstanding stability challenges that have historically limited commercial progress in this field. This accelerated approach has enabled us to achieve remarkable efficiency results and catch up with leading players in the field within a short span of time—showcasing the power of AI to rapidly close technical gaps and drive competitive performance.
We will discuss how these AI systems are being integrated into existing fab environments, including the role of human-in-the-loop collaboration in building trust, ensuring safety, and enhancing decision-making. The session will also cover best practices for transitioning from proof-of-concept pilots to scaled deployment in high-throughput manufacturing settings.
Cosmos Innovation’s work represents a concrete and scalable path toward AI-native semiconductor manufacturing. By blending physics-based reasoning with generative intelligence and adaptive control, we aim to enable a new era of faster, smarter, and more autonomous innovation in next-generation semiconductor technologies.
About the Company Cosmos Innovation is an AI company accelerating semiconductor process development from years to months. The company is backed by leading investors globally and leading minds in AI, including Demis Hassabis, CEO of DeepMind and Nobel Prize Winner: the company was a World Economic Forum Tech Pioneer 2024.
.jpg "Joel Li photo")