Recorded at CDFAM Computational Design Symposium, NYC, October 29-30, 2025
https://cdfam.com/nyc-2025/
Organization:
Presenter:
Gregory Roberts + Qiqi Wang
Real-Time Computer-Aided Optimization (CAO): How GPU-Native CFD Changes the Industry
Presentation Abstract
Computer-aided engineering (CAE) has been a foundational tool in aerospace and photonics design, but slow workflows, high costs, and constrained design exploration limit its potential. Traditional methods rely heavily on intuition and a few simulations to validate designs, leaving vast opportunities untapped. However, a paradigm shift is underway: integrating mathematical optimization techniques like adjoint optimization and inverse design into CAE is redefining what’s possible in engineering.
This modern approach – Computer-Aided Optimization (CAO) – directly leverages advanced mathematical optimization to automate and enhance the design process. CAO replaces intuition-driven, validation-focused methods with a data-driven, goal-oriented workflow by specifying design goals and using algorithms to refine configurations iteratively. Techniques like inverse design, which uses objective functions and gradient-based optimization, and adjoint methods, which enable efficient sensitivity analysis, are central to this transformation.
GPU-native simulations amplify the impact of these methodologies, making it feasible to address industry-scale problems in a fraction of the time previously required. High-performance GPU computing accelerates the iterative optimization process, enabling rapid exploration of vast design spaces with unprecedented fidelity. Applications range from optimizing aerodynamic performance in aerospace to creating innovative photonic devices like metalenses and quantum computing components.
This synergy of mathematical optimization and GPU acceleration positions CAO as the future of engineering design. By reducing costs, accelerating development cycles, and enabling robust design exploration, CAO allows engineers to confidently tackle complex challenges. Whether designing aircraft or photonic circuits, these advancements fundamentally reshape how industries approach innovation, driving breakthroughs across disciplines and unlocking new possibilities for high-performance, efficient design.
Speaker Bio
Greg Roberts is a research scientist at Flexcompute working on building gradient-based inverse design tools for photonic optimizations. He earned his PhD from Caltech in August 2023 on this same topic. His dissertation focused on the inverse design of 3-dimensional structures for advanced and high efficiency mid-infrared imaging applications. By using gradient information, he demonstrated practical design of color and polarization sorting devices that could be tiled on the pixels of focal plane arrays. Using multilayer fabrication via a finely tuned two photon lithography process, he was able to measure these novel devices to confirm their complex, target behavior. Greg followed graduate school with a postdoctoral research role at NYU applying inverse design to enhance contrast in biomedical imaging. Before graduate school, Greg worked as an embedded software engineer at an augmented reality startup called Magic Leap. Here, he optimized computer vision and machine learning algorithms to run at high speeds on a low-power embedded processor.
