Member Of Technical Staff - Extreme-Scale Sparse Linear Algebra, Domain Decomposition GPU Solver Architecture

📋 Description

• • At Vinci, we are at the forefront of building AI-enabled infrastructure designed to accelerate physics decisions with unparalleled confidence for modern hardware programs. Our innovative software delivers manufacturing-resolution physics simulation, achieving verified accuracy at runtimes orders of magnitude faster than traditional tools by entirely bypassing meshing and approximation overhead. We are proud to be deployed or in active validation with a diverse range of Tier-1 ecosystem players, including semiconductor IDMs, foundries, advanced packaging specialists, fabless companies, automotive manufacturers, EMS providers, and energy hardware developers. This engagement means we are tackling real-world solver constraints, not just theoretical benchmarks. The simulation decisions made here directly influence actual hardware outcomes, involving a wide array of operator structures and conditioning regimes.
• • Now, we are embarking on the critical mission of constructing the core solver substrate that must achieve unprecedented scalability, extending beyond billions to trillions of degrees of freedom (DOFs). This substrate must maintain strict determinism and generalize effectively across radically different operator landscapes and distributed environments. This role is centered on the fundamental numerical substrate itself, rather than application wrappers. You will be instrumental in addressing key challenges such as ensuring convergence and robust conditioning at extreme scales, developing advanced domain decomposition techniques for production environments, and implementing scalable, multilevel, and multigrid preconditioning strategies.
• • Your responsibilities will include architecting and delivering production-grade solver infrastructure. This encompasses the design and implementation of Domain Decomposition and Schwarz Methods, including additive and multiplicative Schwarz frameworks, and both overlapping and non-overlapping strategies. You will also focus on scalable coarse space construction and the development of hybrid coarse-fine hierarchies for production meshes. Furthermore, you will tackle Preconditioning at Extreme Scale by developing algebraic and geometric multigrid methods, block physics-aware preconditioners, ILU variants, sparse approximate inverses, and communication-efficient preconditioner designs.
• • The role also involves deep engagement with Krylov and Solver Architecture, including methods like CG, GMRES, FGMRES, and BiCGStab. You will work on pipelined communication-reducing methods and explore mixed-precision strategies with robust guarantees. A significant aspect will be ensuring deterministic reduction ordering over distributed execution environments. Additionally, you will contribute to AI-Augmented Solver Enhancements, focusing on learned augmentations for coarse space discovery, adaptive preconditioner selection, and spectral approximations and operator compression. It's crucial to understand that AI in this context is intended to support and enhance numerical structure, not to replace it.
• • We are seeking individuals with profound expertise in domain decomposition and Schwarz methods, multilevel solvers, and scalable preconditioning for large sparse systems at extreme scales. A strong grasp of parallel numerical stability and conditioning, alongside experience with GPU-accelerated sparse linear algebra (CUDA and HIP), and multi-GPU and distributed execution paradigms is essential. Your thinking should encompass spectral equivalence and coarse space quality, understanding the trade-offs between strong and weak scaling, and balancing communication versus computation. Crucially, you must have a track record of shipping real solver infrastructure, not just theoretical prototypes.
• • From a Systems and Engineering perspective, we expect proficiency in CUDA (HIP is appreciated), kernel-level performance engineering, and multi-GPU scaling experience. Strong disciplines in CI, regression testing, and correctness validation are vital. You should possess a deep understanding of how algorithms map to hardware and can withstand the pressures of production environments. This is an execution-oriented principal engineering role within a startup that has real production deployments. You will architect foundational solver systems, implement and ship solutions into Tier-1 environments, build continuous validation and regression frameworks, and improve throughput and determinism under real-world constraints. We are ambitious, but our focus is on shipping impactful solutions.
• • Vinci offers a unique opportunity to work on problems that are truly architectural in nature, pushing the boundaries of what's possible in deterministic, robust solver substrates that form the heart of future physics infrastructure. We believe AI should augment numerical authority, not override it. This role is designed for those who are driven by these principles and want to contribute to building the solver core that enables deterministic physics infrastructure, validated within real hardware workflows and ready to scale beyond current limits.