Member of Technical Staff - Extreme-Scale Sparse Linear Algebra, Domain Decomposition & GPU Sol

The Mission

At Vinci, we are building the AI-enabled infrastructure that modern hardware programs use to converge on physics decisions with confidence. Our software delivers manufacturing-resolution physics simulation with verified accuracy at orders‑of‑magnitude faster runtimes than traditional tools, bypassing meshing and approximation overhead entirely. We are deployed or in active validation with a broad range of Tier‑1 ecosystem players — across semiconductor IDMs, foundries, advanced packaging, fabless companies, automotive, EMS, and energy hardware development.

This means real solver constraints, not benchmarks. Simulation decisions here drive actual hardware outcomes, with diverse operator structures and conditioning regimes. Now we are building the core solver substrate that must scale beyond billions of DOFs — to trillions, preserve determinism, and generalize across radically different operator landscapes and distributed environments.

• Conditioning and convergence at extreme scale
• Domain decomposition and Schwarz theory at production scale
• Robust, multilevel and multigrid, preconditioning
• Communication-avoiding Krylov and hierarchical solvers
• Deterministic parallel reductions across GPU clusters
• AI-accelerated solver components grounded in numerical rigor

Your work will shape the solver architecture that supports not just a single physics, but a rich operator ecosystem including indefinites, saddle‑point systems, strong coefficient jumps, anisotropy, and tightly coupled multiphysics blocks encountered in real hardware workflows.

You will own the design and delivery of production‑grade solver infrastructure, including:

• Additive and multiplicative Schwarz frameworks
• Overlapping and non-overlapping strategies
• Scalable coarse space construction
• Hybrid coarse/fine hierarchies for production meshes

• Algebraic and geometric multigrid
• Block/physics-aware preconditioners
• ILU variants, sparse approximate inverses
• Communication-efficient preconditioner designs

• CG, GMRES/FGMRES, BiCGStab
• Pipelined/communication-reducing methods
• Mixed-precision strategies with robustness guarantees
• Deterministic reduction ordering over distributed execution

• Learned augmentations for coarse space discovery
• Adaptive preconditioner selection
• Spectral approximations and operator compression

AI here supports numerical structure, not replaces it.

• Domain decomposition and Schwarz methods
• Multilevel solvers and scalable preconditioning
• Large sparse systems at extreme scale
• Parallel numerical stability and conditioning
• GPU-accelerated sparse linear algebra (CUDA + HIP)
• Multi-GPU and distributed execution paradigms

You think about:

• Spectral equivalence and coarse space quality
• Strong/weak scaling tradeoffs
• Communication vs computation balance

You’ve shipped real solver infrastructure — not just prototypes.

• CUDA first, HIP appreciated
• Kernel-level performance engineering
• Multi-GPU scaling experience
• Strong CI, regression, and correctness validation disciplines

You understand how algorithms map to hardware and survive production pressure.

• Architect foundational solver systems
• Implement and ship into Tier‑1 environments
• Build continuous validation and regression frameworks
• Improve throughput and determinism under real constraints

We are ambitious — but we ship solutions that matter.

• Already proven at scale with real validation across Tier‑1 ecosystem participants.
• Physics-first software built on verified methods, not heuristics.
• A small, technically serious team with deep domain expertise.
• High ownership, equity participation
• Production impact — not academic benchmarks

• Trillion‑DOF problems are architectural — not just hardware —
• Deterministic, robust solver substrates are the heart of future physics infrastructure
• AI should augment numerical authority, not override it

We are building the solver core that enables deterministic physics infrastructure — validated inside real hardware workflows and ready to scale beyond today’s limits.

Compensation Range: $100K - $220K