GPUTopOpt: GPU Topology Optimization in Julia for Large-Scale Structural Design

The code implements the SIMP (Solid Isotropic Material with Penalization) formulation with Wang–Lazarov–Sigmund Heaviside projection, solved through a matrix-free multigrid-preconditioned conjugate gradient (PCG) elasticity solver. Design updates use the Method of Moving Asymptotes (MMA).

An optional porous / local-volume infill extension based on the Wu–Aage–Sigmund p-norm formulation is included for lattice and bio-inspired structures (validated on a femur reconstruction case at ~77 M DOFs). The pipeline is fully voxel-based: STL design domains are automatically voxelized; results are exported as STL meshes ready for downstream FEM validation or additive manufacturing. CPU/GPU tiled assembly minimises GPU memory pressure, enabling problems that would otherwise exceed VRAM.