The Semantics of Entity Component System

Entity Component System (ECS) architecture has become a widely adopted foundation for large-scale simulation and game-engine design because it separates identity, state, and behavior in a form that supports data-oriented execution. Among its variants, archetype-based ECS is especially effective because entities with identical component sets are organized into dense columnar tables, enabling regular iteration and efficient memory access. Despite its practical success, the semantics of archetype-based ECS remain only partially formalized, and it is still unclear to what extent its execution discipline can be preserved beyond CPU-oriented runtimes. This paper studies archetype-based ECS from both semantic and architectural perspectives. First, it formalizes the core mechanisms of archetype ECS, including entity creation, component association, system execution, deferred structural mutation, and archetype migration, and presents a type system that prevents unsafe access to archetype storage during execution. Second, it investigates how this model can be extended to GPU-resident execution. To this end, we implement a reference CPU realization and a GPU-resident realization using a Tower Defense simulation. The GPU design preserves stable table iteration, query-driven execution, and deferred structural mutation, while adapting storage and synchronization to device-resident execution through GPU SoA buffers, device-side cardinality tracking, and explicit reconciliation passes. Experimental results show that the GPU-resident realization achieves substantially higher steady-state throughput than the CPU baseline, with approximately 6.4-6.81× speedup across two stress scenarios and low run-to-run variability. These results indicate that the essential execution discipline of archetype-based ECS can be preserved in a constrained GPU-resident setting and can provide a practical foundation for high-throughput simulation.