What I could gather from it is that this is just an implementation of archetype-based ECS. Which just means that entities with the same set of components (that is, the same "archetype") are allocated together.
But I don't know whether they are patenting archetypes in general or just a narrow usage.
Now I have a question: is it possible that things published before the date of the patent (24 march 2020) are infringing on the patent? Or, on the contrary, things published before this date can actually be prior art?
Because, there are Rust ECS libraries that use archetypes, like for example legion https://crates.io/crates/legion/0.1.0 - that had its first release on 9 march 2019. And even at this date, the concept of using archetypes to optimize the memory layout of ECS applications is not novel. More about this here
https://csherratt.github.io/blog/posts/specs-and-legion/
I would say that the google v8's Javascript optimization, where they create a new struct layout for each possible combination of object fields, might as well count as prior art. It doesn't use ECS though: but I don't think that prior art needs to check all technology boxes. I argue that, in light of v8's trick to turn objects with dynamic fields (traditionally stored as a hash table) into objects with statically known fields, then ECS archetypes are not novel enough for a patent.
(And I don't even know if the v8 optimization is itself novel)