Has Google or the TensorFlow team developed — or are they currently developing — any mature and specialized framework for solving differential equations or performing PDE-based scientific simulations (like PINNs)?

I’m looking for something comparable to NVIDIA’s Modulus/PhysicsNeMo, which provides symbolic PDE definitions, domain abstractions, and support for neural operators (FNO, FourierNet, etc.).
If there’s anything official (or even experimental) being worked on inside Google or by the TensorFlow team, I’d love to know.