Mapping the Stack to Hyperion (Breeder)
Layer by layer, how the eight-layer architecture instantiates on the D-T spherical tokamak — from FPGA coil control to neutronics retraining.
The stack on the breeder
On Hyperion the architecture takes on the tokamak's specific physics while keeping every layer's role unchanged. The mapping shows how a general stack becomes a concrete control system for a D-T spherical tokamak at Q_sci 3.076 and 85.0 MW.
Layer by layer
- L0 — Monte Carlo neutronics for tritium breeding across TBR 1.1/1.5/1.8; shape-optimization sweeps for delta -0.30.
- L1 — FPGA control of shaping and vertical-stability coils; quench protection on the 16.84 T magnets.
- L2 — Thomson, Mirnov, flux loops, ECE, REBCO strain, and neutron flux fused into equilibrium features.
- L3 — PINNs solving Grad-Shafranov; anomaly ensembles for disruption precursors; MPC for shape.
- L4-L7 — orchestration, the plasma copilot, control-room overlays, and fuel-cycle scheduling.
The dominant loop
The breeder's signature loop is equilibrium-and-disruption: diagnostics reconstruct the equilibrium, the twin projects it 50-100 ms ahead, the MPC agent plans coil adjustments to hold the negative-triangularity shape and stay clear of disruption boundaries, and L1 executes deterministically within the command budget.
The fuel-cycle overlay
Unique to the breeder is real-time isotope balancing. The fuel cycle produces roughly a 4 kg/yr tritium class output and 1.97 kg/yr helium-3 alongside 14 MeV neutrons. L2 tracks the isotope inventory and L7's operations functions schedule around it — as a logistics problem, never an economic one.
Contrast the burner mapping in mapping to the tandem mirror; the physics is enumerated in breeder control problems.