Computing Library › AI Plasma Control
AI Plasma Control

Layer 1: Deterministic Hardware Reflexes

The fastest control layer: microsecond trip lines and interlocks implemented in fixed logic that protect the machine without waiting on any software decision.

Reflexes, not decisions

The lowest layer does not reason. It watches a small set of critical signals - coil current, quench detectors, arc detectors, out-of-range magnetics - and, when any crosses a hard threshold, it acts in microseconds by fixed rule. There is no model, no optimization, and no network round trip. The logic is simple enough to verify exhaustively and fast enough to beat the fault it guards against.

Why it must be deterministic

Kronos motion — lego machine

A superconducting magnet quench, a power-supply fault, or a plasma-facing arc can damage hardware faster than any general-purpose computer can schedule a response. Deterministic here means: bounded, known worst-case latency with no jitter from operating systems, garbage collectors, or contention. These functions live in field-programmable gate arrays (FPGAs) or hard-wired comparators, not in application code.

Typical reflex functions

The watchdog relationship

Layer 1 also watches the layers above it. If the fast feedback controller stops publishing valid commands within its deadline, the reflex layer assumes the higher layer has failed and drives actuators to a predefined safe state. This is how a slow or crashed digital twin can never endanger the machine: something simpler and faster is always ready to take the wheel.

Verification

Because these functions are safety-critical and small, they are specified formally, tested against fault-injection stimuli, and frozen. Changes are rare and reviewed as hardware changes, not software updates. The design goal is that the reflex layer is boring, predictable, and never the interesting part of an incident report.