Traditional computing separates memory from processing. Data is stored in one place, fetched to another, transformed, and written back. The fetch operation is the bottleneck — the “memory wall” that limits performance regardless of how fast the processor runs.
Physical selection systems — lasers reaching threshold, Ising machines settling into ground states, mode-locked oscillators choosing a phase — collapse this separation (arXiv:2603.17049). When a laser is pumped above threshold, it “selects” a mode to oscillate in. The selection is a computation: the system evaluates a landscape of possible modes and settles on the one with lowest loss. The selection is also a memory access: the system's attractor structure encodes prior information, and the mode it selects is the one that matches the stored pattern.
The key insight is that these systems produce repeatable signatures across trials. Run a laser to threshold multiple times with the same pump configuration, and it selects the same mode — not approximately but with high fidelity. This repeatability turns the attractor into a key: the physical state of the system (the pump pattern) maps to a specific output (the selected mode) through the dynamics of the attractor landscape, just as a memory address maps to stored data through the circuitry of a RAM chip.
But the attractor-keyed memory does something RAM cannot: it combines retrieval and decision into one event. The laser does not first retrieve a stored mode and then decide which one to amplify. It does both simultaneously, in the same physical process. The energy that selects the mode is the same energy that produces the output. There is no fetch step because there is no separation between the stored information and the computation that uses it.
The memory is the computation. The decision is the recall. The wall is gone because there was never a wall — only two operations pretending to be different.