Loading data has an energy price in a new quantum-memory model

A new study puts an energy budget around a step that can disappear in discussions of quantum algorithms: loading ordinary data into a quantum computer. The November 10 preprint models quantum random-access memory, or QRAM, which lets an algorithm query stored data while preserving a quantum address.^[1]

This is a theoretical engineering model, not a working memory or a measured electricity bill. Its favorable assumptions include one-second coherence, the time quantum information survives, and ideal refrigeration efficiency. The calculations exclude static power consumption. The authors place the coherence and refrigeration assumptions roughly a thousandfold beyond reported hardware.^[1]

A branching design called bucket-brigade uses 12 address qubits but 12,298 qubits overall. Those address qubits select among 4,096 locations; they are not a count of the whole machine. At the selected operating point, a query takes 80.2 microseconds and consumes 0.351 millijoules of dynamic energy, including modeled cooling overhead.^[1]

A 24-qubit multiplexer design, which selects data through a different sequence of operations, takes 3.02 milliseconds and uses 0.0595 millijoules. QubitWire’s calculation from those figures makes the larger design about 38 times faster, but about 5.9 times more energy-intensive per query. Their modeled fidelities, measures of how well a query preserves the intended result, also differ, so this is not an equal-accuracy efficiency contest.^[1]

The useful distinction is between making a query fast and making it cheap. For a data-intensive application, neither the address size nor the speed alone describes the supporting machine.

The next useful test is a comparison using measured hardware parameters and a complete power budget, alongside the application’s accuracy requirements.