michael_nielsen

More prosaically, the classical control hardware scales linearly, but currently still has a high per-qubit cost. But it has come down from a 19" rack full per qubit, to an FPGA or two in the last decade.

Martinis's group (I think) even published elements of their FPGA design as open source, if I recall, but I couldn't dig that up, either.

Moreover, while the idea of multiplexing control of qubits on-chip is an automatic thought, it's hard due to the fact that many semiconductor circuits don't work at the millikelvin temps the quantum devices prefer. But people are working that problem, as well.

Sorry I'm not more help on this area, it's just outside my usual trodden path. Let's leave it here.

So, c. building superpositions from classical data. Without some solution to this problem, quantum systems will be limited to problems with a small problem description but large computational space, such as e.g. chess/shogi/go or small molecule simulation.

In particular, quantum computing isn't going to work well for image recognition or any machine learning tasks with dataset-based learning until we fix this (but see caveat a little later).

We want to take a big pile of classical data, put in a quantum superposition as an address, and receive as output a superposition of all of the data.