Adding external random phase noise might help the solver converge on a more optimal solution by preventing it from getting stuck in local energy minima. It also may enable new features involving measuring the second moment of the converged system under noise, as discussed here: https://arxiv.org/pdf/2308.05660.pdf
This feature will likely be implemented with a "noise injection" module that sits in between the output wires from the right/top of the core matrix and the input wires to the left/bottom of the core matrix. I think we'd have a line of coupled cells, with one of the the coupled cell inputs being fed from an LFSR rather than a ring oscillator. The coupling strength between the spins and the noise should be programmable.
zachbe
commented
7 months ago
Adding external random phase noise might help the solver converge on a more optimal solution by preventing it from getting stuck in local energy minima. It also may enable new features involving measuring the second moment of the converged system under noise, as discussed here: https://arxiv.org/pdf/2308.05660.pdf
This feature will likely be implemented with a "noise injection" module that sits in between the output wires from the right/top of the core matrix and the input wires to the left/bottom of the core matrix. I think we'd have a line of coupled cells, with one of the the coupled cell inputs being fed from an LFSR rather than a ring oscillator. The coupling strength between the spins and the noise should be programmable.