* the travelling salesman problem is NP hard, not NP complete (i.e. verification of the optimal solution is not in P).
* An Ising machine is not a turing machine, AFAIU, and it may not be turing complete (but not sure about that). AFAICS that machine merely implements search for an approximate optimum on some quadratic function of discretely valued variables. Maybe similar or even equivalent to the closest lattice point problem (https://en.wikipedia.org/wiki/Lattice_problem#Closest_vector...).
One could optimistically say, that this is a very special purpose machine. Pessimistically I would say this is the only remotely usable "computation" that can be implemented when you only have unreliable (non-deterministic) components. So the fact that they did not implement a proper turing-complete circuit may be more a bug than a feauture.
[edit] one may view their chip as a very fast hardware-accelerated implementation of a very simple (i.e. slow) algorithm.
[edit2] Obviously, if the Ising machine is capable to find an exact solution to an arbitrary NP-hard optimization problem, then that machine must be turing complete (but still may take exponential time even for problems in P). However, with an imperfect Ising machine (that does not implement perfect annealing or retains an error rate even at lowest temperature) it is pretty difficult to assess turing completeness.
* the travelling salesman problem is NP hard, not NP complete (i.e. verification of the optimal solution is not in P).
* An Ising machine is not a turing machine, AFAIU, and it may not be turing complete (but not sure about that). AFAICS that machine merely implements search for an approximate optimum on some quadratic function of discretely valued variables. Maybe similar or even equivalent to the closest lattice point problem (https://en.wikipedia.org/wiki/Lattice_problem#Closest_vector...).
One could optimistically say, that this is a very special purpose machine. Pessimistically I would say this is the only remotely usable "computation" that can be implemented when you only have unreliable (non-deterministic) components. So the fact that they did not implement a proper turing-complete circuit may be more a bug than a feauture.
[edit] one may view their chip as a very fast hardware-accelerated implementation of a very simple (i.e. slow) algorithm.
[edit2] Obviously, if the Ising machine is capable to find an exact solution to an arbitrary NP-hard optimization problem, then that machine must be turing complete (but still may take exponential time even for problems in P). However, with an imperfect Ising machine (that does not implement perfect annealing or retains an error rate even at lowest temperature) it is pretty difficult to assess turing completeness.