A Scalable Architecture for the Calculation of the All-to-All Ising Model Hamiltonian
NP-Hard combinatorial optimization problems offer a large class of problems difficult to solve with traditional computing technology. Recent advances in Ising model computers offer a faster solution to this class of problem. An Ising computer functions by modeling physical spin interactions. A problem mapping step creates interaction strengths for an Ising model with a ground state corresponding to the problem solution. Allowing the model to anneal to its ground state then produces a solution. Many systems for the annealing state have been developed, including traditional static CMOS circuits, analog transistor solutions, and quantum computers. All of these methods require the repeated computation of the current state of the model to determine when a ground state of the system has been reached. Measuring the state of the model requires two vector-matrix products. Increasing the complexity of a problem that can be mapped to the Ising computer increases the size of this matrix. Ising models that allow interactions between every spin (All-to-All) are also desirable for problem mapping, however this requires that the matrix in the spin calculation is a full symmetric matrix instead of the sparse matrix produced by other spin interaction models. Thus an efficient and scalable method for calculating this sum is desirable. By parallelizing computation and exploiting constraints in the matrix and vectors, an efficient architecture is developed for fast calculator of an Ising model state.