Optoelectronic devices for Boltzmann machines and simulated annealing

An optoelectronic device for generating random numbers at high flow rates is presented. Such a device can be exploited for implementating massive arrays of stochastic decision-making elements encountered in simulated annealing and in Boltzmann machine algorithms. It is comprised of an optical random number generator and an array of electronic neuron elements. The random number arrays are derived from the continually changing modal noise of a step-index optical fiber, and the neuron elements include photodiodes and a few associated electronic components. To demonstrate the feasibility of large arrays of neuron elements, a circuit with three neurons has been implemented and tested. The results reveal that 104 neuron elements can be integrated into 1 cm2 and that flow rates of 1010 random numbers per second are possible.