Hardware annealing on DT-CNN using CAM<sup>2</sup>

Abstract

This is a feasibility study of the implementation of discrete time cellular neural network (DT-CNN) annealing on Cellular AutoMata on Content Addressable Memory (CAM <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> ). CAM <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> is a dedicated hardware for cellular automata (CA) and DT-CNN. We propose an annealing method on DT-CNN to solve quadratic assignment problems. This method uses the noise generated by chaotic behavior of class 3 CA. Since CA can be implemented on CAM <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> easily, our proposed method is suitable for hardware implementation. In this paper we evaluate the performance of the hardware annealing. Our experimental results show the network with the CA noise tends to one particular solution under some condition. We also evaluate how the hardware restrictions of CAM <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> affect on the annealing performance. In spite of the hardware restrictions, our experimental results show the hardware annealing can be performed on the existent implementation of the CAM <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> .