论文信息 - A VLSI Implementation of an Analog Neural Network Suited for Genetic Algorithms

A VLSI Implementation of an Analog Neural Network Suited for Genetic Algorithms

The usefulness of an artificial analog neural network is closely bound to its trainability. This paper introduces a new analog neural network architecture using weights determined by a genetic algorithm. The first VLSI implementation presented in this paper achieves 200 giga connections per second with 4096synapses on less than 1 mm2 silicon area. Since the training can be done at the full speed of the network, several hundred individuals per second can be tested by the genetic algorithm. This makes it feasible to tackle problems that require large multi-layered networks.

Johannes Schemmel | Karlheinz Meier | Felix Schürmann

[1] Tadashi Shibata,et al. A neuron-MOS neural network using self-learning-compatible synapse circuits , 1995, IEEE J. Solid State Circuits.

[2] Johannes Schemmel,et al. Towards a Silicon Primordial Soup: A Fast Approach to Hardware Evolution with a VLSI Transistor Array , 2000, ICES.

[3] Carver A. Mead,et al. A single-transistor silicon synapse , 1996 .

[4] K. Meier,et al. A 66 × 66 pixels analog edge detection array with digital readout , 1999, Proceedings of the 25th European Solid-State Circuits Conference.

[5] R. S. Gyurcsik,et al. An analog VLSI neural network with on-chip perturbation learning , 1997, IEEE J. Solid State Circuits.

[6] Anders Krogh,et al. Introduction to the theory of neural computation , 1994, The advanced book program.

[7] Alan H. Kramer. Array-based analog computation , 1996, IEEE Micro.