Title :
A VLSI neuroprocessor for dynamic assignment of resources
Author :
Eberhardt, Silvio ; Daud, Taher ; Thakoor, Anil
Author_Institution :
Jet Propulsion Lab., California Inst. of Technol., Pasadena, CA, USA
Abstract :
An analog processor for dynamic assignment of resources is proposed. The neural-network-inspired processor consists of a matrix of processing elements where columns are associated with resources and rows with consumers. Analog cost-of-association values are programmed into each element. Each row and column is overseen by a winner-take-all network that serves to enforce input and output blocking constraints by dynamically controlling the number of active elements in that row or column. After the processor has settled on a solution, the pattern of active and inactive elements gives the required assignment configuration. The architectural issues, the innovations and optimization introduced, and the optimization processor are discussed. Simulation results show that for a 64-resource problem, on the average, an optimal or near optimal solution can be found in a few hundred microseconds. This is orders of magnitude faster than solutions obtained by sequential computing technology
Keywords :
VLSI; linear integrated circuits; neural nets; optimisation; parallel architectures; resource allocation; analogue VLSI neuroprocessor; dynamic resource assignment; input and output blocking constraints; neural nets; optimization; simulated annealing; winner-take-all network; Computational modeling; Computers; Costs; Integrated circuit interconnections; Neural network hardware; Neural networks; Neurons; Resource management; Space technology; Very large scale integration;
Conference_Titel :
Systems, Man and Cybernetics, 1990. Conference Proceedings., IEEE International Conference on
Conference_Location :
Los Angeles, CA
Print_ISBN :
0-87942-597-0
DOI :
10.1109/ICSMC.1990.142212