Synapse weight accuracy of analog neuro chip
Taiki Kimura, T. Shima · 2005
This paper presents experimental results concerning the influence of synapse weight accuracy and the number of hidden layer neurons during learning for analog neuro chip. A simplified problem of distinguishing the letters T and C was adapted for this purpose. Convergency property is more sensitive to the number of hidden layer neurons than synapse weight accuracy. A solution of synapse weight values does not require such high accuracy as synapse weights require during the learning process. These experimental results provide information significant for design of large-scale network size neural network chips.