A model of neural circuits for programmable VLSI implementation

A new model for neural circuits is introduced which has qualitatively the same dynamic properties as gradient continuous-time feedback neural nets. This model (i) reduces the maximum number of connections to n(n+1)/2, (ii) does not suffer from the synaptic weight problem, i.e. the problem of implementing variable linear resistive elements in large scale, and (iii) is implementable via all MOS elements. Hence, it lends itself naturally to analog MOS VLSI implementation.<
>