초록 ▼

A means and method for training an electronic network of analog cells. A generic universal programmer interface (GUPI) is provided to enable connection of an adaptor to a host computer via a personal computer personal programmer (PCPP) for training an ETANN chip. An opening in the top of the GUPI pr

A means and method for training an electronic network of analog cells. A generic universal programmer interface (GUPI) is provided to enable connection of an adaptor to a host computer via a personal computer personal programmer (PCPP) for training an ETANN chip. An opening in the top of the GUPI provides for installation of the adaptor. The adaptor of the present invention plugs into the GUPI module to provide for connecting either an ETANN chip or a Confidence Test Module (CTM) for access by the user via the host computer. A target socket located on the top surface of the adaptor allows the user to physically connect the ETANN chip or the CTM to adapter of the present invention. The adaptor provides the resources needed for training an ETANN in several modes including: digital to analog conversion for driving the ETANN\s analog input pins; a single analog to digital converter with an 80 channel analog multiplexer for measuring the ETANN\s analog outputs; neuron perturbation circuitry; synapse current measurement circuitry; neuron sum measurement circuitry; VPP management circuitry; a binary searching synapse weight measurement circuit; isolation switches and the logic necessary to connect all these resources to the ETANN and the GUPI base. The present invention also includes host computer software necessary to control the operation of the adapter.

대표청구항 ▼

A device for training a non-volatile neural network, said non-volatile neural network having a plurality of synapses and a plurality of neurons, outputs of said synapses being coupled to inputs of said neurons, said device comprising: a host computer having a personal computer personal programmer (P

A device for training a non-volatile neural network, said non-volatile neural network having a plurality of synapses and a plurality of neurons, outputs of said synapses being coupled to inputs of said neurons, said device comprising: a host computer having a personal computer personal programmer (PCPP) interface; a generic universal programmer interface (GUPI) coupled to said PCPP interface of said host computer; an adapter coupled to said GUPI, said neural network being coupled to said adapter, said adapter including: (a) means for manipulating control lines in said neural network to select a neuron of said plurality of neurons, (a) means for manipulating control lines in said neural network to select a synapse of said plurality of synapses, (b) means for converting digital signals received from said host computer to an analog level for training said selected synapse of said neural network, (c) means for setting a weight of said selected synapse to said analog level, (d) means for converting analog levels received from said selected synapse of said neural network to a digital signal for routing to said host computer, (e) means for measuring said weight existing in said selected synapse; and host computer software operably disposed within said host computer for controlling the operation of said adapter.