Loading patent…
AssigneeInternational Business Machines Corporation
InventorGEOFFREY W. BURR
Estimating the term…
Artificial neural networks (ANNs) are a distributed computing model in which computation is accomplished with many simple processing units, called neurons, with data embodied by the connections between neurons, called synapses, and by the strength of these connections, the synaptic weights. An attractive implementation of ANNs uses the conductance of non-volatile memory (NVM) elements to record the synaptic weight, with the important multiply—accumulate step performed in place, at the data. In this application, the non-idealities in the response of the NVM such as nonlinearity, saturation, stochasticity and asymmetry in response to programming pulses lead to reduced network performance compared to an ideal network implementation. A method is shown that improves performance by periodically inverting the polarity of less-significant signed analog conductance-pairs within synaptic weights that are distributed across multiple conductances of varying significance, upon transfer of weight information between less-significant signed analog conductance-pairs to more-significant analog conductance-pairs.
A method to mitigate undesired contributions to weight induced by fixed device asymmetry in an artificial neural network (ANN), a synaptic weight of each synapse in the ANN being represented by a weighted current flow from multiple conductance-pairs, where a plurality of training examples are serially input to the ANN while observing its output, the method comprising:
A method implemented in an artificial neural network (ANN), the ANN comprising a plurality of neurons arranged in layers with the outputs of one layer connected to the inputs of a plurality of the neurons of a subsequent layer, where neurons are connected to each other via a plurality of synapses, each of the synapses having a synaptic weight encoding a connection strength between two connected neurons, the magnitude of the synaptic weight of each synapse being represented by a weighted current flow from multiple conductance-pairs, where a plurality of training examples are serially input to the ANN while observing its output, where a backpropagation algorithm updates the synaptic weight in response to the difference between the output from a given layer and a desired output from the given layer, the method comprising:
A method implemented in an artificial neural network (ANN), the ANN comprising a plurality of neurons arranged in layers with the outputs of one layer connected to the inputs of a plurality of the neurons of a subsequent layer, where neurons are connected to each other via a plurality of synapses, each of the synapses having a synaptic weight encoding a connection strength between two connected neurons, the magnitude of the weight of each of the synapses represented by a weighted current flow from multiple conductance pairs, each of the multiple conductance pairs representing a joint contribution and having a higher-significance conductance-pair and a lower-significance conductance-pair, where a plurality of training examples are serially input to the ANN while observing its output, where a backpropagation algorithm updates the synaptic weight in response to a difference between the output from a given layer and a desired output from the given layer, the method comprising:
The method of claim 21, wherein one of the two conductances in the conductance-pair of lower significance is reconfigured from a first shared conductance to a second shared conductance.
A method implemented in an artificial neural network (ANN), the ANN comprising a plurality of neurons arranged in layers with the outputs of one layer connected to the inputs of a plurality of the neurons of a subsequent layer, where neurons are connected to each other via a plurality of synapses, each of the synapses having a synaptic weight encoding a connection strength between two connected neurons, the magnitude of the weight of each of the synapses represented by a weighted current flow from multiple conductance pairs, each of the multiple conductance pairs representing a joint contribution and having a higher-significance conductance-pair and a lower-significance conductance-pair, where a plurality of training examples are serially input to the ANN while observing its output, where a backpropagation algorithm updates the synaptic weight in response to a difference between the output from a given layer and a desired output from the given layer, the method comprising:
Dependencies are inferred from the claim text.