Artificial neuron synaptic weights implemented with variable dissolvable conductive paths
A low-power, controllable, and reconfigurable method to control weights in model neurons in an Artificial Neural Network is disclosed. Memristors are utilized as adjustable synapses, where the memristor resistance reflects the synapse weight. The injection of extremely small electric currents (a few nanoamperes) in each cell forces the resistance to drop abruptly by several orders of magnitudes due to the formation of a conductive path between the two electrodes. These conductive paths dissolve as soon as the current injection stops, and the cells return to their initial state. A repeated injection of currents into the same cell results in an almost identical effect in resistance drop. Different, stable resistance values in each cell can be controllably achieved by injecting different current values.
1. A system comprising:
a physical neural network comprising a plurality of neurons, wherein each neuron of the plurality of neurons includes circuitry having at least one memristor and at least one input connection receiving an input signal, wherein the at least one memristor is a nonlinear resistor having a resistance that changes with relation to an electrical current passing through the at least one memristor and the at least one input connection is associated with at least a first synaptic weight, wherein the first synaptic weight comprises a multiplicative gain applied to a first input signal received by the at least one input connection; and
a controller coupled to the physical neural network, the controller configured to execute instructions for:
receiving commands, wherein each command identifies a target resistance value; determining, for each command, a current value corresponding to the target resistance value; and
applying, for each command, the determined current value to the at least one memristor, wherein the at least one memristor is pre-formed;
and wherein the controller is configured to:
access a memristor calibration table, the memristor calibration table including at least one memristor data set, the memristor data set including at least one pair of calibration values, the pair of calibration values including an applied current and a measured resistance; and
use the memristor calibration table to perform the process of determining, for each command, a current value corresponding to the target resistance value.
2. The system of claim 1 , wherein the at least one input connection is associated with at least a first synaptic decay constant, and the first synaptic decay constant comprises an exponential decay time constant applied to the first input signal.
3. The system of claim 1 , wherein the first synaptic weight is determined by a first resistance value of the at least one memristor.
4. The system of claim 1 , wherein the memristor comprises a conductive bridge random access memory element, a resistive random access memory element, or a phase change memory element.
5. The system of claim 1 , wherein applying the determined current value to the at least one memristor creates a temporary conductive path within the at least one memristor, and the temporary conductive path is formed between a top electrode and a bottom electrode of the at least one memristor, the temporary conductive path being formed from a material including metallic positive ions, or a dielectric material manufactured with chalcogenide, tantalum oxide, silicon dioxide, or hafnium oxide.
6. A system, comprising:
a physical neural network including a plurality of neurons, wherein each neuron of the plurality of neurons includes circuitry having at least one memristor, and the at least one memristor is a nonlinear resistor having a resistance that changes with relation to an electrical current passing through the at least one memristor; and
wherein an electrical current passing through the at least one memristor is limited such that the memristor is operated according to a temporary dissolvable conductive path method, such that the current value is limited to be below a known forming threshold in order to avoid creating a conductive filament within the memristor, and such that the memristor resistance returns to a baseline value upon cessation of the electrical current; and
a controller coupled to the physical neural network, the controller configured to execute instructions for;
receiving a command identifying a target resistance value;
determining a current value corresponding to the target resistance value; and
applying the determined current value to the at least one memristor;
accessing a memristor calibration table, the memristor calibration table including at least one memristor data set, wherein the memristor data set includes at least one pair of calibration values, the pair of calibration values including an applied current and a measured resistance; and
using the memristor calibration table to perform the process of determining, for each command, a current value corresponding to the target resistance value in the at least one memristor.
7. The system of claim 6 , wherein each neuron in the plurality of neurons further comprises at least a first input connection associated with at least a first synaptic decay constant, wherein the first synaptic decay constant is associated with an exponential decay time constant applied to the first input signal.
8. The system of claim 6 , wherein at least a first input connection of at least a first neuron of the plurality of neurons is associated with at least a first synaptic weight, wherein the first synaptic weight includes a multiplicative gain applied to a first input signal received by the first input connection; and
wherein the first synaptic weight is determined by a first resistance value of the at least one memristor.
9. The system of claim 6 , wherein the at least one memristor includes a conductive bridge random access memory element, a resistive random access memory element, or a phase change memory element.
10. The system of claim 6 , wherein applying an electrical current value to the at least one memristor creates a temporary conductive path within the at least one memristor, and the temporary conductive path is formed between a top electrode and a bottom electrode of the at least one memristor, the temporary conductive path being formed from a material including metallic positive ions or a dielectric material manufactured with chalcogenide, tantalum oxide, silicon dioxide, or hafnium oxide.
11. A system, comprising:
a physical neural network including a plurality of neurons, each neuron of the plurality of neurons comprising circuitry having at least one memristor and at least one input connection receiving an input signal, wherein the at least one memristor is a nonlinear resistor having a resistance that changes with relation to an electrical current passing through the at least one memristor and at least a first input connection of at least a first neuron of the plurality of neurons is electrically coupled to the at least one memristor, such that the at least one memristor at least partially determines one or more electrical properties of the first input connection; and
a controller coupled to the physical neural network, the controller configured to execute instructions for:
receiving a command identifying a target resistance value;
determining a current value corresponding to the target resistance value; and
applying the determined current value to the at least one memristor,
wherein the at least one memristor is preformed;
and wherein the controller is configured to:
access a memristor calibration table, the memristor calibration table including at least one memristor data set, the memristor data set including at least one pair of calibration values, the pair of calibration values including an applied current and a measured resistance; and
use the memristor calibration table to perform the process of determining, for each command, a current value corresponding to the target resistance value in the at least one memristor.
12. The system of claim 11 , wherein the current value is determined such that the memristor is operated according to a temporary dissolvable conductive path method, such that the current value is limited to be below a known forming threshold and is insufficient to create a conductive filament within the memristor, and such that the memristor resistance returns to a baseline value upon cessation of the current.
13. The system of claim 11 , wherein at least the first input connection further comprises at least a first synaptic decay constant, wherein the first synaptic decay constant comprises an exponential decay time constant applied to the first input signal.
14. The system of claim 11 , wherein the memristor includes a conductive bridge random access memory element, a resistive random access memory element, or a phase change memory element.
15. The system of claim 11 , wherein applying the determined current value to the at least one memristor creates a temporary conductive path within the at least one memristor, and the temporary conductive path is formed between a top electrode and a bottom electrode of the at least one memristor, the temporary conductive path being formed from a material including metallic positive ions or a dielectric material manufactured with chalcogenide, tantalum oxide, silicon dioxide, or hafnium oxide.