Neuromorphic architectures, actuators, and related methods
A neuromorphic architecture is formed from a laminate of non-woven carbon fiber reinforced polymer layers arranged in a plurality of different directions. A plurality of distributed nodes are formed through the laminate via transverse voids, and an encapsulant encapsulates an electrochemical fluid or gel such that the electrochemical fluid or gel may flow within the nodes and around the laminate. Electrical current flowing through the architecture creates reversible metal deposits at various nodes, depending on the path developed through the architecture, with a complexity sufficient for neuromorphic processing, and providing a writable and erasable memory. A neuromorphic actuator may be formed by combining shape memory materials with such a neuromorphic architecture, which may provide desired surface contours and/or actuations based on current in the neuromorphic architecture. Such neuromorphic architectures and actuators may be trained according to various methods, using feed-forward and/or feedback techniques.
1 . A neuromorphic architecture, comprising:
a laminate formed of a plurality of layers of conductive fiber material, wherein each layer of the plurality of layers comprises substantially unidirectional fibers, wherein a respective orientation of the unidirectional fibers of each respective layer of the plurality of layers is different from each other respective orientation of the unidirectional fibers of adjacent respective layers of the plurality of layers, wherein the plurality of layers comprises:
an uppermost layer of the plurality of layers forming an upper surface of the laminate;
a lowermost layer of the plurality of layers forming a lower surface of the laminate; and
one or more intermediary layers of the plurality of layers sandwiched between the uppermost layer of the plurality of layers and the lowermost layer of the plurality of layers;
a plurality of distributed nodes formed through the laminate, wherein each distributed node comprises a void that extends transversely from the upper surface to the lower surface;
an electrochemical fluid comprising a plurality of metal ions;
an encapsulant configured to encapsulate at least a portion of the laminate, the plurality of distributed nodes, and the electrochemical fluid such that the electrochemical fluid is free to flow within the encapsulant and into the plurality of distributed nodes, and wherein the neuromorphic architecture is configured to perform neuromorphic processing.
2 . The neuromorphic architecture according to claim 1 , wherein the conductive fiber material comprises carbon fiber reinforced polymer material.
3 . The neuromorphic architecture according to claim 1 , wherein unidirectional fibers of a first layer of the laminate have a substantially 0 degree orientation, wherein unidirectional fibers of a second layer of the laminate have a substantially 45 degree orientation, wherein unidirectional fibers of a third layer of the laminate have a substantially 90 degree orientation, and wherein unidirectional fibers of a fourth layer of the laminate have a substantially 135 degree orientation, using a longitudinal axis of the laminate to define the 0 degree orientation.
4 . The neuromorphic architecture according to claim 1 , wherein each fiber of the unidirectional fibers of each layer of the plurality of layers of the laminate is interrupted at each distributed node, such that each respective fiber has a first terminal end on a first side of each distributed node, and a second terminal end on a second side of each distributed node, wherein metal ions from the electrochemical fluid are deposited on the first terminal ends of at least some of the unidirectional fibers when a current flows in a first direction with respect to the unidirectional fibers, and wherein metal ions from the electrochemical fluid are deposited on the second terminal ends of at least some of the unidirectional fibers when the current flows in a second direction with respect to the unidirectional fibers.
5 . The neuromorphic architecture according to claim 1 , wherein the neuromorphic architecture has an effective memory from a respective amount of metal ions deposited at each respective node of the plurality of distributed nodes, and wherein metal deposition within each node of the plurality of distributed nodes is reversible.
6 . The neuromorphic architecture according to claim 1 , wherein the plurality of distributed nodes are configured to provide nonlinear, multidimensional connectivity between the plurality of distributed nodes.
7 . The neuromorphic architecture according to claim 1 , wherein the upper surface of the laminate is embedded neuromorphic intelligence that can be trained and that dynamically responds to environmental changes without an external central processing unit, and wherein the lower surface of the laminate is embedded neuromorphic intelligence that can be trained and that dynamically responds to environmental changes without the external central processing unit.
8 . The neuromorphic architecture according to claim 1 , further comprising at least one neural interface coupled to at least one layer of the laminate, wherein the at least one neural interface comprises:
a neural input interface configured to deliver electrical current to the electrochemical fluid; and
a neural output interface configured to receive an output from the neuromorphic architecture.
9 . The neuromorphic architecture according to claim 8 , wherein the neural output interface feeds back into the neural input interface, thereby creating a feedback loop.
10 . The neuromorphic architecture according to claim 1 , comprising a plurality of laminates, wherein the neuromorphic architecture comprises one or more dielectric insulation layers, with one dielectric insulation layer positioned between each adjacent pair of laminates of the plurality of laminates, such that the neuromorphic architecture is configured for increased node density by virtue of each node comprising multiple segments due to placement of the one or more dielectric insulation layers intersecting each node.
11 . The neuromorphic architecture according to claim 1 , wherein the neuromorphic architecture is selectively configurable to be a feed-forward architecture or a feedback architecture.
12 . The neuromorphic architecture according to claim 1 , further comprising a plurality of pins, wherein a respective pin of the plurality of pins is positioned in each respective node of the plurality of distributed nodes, wherein the plurality of pins are configured such that ions are deposited on a respective pin in a respective node when current flows through the respective node, and wherein the plurality of pins are segmented such that each pin comprises a plurality of segments, with each segment of the plurality of segments being electrically insulated from each of the other segments of the plurality of segments, and wherein each respective segment corresponds to a different respective layer of the plurality of layers of the laminate.
13 . A neuromorphic actuator, comprising:
a first neuromorphic architecture, comprising:
a first laminate formed of a first plurality of layers of non-woven carbon fiber reinforced polymer (CFRP) material, the first laminate having a first upper surface and a first lower surface; and
a first plurality of distributed nodes formed through the first laminate, wherein each node of the first plurality of distributed nodes comprises a void that extends transversely from the first upper surface to the first lower surface;
a second neuromorphic architecture, comprising:
a second laminate formed of a second plurality of layers of non-woven CFRP material, the second laminate having a second upper surface and a second lower surface; and
a second plurality of distributed nodes formed through the second laminate, wherein each node of the second plurality of distributed nodes comprises a void that extends transversely from the second upper surface to the second lower surface;
an electrochemical fluid comprising a plurality of metal ions, wherein the electrochemical fluid is free to flow into the voids of the first plurality of distributed nodes and into the voids of the second plurality of distributed nodes;
a dielectric insulation layer positioned between the first neuromorphic architecture and the second neuromorphic architecture, separating the first lower surface from the second upper surface, wherein the dielectric insulation layer is configured to electrically insulate the first neuromorphic architecture from the second neuromorphic architecture;
a first shape memory alloy (SMA) layer coupled to the first upper surface of the first neuromorphic architecture; and
a second SMA layer coupled to the second lower surface of the second neuromorphic architecture.
14 . The neuromorphic actuator according to claim 13 , wherein the neuromorphic actuator is configured to actuate in response to an applied current, via different movements of the first SMA layer and the second SMA layer.
15 . The neuromorphic actuator according to claim 13 , wherein the neuromorphic actuator is configured to be trained via external forces acting on the first SMA layer and the second SMA layer, wherein the neuromorphic actuator is configured such that external forces on the first SMA layer and/or the second SMA layer results in changes in connectivity amongst the first neuromorphic architecture and the second neuromorphic architecture.
16 . A method, comprising:
actuating and/or shaping a surface by applying an electrical current to a neuromorphic actuator, wherein the neuromorphic actuator is the neuromorphic actuator according to claim 13 ;
varying the electrical current to obtain a desired contour and/or a desired movement in the surface; and
evaluating local deflection around one or more nodes of the plurality of distributed nodes, wherein the local deflection around a respective node of the plurality of distributed nodes changes its respective weight with respect to neural network pathways, and wherein the local deflection around a respective node of the plurality of distributed nodes changes a current distribution at fibers ends within the respective node.