Synaptic memory and memory array using Fowler-Nordheim timers
An analog memory device includes a first node and a second node. The first node includes a first floating gate, a second floating gate, and a capacitor. The first node first floating gate is connected to the first node second floating gate via the capacitor. The second node includes a first floating gate, a second floating gate, and a capacitor. The second node first floating gate is connected to the second node second floating gate via the capacitor. The second node is connected to the first node, and an analog state of the first node and an analog state of the second node continuously and synchronously decay with respect to time.
1 . An analog memory device comprising:
a first node comprising a first floating gate, a second floating gate, and a capacitor, where the first floating gate is connected to the second floating gate via the capacitor; and
a second node comprising a first floating gate, a second floating gate, and a capacitor, where the first floating gate is connected to the second floating gate via the capacitor, and the second node is connected to the first node, and wherein an analog state of the first node and an analog state of the second node continuously and synchronously decay with respect to time, wherein the first node and the second node are configured such that a charge applied the first or second floating gate of the first or second nodes modifies a shape of a barrier of the associated first or second floating gate to enable Fowler-Nordheim quantum tunneling of electrons through the barrier.
2 . The analog memory device of claim 1 , wherein the first node is operable for a SET operation and the second node is operable for a RESET operation.
3 . The analog memory device of claim 1 , wherein the first floating gates of the first and second nodes are tunneling gates and the second floating gates of the first and second nodes are readout gates.
4 . The analog memory device of claim 1 , wherein the charge is applied by a combination of tunneling and hot electron injection.
5 . The analog memory device of claim 1 , wherein the first node and the second node are configured such that application of a voltage difference between the first and second floating gates of the first node and the second node prevents hot electron injection onto the second floating gates of the first node and the second node during readout operation.
6 . The analog memory device of claim 1 , wherein the analog memory device is one memory device of a plurality of analog memory devices configured to implement machine learning or neural network training.
7 . A memory array comprising a plurality of analog memory devices of claim 1 .
8 . An adaptive synaptic array comprising:
a plurality of analog memory devices driven by Fowler-Nordheim quantum tunneling, wherein each analog memory device of the plurality of the analog memory devices comprises:
a first node comprising a first floating gate, a second floating gate and a capacitor, where the first floating gate is connected to the second floating gate via the capacitor; and
a second node comprising a first floating gate, a second floating gate and a capacitor, where the first floating gate is connected to the second floating gate via the capacitor, and the second node is connected to the first node, and wherein an analog state of the first node and an analog state of the second node continuously and synchronously decay with respect to time, wherein the first node and the second node of each analog memory device are configured such that a charge applied the first or second floating gate of the first or second nodes modifies a shape of a barrier of the associated first or second floating gate to enable Fowler-Nordheim quantum tunneling of electrons through the barrier.
9 . The adaptive synaptic array of claim 8 , wherein the first node of each analog memory device is operable for a SET operation and the second node of each analog memory device is operable for a RESET operation.
10 . The adaptive synaptic array of claim 8 , wherein the first floating gates of the first and second nodes of each analog memory device are tunneling gates and the second floating gates of the first and second nodes are readout gates.
11 . The adaptive synaptic array claim 8 , wherein the charge is applied by a combination of tunneling and hot electron injection.
12 . The adaptive synaptic array claim 8 , wherein the first node and the second node of each analog memory device are configured such that application of a voltage difference between the first and second floating gates of the first node and the second node prevents hot electron injection onto the second floating gates of the first node and the second node during readout operation.
13 . The adaptive synaptic array of claim 8 , wherein the analog memory devices are configured to implement machine learning or neural network training.
14 . A method of machine learning using an adaptive synaptic array comprising a plurality of analog memory devices, each analog memory device including a first node comprising a first floating gate, a second floating gate and a capacitor, where the first floating gate is connected to the second floating gate via the capacitor, and a second node comprising a first floating gate, a second floating gate and a capacitor, where the first floating gate is connected to the second floating gate via the capacitor, and the second node is connected to the first node, and wherein an analog state of the first node and an analog state of the second node continuously and synchronously decay with respect to time, and the first node and the second node of each analog memory device are configured such that a charge applied the first or second floating gate of the first or second nodes modifies a shape of a barrier of the associated first or second floating gate to enable Fowler-Nordheim quantum tunneling of electrons through the barrier the method comprising:
selecting a network to be trained, the selected network having a plurality of learnable parameters;
assigning each learnable parameter to a different analog memory device of the adaptive synaptic array;
storing a weight for each learnable parameter in its assigned analog memory device through application of a charge to one or both of the first and second nodes to modify the shape of the barrier of the associated first or second floating gate of the one or both of the first and second nodes; and
updating the weight of each learnable parameter in its assigned memory device in each training iteration of the selected network.
15 . The method of claim 14 , further comprising:
allowing the updated weight of each learnable parameter stored in its assigned memory device to decay for a period of time; and
mapping the decayed weights into the network.
16 . The method of claim 14 , further comprising repeating the updating, allowing, and mapping until a desired number of training iterations has been completed.