Binary neural network hardware apparatus
AssigneeElectronics and Telecommunications Research Institute
InventorsSung Min Hwang, Dong Woo Suh, Wang Joo Lee, Jeong Woo Park
Life of this patent
Estimating the term…
Abstract
Disclosed is a binary neural network hardware apparatus. The binary neural network hardware apparatus includes a sense amplifier configured to compare a bit line voltage of a bit line with a predetermined reference voltage; an input unit configured to input the bit line voltage to the sense amplifier; and a threshold voltage regulator configured to be connected to an artificial intelligence binary synapse through the bit line and change the bit line voltage in multiple levels.
Claims9
A binary neural network hardware apparatus comprising:
- a sense amplifier configured to compare a bit line voltage of a bit line with a predetermined reference voltage;
- an input unit configured to input the bit line voltage to the sense amplifier;
- and
- a threshold voltage regulator configured to be connected to an artificial intelligence binary synapse through the bit line and change the bit line voltage in multiple levels,
- wherein the threshold voltage regulator includes a multi-level synapse configured to vary conductance in multiple levels to change synthetic conductance with the artificial intelligence binary synapse.
A binary neural network hardware apparatus comprising:
- a multi-level synapse configured to vary conductance in multiple levels to change synthetic conductance with the artificial intelligence binary synapse,
- wherein the multi-level synapse includes
- a substrate;
- a gate electrode applied to the substrate;
- an Al2O3 layer applied on the substrate and the gate electrode;
- a TiO2 layer applied to the Al2O3 layer;
- a source electrode applied to be spaced apart from the TiO2 layer and the Al2O3 layer;
- and
- a drain electrode applied to be spaced apart from the TiO2 layer and the Al2O3 layer, and
- wherein the source electrode and the drain electrode are spaced apart from each other, and a gate voltage applied to the gate electrode is controlled to vary conductance in multiple levels based on electrons and holes trapped at an interface between the TiO2 layer and the Al2O3 layer.
Dependencies are inferred from the claim text.