Capacitor Synapse Cell Using Non-Volatile Capacitance States
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing capacitor-based synapse devices in binary neural networks suffer from reduced accuracy due to threshold voltage dispersion, leading to inefficiencies in power consumption and performance.
Innovation Solution
A capacitor device structure with a gate electrode and a first doping region, utilizing a gate insulator stack for non-volatile information storage, allows for binary operations through capacitance-based states, enhancing reliability and reducing power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If conventional capacitor-based synapse devices are used in binary neural networks, then power consumption is reduced and area is minimized, but accuracy deteriorates due to threshold voltage dispersion
Solution Approach 1:
The patent changes the fundamental operating parameter from voltage-based to capacitance-based states. By utilizing a gate insulator stack with memory function that stores charge information, the device operates in distinct capacitance states (high/low) rather than continuous voltage levels. This parameter change eliminates threshold voltage dispersion effects while maintaining binary operation, thereby improving accuracy without increasing power consumption.
2Adaptability or versatility
If conventional synapse devices with multiple weight levels are used, then network expressiveness is improved, but accuracy deteriorates due to threshold voltage distribution
Solution Approach 1:
The patent transitions from voltage-level weighting to capacitance-state weighting. The gate insulator stack memory function creates stable, discrete capacitance states that serve as weight representations. This parameter change allows for reliable binary weight storage (0 or 1) that can be combined through capacitive operations, maintaining network expressiveness while achieving high accuracy through stable state retention.
3Area of stationary object
If binary neural networks are implemented with capacitor devices, then area and power are reduced by removing multipliers, but reliability deteriorates due to device variability
Solution Approach 1:
The patent changes the state representation from voltage to capacitance, where the gate insulator stack memory function creates robust, binary capacitance states. These states are inherently more reliable against variability because they depend on stored charge presence/absence rather than precise voltage levels. The capacitance-based operation maintains binary simplicity for area reduction while achieving higher reliability through stable state definition.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The proposed solution improves accuracy and reduces power consumption in binary neural networks by stabilizing capacitance states, enabling efficient binary operations and high integration density.
Implementation Method 1
a capacitance between the gate electrode and the first doping region which is determined according to information stored in the gate insulator stack having a memory function
Implementation Method 2
gate insulator stack having a memory function capable of storing information in a non-volatile form
Data Source
AI summary
Provided is a capacitor device, a unit synapse using the capacitor device, a synapse array using the unit synapses. The capacitor device comprises a semiconductor layer which include first and second doping regions formed to be spaced apart from each other and a body region formed between the first and second doping regions; a gate electrode provided above the body region; and a gate insulator stack to have a memory function and disposed between the gate electrode and the semiconductor layer. The capacitance between the gate electrode and the first doping region is determined according to information stored in the gate insulator stack, and the state of the capacitor device is determined according to the capacitance to be one of two preset states. The unit synapse comprises a pair of capacitor devices to perform an XNOR operation.