Gaussian Synapse Device for Probabilistic Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional artificial neural networks (ANNs) face challenges in energy and size scaling, slow learning, incremental adaptation, and false convergence, limiting complexity scaling in brain-inspired neuromorphic computing.
Innovation Solution
A Gaussian synapse device using heterostructures of atomically thin two-dimensional materials like MoS2 and black phosphorus in dual-gated field-effect transistors, enabling probabilistic computational devices for hardware implementation of statistical neural networks with tunable amplitude, mean, and standard deviation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If traditional von Neumann architecture is used for artificial neural networks, then computational functionality is achieved, but energy consumption increases and size scaling deteriorates
Solution Approach 1:
The patent replaces the traditional von Neumann architecture with a neuromorphic architecture that uses memristive devices to emulate synaptic behavior. This substitution eliminates the separation between storage and processing, allowing analog computation to occur directly in the memory devices, thereby reducing energy consumption and improving scalability.
Solution Approach 2:
The patent utilizes the resistive switching characteristics of memristive devices to dynamically change their resistance states, which represent synaptic weights. By programming these resistance values, the system can implement learning and adaptation functions, enabling energy-efficient analog computation that scales better than traditional digital architectures.
2Device complexity
If deterministic digital logic or analog devices are used to emulate neural functions, then complexity scaling is achieved, but energy efficiency deteriorates
Solution Approach 1:
The patent replaces deterministic digital logic with probabilistic computing elements that operate in the subthreshold regime. These devices use random telegraph noise and stochastic switching to perform probabilistic inference, which is inherently more energy-efficient than deterministic computation while maintaining the ability to scale in complexity.
Solution Approach 2:
The patent operates computing devices in the subthreshold voltage regime, where transistors operate with gate voltages below the threshold voltage. This parameter change enables ultra-low power operation while still achieving functional computation through probabilistic mechanisms, resolving the trade-off between complexity and energy efficiency.
3Productivity
If conventional neural network devices are used, then computational capability is achieved, but area efficiency deteriorates
Solution Approach 1:
The patent merges multiple functions into single devices: memristive elements simultaneously perform storage, computation, and synaptic emulation functions. This consolidation eliminates the need for separate processing and memory units, dramatically reducing the area required per computational operation while maintaining high computational capability.
Solution Approach 2:
The patent designs universal neuromorphic devices that can perform multiple neural network operations including synaptic weight storage, forward propagation, and learning updates. This multi-functionality allows a single device type to replace what would traditionally require multiple specialized components, improving area efficiency.
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 Gaussian synapse device improves area and energy efficiency, enabling seamless classification of brainwave patterns and facilitating complexity scaling without compromising energy and size scaling, while operating in a low-power subthreshold regime.
Implementation Method 1
The device has two channels formed between the source contact and the drain contact. The source contact connects to a p-type channel. The drain contact connects to an n-type channel. The p-type and n-type channels are connected in series.
Data Source
AI summary
Embodiments relate to a Gaussian synapse device configured so that its transfer characteristics resemble a Gaussian distribution. Embodiments of the Gaussian synapse device include an n-type field-effect transistor (FET) and p-type FET with a common contact so that the two FETs are connected in series. Some embodiments include a global back-gate contact and separate top-gate contact to obtain dual-gated FETs. Some embodiments include two different 2D materials used in the channel to generate the two FETs, while some embodiments use a single ambipolar transport material. In some embodiments, the dual-gated structure is used to dynamically control the amplitude, mean and standard deviation of the Gaussian synapse. In some embodiments, the Gaussian synapse device can be used as a probabilistic computational device (e.g., used to form a probabilistic neural network).