Spike Neural Network Circuit Charge Sharing Synaptic Design
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current spike neural network circuits lack the ability to effectively mimic biological neural networks, particularly in terms of simulating synapse short-term plasticity and membrane potential dynamics, which are crucial for accurate pulse-based operations.
Innovation Solution
A spike neural network circuit comprising a weight storage, charge sharing synaptic circuit, switched capacitor circuit, voltage-to-current conversion circuit, and neuron circuit that naturally discharges synaptic voltage, mimicking biological waveforms by using capacitors and MOSFETs to manage membrane potential and output spikes based on threshold voltages.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a conventional semiconductor circuit is used to implement spike neural network operations, then the circuit can perform basic computational functions, but it fails to accurately mimic biological neural network behaviors such as synapse short-term plasticity and membrane potential dynamics
Solution Approach 1:
The patent replaces conventional voltage-based neural network circuits with a charge-based system. The weight storage uses charge to represent synaptic weights, the charge sharing synaptic circuit uses charge transfer to simulate synapse plasticity, and the membrane capacitor accumulates charge to represent membrane potential. This substitution of charge-based mechanisms for voltage-based operations enables accurate biological mimicry while maintaining semiconductor implementation.
2Reliability
If charge-based mechanisms are introduced to accurately simulate synapse plasticity and membrane potential, then biological fidelity is improved, but the circuit design and operation become more complex
Solution Approach 1:
The patent segments the neural network circuit into distinct functional blocks, each handling a specific charge-based operation: weight storage for synaptic weights, charge sharing synaptic circuit for plasticity simulation, switched capacitor circuit for timing control, voltage-to-current conversion circuit for signal transformation, and neuron circuit for spike generation. This segmentation manages complexity by localizing charge management functions to specific modules.
Solution Approach 2:
The patent introduces a membrane capacitor as an intermediary element that accumulates charge from multiple synapses and converts it to membrane potential. This intermediary enables the separation of charge accumulation (synaptic integration) from spike generation (action potential), simplifying the overall charge management while maintaining biological fidelity.
3Reliability
If the circuit uses natural discharge of synaptic voltage through switched capacitor circuits, then biological realism is enhanced, but the precision of timing and voltage control becomes more difficult to maintain
Solution Approach 1:
The patent employs periodic clock signals to control the switched capacitor circuits, which periodically charge and discharge the synaptic capacitors. This periodic action creates natural exponential decay of synaptic voltage that mimics biological synapse behavior, while the regular timing ensures precise control despite the natural discharge process.
4Measurement precision
If the neuron circuit compares membrane voltage with threshold voltage to generate output spikes, then spike generation accuracy is improved, but the circuit's ability to handle continuous membrane potential dynamics is reduced
Solution Approach 1:
The patent implements dynamic threshold comparison where the membrane capacitor continuously accumulates charge from incoming synapses, creating a dynamically changing membrane potential. The neuron circuit continuously monitors this potential against the threshold, enabling both precise spike generation and adaptability to varying input patterns. The system transitions from static voltage levels to dynamic charge accumulation and discharge processes.
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 solution enables a semiconductor-based spike neural network that replicates biological neural network behavior, including synapse short-term plasticity, ensuring consistent operation and reliable simulation of biological waveforms, thus enhancing the accuracy of pulse-based operations.
Implementation Method 1
a switched capacitor circuit that naturally discharges the generated synaptic voltage
Implementation Method 2
The current or charge amount accumulated in a membrane capacitor connected to an input of a neuron circuit forms a membrane potential
Data Source
AI summary
Disclosed is a spike neural network circuit including a weight storage that receives an input spike signal and outputs data based on a weight, a charge sharing synaptic circuit that generates a synaptic voltage based on the output data, a switched capacitor circuit that naturally discharges the generated synaptic voltage, a voltage-to-current conversion circuit that receives the synaptic voltage and generates a membrane voltage, and a neuron circuit that receives the membrane voltage and a threshold voltage and generates an output spike signal based on the received membrane voltage and the received threshold voltage.