Spiking Neuron Cutoff Circuit for Refractory-Period Signal Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional spiking neural networks implemented on semiconductor chips experience information loss due to the failure in transmitting synaptic current during the membrane potential's refractory period, leading to inaccuracies when performing arithmetic operations compared to digital operation circuits.
Innovation Solution
A neural network device with synapse and neuron circuits that utilize charge accumulation and cutoff mechanisms to manage synaptic currents, ensuring accurate transmission and reducing information loss by controlling synaptic current flow during a refractory period.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If conventional spiking neural networks are implemented on semiconductor chips, then energy consumption is reduced compared to digital operation circuits, but information loss occurs due to failure in transmitting synaptic current during the refractory period
Solution Approach 1:
The cutoff circuit proactively stops the supply of synaptic current to the charge accumulation circuit during the refractory period before information loss can occur. This preliminary action prevents the problematic accumulation of charge during periods when the neuron cannot properly process signals, thereby eliminating information loss while maintaining the energy efficiency of spiking neural networks.
Solution Approach 2:
The cutoff circuit acts as an intermediary component between the synapse circuit and the charge accumulation circuit. It controls the flow of synaptic current, allowing it to pass through during active periods while blocking it during the refractory period. This intermediary mechanism resolves the contradiction by enabling precise control over when charge accumulation occurs, preventing information loss without sacrificing energy efficiency.
2Productivity
If synaptic current is continuously supplied to the charge accumulation circuit, then the neuron can maintain continuous learning capability, but information accuracy deteriorates during the refractory period
Solution Approach 1:
The cutoff circuit implements periodic control of synaptic current supply by enabling it during active periods and disabling it during refractory periods. This periodic action pattern matches the natural firing behavior of neurons, allowing continuous learning capability while maintaining information accuracy by synchronizing charge accumulation with periods when the neuron is ready to process signals.
Solution Approach 2:
The cutoff circuit preemptively stops charge accumulation during the refractory period before accuracy degradation can occur. By anticipating the refractory period and preventing charge accumulation in advance, the system maintains both continuous learning capability and high information accuracy.
3Duration of action of stationary object
If the neuron accumulates charge from synaptic current during the refractory period, then charge accumulation continues without interruption, but the membrane potential calculation becomes inaccurate
Solution Approach 1:
The cutoff circuit extracts or removes the problematic charge accumulation process during the refractory period. By taking out this specific portion of the charge accumulation that occurs during inaccurate periods, the system maintains overall charge accumulation continuity while eliminating the source of membrane potential calculation errors.
Solution Approach 2:
The cutoff circuit preemptively prevents charge accumulation during the refractory period before membrane potential inaccuracy can develop. This preliminary intervention ensures that charge accumulation only occurs during accurate processing periods, maintaining both continuity and precision.
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 neural network device achieves high-accuracy spiking neural network operations with reduced energy consumption, enabling tasks like image recognition and classification without the need for CPUs or GPUs.
Implementation Method 1
a charge accumulation circuit configured to accumulate charge corresponding to the synaptic current and generate a membrane potential corresponding to the accumulated charge
Implementation Method 2
a cutoff circuit configured to stop the supply of the synaptic current from the first terminal to the charge accumulation circuit during a cutoff period being a predetermined period of time after the output of the spike signal
Data Source
AI summary
A neural network device according to an embodiment includes a plurality of synapse circuits and a plurality of neuron circuits. In a first neuron circuit out of the neuron circuits, a synaptic current is supplied to a first terminal from each of one or more first synapse circuits out of the synapse circuits. The first neuron circuit includes a charge accumulation circuit, a spike output circuit, and a cutoff circuit. The charge accumulation circuit accumulates charge corresponding to the synaptic current and generates a membrane potential corresponding to the accumulated charge. The spike output circuit outputs a spike signal when the membrane potential is higher than a preset threshold potential. During a cutoff period that is a predetermined period of time after the output of the spike signal, the cutoff circuit stops the supply of the synaptic current from the first terminal to the charge accumulation circuit.