Spiking Neural Network Calibration for Hardware Variability

Solution Overview

Problem

Existing spiking neural networks (SNNs) face challenges in achieving consistent performance across different hardware implementations due to manufacturing tolerances and environmental variations, with current calibration methods being time-consuming and ineffective in addressing variability caused by temperature and voltage changes.

Innovation Solution

A calibration strategy for SNNs that adjusts circuit parameters using observable parameters to deduce values for non-observable parameters, involving offset current generators, synapses, neurons, and analog-to-digital converters, allowing for precise calibration of components like input processing circuits, synapses, and neurons, with a focus on reducing variability and ensuring consistent performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If training is performed for each individual SNN copy to compensate for manufacturing errors, then manufacturing precision is improved, but productivity deteriorates due to time-consuming training

Engineering Contradiction:

Improveperformance consistencyVSAvoidcalibration time

Core Design Contradiction:

Manufacturing precisionVSProductivity

Solution Approach 1:

The patent performs a reduced training step on each individual copy of the SNN to adjust weights to the SNN's particular properties. This preliminary calibration action compensates for manufacturing variations without requiring full training, thereby improving performance consistency while reducing calibration time compared to complete training of each copy.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent changes the training parameters by performing a reduced training step with limited epochs and simplified data processing. This parameter adjustment allows faster calibration that still achieves acceptable performance consistency across different hardware copies, balancing manufacturing precision and productivity.

Inventive Principle:

Principle #35Parameter changes

2Manufacturing precision

If reduced training is performed on each individual SNN copy to adjust weights, then manufacturing precision is improved, but productivity deteriorates due to long calibration time

Engineering Contradiction:

Improveweight accuracyVSAvoidcalibration time

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent performs a reduced training step on each individual copy of the SNN to adjust weights to the SNN's particular properties. This preliminary calibration action compensates for manufacturing variations without requiring full training, thereby improving performance consistency while reducing calibration time compared to complete training of each copy.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent applies partial training action by performing only a reduced training step rather than complete training. This partial action is sufficient to adjust weights to individual hardware properties and achieve acceptable weight accuracy, while significantly reducing the time loss compared to full training.

Inventive Principle:

Principle #16Partial or excessive action

3Reliability

If full training is performed on each SNN copy to ensure convergence, then reliability is improved, but productivity deteriorates due to time consumption

Engineering Contradiction:

Improveconvergence guaranteeVSAvoidcalibration throughput

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The patent applies partial training action by performing only a reduced training step rather than complete training. This partial action is sufficient to adjust weights to individual hardware properties and achieve acceptable weight accuracy, while significantly reducing the time loss compared to full training.

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The patent changes the training parameters by performing a reduced training step with limited epochs and simplified data processing. This parameter adjustment allows faster calibration that still achieves acceptable performance consistency across different hardware copies, balancing manufacturing precision and productivity.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20250378314A1Calibration of spiking neural networks

Publication Date: 2025.12.11 INNATERA NANOSYSTEMS BV

AI summary

A spiking neural network comprising a plurality of input processing circuits, each input processing circuit having an input for receiving a spiking neural network input signal and being configured to apply a transfer function to the input signal to generate a processed input signal; a plurality of offset current generators, each offset current generator configured to generate an offset current signal at a predetermined level; a plurality of synapses, each synapse connected to receive a processed input signal from one of the input processing circuits and configured to apply a predetermined weight to the processed input signal to generate a synapse output signal; a plurality of neurons, each neuron connected to receive synapse output signals from a subset of the synapses and an offset current signal from one of the offset current generators, and each neuron configured to generate a neuron output signal in response to the received synapse output signals and offset current signal; and an analog-to-digital converter having an input, the input being connectable to receive an offset current signal from one of the offset current generators, and being configured to convert the received offset current signal to a corresponding digital output signal.