Simplified Sigmoid Circuit for Neuromorphic Processing

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The increasing complexity of artificial neural networks leads to a significant increase in computation requirements, making it difficult to maintain data processing speed, despite methods like tensor decomposition, network pruning, and quantization, which struggle to markedly reduce computation.

Innovation Solution

A simplified sigmoid function circuit is introduced, which transforms the sigmoid function into a logarithmic region and applies variational transformations, using multiplexers and multipliers to select and combine coefficients for efficient computation based on input data signs, implemented in a neuromorphic processor.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If the complexity of artificial neural network increases, then the capability of AI processing is improved, but the computation requirement increases significantly

Engineering Contradiction:

ImproveAI processing capabilityVSAvoidcomputation requirement

Core Design Contradiction:

Adaptability or versatilityVSPower

Solution Approach 1:

The patent transforms the sigmoid function from the real region to the logarithmic region and applies variational transformation to approximate the function. This parameter change allows the neural network to maintain high AI processing capability while significantly reducing computation requirements, as the variational transformation provides an efficient approximation of the sigmoid function with fewer computational operations.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent uses a simplified sigmoid function that applies variational transformation only to the logarithmic region rather than computing the full sigmoid function across all regions. This partial action approach maintains sufficient accuracy for neural network processing while reducing the overall computation burden, achieving the desired balance between capability and computational efficiency.

Inventive Principle:

Principle #16Partial or excessive action

2Productivity

If the amount of computation is reduced, then the data processing speed is maintained, but the accuracy of neural network processing may deteriorate

Engineering Contradiction:

Improvedata processing speedVSAvoidprocessing accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent transforms the sigmoid function to the logarithmic region and applies variational transformation with specific coefficients (first and second coefficients) to approximate the function. This parameter change maintains processing accuracy by providing a mathematically sound approximation that preserves the essential characteristics of the sigmoid function while enabling faster computation and higher data processing speed.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent creates a simplified version of the sigmoid function by copying its essential mathematical properties into the logarithmic region through variational transformation. This copying approach retains the functional characteristics needed for accurate neural network processing while using fewer computational resources, thus maintaining both accuracy and processing speed.

Inventive Principle:

Principle #26Copying

3Power

If a simplified sigmoid function is used, then the computation is reduced, but the complexity of the circuit implementation increases

Engineering Contradiction:

Improvecomputation requirementVSAvoidcircuit implementation complexity

Core Design Contradiction:

PowerVSDevice complexity

Solution Approach 1:

The patent segments the sigmoid function computation into distinct circuits: a first circuit for positive input data, a second circuit for negative input data, and a multiplexer to select between them. This segmentation reduces the overall computation requirement by handling different input regions with specialized simple circuits, while the modular structure keeps the circuit implementation manageable despite the increased number of components.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent implements a dynamic circuit selection mechanism using a multiplexer that adapts its behavior based on the sign of the input data. The circuit dynamically switches between the first and second computation paths depending on whether the input is positive or negative, optimizing computation efficiency for each region while maintaining a relatively simple overall circuit structure through this adaptive design.

Inventive Principle:

Principle #15Dynamics

Data Source

PatentUS20220172029A1Circuit for implementing simplified sigmoid function and neuromorphic processor including the circuit

Publication Date: 2022.06.02 ELECTRONICS & TELECOMM RES INST

AI summary

Disclosed is a simplified sigmoid function circuit which includes a first circuit that performs a computation on input data based on a simplified sigmoid function when a sign of a real region of the input data is positive, a second circuit that performs the computation on the input data based on the simplified sigmoid function when the sign of the real region of the input data is negative, and a first multiplexer that selects and output one of an output of the first circuit and an output of the second circuit, based on the sign of the input data. The simplified sigmoid function is obtained by transforming a sigmoid function of a real region into a sigmoid function of a logarithmic region and performing a variational transformation for the sigmoid function of the logarithmic region.