Neuron Current-to-Time Pulse Conversion for Leakage-Free VMM Transfer

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing systems face challenges in accurately measuring and transferring the output current of vector-by-matrix multiplication (VMM) arrays in artificial neural networks due to issues like loss of information through leakage current.

Innovation Solution

The system converts neuron current output by a VMM array into neuron current-based time pulses using CMOS technology and non-volatile memory arrays, allowing these pulses to be input to another VMM array within an artificial neural network.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Use of energy by moving object

If neuron current is directly transferred between VMM arrays, then energy consumption is reduced, but information loss occurs through leakage current

Engineering Contradiction:

Improveenergy consumptionVSAvoidinformation loss through leakage current

Core Design Contradiction:

Use of energy by moving objectVSLoss of information

Solution Approach 1:

The patent introduces time pulses as an intermediary carrier to transfer information between VMM arrays. The neuron current is converted into time pulses with widths proportional to the current magnitude, which then serve as inputs to the next VMM array. This intermediary representation eliminates direct current transfer and its associated leakage problems while preserving the computational information.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent transforms the physical parameter of neuron current into a different parameter representation - time pulse width. By converting the continuous current signal into discrete time-domain pulses where the width encodes the current magnitude, the system changes the parameter domain to avoid leakage current while maintaining information fidelity.

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If separate multiplication and addition logic circuits are used, then computational accuracy is improved, but device complexity increases

Engineering Contradiction:

Improvecomputational accuracyVSAvoiddevice complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent merges the multiplication and addition operations into a single hardware stage by utilizing the inherent properties of the VMM array and the time pulse representation. The VMM array performs multiplication of inputs by weights, and the resulting neuron currents are directly converted to time pulses that are fed to the next stage, eliminating the need for separate adder circuits while maintaining computational accuracy through the continuous analog representation.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The VMM array is designed to perform both multiplication and accumulation functions simultaneously. The same hardware structure that multiplies inputs by weights also sums the results through the natural current summation property, making the circuit multi-functional and eliminating the need for dedicated addition logic circuits.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentEP4235671B1System for converting neuron current into neuron current-based time pulses in an analog neural memory in a deep learning artificial neural network

Publication Date: 2026.02.11 SILICON STORAGE TECHNOLOGY INC

Figure 1
Figure 2
Figure 3

AI summary

Numerous embodiments are disclosed for converting neuron current output by a vector-by-matrix multiplication (VMM) array into neuron current-based time pulses and providing such pulses as an input to another VMM array within an artificial neural network. Numerous embodiments are disclosed for converting the neuron current-based time pulses into analog current or voltage values if an analog input is needed for the VMM array.