Programmable Output Blocks for VMM Current Conversion Accuracy

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing systems face challenges in accurately measuring and transferring the output current from a Vector-by-Matrix Multiplication (VMM) array in artificial neural networks, often resulting in information loss due to leakage current.

Innovation Solution

The development of programmable output blocks that can configure the gain and resolution of an Analog-to-Digital Converter (ADC) within the output block, allowing for improved conversion and transfer of output current from the VMM array.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If output current from VMM array is measured and transferred using existing systems, then the transfer process is simple, but information loss due to leakage current occurs

Engineering Contradiction:
Improveinformation lossVSAvoidoutput block complexity
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary Analog-to-Digital Converter (ADC) between the VMM array output and the digital processing stage. This ADC acts as a mediator that converts the analog output current to digital values, preventing information loss from leakage current while managing the complexity through integrated design

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent applies parameter changes by making the ADC gain and resolution programmable and configurable. This allows optimization of the conversion process to minimize information loss while adapting to different operational requirements, balancing accuracy needs with device complexity

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If the gain and resolution of ADC are made programmable and configurable, then conversion accuracy is improved, but device complexity increases

Engineering Contradiction:
Improveconversion accuracyVSAvoidoutput block complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent implements dynamic configurability where the ADC gain and resolution can be programmed and adjusted based on operational requirements. This dynamic adaptation allows the system to optimize measurement precision for different scenarios while managing complexity through systematic control mechanisms

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The output block is designed with multi-functional capabilities, where a single configurable ADC structure can adapt to different conversion requirements by changing its parameters. This universality reduces the need for multiple specialized components, thereby managing overall device complexity while maintaining high conversion accuracy

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP4367665B1Programmable output blocks for analog neural memory in a deep learning artificial neural network
Publication Date: 2025.05.07 SILICON STORAGE TECHNOLOGY INC
  • EP4367665B1 patent drawingFigure 1
  • EP4367665B1 patent drawingFigure 2
  • EP4367665B1 patent drawingFigure 3

AI summary

Numerous embodiments are disclosed for programmable output blocks for use with a VMM array within an artificial neural network. In one embodiment, the gain of an output block can be configured by a configuration signal. In another embodiment, the resolution of an ADC in the output block can be configured by a configuration signal.