Configurable Analog Neural Memory Hardware for Variable Array Sizes

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing analog neural memory systems require customized hardware for each system, which is costly and time-consuming, especially when dealing with arrays of different sizes that need varying levels of supporting circuitry.

Innovation Solution

A configurable hardware system for analog neural memory systems that can provide various layers of vector-by-matrix multiplication arrays of different sizes, along with customizable supporting circuitry, allowing the same hardware to be used across different system requirements.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If customized hardware is designed for each analog neural memory system with specific array sizes, then the system can meet specific computational requirements, but the development cost and time increase significantly

Engineering Contradiction:
Improveadaptability to different array sizesVSAvoidhardware customization complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent implements a universal hardware platform that can be configured to support multiple array sizes (e.g., 1Kx1K, 2Kx2K, 4Kx4K) through programmable control logic. The same physical hardware can be reconfigured via software to meet different computational requirements, eliminating the need for separate customized hardware designs for each array size while maintaining full functionality.

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

2Reliability

If separate supporting circuitry is provided for each array size, then the system can be optimized for specific performance requirements, but the manufacturing cost and design time increase

Engineering Contradiction:
Improvesystem performance optimizationVSAvoidmanufacturing cost
Core Design Contradiction:
ReliabilityVSEase of manufacture

Solution Approach 1:

The patent employs a set of reusable supporting circuits including voltage generation circuits, sense amplifiers, and control logic that can serve multiple array configurations. These circuits are designed with programmable parameters and configurable connections, allowing them to be adapted to different array sizes without requiring separate physical circuit implementations, thereby reducing manufacturing costs while maintaining performance optimization.

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

Solution Approach 2:

The supporting circuitry incorporates dynamically reconfigurable elements such as programmable voltage references, adjustable sense amplifier thresholds, and configurable control signals that can be tuned via software to optimize performance for different array sizes. This dynamic adaptability allows the same hardware to be optimized for specific performance requirements without requiring physical redesign or separate manufacturing processes.

Inventive Principle:
Principle #15Dynamics

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

Enables efficient and cost-effective implementation of analog neural memory systems with different requirements by providing a flexible hardware solution that can adapt to various array sizes and supporting circuitry needs.

Implementation Method 1

The memory cell is configured to multiply the input by the stored weight value to generate an output

Methodology Applied
Scientific EffectOhm's Law: Ohm's Law

Data Source

PatentEP4220489B1Configurable analog neural memory system for deep learning neural network
Publication Date: 2025.02.12 SILICON STORAGE TECHNOLOGY INC
  • EP4220489B1 patent drawingFigure 1
  • EP4220489B1 patent drawingFigure 2
  • EP4220489B1 patent drawingFigure 3

AI summary

Numerous embodiments are disclosed for a configurable hardware system for use in an analog neural memory system for a deep learning neural network. The components within the configurable hardware system that are configurable can include vector-by-matrix multiplication arrays, summer circuits, activation circuits, inputs, reference devices, neurons, and testing circuits. These devices can be configured to provide various layers or vector-by-matrix multiplication arrays of various sizes, such that the same hardware can be used in analog neural memory systems with different requirements.