Non-volatile Memory Arithmetic Circuitry for Neural Network Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current neural network systems rely solely on memory controllers for arithmetic operations, which limits processing efficiency and speed, as they do not leverage the capabilities of non-volatile memory devices for parallel processing.

Innovation Solution

Incorporating non-volatile memory devices with arithmetic circuitry and control logic to perform neural network operations, allowing for parallel processing of internal and input data, thereby reducing operation time and enhancing processing speed.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If neural network systems rely solely on memory controllers for arithmetic operations, then device complexity is reduced, but processing efficiency and speed deteriorate

Engineering Contradiction:
Improveprocessing efficiencyVSAvoiddevice complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent merges the memory device and arithmetic circuitry into a single integrated system. The non-volatile memory device includes both storage cells and arithmetic processing units that can perform neural network operations directly within the memory architecture, eliminating the need for separate memory controllers and enabling parallel processing of multiple operations simultaneously.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The memory device is designed to perform multiple functions: it can store data in non-volatile memory cells and simultaneously perform arithmetic operations for neural network processing. The arithmetic circuitry can handle various operations including multiplication, accumulation, and activation functions, making the memory device a multi-functional component that replaces both memory storage and separate processing units.

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

2Speed

If memory devices perform arithmetic operations concurrently with data retrieval, then processing speed increases, but device complexity increases

Engineering Contradiction:
Improveprocessing speedVSAvoiddevice complexity
Core Design Contradiction:
SpeedVSDevice complexity

Solution Approach 1:

The arithmetic circuitry is pre-configured within the memory device structure, with processing units positioned to receive and process data as it is being retrieved from memory cells. This preliminary arrangement of processing components enables immediate computation on retrieved data without requiring additional data movement or separate processing stages.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent transitions from a sequential processing model (retrieve data then process) to a parallel processing model by adding the arithmetic dimension within the memory architecture. Multiple arithmetic operations can be performed simultaneously across different memory blocks, effectively utilizing the third dimension of parallelism in the memory structure to achieve concurrent data retrieval and processing.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentUS11741347B2Non-volatile memory device including arithmetic circuitry for neural network processing and neural network system including the same
Publication Date: 2023.08.29 SAMSUNG ELECTRONICS CO LTD
  • US11741347B2 patent drawing
  • US11741347B2 patent drawing
  • US11741347B2 patent drawing

AI summary

A non-volatile memory device includes a memory cell array to which an arithmetic internal data is written; and an arithmetic circuitry configured to receive an arithmetic input data and the arithmetic internal data for an arithmetic operation of a neural network with the arithmetic internal data and the arithmetic input data in response to an arithmetic command, perform the arithmetic operation using the arithmetic internal data and the arithmetic input data to generate an arithmetic result data, and output the arithmetic result data of the arithmetic operation of the neural network.