Memory Weight Adjustment for Stable Neuromorphic MAC Operations

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Neuromorphic devices face issues with undesired changes in weight data values stored in memory cell arrays, leading to inaccurate MAC operations due to varying rates of change in data levels over time.

Innovation Solution

A semiconductor memory apparatus with a data adjusting circuit that compares weight data values to a reference value, generates flag information, and adjusts the data values to minimize changes, while a data calculation circuit recovers the adjusted values for reliable MAC operations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If weight data values are stored directly in the memory cell array circuit, then the MAC operation can be performed, but the weight data values change over time to undesired data level values due to rate of change variations

Engineering Contradiction:
ImproveMAC operation capabilityVSAvoidweight data value stability
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent applies preliminary action by adjusting the weight data values before storing them in the memory cell array circuit. The data adjusting circuit modifies the weight data values in advance to account for expected rate of change variations, ensuring that even as the values change over time, they remain within acceptable ranges for accurate MAC operations.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements feedback through the data calculation circuit that recovers the adjusted weight data values from the memory cell array circuit and uses this information to correct or compensate for changes. This feedback mechanism ensures that the MAC operation uses accurate weight data values despite temporal variations in the stored data.

Inventive Principle:
Principle #23Feedback

2Adaptability or versatility

If the memory cell array circuit stores multi-level data values, then neuromorphic computing functions are enabled, but the data levels change at different rates leading to calculation errors

Engineering Contradiction:
Improveneuromorphic computing capabilityVSAvoiddata level accuracy
Core Design Contradiction:
Adaptability or versatilityVSManufacturing precision

Solution Approach 1:

The patent applies local quality by treating different data levels differently through the data adjusting circuit. Each weight data value is individually adjusted based on its specific characteristics and expected rate of change, rather than applying a uniform approach to all multi-level data values. This localized adjustment maintains precision for each data level while preserving the multi-level capability.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent implements parameter changes by modifying the weight data values to compensate for rate of change variations. The data adjusting circuit changes the parameters of the stored data (the weight values themselves) to predictively counteract expected drift, thereby maintaining data level accuracy despite the inherent variations in multi-level storage.

Inventive Principle:
Principle #35Parameter changes

3Reliability

If weight data values are adjusted to minimize rate of change, then data stability is improved, but additional circuits and operations are required

Engineering Contradiction:
Improveweight data value stabilityVSAvoidcircuit structure
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies merging by integrating the data adjusting circuit and data calculation circuit into the existing memory architecture. Rather than adding completely separate systems, the adjustment and recovery functions are combined with the memory cell array circuit to form a unified structure that performs both storage and correction functions.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent implements self-service through the data calculation circuit that automatically recovers and corrects the weight data values using the flag information. The system performs its own correction without requiring external intervention, making the complexity self-contained within the memory apparatus while maintaining high reliability.

Inventive Principle:
Principle #25Self-service

Data Source

PatentUS12474890B2Semiconductor memory apparatus and operating method thereof
Publication Date: 2025.11.18 SK HYNIX INC
  • US12474890B2 patent drawing
  • US12474890B2 patent drawing
  • US12474890B2 patent drawing

AI summary

A semiconductor memory apparatus may include: a data adjusting circuit configured to conditionally adjust a weight data value for a MAC (Multiplication and ACcumulation) operation based on comparing the weight data value to a reference data value, and generate flag information indicating whether the weight data value has been adjusted; a memory cell array circuit configured to store the adjusted weight data value outputted from the data adjusting circuit; and a data calculation circuit configured to recover, on the flag information, a result based on the weight data value from a result based on the adjusted weight data value to perform the MAC operation on an input data value and the weight data value.