Analog Memory Cell Temperature and Leakage Compensation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Analog neural memory systems face challenges in maintaining precision and accuracy due to temperature changes and leakage effects, which affect the current-voltage response characteristics of memory cells, requiring effective compensation methods.

Innovation Solution

A method is disclosed to generate a bias voltage for memory cells in an array, employing temperature compensation techniques such as discrete or continuous slope adjustment and leakage reduction by applying compensation currents or voltages to the control gate, word line, or source line, using temperature sensors and controllers to adjust the operating conditions based on temperature readings.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If memory cells operate in analog mode for neural network computation, then computational parallelism and energy efficiency are improved, but precision and accuracy deteriorate due to temperature changes and leakage effects

Engineering Contradiction:
Improvecomputational parallelismVSAvoidprecision
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent applies parameter changes by dynamically adjusting the bias voltage applied to memory cells based on temperature readings from temperature sensors. This compensation mechanism modifies the electrical parameters (voltage levels) to counteract temperature-induced drift in the current-voltage characteristics, thereby maintaining precision despite operating in analog mode for high computational parallelism

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent implements feedback control through temperature sensors that continuously monitor the temperature of memory cells and feed this information to control circuits. These control circuits adjust the bias voltages in real-time based on the temperature feedback, creating a closed-loop system that maintains precision while enabling analog computation for high productivity

Inventive Principle:
Principle #23Feedback

2Productivity

If memory cells operate at higher temperatures for sustained computation, then productivity is improved, but leakage current increases causing accuracy to deteriorate

Engineering Contradiction:
Improvesustained computation capabilityVSAvoidleakage current
Core Design Contradiction:
ProductivityVSObject-generated harmful factors

Solution Approach 1:

The patent applies preliminary anti-action by proactively applying compensation voltages to counteract leakage current before it significantly degrades computation accuracy. The temperature compensation mechanism predicts and counteracts the harmful effects of thermal leakage by adjusting bias conditions in advance, allowing sustained computation at higher temperatures without accumulating error

Inventive Principle:
Principle #9Preliminary anti-action

3Measurement precision

If compensation circuits are added to correct temperature and leakage effects, then precision is improved, but device complexity increases

Engineering Contradiction:
Improvecomputation accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies universality by designing compensation circuits that serve multiple functions: temperature sensing, leakage compensation, and bias voltage regulation are integrated into a unified control system. This multi-functional approach improves computation accuracy while minimizing the increase in device complexity by avoiding separate dedicated circuits for each compensation function

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

Data Source

PatentEP4138079B1Temperature and leakage compensation for memory cells in an analog memory system for use in a deep learning neural network
Publication Date: 2024.08.28 SILICON STORAGE TECHNOLOGY INC
  • EP4138079B1 patent drawingFigure 1
  • EP4138079B1 patent drawingFigure 2
  • EP4138079B1 patent drawingFigure 3

AI summary

Numerous embodiments are disclosed for providing temperature compensation and leakage compensation for an analog neuromorphic memory system used in a deep learning neural network. The embodiments for providing temperature compensation implement discreet or continuous adaptive slope compensation and renormalization for devices, reference memory cells, or selected memory cells in the memory system. The embodiments for providing leakage compensation within a memory cell in the memory system implement adaptive erase gate coupling or the application of a negative bias on a control gate terminal, a negative bias on a word line terminal, or a bias on a source line terminal.