NAND Flash Read Voltage Shift Estimation With Adaptive Quantization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memory systems face challenges in accurately estimating read voltages for NAND flash memory due to changes caused by stress conditions, leading to potential data reading errors and inefficiencies.
Innovation Solution
A memory system utilizing a quantized neural network to estimate the shift value of read voltages based on the number of on-cells, adjusting the quantization range according to stress conditions, ensuring accurate estimation with reduced memory capacity, latency, and circuit size.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a traditional neural network is used to estimate read voltage shifts, then estimation accuracy can be achieved, but memory capacity requirements, circuit size, and processing latency increase significantly
Solution Approach 1:
The patent applies quantization to transform the neural network parameters from continuous floating-point values to discrete integer values. Specifically, weight values are quantized from high-precision floating-point numbers to low-precision integers (e.g., 8-bit or lower), and activation functions are replaced with quantized versions. This parameter transformation maintains sufficient estimation accuracy while dramatically reducing memory capacity requirements and circuit complexity, enabling deployment in resource-constrained memory controller environments
Solution Approach 2:
The patent replaces the traditional floating-point arithmetic operations with integer arithmetic operations in the neural network computation. By substituting the computational mechanism from floating-point units to integer units, the system achieves the same estimation function with reduced hardware complexity, lower power consumption, and faster processing speed, directly addressing the contradiction between accuracy and device complexity
2Ease of manufacture
If the quantization range is fixed, then circuit implementation is simplified, but estimation accuracy degrades under varying stress conditions
Solution Approach 1:
The patent introduces dynamic adaptability to the quantization range by making it variable based on operating conditions. The system monitors stress conditions (such as retention time, program/erase cycles, and temperature) and dynamically adjusts the quantization range parameters to match the current operating context. This dynamic adjustment ensures high estimation accuracy across varying stress conditions while maintaining reasonable circuit complexity through structured adaptation mechanisms
Solution Approach 2:
The patent implements feedback mechanisms where the system continuously monitors actual read voltage shifts and comparison results, then uses this information to refine and adjust the quantization range parameters. The feedback loop enables the system to learn from past performance and optimize the quantization range for different stress conditions, maintaining high accuracy without requiring overly complex fixed-range designs
3Measurement precision
If high-precision floating-point operations are used, then estimation accuracy is maintained, but processing latency and power consumption increase
Solution Approach 1:
The patent systematically replaces floating-point arithmetic operations with integer arithmetic operations throughout the neural network computation pipeline. This substitution includes quantized matrix multiplications, quantized activation functions, and quantized accumulation operations. Integer operations execute significantly faster on standard digital logic and require simpler hardware, directly reducing processing latency while maintaining sufficient estimation accuracy through careful quantization design
Data Source
AI summary
A memory system includes an input data acquiring unit, a quantization range determining unit, a parameter setting unit, an estimating unit, and a voltage setting unit. The input data acquiring unit acquires input data indicating a relationship between the number of on-cells and a plurality of read voltages. The quantization range determining unit determines a quantization range of a shift value of the read voltages. The parameter setting unit sets a parameter of a quantized neural network. The estimating unit estimates a shift value of the read voltages from the input data with the quantized neural network using the parameter set by the parameter setting unit. The voltage setting unit sets a read voltage that is used during a read operation of a semiconductor storage device based on the shift value of the read voltages.