Multilevel Memory Cell for Feature Vector Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current memory technologies, particularly CAM devices, require complex comparison circuitry for searching feature vectors, which is inefficient for matching within a range of values, leading to a need for a more streamlined approach.
Innovation Solution
The use of multilevel cells (MLC) in memory arrays allows for the storage of multiple data states per cell, enabling efficient comparison and matching of feature vectors by programming specific threshold voltage ranges, enabling exact and inexact matches within a range using a single pair of memory cells.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If CAM devices use dedicated comparison circuitry for each memory cell to enable searching, then searching capability is improved, but device complexity increases significantly
Solution Approach 1:
The patent merges the comparison function into the memory cell itself by using the natural threshold voltage comparison property of flash memory cells. Instead of having separate comparison circuits for each cell, the memory cell's threshold voltage directly performs the comparison operation, combining storage and comparison functions into a single element.
Solution Approach 2:
The memory cell performs its own comparison operation autonomously by comparing the input voltage against its stored threshold voltage. The cell self-determines whether to pass current based on this comparison, eliminating the need for external comparison circuitry and reducing overall system complexity.
2Quantity of substance
If MLC is used to increase memory density, then memory density is improved, but manufacturing precision requirements increase
Solution Approach 1:
The patent changes the operational parameter from binary threshold detection to multi-level threshold voltage range detection. By programming memory cells to specific threshold voltage ranges corresponding to different data states, the system achieves higher density while managing manufacturing variations through range-based classification rather than precise single-point thresholds.
Solution Approach 2:
The patent segments the threshold voltage spectrum into multiple discrete ranges, each representing a different data state. This segmentation approach allows MLC to store multiple bits per cell by dividing the analog threshold voltage domain into distinct zones, effectively managing the complexity of analog variations through structured categorization.
3Measurement precision
If exact matching is performed using traditional memory structures, then matching precision is improved, but productivity decreases due to inability to perform range searches
Solution Approach 1:
The patent introduces dynamic search capabilities by allowing the search operation to adapt between exact matching and range matching modes. The system can dynamically adjust the search criteria based on the application requirements, enabling both precise exact matches and flexible range searches without sacrificing either precision or efficiency.
Solution Approach 2:
The patent creates a universal search mechanism that handles both exact matching and range matching through the same hardware structure. The memory cell's threshold voltage comparison property serves multiple functions: it can perform exact equality comparison when needed, or range comparison when broader matching is required, eliminating the need for separate circuitry for different search types.
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
This approach reduces the complexity of comparison circuitry and increases memory density, allowing for efficient searching of feature vectors within a range of values, enhancing the performance and cost-effectiveness of memory systems.
Implementation Method 1
Multilevel cells (which are also sometimes referred to herein as multiple level cells) can take advantage of the analog nature of a traditional charge storage cell by assigning a particular bit pattern to a specific threshold voltage (Vt) range stored on the cell.
Data Source
AI summary
Apparatus and methods of operating a memory include storing a value of an attribute of a feature vector to a pair of memory cells by programming each of the memory cells to a respective data state of three or more data states, searching for an exact match to a particular value of the attribute by applying respective voltage levels to control gates of the memory cells to activate both memory cells only when the value of the attribute is the particular value, and searching for an inexact match to the particular value of the attribute by applying respective voltage levels to control gates of the memory cells to activate both memory cells when the value of the attribute is within a range of possible values of the attribute including the particular value.