Multi-Level Phase-Change Memory Read Method
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is a need for a reliable, simple, and fast method to read multi-level non-volatile memories, which store more than two logic values, to efficiently manage and retrieve data in high-density memory applications such as artificial intelligence systems.
Innovation Solution
A method using a dichotomous algorithm to compare the read current output by a multi-level non-volatile memory cell with successively selected reference currents, employing a structurally identical reference memory cell or an adjustable current source to determine the stored data, allowing for precise programming and verification of the memory cell levels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If multi-level memory stores more than two logic values to increase storage density, then storage capacity is improved, but reading reliability deteriorates due to difficulty in distinguishing multiple current levels
Solution Approach 1:
The reading process is segmented into multiple comparison steps using a dichotomous algorithm. Instead of attempting to directly distinguish all N current levels in one step, the method divides the comparison into successive binary decisions, comparing the read current with reference currents that split the remaining range in half at each step. This segmentation transforms a complex N-level discrimination problem into a series of simpler binary comparisons, improving reading reliability while maintaining multi-level storage capacity.
Solution Approach 2:
Reference currents serve as intermediaries between the read current and the decision logic. These reference currents, generated by programming reference memory cells to specific conductance levels, act as mediator values that facilitate the comparison process. The sense amplifier uses these intermediary reference currents to indirectly determine the stored data level without requiring direct N-level discrimination, thereby improving reading reliability.
2Ease of operation
If a simple reading method is used for multi-level memories, then ease of operation is improved, but measurement precision deteriorates due to difficulty in accurately distinguishing multiple logic values
Solution Approach 1:
The reading method dynamically adapts the reference current selection based on previous comparison results. The dichotomous algorithm adjusts which reference currents are used in subsequent comparisons based on the outcomes of earlier comparisons. This dynamic approach maintains reading simplicity while improving measurement precision, as the system automatically narrows down the possible current levels through adaptive binary decisions rather than requiring static, complex multi-level discrimination circuitry.
3Productivity
If a fast reading method is used for multi-level memories, then productivity is improved, but measurement precision deteriorates due to reduced time for accurate current level distinction
Solution Approach 1:
Reference memory cells are pre-programmed to specific conductance levels that correspond to the reference currents needed for the dichotomous algorithm. This preliminary action prepares the reference currents in advance, eliminating the need for complex real-time generation during the reading process. The pre-prepared reference currents enable fast reading while maintaining measurement precision, as the comparison can proceed quickly without sacrificing accuracy.
4Measurement precision
If successive comparisons with multiple reference currents are performed to improve reading accuracy, then measurement precision is improved, but device complexity increases due to additional reference currents and comparison circuits
Solution Approach 1:
Reference memory cells serve multiple functions: they store reference data patterns and simultaneously generate reference currents for comparison. This multi-functionality eliminates the need for separate reference current generation circuits, reducing device complexity while maintaining the capability for successive comparisons with multiple reference currents. The same memory structure that stores user data also provides the reference signals needed for accurate reading.
Solution Approach 2:
Instead of using complex analog reference current sources, the patent uses digital copies of data patterns stored in reference memory cells to generate the necessary reference currents. This copying approach simplifies the reference circuit by using discrete, programmable reference values rather than continuous analog generation, reducing device complexity while maintaining measurement precision through the dichotomous comparison algorithm.
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 enables efficient and accurate reading of multi-level non-volatile memory cells, ensuring reliable data retrieval and programming across various logic values, enhancing the performance of high-density memory devices used in AI applications.
Implementation Method 1
heating the phase-change material above a melting temperature of the phase-change material and then rapidly cooling the phase-change material to a temperature below a glass transition temperature of the phase-change material to form an amorphous region
Implementation Method 2
heating the phase-change material above a melting temperature of the phase-change material
Data Source
AI summary
A multi-level non-volatile memory cell has N levels, N being even and greater than two, corresponding respectively to N logical data that can be stored in the memory cell and to N corresponding read current ranges. A datum stored in the memory cell is read by performing successive comparisons of a read current output by the memory cell with reference currents selected from a set of N-1 reference currents having values respectively lying between two different successive ranges using a dichotomous algorithm starting with the reference current having the median value.