Memory Array Diagonal Isolation for Ising Model Computation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex electronic devices and semiconductor components require time-consuming and energy-consuming model computations, especially for large-scale Ising models, which can lead to computational errors due to hardware defects.
Innovation Solution
A memory device with a memory array and calculation units processes model computations by generating source currents from input values and interact coefficients, calculating local field energies, and using electrically isolated memory cells to efficiently perform Ising model computations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the dimension of the Ising model is increased to handle more parameters, then the model can represent more complex problems, but it requires more computing resources and longer computing time
Solution Approach 1:
The memory array is divided into multiple memory sub-arrays, with each sub-array handling a portion of the Ising model computation. The memory cells are arranged in groups where each group processes specific spin states, allowing parallel computation across multiple sub-arrays. This segmentation enables the system to handle larger model dimensions without proportionally increasing computation time.
Solution Approach 2:
The patent utilizes the two-dimensional structure of the memory array (rows and columns) to map the Ising model parameters. By arranging memory cells in a grid pattern and using row/column signal lines, the system transforms the computational problem into a spatial arrangement that enables parallel processing. The diagonal electrical isolation further partitions the computation space to reduce interference.
2Productivity
If fully-connected Ising model is used for parallel computation, then the anneal computation speed increases, but hardware defects lead to computational errors
Solution Approach 1:
The memory array is divided into multiple independently isolated memory sub-arrays. Each sub-array processes a specific portion of the spin states, and the diagonal electrical isolation ensures that defects in one sub-array do not propagate to others. This segmentation maintains parallel computation speed while improving reliability through isolation.
Solution Approach 2:
Different regions of the memory array are electrically isolated from each other using diagonal isolation structures. This creates locally independent computation zones where each sub-array can operate without interference from adjacent regions, reducing the impact of hardware defects on overall computation accuracy.
3Measurement precision
If more computing resources are allocated to handle larger Ising models, then the computation accuracy improves, but the energy consumption increases
Solution Approach 1:
The patent combines multiple functions into the memory device itself. The memory cells store Ising model parameters (weights and biases) and simultaneously perform analog computation by generating currents proportional to spin states. The calculation units integrated with the memory array compute local field energies directly from the stored parameters, eliminating the need for separate computing hardware and reducing overall energy consumption.
Solution Approach 2:
The memory device performs computation using its own stored data without requiring external processing. The memory cells automatically generate currents based on their stored parameters and the applied input signals, and the calculation units directly compute the local field energies from these currents. This self-service approach reduces energy consumption by eliminating data transfer and separate processing steps.
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
The solution significantly reduces computation time and energy consumption while minimizing errors, enabling faster and more efficient execution of complex model computations, even for larger Ising models with many spin states.
Implementation Method 1
the memory cells generate a plurality of source currents, the source currents flowing through the first signal lines to generate a plurality of common source currents
Data Source
AI summary
The application provides a memory device and an operation method thereof. The memory device includes: a memory array, for processing model computation having a plurality of input values and a plurality of interact coefficients; and at least one calculation unit. In receiving the input values, a first part and a second part of the memory cells generate a first part and a second part of the common source currents, respectively. The first part of the memory cells is electrically isolated from the second part of the memory cells based on a diagonal of the memory array. The at least one calculation unit calculates a first part and a second part of a local field energy of the model computation based on the first part and the second part of the common source currents.