In-Memory Dot Product Circuit for Low-Transfer Neural Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional computer-based computations for operations like forward and backward propagation in neural networks are processor and memory intensive, requiring extensive data transfer between compute cores and memory arrays, which can be inefficient in terms of performance and power usage.
Innovation Solution
Implementing processing-in-memory (PIM) operations within a memory device, where a processor is integrated near or on the same chip as the memory array, allowing for dot product operations to be performed internally without external data transfer, using a memory array with bit lines, word lines, and a summing circuit to generate the product of two numbers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data is transferred between external memory arrays and compute cores, then processing operations can be performed, but performance and power usage are degraded due to extensive data transfer requirements
Solution Approach 1:
The patent combines memory storage and processing functions into a single integrated device. The memory array is directly coupled with processing circuitry including sense amplifiers that can perform computational operations, eliminating the need for separate external memory arrays and compute cores. This merging reduces data transfer distances and associated power consumption while maintaining processing capability.
2Productivity
If data is transferred between external memory arrays and compute cores, then processing operations can be performed, but extensive external communications reduce processing efficiency
Solution Approach 1:
The patent integrates memory and processing functions within a single device, reducing system complexity by eliminating the need for separate external memory arrays and compute cores. The memory array is directly coupled with processing circuitry, allowing operations to be performed in-place without complex external communication interfaces and data transfer protocols.
3Speed
If dot product operations are performed within the memory device, then processing speed is improved, but additional circuitry is required within the memory array
Solution Approach 1:
The sense amplifiers in the memory device are designed to perform multiple functions: traditional memory read operations and computational operations such as dot product calculations. This multi-functionality allows the same circuitry to handle both storage retrieval and processing tasks, reducing the need for additional dedicated processing hardware while enabling in-memory computation.
Data Source
AI summary
Methods, apparatuses, and systems for in-or near-memory processing are described. Bits of a first number may be stored on a number of memory elements, wherein each memory element of the number of memory elements intersects a bit line and a word line of a number of word lines. A number of signals corresponding to bits of a second number may be driven on the number of word lines to generate a number of output signals. A value equal to a product of the first number and the second number may be generated based on the number of output signals.