Transistor String In-Memory Multiply-Accumulate Circuit
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In machine learning, frequent access to semiconductor memory for multiply-accumulate operations hinders high-speed performance due to the need for frequent CPU access to stored weighting factors.
Innovation Solution
An information processing device with a string of transistors and wiring lines, where the first transistor has a threshold voltage corresponding to first data, and the second transistor has a resistance value corresponding to second data, allowing a current corresponding to the product of the first and second data to flow through the string.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If weighting factors are stored in semiconductor memory and accessed by CPU, then data storage is achieved, but processing speed decreases due to frequent access requirements
Solution Approach 1:
The patent merges the storage function and computation function into a single integrated structure. Transistors are configured to simultaneously store weighting factors (via threshold voltage or resistance values) and perform multiply-accumulate operations, eliminating the need for separate CPU access to memory and enabling direct in-memory computation that achieves both storage and high-speed processing
Solution Approach 2:
The patent introduces a specialized transistor circuit configuration as an intermediary between traditional memory and CPU. This intermediate structure enables data to be processed directly where it is stored, acting as a bridge that allows weighting factors to be utilized for computation without requiring retrieval by the CPU, thus resolving the speed bottleneck
2Productivity
If CPU frequently accesses semiconductor memory for multiply-accumulate operations, then computation is performed, but processing efficiency deteriorates
Solution Approach 1:
The patent performs preliminary configuration of transistors to store weighting factors in advance within the transistor parameters (threshold voltage or resistance). This preliminary setup allows the computation to proceed directly using pre-configured elements, eliminating the need for time-consuming data retrieval during the actual multiply-accumulate operations and significantly improving computational throughput
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 configuration enables faster information processing, such as multiply-accumulate operations, by reducing the need for frequent CPU access to semiconductor memory and improving the determination of first and second data from the current flowing through the string.
Implementation Method 1
the first transistor having a threshold voltage corresponding to first data
Implementation Method 2
the second transistor having a drain and a source, a resistance value between the drain and the source corresponding to second data
Implementation Method 3
a current corresponding to a product of the first data and the second data flows through the string
Data Source
AI summary
An information processing device includes a string including a first transistor and a second transistor, and a first wiring line connected to an end of the string. The first transistor has a threshold voltage corresponding to first data. The second transistor has a drain and a source, and a resistance value between the drain and the source corresponds to second data. A current corresponding to a product of the first data and the second data flows through the string.