Compact Arithmetic Processing Elements for Low-Precision Parallel Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional computing systems inefficiently utilize transistors due to a focus on high precision arithmetic, limiting the ability to harness the full computing power of modern silicon chips, despite the potential for performing a significant fraction of a billion computational operations per cycle.
Innovation Solution
The development of processors with low precision high dynamic range (LPHDR) processing elements that perform arithmetic operations with a precision of about 0.1%, allowing for a greater number of operations per unit of time or power, and enabling massively parallel computation with a smaller number of transistors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional high precision arithmetic is used, then measurement precision is improved, but productivity deteriorates
Solution Approach 1:
The patent changes the precision parameter from conventional high precision (e.g., 32-bit or 64-bit floating point) to low precision (e.g., 8-bit or 16-bit integers or floating point). This parameter change enables significantly more arithmetic operations to be performed per cycle, as the reduced precision requires fewer transistors per operation, thereby resolving the contradiction between precision and productivity
Solution Approach 2:
The patent segments the computational task into multiple low precision operations that can be executed in parallel. Instead of performing fewer high precision operations sequentially, the system performs many low precision operations simultaneously, achieving comparable or superior effective precision while dramatically increasing throughput and productivity
2Measurement precision
If high precision arithmetic elements are used, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent changes the precision parameter to low precision, which directly reduces the transistor count required per arithmetic element. Low precision arithmetic elements use fewer transistors because they require less circuitry for maintaining high precision intermediate calculations and fewer bits for data storage and transmission, thereby reducing overall device complexity
Solution Approach 2:
The patent uses multiple copies of simple low precision arithmetic elements instead of fewer complex high precision elements. The replicated simple structures are more efficient to manufacture and require less total transistor count while achieving the same effective computational capability through parallel execution
3Productivity
If the number of arithmetic elements is increased, then productivity is improved, but loss of substance increases
Solution Approach 1:
The patent changes the precision parameter to low precision, which reduces the substance (transistor count) required per arithmetic element. This enables a much larger number of arithmetic elements to be packed into the same physical area, increasing productivity without proportionally increasing total transistor count
Solution Approach 2:
The patent segments the computational workload across many simple low precision arithmetic elements rather than using fewer complex elements. This segmentation allows efficient utilization of available transistors, as each simple element can be implemented with minimal transistor count while the collective array achieves high productivity
Data Source
AI summary
Low precision computers can be efficient at finding possible answers to search problems. However, sometimes the task demands finding better answers than a single low precision search. A computer system augments low precision computing with a small amount of high precision computing, to improve search quality with little additional computing.