Compact Arithmetic Elements Using Low-Precision Parallel Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional computing systems inefficiently utilize transistors, as they are designed for high precision arithmetic, limiting the ability to harness the full computing power of modern silicon-based microprocessors, despite having the potential for performing many more operations due to their large number of transistors.
Innovation Solution
Implementing low precision high dynamic range (LPHDR) processing elements that can perform arithmetic operations with a precision of about 0.1% error, allowing for a greater number of operations per unit time and area, and using logarithmic or analog representations to optimize circuit design.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional high precision arithmetic processing elements are used, then measurement precision is improved, but productivity deteriorates due to limited number of operations per unit time
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 fixed point or floating point). This parameter change allows significantly more arithmetic processing elements to be integrated into the same silicon area, thereby increasing the number of operations that can be performed in parallel per unit time, while still achieving adequate computational accuracy for many applications
Solution Approach 2:
The patent divides the computational task into multiple segments by using many low precision processing elements instead of fewer high precision ones. Each processing element performs simpler arithmetic operations, but the collective ensemble of many such elements can perform complex computations through parallel execution, effectively segmenting the overall computational workload across numerous independent units
2Measurement precision
If conventional high precision arithmetic processing elements are used, then measurement precision is improved, but device complexity increases due to transistor requirements
Solution Approach 1:
The patent changes the precision parameter to low precision arithmetic, which dramatically reduces the transistor count required per processing element. Low precision arithmetic elements require far fewer transistors compared to high precision elements, allowing the same silicon area to accommodate many more processing elements, thereby reducing overall device complexity while maintaining adequate computational capability
Solution Approach 2:
The patent employs many inexpensive low precision processing elements instead of a few expensive high precision ones. Each low precision element is computationally simpler and requires fewer resources (transistors, power, area), making them 'cheaper' in terms of hardware resources. Although individually less precise, the collective computational power of many such elements achieves the required performance
3Measurement precision
If conventional high precision arithmetic processing elements are used, then measurement precision is improved, but use of energy deteriorates due to power consumption
Solution Approach 1:
The patent changes the precision parameter to low precision arithmetic, which reduces the power consumption per processing element. Low precision arithmetic operations require less switching activity, smaller circuit capacitances, and fewer transistors, all of which contribute to lower dynamic power consumption. This allows the system to perform more operations with the same power budget or consume less power for the same computational workload
Solution Approach 2:
The patent segments the computational workload across many low precision processing elements that can operate in parallel with lower individual power consumption. The cumulative power consumption of many low precision elements performing simple operations is less than that of fewer high precision elements performing the same overall computational task, due to the quadratic relationship between precision and power consumption in conventional architectures
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.