Reconfigurable SIMD Units Exchange Intermediate Results
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Contemporary chip/core designs face challenges in balancing high precision and low precision arithmetic requirements, leading to inefficiencies in processing emerging workloads like cognitive computing, as they struggle to strike a balance between a few high precision units and a sea of low precision units, resulting in performance penalties and software overhead.
Innovation Solution
A system comprising reconfigurable SIMD units that exchange intermediate results to compute a combined high precision complex arithmetic operation, allowing for efficient execution of both high and low precision calculations by combining partial results from multiple units, thereby enabling high precision operations while minimizing wiring overhead and performance delays.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a few high precision units are used, then high precision computation is improved, but low precision throughput deteriorates
Solution Approach 1:
The patent segments the arithmetic units into multiple reconfigurable units, each capable of performing both high and low precision operations. This segmentation allows the system to distribute computational tasks across multiple units, maintaining high precision capability while improving overall throughput through parallel processing.
Solution Approach 2:
The patent implements dynamic reconfiguration of the arithmetic units, allowing them to switch between high precision and low precision modes based on workload requirements. This dynamic capability enables the system to optimize performance for different computational tasks, resolving the contradiction between precision and throughput.
2Productivity
If a sea of low precision units is used, then low precision throughput is improved, but high precision computation deteriorates due to software overhead
Solution Approach 1:
The patent designs universal reconfigurable arithmetic units that can perform both high precision and low precision operations. This multi-functionality eliminates the need for separate dedicated units, reducing software overhead while maintaining the ability to execute both precision levels efficiently.
Solution Approach 2:
The patent changes the operational parameters of the arithmetic units dynamically, allowing them to switch between different precision modes. This parameter change approach simplifies the system architecture compared to having separate units, while maintaining high precision computation capability through hardware-level configuration changes.
3Measurement precision
If reconfigurable units exchange intermediate results, then high precision complex arithmetic is improved, but wiring overhead increases
Solution Approach 1:
The patent merges multiple reconfigurable units into a unified computational structure where intermediate results are exchanged through integrated pathways. This merging approach reduces wiring overhead by eliminating the need for separate communication channels between units, while maintaining the capability to perform high precision complex arithmetic operations.
Data Source
AI summary
A circuit includes reconfigurable units that are reconfigurable to compute a combined result. A first intermediate result of a first reconfigurable unit of the reconfigurable units is exchanged with a second intermediate result of the second reconfigurable unit of the reconfigurable units. The first reconfigurable unit computes a first portion of the combined result utilizing the second intermediate result. The second reconfigurable unit of the reconfigurable units computes a second portion of the combined result utilizing the first intermediate result.