Compact Arithmetic Processing Elements for Low-Precision 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-based processors, despite the potential for performing many more operations per cycle.
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 of time and space, and using logarithmic or analog representations to reduce circuit size and increase parallel processing capabilities.
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 (32-bit or 64-bit floating point) to low precision (about 0.1% error, roughly 10-bit equivalent). This parameter change allows dramatically more arithmetic operations to be performed per cycle - potentially thousands or millions instead of just a few - while still achieving adequate computational results for many applications through techniques like randomized rounding and error compensation.
2Productivity
If more arithmetic elements are implemented, then productivity is improved, but device complexity increases
Solution Approach 1:
The patent extracts and eliminates the complex control logic and precision management circuitry that characterizes conventional high-precision processors. By using simple low-precision arithmetic elements with about 0.1% error tolerance, the design removes the need for complex rounding, normalization, and exception handling circuits, thereby achieving high productivity with reduced device complexity.
Solution Approach 2:
The patent employs numerous inexpensive low-precision arithmetic elements that can be easily fabricated in large numbers on silicon chips. These simple computing elements trade individual precision for collective computational power, with each element being computationally 'disposable' in the sense that their individual errors are tolerable and can be compensated for through ensemble computation and randomized techniques.
3Productivity
If low precision arithmetic is used, then productivity is improved, but measurement precision deteriorates
Solution Approach 1:
The patent incorporates feedback mechanisms where computational results are repeatedly processed through the low-precision arithmetic elements. By iterating computations and using techniques like randomized rounding with compensation, the system gradually reduces accumulated errors and converges toward more accurate results, thereby maintaining adequate precision while benefiting from high-speed low-precision arithmetic operations.
Solution Approach 2:
The patent performs more arithmetic operations than strictly necessary - using excessive computation with low-precision elements - and then selects or averages results to achieve the desired effective precision. This approach trades computational redundancy for individual operation simplicity, allowing high productivity while achieving adequate measurement precision through statistical and algorithmic techniques.
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.