ML Accelerator Fixed-Point Conversion Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
General purpose processors are inefficient in processing large loads, leading to delays in machine learning tasks such as training artificial intelligence models, due to their limited operating speed.
Innovation Solution
A computation device and method that includes a controller unit, operation unit, and conversion unit, utilizing a primary processing circuit and secondary processing circuits to perform machine learning computations by converting input data into fixed-point data and executing operations to improve processing speed and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a general purpose processor is used to process machine learning data, then the device complexity is low, but the processing speed and efficiency are insufficient for large load computations
Solution Approach 1:
The computing device is segmented into distinct functional units: a controller unit for instruction parsing and data conversion, and an operation unit with multiple processing circuits for executing computations. This segmentation allows each component to be optimized for its specific function, improving overall processing speed while maintaining manageable device complexity through modular design.
Solution Approach 2:
The computing device is designed as a universal machine learning accelerator that can handle various types of machine learning computations through its multi-functional processing circuits. The device provides universal support for different data types and operations while maintaining high processing speed through its specialized architecture.
2Productivity
If a general purpose processor operates at higher speeds, then the processing speed improves, but the efficiency in handling large load computations decreases
Solution Approach 1:
The device changes the fundamental parameter of data representation by converting floating-point data to fixed-point data for machine learning computations. This parameter change enables more efficient processing of large load computations, significantly reducing computational time while improving overall efficiency through optimized data handling.
Solution Approach 2:
The controller unit performs preliminary actions by parsing computation instructions and converting data to fixed-point format before the actual computation begins. This preliminary processing prepares the data and instructions in an optimized format, enabling the processing circuits to execute computations more efficiently and reduce overall computational time.
3Measurement precision
If data conversion is performed for each computation, then the processing accuracy is improved, but the processing time increases
Solution Approach 1:
The controller unit performs data conversion to fixed-point format as a preliminary action before the computation is executed. By completing the data conversion in advance, the processing circuits can focus solely on executing the computation without time-consuming conversions during execution, thus maintaining high accuracy while minimizing the time overhead of data conversion.
Data Source
AI summary
The present disclosure provides a computation device. The computation device is configured to perform a machine learning computation, and includes an operation unit, a controller unit, and a conversion unit. The storage unit is configured to obtain input data and a computation instruction. The controller unit is configured to extract and parse the computation instruction from the storage unit to obtain one or more operation instructions, and to send the one or more operation instructions and the input data to the operation unit. The operation unit is configured to perform operations on the input data according to one or more operation instructions to obtain a computation result of the computation instruction. In the examples of the present disclosure, the input data involved in machine learning computations is represented by fixed-point data, thereby improving the processing speed and efficiency of training operations.