Fixed-Point ML Processor Architecture for Large-Load Computation
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 efficiently, with a DMA unit and storage unit for data management.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If general purpose processors are used to process large loads, then versatility is maintained, but processing speed and efficiency deteriorate
Solution Approach 1:
The processor is divided into a control unit and multiple processing units (first processing unit, second processing unit, third processing unit). Each processing unit is dedicated to specific data types (floating-point, fixed-point, integer), allowing parallel processing while maintaining versatility through coordinated operation under the control unit.
Solution Approach 2:
The patent introduces a new dimension of processing by adding processing units that operate in parallel with the traditional floating-point unit. This multi-dimensional processing architecture handles different data types simultaneously, increasing overall throughput without sacrificing the ability to handle diverse workloads.
2Productivity
If general purpose processors operate at high speed, then processing efficiency improves, but handling large loads becomes limited
Solution Approach 1:
The processing capacity is segmented across multiple specialized units, each optimized for specific operations. The first processing unit handles floating-point operations, the second handles fixed-point operations, and the third handles integer operations, collectively increasing load capacity while maintaining high efficiency for each data type.
Solution Approach 2:
Multiple processing units with different specializations are merged into a single processor system. This combination allows the processor to handle diverse and large loads by distributing work across units optimized for different data types, thereby increasing overall load capacity without compromising processing efficiency.
3Adaptability or versatility
If data type conversion is performed, then processing compatibility improves, but conversion overhead increases
Solution Approach 1:
The control unit pre-processes operation instructions to identify and prepare data conversion requirements before execution. By analyzing the instruction stream in advance and pre-converting data where possible, the system reduces conversion overhead during actual processing while maintaining compatibility across different data types.
Solution Approach 2:
The control unit acts as an intermediary that manages data type conversions between different processing units. It translates operations between floating-point, fixed-point, and integer formats, reducing conversion overhead by handling conversions centrally rather than requiring each processing unit to perform conversions independently.
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.