Configurable Neural Network Engine Optimizing MAC Operations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning systems face challenges in efficiently performing deep learning network operations due to the high demand for multiply-accumulate (MAC) operations, which are space-intensive and require specialized hardware, especially in system-on-a-chip (SoC) devices, where space is limited, and current open-sourced AI engines are not compatible with various field-specific needs.
Innovation Solution
A configurable hardware implementation for neural networks using a controller, configuration buffer, data buffer, and multiple computational layers, including a MAC layer with multiple MAC units, that can load and apply configurations and parameters to perform computations for fully-connected neural networks (FNN) and convolutional neural networks (CNN) operations, optimizing MAC operations and supporting different network topologies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a large amount of hardware components are placed to perform MAC operations, then the computational capability for neural network operations is improved, but the space consumption on SoC chip increases
Solution Approach 1:
The patent implements a configurable hardware neural network engine where a single hardware platform can be reconfigured through software loading to support different neural network architectures and computational tasks. The same physical hardware components can perform various MAC operations for different applications by loading appropriate configuration data, eliminating the need for dedicated hardware for each specific computational task.
Solution Approach 2:
The patent changes the operational parameters of the hardware neural network engine by loading different configuration data and weights into the computational layers. This allows the same hardware circuitry to be dynamically reconfigured to perform different computational operations, effectively changing its functionality without physical modification.
2Ease of operation
If open-sourced AI engines are used, then the ease of implementation is improved, but the adaptability to field-specific needs deteriorates
Solution Approach 1:
The patent creates a dynamic hardware neural network engine that can be reconfigured through software loading to adapt to different field-specific requirements. The system transitions from a static, fixed-function hardware design to a dynamic platform where computational layers can be programmed with different architectures and parameters to match specific application needs.
Solution Approach 2:
The configurable hardware engine serves as a universal platform that can be adapted to various field-specific applications by loading appropriate configurations. This single hardware system can support multiple neural network architectures and computational tasks, providing both ease of implementation and high adaptability.
Data Source
AI summary
Systems, apparatus and methods are provided for performing computations of a neural network using hardware computational circuitry. An apparatus may include a controller, a configuration buffer and a data buffer. The controller may be configured to dispatch computing tasks of a neural network, load configurations into the configuration buffer and load input data and parameters including weights and biases into the data buffer. The apparatus may also include a multiply-accumulate (MAC) layer. The configurations may include at least one FNN configuration. The MAC layer may apply the at least one FNN configuration, which includes settings for a FNN operation topology for the MAC layer to perform computations for at least one FNN layer. Optionally, the neural network may be a CNN and the configurations may further include at least one CNN configuration for the MAC layer to perform computations for at least one CNN layer.