Virtualized AI Hardware Framework for Edge Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional AI frameworks rely on virtual machine operating systems and software frameworks, leading to inefficient and costly AI solution model processing due to the need for multiple processors, GPUs, and extensive hardware resources, resulting in high power consumption, complexity, and time-consuming training processes.
Innovation Solution
A virtualized AI system with a multilane parallel hardware framework that executes AI solution models without a software processing framework, utilizing a secure AI solution hardware processing concept, eliminating the need for CUDA or Tensorflow, and enabling real-time, continuous training and inference with dynamic creation and management of AI system lanes and multilanes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If conventional virtualized AI frameworks use VM operating systems and software processing frameworks, then AI solution models can be executed with flexible software environments, but power consumption increases and processing efficiency decreases due to multiple processors and GPUs being required
Solution Approach 1:
The patent replaces the software-based processing framework with a hardware-based processing framework. Specifically, it substitutes the VM/OS/software stack with dedicated AI processing lanes that execute AI solution models directly in hardware, eliminating the need for multiple processors and GPUs while reducing power consumption and improving processing efficiency
Solution Approach 2:
The patent creates a universal hardware processing framework where a single AI processing lane can execute multiple different AI solution models through dynamic configuration. The lane composer and lane maintainer enable the same physical hardware to adaptively process various AI workloads, replacing the need for dedicated hardware for each model
2Power
If conventional AI frameworks use multiple processors and GPUs, then AI solution models can be processed with sufficient computational power, but device complexity increases and processing becomes more time-consuming
Solution Approach 1:
The patent segments the AI processing function into discrete, manageable lanes, each capable of independent execution. The lane composer divides AI solution models into segments that can be processed by individual lanes, and the lane maintainer manages the coordination. This segmentation allows complex AI workloads to be handled by simpler, dedicated hardware units rather than requiring complex multi-processor systems
Solution Approach 2:
The patent introduces a new dimension of parallelism through multiple AI processing lanes operating simultaneously. Instead of increasing computational power within a single processor or GPU, the system creates multiple independent processing dimensions, each handling a portion of the AI workload, thereby maintaining high computational power while reducing individual device complexity
3Ease of operation
If conventional AI frameworks use VM and OS in the execution path, then software compatibility and ease of operation are improved, but processing speed decreases and training becomes more time-consuming
Solution Approach 1:
The patent extracts and removes the VM and OS layers from the AI solution model execution path. By eliminating these software intermediaries, the system achieves direct hardware execution of AI models, dramatically improving processing speed while the lane composer and configurability maintain software compatibility through hardware-level abstraction
4Quantity of substance
If conventional AI systems use extensive hardware resources, then AI solution models can be trained and executed with sufficient capacity, but loss of time increases during training processes
Solution Approach 1:
The patent enables continuous AI training and inference by eliminating the idle time associated with software framework overhead and VM context switching. The hardware processing lanes maintain continuous operation, processing AI solution models without interruption, thereby reducing training time while utilizing hardware resources efficiently
Data Source
AI summary
An artificial intelligence (AI) system is disclosed. The AI system provides an AI system lane processing chain, at least one AI processing block, a local memory, a hardware sequencer, and a lane composer. Each of the at least one AI processing block, the local memory coupled to the AI system lane processing chain, the hardware sequencer coupled to the AI system lane processing chain, and the lane composer is coupled to the AI system lane processing chain. The AI system lane processing chain is dynamically created by the lane composer.


