Cloud AI Processor Simulation for Pre-Hardware Algorithm Debugging

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current AI processor technologies require hardware availability for debugging AI algorithms, leading to delayed product releases and missed market opportunities due to the inability to debug algorithms before the AI processor has taped out.

Innovation Solution

A data processing method that uses a cloud AI processing platform to simulate an AI processor, allowing for debugging and testing of AI algorithms without physical hardware, by generating binary instructions based on device information and executing AI learning tasks to generate offline running files that can be deployed on compatible SoC chips.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If AI processor hardware is required for debugging AI algorithms, then algorithm testing can be performed on actual hardware, but product development cycle is extended and market opportunities are missed

Engineering Contradiction:
Improvealgorithm testing accuracyVSAvoidproduct development cycle
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent creates a virtual simulation system that copies the essential characteristics of the AI processor hardware. The simulation platform includes a virtual AI processor core, virtual memory system, and virtual peripheral devices that replicate the target hardware's architecture and behavior, enabling algorithm debugging before physical hardware is available.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent enables preliminary debugging and testing of AI algorithms on the simulated hardware platform before the actual AI processor is tape-out or physically available. This preliminary action allows developers to verify algorithm correctness, performance, and compatibility in advance, eliminating the waiting period that previously existed between hardware design and algorithm development.

Inventive Principle:
Principle #10Preliminary action

2Manufacturing precision

If separate development environments are created for each SoC chip, then hardware-specific optimization can be performed, but device complexity and development cost increase

Engineering Contradiction:
Improvehardware optimizationVSAvoiddevelopment environment complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent designs a universal simulation platform that can emulate multiple different AI processor architectures and configurations through a single system. The virtualization layer allows one development environment to serve multiple SoC chip types by dynamically configuring virtual hardware parameters, eliminating the need for separate dedicated development environments for each chip architecture.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent introduces a virtualization layer as an intermediary between the algorithm software and the physical hardware. This virtualization layer abstracts the hardware details and provides a unified interface for algorithm development, while still allowing hardware-specific optimizations to be tested through configuration of virtual hardware parameters without requiring separate development environments for each target device.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentUS11934940B2AI processor simulation
Publication Date: 2024.03.19 CAMBRICON TECH CO LTD
  • US11934940B2 patent drawing
  • US11934940B2 patent drawing
  • US11934940B2 patent drawing

AI summary

The present disclosure discloses a data processing method and related products, in which the data processing method includes: generating, by a general-purpose processor, a binary instruction according to device information of an AI processor, and generating an AI learning task according to the binary instruction; transmitting, by the general-purpose processor, the AI learning task to the cloud AI processor for running; receiving, by the general-purpose processor, a running result corresponding to the AI learning task; and determining, by the general-purpose processor, an offline running file according to the running result, where the offline running file is generated according to the device information of the AI processor and the binary instruction when the running result satisfies a preset requirement. By implementing the present disclosure, the debugging between the AI algorithm model and the AI processor can be achieved in advance.