AI Accelerator Primitive Testing With Seed-Based Coverage Expansion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional test generation techniques for AI hardware accelerators are inefficient and time-consuming, leading to prohibitive test coverage and error analysis challenges due to the large space and time required for system level testing, making effective error detection and analysis impossible.
Innovation Solution
A method involving the use of seed units and random units to generate test cases, combined with hardware execution of daisy-chained function codes and software simulation for result prediction, reduces test data build time and execution time, while enabling effective test coverage and error analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional test generation techniques are used for AI hardware accelerator testing, then test coverage can be achieved, but test generation time and space consumption become prohibitively large
Solution Approach 1:
The patent pre-generates a limited set of seed test data that covers critical functionality, then uses this seed data to systematically generate additional test cases. This preliminary action reduces the overall test generation time by avoiding random exhaustive generation while maintaining effective coverage.
Solution Approach 2:
The patent segments the test data generation process into multiple stages: generating seed test data, expanding to additional test cases, and prioritizing test execution. This segmentation allows the system to manage test generation time by processing tests in phases rather than requiring all tests to be generated and executed simultaneously.
2Reliability
If conventional test generation techniques are used for AI hardware accelerator testing, then test coverage can be achieved, but space consumption becomes prohibitively large
Solution Approach 1:
The patent segments test data into seed data and generated test cases, storing only essential information at each stage. This segmentation reduces space consumption by avoiding storage of redundant intermediate data while maintaining the ability to generate complete test suites.
Solution Approach 2:
The patent uses seed test data as templates to generate additional test cases through systematic variations rather than storing all possible test data. This copying approach reduces space requirements by generating test data on-demand from compact seed representations.
3Measurement precision
If comprehensive test data is generated for error detection, then error analysis accuracy improves, but error printing and analysis time becomes prohibitive
Solution Approach 1:
The patent extracts and prioritizes only the most relevant test cases for execution based on risk assessment and coverage requirements. By taking out only the essential tests rather than running all possible tests, the system maintains error detection capability while significantly reducing error analysis time.
Solution Approach 2:
The patent implements feedback mechanisms that use test results to dynamically adjust subsequent test execution and error analysis focus. This feedback allows the system to concentrate error printing efforts on the most likely failure modes, reducing overall analysis time while maintaining accuracy.
Data Source
AI summary
Methods, system, and computer program product are disclosed for enhanced system level testing of an AI hardware accelerator comprising Artificial Intelligence (AI) primitives. The system level testing provides uncompromised test coverage of the AI hardware accelerator while enabling efficient and effective testing performance. Enhanced test data generation, result prediction, and test execution are provided for system level testing of AI hardware accelerator including the AI primitives.