NPU Runtime Self-Testing for Scan-Based Defect Isolation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural processing units (NPUs) in critical systems like autonomous vehicles and drones can fail due to minor defects not detected during pre-shipment testing, leading to unpredictable AI operations and potential safety hazards.
Innovation Solution
A method for performing runtime tests on NPUs, using a component tester to scan test functional components, including memory instances and processing elements, to detect defects through scan chains and monitor states, allowing for the isolation or deactivation of faulty components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If pre-shipment testing is performed on NPU components, then manufacturing defects can be detected, but minor or minute defects may still be present and undetected
Solution Approach 1:
The patent applies preliminary action by performing scan tests on functional components during runtime before defects can cause system failure. The component tester continuously monitors and tests components (memories, processing elements, controllers) during operation, detecting minor defects before they amplify into critical failures. This proactive testing approach complements pre-shipment testing by catching defects that escape initial manufacturing checks.
2Productivity
If NPU operates continuously in mission-critical systems, then productivity is maintained, but defects may be amplified over time due to fatigue stress and physical stress
Solution Approach 1:
The patent implements feedback by continuously monitoring component states during runtime and using this information to detect defects. The component tester receives feedback signals from functional components and analyzes test results to identify defective components. This closed-loop monitoring system allows the NPU to maintain productivity while continuously assessing component health and detecting defects before they lead to system failure.
Solution Approach 2:
The NPU performs self-testing through the integrated component tester that monitors its own functional components during runtime. The system uses scan chains and test patterns to automatically detect defects in memories, processing elements, and controllers without external intervention, enabling continuous operation while maintaining reliability through self-diagnosis.
3Reliability
If scan test is performed on first group of memory instances, then defects can be detected, but second group of memory instances must be available for operation
Solution Approach 1:
The patent applies segmentation by dividing memory instances into multiple groups (first group and second group). During runtime, the component tester can select one group for scan testing while the other group remains available for normal NPU operations. This segmentation allows defect detection in one memory group without completely halting system functionality, as the other group continues to serve operational needs.
Solution Approach 2:
The patent implements partial action by testing only one group of memory instances at a time rather than requiring all memory instances to be tested simultaneously. The component tester selectively applies scan tests to the first group of memory instances while the second group remains operational, providing sufficient defect detection coverage without completely stopping system productivity.
4Reliability
If component tester monitors each state of functional components, then defects can be detected during runtime, but system complexity increases
Solution Approach 1:
The patent merges the component tester functionality with the NPU's existing operational infrastructure. The tester integrates with functional components (memories, processing elements, controllers) and uses the same interconnect structures and control mechanisms already present in the NPU. This merging approach enables runtime defect monitoring without adding substantial external complexity, as the testing capability is built into the existing system architecture.
Data Source
AI summary
A neural processing unit (NPU) for testing a component during runtime is provided. The NPU may include a plurality of functional components including a first functional component and a second functional component. At least one of the plurality of functional components may be driven for calculation of an artificial neural network. Another one of the plurality of functional components may be selected as a component under test (CUT). A scan test may be performed on the at least one functional component selected as the CUT. A tester for detecting a defect of an NPU is also provided. The tester may include a component tester configured to communicate with at least one functional component of the NPU, select the at least one functional component as a CUT, and perform a scan test for the selected CUT.