NPU Runtime Scan Testing for Faulty Component Isolation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural processing units (NPUs) in critical systems like autonomous vehicles and drones may operate erratically due to minor defects not detected during pre-shipment testing, which can amplify over time, leading to unpredictable AI behavior and potential safety issues.
Innovation Solution
Implementing a runtime testing system that includes a component tester to perform scan tests on NPUs, connecting flip-flops to form scan chains, applying test inputs, and analyzing the results to detect defects, allowing for the isolation or deactivation of faulty components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If pre-shipment testing is performed on NPU components, then manufacturing defects can be detected, but minor or minute defects may remain undetected and cause erratic operation during runtime
Solution Approach 1:
The patent implements scan chains and test modes that enable runtime testing of NPU components, allowing defects to be detected during operational periods rather than only during pre-shipment testing. This preliminary action of preparing test infrastructure enables continuous monitoring and detection of minor defects that would otherwise remain undetected.
Solution Approach 2:
The component tester continuously monitors functional components during runtime and provides feedback about their operational status. This feedback mechanism allows the system to detect defects as they occur or develop, enabling real-time identification of components that have degraded or failed, thereby improving overall reliability.
2Ease of manufacture
If increased density of semiconductor devices is used to reduce manufacturing cost, then production becomes cheaper, but defect rate increases due to increased complexity
Solution Approach 1:
The NPU is divided into multiple functional components (processing elements, memory blocks, controllers) that can be independently tested. Scan chains are segmented to test specific groups of flip-flops and components separately. This segmentation allows comprehensive testing of complex high-density devices without requiring complete system disassembly, making testing feasible for cost-effective high-density manufacturing.
Solution Approach 2:
The NPU includes built-in test infrastructure (scan chains, test modes, component testers) that enable self-testing of the device. This self-service capability allows the complex high-density NPU to test its own components without requiring external specialized equipment, thereby maintaining manufacturing cost effectiveness while improving defect detection capability.
3Reliability
If runtime testing is implemented to detect defects, then NPU reliability improves, but system complexity increases due to additional testing infrastructure
Solution Approach 1:
The test infrastructure is merged with the normal operational infrastructure of the NPU. Scan chains use the same flip-flops and logic paths that are already present in the functional design. The component tester integrates with the existing control and data flow structures. This merging approach enables runtime testing without adding significant external complexity, as the testing capability is embedded within the existing architectural framework.
Solution Approach 2:
The scan chains and test modes serve multiple functions: they enable both normal operational data flow and test data flow through the same hardware structures. The flip-flops serve dual purposes of storing operational data and facilitating scan testing. This multi-functionality reduces the need for separate dedicated test hardware, thereby limiting the increase in overall system complexity while enabling comprehensive runtime testing.
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.