TL;DR
This paper introduces a novel IEEE 1500-compatible test wrapper for Network-on-Chip (NoC) based Systems-on-Chip (SoCs). This wrapper standardizes testing of diverse embedded IP cores, incorporates a response comparator for immediate fault detection, and supports both unicast and multicast testing modes, significantly improving test efficiency and flexibility.
What Problem Does This Paper Address?
As SoCs increasingly integrate pre-designed, reusable Intellectual Property (IP) cores, testing these complex systems becomes a major challenge. Specifically, testing embedded IP cores within a Network-on-Chip (NoC) architecture presents difficulties due to limited external observability and controllability. Traditional test methods are inefficient for the large number of cores in an NoC. Furthermore, the diversity of IP cores and testing strategies complicates test integration. The paper aims to address this by proposing a unified test wrapper that makes different IP cores appear homogeneous for test purposes, enabling efficient and flexible testing within the NoC environment, including support for both unicast and the more efficient multicast testing.
Methods
The core contribution is the design of a compatible IEEE 1500 standard test wrapper. This wrapper acts as an interface between the Test Access Mechanism (TAM) and the IP cores. Key features include:
- IEEE 1500 Compatibility: Adheres to the IEEE 1500 standard for embedded core testing, ensuring interoperability and simplifying system-level test integration.
- Homogenization of IP Cores: Makes diverse IP cores appear identical from a test integration perspective, simplifying the overall test architecture.
- Test Response Comparator: A novel addition that allows for direct, on-chip comparison of the IP core's actual test response against the expected response. This immediate verification improves testing efficiency and reduces the need to transfer all test results off-chip.
- Multi-mode Testing Support: The wrapper is designed to support various test modes, including unicast and multicast testing.
- NoC Integration: The wrapper is designed to leverage the NoC's communication infrastructure for test data transfer, rather than relying solely on traditional buses.
- Wrapper Interface: Includes standard IEEE 1500 ports (WSI, WSO, WPI, WPO, WIP) and registers (WIR, WBR, WBY) for control and data transfer.
- Wrapper Boundary Cells (WBCs): Used to provide observability and controllability for the IP core, enabling different test modes.
The paper also details the NoC architecture, a 2D Mesh, and discusses unicast and multicast data transmission modes within the NoC context. For multicast testing, a specific transmission mode based on paths is adopted, and routing schemes are proposed to avoid deadlock.
Datasets
The paper uses ISCAS '89 benchmark circuits (S386, S444, S526, and S1238) as representative IP cores for evaluation. These circuits were integrated into a 2D Mesh NoC for testing. No new benchmarks were introduced.
Key Insights and Ideas
- On-chip comparison is key: Integrating a test response comparator directly into the wrapper significantly boosts testing efficiency by enabling immediate fault detection without needing to send all test results to an external test sink.
- IEEE 1500 is adaptable: The standard IEEE 1500 wrapper concept can be effectively adapted for NoC environments, provided modifications are made to leverage the network architecture for test access.
- Multicast testing offers substantial benefits: Multicast testing, when implemented efficiently, can drastically reduce test time and power consumption compared to unicast testing, especially as the number of cores increases.
- NoC structure can be reused for testing: Utilizing the existing NoC communication infrastructure for test data transfer is a practical and efficient approach to overcome the challenges of testing embedded cores in NoCs.
- Wrapper design impacts overhead: While wrappers add area overhead, the benefits in terms of improved testability, flexibility, and efficiency often outweigh this cost, especially for complex SoCs.
Key Results
- Wrapper Functionality: The designed wrapper was simulated and verified to function correctly across various test modes.
- Size Overhead: The wrapper adds an average of approximately 33% to the original IP core size. This overhead decreases as the original IP core size increases, suggesting it's more acceptable for larger cores.
- Test Time and Power:
- Multicast testing demonstrated significantly lower test time and power consumption compared to unicast testing (which was implemented using subnet partitioning).
- For example, in one comparison, multicast testing showed a substantial reduction in test time compared to the most efficient unicast (subnet) scheme.
- The paper provides a table (Table 1) comparing hop counts and port numbers for different unicast (subnet) and multicast schemes, indicating that multicast generally requires fewer hops and ports for larger groups, leading to better efficiency.
- Adaptability: The wrapper showed good adaptability to support various NoC test strategies, including unicast and multicast testing on a 2D Mesh NoC.
Limitations and Future Work
- Limitations:
- The paper acknowledges that the current multicast routing schemes might still face challenges with deadlock avoidance, although proposed solutions aim to mitigate this.
- While the wrapper size overhead is deemed acceptable, further optimization for minimal area impact could be explored.
- Future Work:
- The authors state that further enhancing parallelism is a goal for wrapper optimization.
- Continued research on improving multicast communication efficiency and further refining routing algorithms for deadlock avoidance in NoC testing is suggested.
Summary updated