Affiliation:
1. University of Ferrara, Italy
2. Universidad Politécnica de Valencia, Spain
Abstract
Source synchronous links for use in multi-synchronous networks-on-chip (NoCs) are becoming the most vulnerable points for correct network operation and must be safeguarded against intra-link delay variations and signal misalignments. The intricacy of matching link net attributes during placement and routing and the growing role of process parameter variations in nanoscale silicon technologies are the root causes for this. This article addresses the challenge of designing a process variation and layout mismatch tolerant link for synchronizer-based GALS NoCs by implementing a self-calibration mechanism. A variation detector senses the variability-induced misalignment between data lines with themselves and with the transmitter clock routed with data in source synchronous links. A suitable delayed replica of the transmitter clock is then selected for safe sampling of misaligned data. The manuscript proves robustness of the link in isolation with respect to a detector-less link, but also assesses integration issues with the downstream synchronizer and switch architecture, proving the benefits in a realistic experimental setting for cost-effective NoCs.
Reference36 articles.
1. Intrinsic threshold voltage fluctuations in decanano MOSFETs due to local oxide thickness variations
2. Beer, S., Ginosar, R., Priel, M., Dobkin, R., & Kolodny, A. (2010). The devolution of synchronizers. In Proceedings of the IEEE Symposium on Asynchronous Circuits and Systems (pp. 94-103).
3. Bonesi, S., Bertozzi, D., Benini, L., & Macii, E. (2008). Process variation tolerant pipeline design through a placement-aware multiple voltage island design style. Design, Automation and Test in Europe, 967-972.
4. Borkar, S., Karnik, T., & Vivek, D. (2004). Design and reliability challenges in nanometer technologies. In Proceedings of the 41st Design Automation Conference (p. 75).
5. Impact of die-to-die and within-die parameter fluctuations on the maximum clock frequency distribution for gigascale integration