透過您的圖書館登入
IP:18.191.175.159
  • 學位論文

適用於溫度感知三維晶片內網路之交通與溫度平衡的可適性環城路由演算法和架構

Traffic- and Thermal- balanced Adaptive Beltway Routing Algorithm and Architecture Design for Thermal-Aware 3D NoC Systems

指導教授 : 吳安宇

摘要


本文針對熱感知三維晶片內網路交通與溫度上的不均所導致的效能降低問題做出演算法和架構設計。因為目前在三維晶片內網路的最小路徑路由演算法會造成三維晶片內網路上交通的不均衡,進一步引發溫度上的不均。此種情形可能會造成及時溫度控制機制的啟動。此一機制會針對快超過臨界溫度的路由器來啟動壓制以便降溫,卻會造成系統拓撲變成非穩態不規則狀網狀拓撲,引發更嚴重的交通不均衡,進一步造成溫度分布更加嚴峻,使得系統效能進入一個惡性的循環。本文提出一個創新的環城路由演算法,設法解決在三維晶片內網路中水平面與垂直面交通不均的問題,而後解決在水平面溫度上的不均來最大化系統吞吐量以及改善溫度分布情形。在水平面加入非最短環城路徑來平衡交通,並且在非穩態不規則狀網狀拓撲中使用串接路由來增進水平面上的路徑多樣性來提升水平面的交通量與水平方向的可達性。在垂直面上則是使用了固定比例下傳階層路由演算法來紓解在晶片最底層因傳統下傳演算法所造成的交通壅塞問題,達到在垂直面上的交通均衡來最大化系統吞吐量。綜合以上在水平面與垂直面的技術,使得網路上在水平與垂直的交通與溫度更加平均,使系統效能能夠穩定且最大化吞吐量。與根據實驗結果,我們提出的演算法能有效地增加效能,以及使交通負擔量更加平衡。我們基於低成本的記憶體縮減實現技術,相較於無法處理非穩態不規則狀網狀拓撲的硬體架構,我們只需多26.7%的實現成本,就可以在非穩態不規則狀網狀拓撲下平均獲得 2.7倍的吞吐量提升。面積效能比也達到了2.1倍,意即在同樣的面積底下,我們的演算法與其他相關成果相比,可以得到2.1倍的效能提升。

並列摘要


In this thesis, we proposed Traffic- and Thermal- balanced Adaptive Beltway Routing (TTABR) algorithm and architecture design for performance reduction due to the traffic load and thermal distribution imbalance in thermal-aware 3D network-on-chip (NoC). The minimal path routing on 3D NoC cause the unbalance traffic load, which also induce the thermal distribution imbalance. To ensure thermal safety and avoid huge performance back-off from the temperature constraint, run time thermal management is required. However the regulation of temperature requires throttling of the near-overheated router, which makes the topology become Non-Stationary Irregular Mesh (NSI-mesh). It still has performance degradation, and the traffic load imbalance gets worse. Hence the thermal distribution might also become worse and trigger more routers to be throttled. We manage to break this loop to get a better performance and stable 3D NoC systems. TTABR aims to balance the lateral traffic load. It has providing the non-minimal path to increase path diversity and using novel cascade routing to heave the lateral traffic. TTABR also proposed to solve the traffic load imbalance in the vertical direction. Based on the experimental results, the proposed routing scheme can significantly improve the performance and balance traffic load. For low cost implementation, we also propose memory reduction techniques, and we gain 2.7x throughput improvement for only 26.7% area overhead. The throughput per area of our proposed algorithm is 2.1x compared to other related work.

參考文獻


[2] R. Marculescu, U. Y. Ogras, P. Li-Shiuan, N. E. Jerger, and Y. Hoskote, "Outstanding Research Problems in NoC Design: System, Microarchitecture, and Circuit Perspectives," Computer-Aided Design of Integrated Circuits and Systems, IEEE Transactions on, vol. 28, pp. 3-21, 2009.
[3] ITRS, International Technology Roadmap for Semiconductors, [Online]. Available: http://public.itrs.net.
[5] D. Sylvester and K. Keutzer, “A global wiring paradigm for deep submicron design,” IEEE Trans. CAD/ICAS, vol. 19, pp. 242-252, Feb. 2000.
[6] L. Benini and G. De Micheli, “Networks on chips: A new SoC paradigm,” IEEE Computer, vol. 35, pp. 70–78, Jan. 2002.
[7] A. W. Topol, D. C. La Tulipe, Jr., L. Shi, D. J. Frank, K. Bernstein, S. E. Steen, A. Kumar, G. U. Singco, A. M. Young, K. W. Guarini and M. Ieong, “Three-dimensional integrated circuits,” IBM J. Res. Develop., vol. 50, no. 4/5, pp. 491-506, 2006.

延伸閱讀