Performance Analysis of Temperature Management Approaches in Networks-on-Chip

With the progress of deep submicron technology, power consumption and temperature related issues have become dominant factors for chip design. Therefore, very large-scale integrated systems like Systems-on-Chip (SoCs) are exposed to an increasing thermal stress. On the one hand, this necessitates effective mechanisms for thermal management. On the other hand, application of thermal management is accompanied by disturbance of system integrity and degradation of system performance. In this paper the authors propose to precompute and proactively manage on-chip temperature of systems based on Networks-on-Chip (NoCs). Thereby, traditional reactive approaches, utilizing the NoC infrastructure to perform thermal management, can be replaced. This results not only in shorter response times for application of management measures and a reduction of temperature and thermal imbalances, but also in less impairment of system integrity and performance. The systematic analysis of simulations conducted for NoC sizes ranging from 2x2 to 4x4 proves that under certain conditions the proactive approach is able to mitigate the negative impact of thermal management on system performance while still improving the on-chip temperature profile. DOI: 10.4018/jertcs.2012100102 20 International Journal of Embedded and Real-Time Communication Systems, 3(4), 19-41, October-December 2012 Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. the integrity of Integrated Circuits (ICs) and have major influence on operability, lifetime and performance. The relationship between temperature and deterioration is illustrated by the Arrhenius model (Srinivasan & Adve, 2003) describing the influence of temperature on the velocity of chemical reactions. This model originates from the Van’t Hoff rule also known as the reaction rate temperature rule (or RGT rule), saying that chemical reactions take place twice as fast when temperature is increased by 10 K. As a rule of thumb, this also can be interpreted as a bisection of lifetime of ICs with every 10 K temperature increase. For this reason, monitoring and control of on-chip temperature distribution are important tasks to secure system functionality and ensure high performance. Typically, monitoring of on-chip temperature is performed by collecting temperaturerelated data (e.g., by using integrated diodes). In order to react to undesirable temperatures this data has to be transferred to a component responsible for data evaluation and determination of appropriate reactions (i.e., thermal management). Then instructions are sent to the concerned components. For NoC-based systems, commonly the NoC infrastructure is used for this communication. Despite the importance of thermal management, reactive approaches impair system performance, since the utilization of the NoC presents an intrusion into the system and the induced traffic curtails the availability of the NoC for regular communication. Another drawback is the comparatively long response time in the order of several milliseconds of thermal management caused by the time that passes between the occurrence of a switching activity and the corresponding temperature change (see end of section Reactive Management for an example calculation of heat propagation delay) as well as the transmission delay (see DP in Table 7 for an uncongested example NoC), which is induced when using the NoC for reporting that change. Since two transmissions (i.e., reporting temperature and sending instructions) are necessary, an already highly congested NoC additionally exacerbates thermal management. This results in possibly delayed reactions to undesired temperatures and unnecessary long periods of reduced performance. Hence, we propose to predict the on-chip temperature profile based on a model that is realized as part of a Thermal Management Unit (TMU), instead of only reverting to physical sensors. This model relies on switching activities in order to calculate the temperature profile. By means of the made predictions, the TMU is able to immediately initiate execution of instructions for thermal management. Such a TMU can be implemented in software running on a core of the SoC or it is an inherent part of a core implemented in hardware. Thereby, response time for thermal management is shortened by avoiding the waiting period from switching activity to temperature change and by reducing the number of monitoring-related transmissions (i.e., only switching activities are reported instead of temperature rises and drops) to the TMU. Additionally, this reduces the traffic load of the NoC and frees up communication capacities for regular data traffic. Prerequisites are that predictions can be accomplished rather fast without generating too much additional heat induced by the logic necessary for the activity counters and the calculation process for the model. To ascertain to which extent proactive thermal management influences system performance and on-chip temperature distribution, this approach is compared to a setup reverting to reactive management and to a setup without any thermal management. Briefly summarized, the contributions of this article are the following: • Introduction of a software-based temperature model for ICs; • Application of this model for proactive temperature management of NoCs; • Evaluation of proactive management and comparison to conventional reactive methods. The remainder of this paper is organized as follows. In section Related Work an overview over existing work regarding modeling of onInternational Journal of Embedded and Real-Time Communication Systems, 3(4), 19-41, October-December 2012 21 Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. chip temperature and approaches for reactive and proactive management strategies is given. In section Simulation Environment the environment for the simulation of proactive and reactive thermal management of NoC-based systems is introduced. In section Experiments and Results experiments focusing on the impact of proactive and reactive management on system performance, effort for thermal management and temperature are conducted. Finally, a conclusion and an outlook for future work are given.