On-line Distributed Thermal Sensing and Monitoring of Multicore Systems

As the number of cores increases thermal challenges increase thereby degrading performance and reliability. We approach this challenge with an online distributed thermal sensing and monitoring method which is based on the use of thermal sensors. In this work, we propose an approach for the strategic placement of thermal sensors in the multicore systems which takes into consideration the thermal effect of one core on the other. Since leakage currents are sensitive to temperature and increase with scaling we propose to use a leakage current based thermal sensing for monitoring purposes. We investigate and simulate an existing leakage current based thermal sensor at 65 nm CMOS technology and show that performance and delay still improves with temperature. A novel sensing interconnection network structure based on self-timed signaling model, comprising of an encoder, transmitter and receiver is described. Furthermore we have identified three different ways for thermal monitoring of multicore systems: core based, cluster based and centralized monitoring and have proposed an hybrid monitoring approach which offers an optimal trade-off between performance and overhead. We have also suggested an hybrid monitoring algorithm with mixed granularities for the same.