| As integrated circuit (IC) technologies step into nanometer regime, modeling and simulation of parasitic interconnects, which dominate the circuit performance, become a challenging problem. Model order reduction has been proved to be an efficient technique to reduce the complexity of interconnect circuits. However, existing model order reduction approaches loss their efficiency when the number of ports is large. One way to mitigate this long-standing problem is by means of combined terminal reduction and model order reduction. In this thesis, we first propose a new combined terminal and model reduction method, ESVDMOR, which utilizes higher order moment information during the terminal reduction and applies singular value decomposition low-rank approximation on input and output terminals separately. Second, a clustering based terminal reduction method, TermMerg, is proposed to find the dominant terminals of the circuit. TermMerg uses high order moments to build the input/output moment matrix, apply K-means clustering method to partition all terminals into different clusters, and select a representative terminal for each cluster. Finally, we propose a new decoupled terminal reduction strategy, MIRST, which considers the spatial correlation of the terminals and the temporal correlation of the input signals from design. Using this method, the original system can be partitioned into several subsystems, which only include fewer number of ports.; Owning to increasing power consumption and the rapid heat generation on a die, efficient on-chip temperature regulation becomes imperative for today's high performance microprocessors. Temperature tracking based on software sensors is more flexible and comprehensive than on-chip physical thermal sensors, where temperature of any location is computed based on realtime power information. In this thesis, we propose three very efficient new thermal analysis methods, which are suitable for fast temperature tracking and runtime thermal regulation. The first method, TMMSpectrum, exploits the periodic patterns in power consumptions of the architecture modules in microprocessors and embedded high-performance systems. We propose to use spectrum analysis in frequency domain to compute the periodic responses of temperatures. The second method, TMMPWC, speeds up the thermal analysis by using piecewise constant average power inputs that determine the trend of temperature changes. The third algorithm, FEKIS, combines the Krylov sub-space reduction and the fixed large-step integration, to future speed up the simulation for the long power traces, which is specially suitable for runtime temperature tracking. |
