Development of a thermal management method to reduce hotspots and to balance the temperature distribution has become an important issue. In this paper, we propose a static thermal management technique at compiler level. The target machine is a VLIW architecture where the compiler is required to schedule instructions to achieve instruction level parallelism (ILP). Two technique are proposed. The first one is register binding to balance the temperature of the register file by taking both spatial and temporal thermal information into consideration. The second one is forwarding methods including forwarding-aware architecture and instruction scheduling to reduce the access count of register file. The experimental results show that by combining the two techniques, the peak temperature reduction can reach 7.89 ( o C) in the best case and 7.22 ( o C) in average with only 0.9% performance penalty in average.
In deep sub-micron technology, the crosstalk effect between adjacent wires has become an important issue, especially between long on-chip buses. This effect leads to the increase in delay, in power consumption, and in worst case, to incorrect result. In this paper, we propose a de-assembler/assembler structure to eliminate undesirable crosstalk effect on bus transmission. By taking advantage of the prefetch process where the instruction/data fetch rate is always higher than instruction/data commit rate in high performance processors, the proposed method would hardly reduce the performance. In addition, the required number of extra bus wires is only 7 as compared with 85 needed in [6] when the bus width is 128 bits.
IR-drop problem during test mode exacerbates delay defects and results in false failures. In this paper, we take the Xfilling approach to reduce IR-drop effect during at-speed test. The main difference between our approach and the previous X-filling methods [7]-[9] lies in two aspects. The first one is that we take the spatial information into consideration in our approach. The second one is how X-filling is performed. We propose a backward-propagation approach instead of a forwardpropagation approach taken in previous work. The experimental results show that we have 42.81% reduction for the worst IRdrop and 45.71% reduction in the average IR-drop as compared to random fill method.978-3-9810801-5-5/DATE09
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.