The module-wise dynamic voltage and frequency scaling (MDVFS) scheme is applied to a single-chip H.264/MPEG-4 audio/visual codec LSI. The power consumption of the target module with controlled supply voltage and frequency is reduced by 40% in comparison with the operation without voltage or frequency scaling. The consumed power of the chip is 63 mW in decoding QVGA H.264 video at 15 fps and MPEG-4 AAC LC audio simultaneously. This LSI keep operating continuously even during the voltage transition of the target module by introducing the newly developed dynamic de-skewing system (DDS) which watches and control the clock edge of the target module.

In order to explore the feasibility of large-scale subthreshold logic circuits and to clarify the lower limit of supply voltage (VDD) for logic circuits, the dependence of the minimum operating voltage (VDD min ) of CMOS logic gates on the number of stages, gate types and gate width is systematically measured with 90 nm CMOS ring oscillators (RO's). The measured average VDD min of inverter RO's increased from 90 mV to 343 mV when the number of RO stages increased from 11 to 1 Mega, which indicates the difficulty of VDD scaling in large-scale subthreshold logic circuits. The dependence of VDD min on the number of stages is calculated using the subthreshold current model with random threshold voltage (VTH) variations and compared with the measured results, and the tendency of the measurement is confirmed. The effect of adaptive body bias control to compensate purely random VTH variation is also investigated. Such compensation would require impractical inverter-by-inverter adaptive body bias control.

This paper describes an efficient SPICE netlist reduction method, which enables collective simulation of large circuits. The method reduces a SPICE netlist to only those devices which affect the simulation results. Parts of the netlist can be significantly reduced in size, with relatively discrepancies arising between the original SPICE simulation and the reduced SPICE simulation. The authors' reduction method is more general than previous works, since it reduces circuits using the features of MOS transistors. According to experimental results, reduction rates can range from 1/2 to 1/223. Depending on the reduction, the time taken time to run a SPICE simulation was reduced by between one and two oder of magnitude. Using this method and working on the reduced netlist, SPICE could even handle netlist for very large circuits which it could not ordinarily handle. The simulation error between the original SPICE simulation and the reduced SPICE simulation was about 3.5%.