Hiroshige Fujii

Interleaving based variable ordering methods for ordered binary decision diagrams

Ordered binary decision diagrams (OBDDs) are efficient representations of Boolean functions and have been widely used in various computer-aided design tools. Since the size of an OBDD depends on variable ordering, it is important to find a good variable order for the efficient manipulation of OBDDs. In particular, it is important to find the same good variable order for multiple functions, since multiple functions are handled at the same time in most computer-aided design tools. The paper describes new variable ordering algorithms for multiple output circuits. The new algorithms use variable interleaving, while conventional algorithms use variable appending. For some benchmark circuits, OBDDs have been successfully generated by using the new algorithms, while they have not been generated by using conventional algorithms. Consequently, the new variable ordering algorithms are effective and allow us to apply OBDD-based CAD tools to wider classes of circuits.

A 0.5-V power-supply scheme for low-power system LSIs using multi-V/sub th/ SOI CMOS technology

This paper proposes a novel power-supply scheme suitable for 0.5-V operating silicon-on-insulator (SOI) CMOS circuits. The system contains an on-chip buck DC-DC converter with over 90% efficiency, 0.5-V operating logic circuits, 100-MHz operating flip-flops at 0.5-V power supply, and level converters for the interface between the 0.5-V operating circuit and on-chip digital-to-analog (D/A) converters or external equipment. Based on the theory, the values of on-resistance and threshold voltage of SOI transistors are clarified for the 0.5-V/10-mW output DC-DC converter, which satisfies both high efficiency and low standby power. The proposed flip-flop can hold the data during the sleep with the use of the external power supply, while maintaining high performance during the active. The level converter comprises dual-rail charge transfer gates and a CMOS buffer with a cross-coupled nMOS amplifier to operate with high speed even in a conversion gain of higher than 6, where the conversion gain is defined as the ratio of the output and input signal swings. The test chip was fabricated for the 0.5-V power supply scheme by using multi-V/sub th/ SOI CMOS technology. The experimental results showed that the buck DC-DC converter achieved a conversion efficiency of 91% at 0.5-V/10-mW output with stable recovery characteristics from the sleep, and that the dual-rail level converter operated with a maximum data rate of 300 Mb/s with the input signal swing of 0.5 V.

A floating-point cell library and a 100-Mflops image signal processor

A new floating point macro cell library, suitable for logic synthesis of image signal processor, has been developed. A floating point Arithmetic Logic Unit(ALU), a floating point multiplier(MPY), an instruction RAM and a data register file are included in the library. The ALU and MPY can support not only IEEE754 floating point operations, but also fixed point operations and logical operations which are often used in image signal processing. They can operate at 33MHz clock cycle with three stage pipeline configuration. A new algorithm for calculation of absolute value is implemented in the ALU to get such a high speed operation. A new type of Vector Processor was synthesized with logic synthesis system. It has peak performance of 100MFLOPS at 33MHz, and it is suitable for large scale image processing, such as FFT, DCT, VQ and so on. High Speed 1.2 micron CMOS fabrication technology was used. In this paper, the details of the new algorithm used in the floating point ALU are discussed. Each component of the library and the synthesized Vector Processor is also described.

A new reduced clock-swing flip-flop: NAND-type keeper flip-flop (NDKFF)

A new reduced clock-swing flip-flop, named NAND-type Keeper Flip-Flop (NDKFF) is proposed. Compared with other conventional reduced clock-swing flip-flops such as HSFF and RCSFF, NDKFF features a simple configuration, which does not have additional clock drivers or does not have additional nand/or p-wells. Compared with the hybrid-latch flip-flop, 52% of the flip-flop power and 64% of the clocking power are saved in the case of 0.25 /spl mu/m CMOS technology. Moreover CLK-to-Q delay is comparable to that of conventional C2MOS-type master-slave flip-flop.