A 16-Mb CMOS SRAM using 0.4-µm CMOS technology has been developed. This SRAM features common-centroid-geometry (CCG) layout sense amplifiers which shorten the access time by 2.4 ns. A flexible redundancy technique achieves high efficiency without any access penalty. A memory cell with stacked capacitors is fabricated for high soft-error immunity. A 16-Mb SRAM with a chip size of 215 mm2 is fabricated and an address access time of 12.5 ns has been achieved.

By refining the deconvolution technique proposed previously by the same authors for determining the impulse response of optical fibers, the time resolution has been improved. In some cases an impulse response consisting of several independent pulses is obtained; probably these pulses correspond to separate LP-mode groups.

An all-CMOS output buffer has been developed. The output buffer is composed of a voltage-follower and a source-follower circuit. The performance of the output buffer is characterized by a low-voltage operation of 1.7 V, a short delay of 1 ns, availability for the wired-OR connection, and adjustability to TTL, ECL, and a reduced swing level (RSL). The output buffer is incorporated into a 64-kb CMOS SRAM. This SRAM has achieved an access time of 4.3 ns at a supply voltage of -3.6 V.

Recent advances in VLSI memories have enabled integration of 10 to 30 million devices on prototype chips for 16 Mbit DRAMs, 16 Mbit EPROMs, and 4 Mbit SRAMs. An experimental 64 Mbit DRAM recently reported clearly shows that an integration density of more than 100 million devices on a chip will be feasible in the near future. These advances have been made not only by progress in fine processing technology, but also by the development of three-dimensional memory cells such as trench capacitor cells and stacked capacitor cells for DRAMs and polysilicon PMOS load cells for SRAMs. Various circuit innovations to increase the signal-to-noise ratio and circuit speed have been, and will continue to be, essential. Future circuits will be required to operate at very low voltages, and the prototype 64 Mbit DRAM has shown that an operating voltage as low as 1.5 volts is feasible. Improvements in packaging technology for reducing package volume and footprint area, as well as for production of multipackage modules, are also becoming more and more important.

We propose a half-swing clocking scheme that allows us to reduce power consumption of clocking circuitry by as much as 75%, because all the clock signal swings are reduced to half of the LSI supply voltage. The new clocking scheme causes quite small speed degradation, because the random logic circuits in the critical path are still supplied by the full supply voltage. We also propose a clock driver which supplies half-swing clock and generates half VDD by itself. We confirmed that the half-swing clocking scheme provided 67% power saving in a test chip fabricated with 0.5 µm CMOS device, ideally 75%, in the clocking circuitry, and that the degradation in speed was only 0.5 ns by circuit simulation. The key to the proposed clocking scheme is the concept that the voltage swing is reduced only for clocking circuitry, but is retained for all other circuits in the chip. This results in significant power reduction with minimal speed degradation.

The dual-sensing-latch circuit proposed here can solve the synchronization problem of the conventional wave-pipelined SRAM, and the proposed source-biased self-resetting circuit reduces both the cycle and access time of cache SRAM's. A 16-kb SRAM using these circuit techniques was designed, and was fabricated with 0.25-µm CMOS technology. Simulation results indicate that this SRAM has a typical clock access time of 2.6 ns at 2.5-V supply voltage and a worst minimum cycle time of 2.6 ns.

High level power estimation model is indispensable to optimize architecture and programs in terms of power consumption systematically. This paper describes power analysis results of a general purpose programmable DSP using switch level and cell based power simulation. The analysis results help to know characteristics of the DSP components and to establish a power estimation model. A compensation method for the lack of intra-cell capacitance in a cell based simulation is proposed to improve the simulation accuracy within -20% of error against the measured power. We considered that the error is caused by ignoring short circuit current and is accurate enough for this level of simulation. Through a result of the power breakdown by modules of the DSP obtained by the simulation, it was found that bus power is much less than generally expected (the simulated power is less than 5% of the total), and that the data operation power dominates the chip power (up to 33%) and is strongly data dependent. The reason that the bus power is low is because both of the load capacitance and the activity are low. Some correlation between the number of input signal transitions and the power consumption is found in each of an ALU, register file, and multiplier through further investigation on the data operation modules. The correlation is worthwhile to establish a power estimation model and is eventually useful to optimize a DSP architecture and DSP programs. The importance of power estimation model is demonstrated by showing an example in which we optimize an FIR filter program based upon the analysis results and proposed a direction of architecture optimization.