Fumihiko Tachibana

A 9.7mW AAC-Decoding, 620mW H.264 720p 60fps Decoding, 8-Core Media Processor with Embedded Forward-Body-Biasing and Power-Gating Circuit in 65nm CMOS Technology

A AAC-decoding, H.264 decoding, media processor with embedded forward-body-biasing and power-gating circuit in CMOS technology is proposed. Since all the components necessary for the scheme are simple MOS circuits requiring no extra supply voltages, they can be placed and routed by a commercial CAD tool. A data-mapping flip-flop was proposed as a high performance and low-power flip-flop. It is concluded that the power dissipation in H.264 720p 60fps decoding of 620mW at the process fast corner is the lowest among the processor-based solutions.

A Digitized Replica Bitline Delay Technique for Random-Variation-Tolerant Timing Generation of SRAM Sense Amplifiers

A digitized replica bitline delay technique has been proposed for random-variation-tolerant timing generation of SRAM sense amplifiers. The sense timing variation attributable to the random variation of transistor threshold voltage is reduced by sufficient count of multiple replica cells, and replica bitline delay is digitized and multiplied for adjusting it to the target sense timing. The variation of the generated timing was 34% smaller than that with a conventional technique and cycle time was reduced by 16% at the supply voltage of 0.6V in 40nm CMOS technology with this scheme.

A process variation compensation scheme using cell-based forward body-biasing circuits usable for 1.2V design

A cell-based forward body-biasing technique to suppress the global process variation and its design flow are proposed. Latch-up free operation is guaranteed by embedded current source cells and limiter cells even when supply voltage is 1.2 V with small area overhead. By applying this technique to a media processor, the worst-case delay is reduced by 20% without sacrificing the maximum leakage spec.

A 47% access time reduction with a worst-case timing-generation scheme utilizing a statistical method for ultra low voltage SRAMs

A variation tolerant sense amplifier timing generator which utilizes a statistical method is proposed. The circuit monitors all the bitline delays and generates the worst timing from the delay distribution. The proposed timing generators have been implemented in 28nm and 40nm SRAMs. The 47% access time reduction has been confirmed in measured results.

A 187.5µVrms-read-noise 51mW 1.4Mpixel CMOS image sensor with PMOSCAP column CDS and 10b self-differential offset-cancelled pipeline SAR-ADC

Low-power and small-area implementations are essential in the mobile-phone market. Serial signal-processing architectures, in which signal-processing circuits such as a programmable-gain amplifier (PGA) and an ADC can be shared by column-level correlated double sampling (CDS) circuits, promise to reduce chip size and power consumption. However, conventional column CDS circuits composed of linear capacitors or NMOS capacitors (NMOSCAPs) with output buffers (OBUFs) still occupy a large footprint. In this work, to reduce the area and the power consumption of column CDS circuits while keeping high linearity, 1.5V PMOS capacitors (PMOSCAPs) are employed. These capacitors work as low-cost sample-and-hold (S/H) capacitors as well as voltage level-shifters by using body-terminal control. To reduce the power consumption of the ADC, instead of a conventional pipeline ADC, we propose a 10b self-differential offset-cancelled pipeline SAR-ADC. It operates with the reference voltage of the ADCs half full-scale voltage (Vfs), leading to a reduction of 80% switching power and 50% capacitor DAC (CDAC) area in the ADC. A black-level correction function is built in the ADC without any additional DACs. The proof-of-concept circuits are implemented in a 1.4Mpixel CMOS image sensor that consumes 51.0mW with a frame rate of 17fps and a read noise of 187.5μVrms at 8.1× analog gain.

Energy efficiency deterioration by variability in SRAM and circuit techniques for energy saving without voltage reduction

The transistor variability deteriorates the energy consumption in SRAM. Especially it increases the energy consumed at the bitlines, which is the major portion of the total energy. The influence of the variation is enhanced at lower supply voltage, thus the voltage reduction sometimes degrades the energy efficiency. In this paper, we present circuit techniques that can reduce the SRAM energy consumption without the supply voltage scaling. An energy-efficient hierarchical bitline scheme can save energy consumption used for the bitline precharge. An energy-efficient offset-cancelling circuit and a process-variability-robust timing-generating circuit are also proposed.

A trimless, 0.5V–1.0V wide voltage operation, high density SRAM macro utilizing dynamic cell stability monitor and multiple memory cell access

A low power SRAM operating at the logic supply voltage of 0.5V-1.0V without chip by chip trimming has been developed. A Dynamic Cell Stability Monitor controls wordline level adaptively by sensing the data flip in reference memory cells. The cell failure rate in every process corner is improved. A Modulated Wordline Level Scheme for Replica Cell optimizes sense timing and the operating frequency is improved by 18% at 1.0V. A Multiple Memory Cell Access Mode pushes the minimum operating cell supply voltage down to 0.5V. A 40nm 2Mb SRAM test chip with 0.24um2 cell has demonstrated 0.5V operation.

A digitized replica bitline delay technique for random-variation-tolerant timing generation of SRAM sense amplifiers

A digitized replica bitline delay technique has been proposed for random-variation-tolerant timing generation of SRAM sense amplifiers. The sense timing variation attributable to the random variation of transistor threshold voltage is reduced by sufficient count of multiple replica cells, and replica bitline delay is digitized and multiplied for adjusting it to the target sense timing. The variation of the generated timing was 34% smaller than that with a conventional technique and cycle time was reduced by 16% at the supply voltage of 0.6V in 40nm CMOS technology with this scheme.

A 27% active and 85% standby power reduction in dual-power-supply SRAM using BL power calculator and digitally controllable retention circuit

This paper presents SRAM circuit techniques to reduce both active and standby mode power especially at room temperature (RT) where actual power consumption is dominant. A bit line power calculator is used to adaptively set the cell supply voltage (VCS) in the active mode. A digitally controllable retention circuit regulates VCS in the standby mode with small control power. These circuits are implemented in a dual-power-supply SRAM in 28 nm CMOS technology. Compared with the conventional scheme, the power consumption in the active and standby mode at 25°C is reduced by 27% and 85%, respectively.

A low-power multi-core media co-processor for mobile application processors

A multi-core co-processor for mobile application processors is introduced. It provides low-power, high-throughput, fully software-based acceleration of multimedia processing. The test chip fabricated in a 65nm CMOS technology consumes 620mW in H.264 720p 60fps decoding and 9.7mW in MPEG-4 AAC decoding. In the maximum workload of H.264 decoding, a symmetrical parallelization achieves 7.5× performance enhancement by 8 cores. The shared L2 cache reduces the required rate of main memory access to 310MB/s. In the minimum workload of AAC decoding, three low-power circuit techniques reduce 98% of leakage. On-chip regulators, which also work as power-gating switches, lower the supply voltage of processing cores. Embedded forward body-biasing circuit reduces Vt variations. A low-power and fast data-mapping F/F relaxes the timing constraint, which enables a reduction in the number of low-Vt transistors.