A low-voltage power supply circuit is developed for micro-power CMOS LSI applications, especially for microwatt smart-sensor LSIs. The circuit consists of a switched-capacitor voltage converter and a series regulator. The switched converter lowers battery voltage 1.5-3V to a low voltage of 0.5-0.7V to drive the series regulator, and the series regulator provides LSI logic gates with a power voltage of 0.4-0.6V such that the logic gates operate in the subthreshold region. In a sample circuit designed to produce a supply voltage of 0.6V, the switched converter lowered a 1.5-V battery voltage to 0.68V, and the series regulator lowered the 0.68V to 0.6V. The power conversion efficiency of the switched converter was 83%, and the total efficiency was 73%, with a 13-µA output current.

View full abstract

A polarization-maintained ultranarrow FBG filter with a linewidth of 1.3GHz has been successfully fabricated by using a 15cm-long phase mask. A maximum reflectance of 65% and a side lobe suppression of 13dB were obtained by employing an apodization technique with slits.

View full abstract

A fully-integrated, dual-band, high gain, low noise amplifier designed for 802.16a WMAN standard is presented. The targeted frequency bands include un-licensed band UNII 5GHz and licensed band Int'l 10GHz. Our LNA design adopts wide band input matching scheme with adjustable load to achieve an area-efficient dual-band low noise amplifier. The LNA has also achieved the gain and noise figure of 16.1dB and 3.6dB at 5GHz, and 15.7dB and 7.6dB at 10GHz. Additionally, the input reflection coefficient is -7dB and -16.5dB, and the IIP₃ is 1.8dBm and 1.1dBm at 5GHz and 10GHz, respectively. Utilizing the 0.18µm CMOS process, the LNA dissipates 14.4mW using a 1.8V supply voltage.

View full abstract

In this paper, two efficient accelerating schemes for decreasing the simulation time of the substrate coupling modelling and analysis are proposed. The first method is based on partitioning the substrate to two parts, including two different boundary conditions, so that to apply FDM for simulation of two parts independently. This partitioning is performed using a MLP neural network to approximate the potential of the middle surface of the substrate. Using parallel processing reduces the simulation time by 57 percent for 50 simulation points. The second method is based on applying weights to the results of each stage of Multi-Grid processing. Then the results of the two previous stages are used for obtaining the magnitudes of the new stage variables. We show that, this method with the best weight factors, results in 75% reduction in the simulation time for the substrate modeling compared to the simple Multi-Grid method.

View full abstract

An erbium-doped fiber laser with a high Er-ion concentration generally causes output power fluctuation and self-pulsing resulting from relaxation oscillation. In this work, we employed a new EDF with high Er- and Al-ion concentrations in order to suppress the relaxation oscillation and increase the output power of the laser. As a result, the output power was more than doubled and the power fluctuation halved. The linewidth was narrowed to 4kHz. Furthermore the frequency stability of the ¹³C₂H₂ frequency-stabilized erbium fiber laser reached as high as 1.3 × 10^-11 for τ = 1s and 2.0 × 10^-11 for τ = 10²s. The short-term stability of τ = 1s was also improved.

View full abstract