Robust Design With Increasing Device Variability In Sub-Micron Cmos And Beyond: A Bottom-Up Framework
My Ph.D. research develops a tiered systematic framework for designing process-independent and variability-tolerant integrated circuits. This bottom-up approach starts from designing self-compensated circuits as accurate building blocks, and moves up to sub-systems with negative feedback loop and full system-level calibration. a. Design methodology for self-compensated circuits My collaborators and I proposed a novel design methodology that offers designers intuitive insights to create new topologies that are self-compensated and intrinsically process-independent without external reference. It is the first systematic approaches to create "correct-by-design" low variation circuits, and can scale beyond sub-micron CMOS nodes and extend to emerging non-silicon nano-devices. We demonstrated this methodology with an addition-based current source in both 180nm and 90nm CMOS that has 2.5x improved process variation and 6.7x improved temperature sensitivity, and a GHz ring oscillator (RO) in 90nm CMOS with 65% reduction in frequency variation and 85ppm/oC temperature sensitivity. Compared to previous designs, our RO exhibits the lowest temperature sensitivity and process variation, while consuming the least amount of power in the GHz range. Another self-compensated low noise amplifiers (LNA) we designed also exhibits 3.5x improvement in both process and temperature variation and enhanced supply voltage regulation. As part of the efforts to improve the accuracy of the building blocks, I also demonstrated experimentally that due to "diversification effect", the upper bound of circuit accuracy can be better than the minimum tolerance of on-chip devices (MOSFET, R, C, and L), which allows circuit designers to achieve better accuracy with less chip area and power consumption. b. Negative feedback loop based sub-system I explored the feasibility of using high-accuracy DC blocks as low-variation "rulers-on-chip" to regulate high-speed high-variation blocks (e.g. GHz oscillators). In this way, the trade-off between speed (which can be translated to power) and variation can be effectively de-coupled. I demonstrated this proposed structure in an integrated GHz ring oscillators that achieve 2.6% frequency accuracy and 5x improved temperature sensitivity in 90nm CMOS. c. Power-efficient system-level calibration To enable full system-level calibration and further reduce power consumption in active feedback loops, I implemented a successive-approximation-based calibration scheme in a tunable GHz VCO for low power impulse radio in 65nm CMOS. Events such as power-up and temperature drifts are monitored by the circuits and used to trigger the need-based frequency calibration. With my proposed scheme and circuitry, the calibration can be performed under 135pJ and the oscillator can operate between 0.8 and 2GHz at merely 40[MICRO SIGN]W, which is ideal for extremely power-and-cost constraint applications such as implantable biomedical device and wireless sensor networks.
Circuit Design; Variability; Robust System
Apsel, Alyssa B.
Afshari, Ehsan; Samorodnitsky, Gennady
Ph.D. of Electrical Engineering
Doctor of Philosophy
Dissertation or Thesis