Robust, Low-Power IC Design Through Reconfigurable References and Biases: A Focus on Environmental Temperature Variation and User Diversity in Health Sensing

Author: ORCID icon
Ownby, Natalie, Electrical Engineering - School of Engineering and Applied Science, University of Virginia
Calhoun, Benton, EN-Elec & Comp Engr Dept, University of Virginia

The growth of the Internet of Things (IoT) coincides with industry trends toward low-power systems, and smaller technology nodes. With these trends, the robustness of IoT nodes becomes an increasing concern. Combating environmental and other variations places stringent design requirements on the integrated circuits used in these nodes and, as a result, increases the design effort, power and area necessary to meet these requirements. This thesis examines two specific types of variation: operating temperature and physiological diversity in users of wearable health sensing systems, and catalogs the impact these sources of variation have on integrated circuit components. It proposes a digitally assisted analog approach to compensate for this using programmable and reconfigurable voltage and current references. This allows for compensation on a system level, reducing the precision and power required for individual components.
This thesis presents a novel method for generating voltages and currents that have a designer-determined piecewise linear (PWL) temperature dependence for use in system-level temperature compensation. This approach has a low power and area overhead and reduces the need for component-level compensation which, in turn, reduces the design time and effort for temperature-robust systems. The thesis highlights how this approach can provide compensation at the component level and discusses how this approach scales to systems on chips (SoCs). It then examines how this architecture can be used in both feedforward and feedback loops to compensate for other on-chip variations due to process and supply variation, circuit degradation or changes in the sensing environment.
Next, this thesis presents a model of how user diversity, specifically skin tone and BMI, affect low power circuit performance in health sensing and how adaptable biasing can increase signal-to-noise ratio (SNR) of these systems. Finally, this thesis examines how voltage references can be broken down into smaller unit cells for use in automated reference generation. These approaches allow designers to prioritize power at the component and signal chain level and provide a wrap-around approach to improve their robustness.

PHD (Doctor of Philosophy)
Internet of Things, Low-Power VLSI, Wearable Health, Temperature Robust Circuits
All rights reserved (no additional license for public reuse)
Issued Date: