A voltage-controlled oscillator (VCO)-based analog-to-digital converter (ADC) is an attractive architecture to achieve all-digital analog-to digital conversion due to favorable properties like shaping of the quantization error, inherent anti-alias filtering etc. Here a VCO operates as a signal integrator as well as a quantizer. A converter employing a ring oscillator as the VCO lends itself to an all-digital implementation.
In this dissertation, we explore the design of VCO-based ADCs synthesized using digital standard cells with the long-term goal of achieving high performance data converters built from low accuracy switch components. In a first step, an ADC is designed using vendor supplied standard cells and fabricated in a 65 nm CMOS process. The converter delivers an 8-bit ENOB over a 25 MHz bandwidth while consuming 3.3 mW of power resulting in an energy efficiency of 235 fJ/step (Walden FoM). Then we utilize standard digital CAD tools to synthesize converter designs that are fully described using a hardware description language. A polynomial-based digital post-processing scheme is proposed to correct for the VCO nonlinearity. In addition, pulse modulation schemes like delta modulation and asynchronous sigma-delta modulation are used as a signal pre-coding scheme, in an attempt to reduce the impact of VCO nonlinearity on converter performance. In order to investigate the scaling benefits of all-digital data conversion, a VCO-based converter is designed in a 28 nm CMOS process. The design delivers a 13.4-bit ENOB over a 5 MHz bandwidth achieving an energy efficiency of 4.3 fJ/step according to post-synthesis schematic simulation, indicating that such converters have the potential of achieving good performance in deeply scaled processes by exploiting scaling benefits. Furthermore, large conversion errors caused by non-ideal sampling of the oscillator phase are studied. An encoding scheme employing ones counters is proposed to code the sampled ring oscillator output into a number, which is resilient to a class of sampling induced errors modeled by temporal reordering of the transitions in the ring. The proposed encoding reduces the largest error caused by random reordering of up to six subsequent bits in the sampled signal from 31 to 2 LSBs. Finally, the impact of process, voltage, and temperature (PVT) variations on the performance while operating the converter from a subthreshold supply is investigated. PVT-adaptive solutions are suggested as a means to achieve energy-efficient operation over a wide range of PVT conditions.