A time-based energy-efficient analog-to-digital converter

Abstract

Dual-slope converters use time to perform analog-to-digital conversion but require 2N+1 clock cycles to achieve N bits of precision. We describe a novel algorithm that also uses time to perform analog-to-digital conversion but requires 5N clock cycles to achieve N bits of precision via a successive sub-ranging technique. The algorithm requires one asynchronous comparator, two capacitors, one current source, and a state machine. Amplification of two is achieved without the use of an explicit amplifier by simply doing things twice in time. The use of alternating Voltage-to-Time and Time-to-Voltage conversions provides natural error cancellation of comparator offset and delay, 1/f noise, and switching charge-injection. The use of few components and an effcient mechanism for amplification and error cancellation allow for energy-effcient operation: In a 0.35 [mu]m implementation, we were able to achieve 12 bits of DNL limited precision or 11 bits of thermal noise-limited precision at a sampling frequency of 31.25kHz with 75 [mu] W of total analog and digital power consumption. These numbers yield a thermal noise-limited energy-efficiency of 1.17pJ per quantization level making it one of the most energy-effcient converters to date in the 10 to 12 bit precision range.

(cont.) This converter could be useful in low-power hearing aids after analog gain control has been performed on a microphone front-end. An 8 bit audio version of our converter in a 0.18 [mu] m process consumes 960nW and yields an energy-efficiency of 0.12pJ per quantization level, perhaps the lowest ever reported. This converter may be useful in biomedical and sensor-network applications where energy-efficiency is paramount. Our algorithm has inherent advantages in time-to-digital conversion. It can be generalized to easily digitize power-law functions of its input, and it can be used in an interleaved architecture if higher speed is desired.