Variation Resolutions for CMOS Sensing Network

Variation and variability have become the main concerns for reliable design methodology in CMOS sensing networks. On one hand, the variation can originate from the sensor tag itself, where the performance can be compromised by the uncontrollable variations. Process variation is an unavoidable consequence from the continuous scaling of modern CMOS technologies, which has increased the transistor count from thousands to billions and improved the microprocessor operating frequency from MHz to GHz. Process variation will reduce the sensor tag sensitivity and distort the transduction accuracy. In this dissertation, we will use the process variation effects on the RF-to-DC rectifiers with process variation as an illustration, as many passive sensors need such units to scavenge the ambient energy to accomplish the sensing functions. To counter such process variation, in this dissertation, a novel tunable-Vth rectifier based on floating-gate MOS diodes is proposed and implemented in logic CMOS foundry technology. The proposed tunable-Vth rectifier can have each constituent diode tuned to its optimal threshold at the wafer testing stage to maximize the operational output voltage at various loads and to compensate device-level variations. An optimization algorithm was implemented to automatically take the output voltage as feedback for system calibration in a short duration. The measurements of the tunable-threshold rectifier show > 4dB improvement in input sensitivity compared to the rectifier built by foundryprovided zero-threshold transistors. The proposed circuits can be combined with other techniques including high-Q impedance matching for input voltage boosting and hierarchical tandem stages for further improvement on operating conditions. With a Q = 10 matching network, -27 dBm sensitivity and 22% efficiency can be achieved for about 0.5 V DC output to a 500 k load at 570 MHz.;On the other hand, the sensing variation can originate from the targeted biological sensing signal, which complicates both the sensor system design and the associated signal analysis. We will illustrate a spike-sorting method to reliably classify the enteric neural signals which have unique waveform features but large variation in magnitude, timing and duration. The proposed fastDTW spike classification algorithm provides improvements in accuracy and computational cost in comparison with Cross-correlation based template matching and PCA + k-means clustering without time warping. When appled to mouse ENS neurons in high noise and high variability environment, fastDTW successfully recognized spikes with variability is as large as 1.2 ms in width and a few millivolt in magnitude. The captured waveform features are used for variation correlation analyses to better understand the operating principles of enteric nervous system.;Although other variation sources can also affect the sensor system design, our approaches of device compensation based on operational feedback and signal tolerance based on time warping are able to give illustrations for sensor designers to successfully countermeasure uncontrollable variation sources.