Marc Pastre
This thesis presents a complete methodology for the digital calibration of analog circuits. It shows how to relax the extreme design constraints in analog circuits, allowing the realization of high-precision systems even with low-performance components. In addition, three applications of the methodology are proposed. To start with, an in-depth analysis of existing compensation techniques for analog circuit imperfections is carried out. The M/2+M sub-binary digital-to-analog converter is thoroughly studied, and the use of this very low-area circuit in conjunction with a successive approximations algorithm for digital compensation is described. A complete methodology based on this compensation circuit and algorithm is then proposed. The detection and correction of analog circuit imperfections is studied, and a simulation tool allowing the transparent simulation of analog circuits with automatic compensation blocks is introduced. An additional up/down converter is suggested for the compensation of continuoustime analog signal processing circuits. The first application shows how the sub-binary M/2+M structure can be employed as a conventional digital-to-analog converter if two calibration and radix conversion algorithms are implemented. The second application, a SOI 1T DRAM, is then presented. An automatic reference current generation for memory reading is described. A digital algorithm chooses a suitable value which compensates several circuit imperfections together, from the sense amplifier offset to the dispersion of the memory read currents. The third application is the calibration of the sensitivity of a current measurement microsystem based on a Hall magnetic field sensor. An integrated reference generates a magnetic field for calibration. Using a variant of the chopper modulation, the spinning current technique, combined with a second modulation of the reference signal, the sensitivity of the complete system is continuously measured without interrupting normal operation. Modulation and demodulation schemes allowing the joint processing of both external and reference magnetic fields are proposed. Additional circuit techniques for extracting the very low reference signal are presented. The implementation of the microsystem is then discussed. The realization of the blocks composing the system is detailed, and circuit issues are presented. Finally, measurements validate the calibration principle. A thermal drift lower than 50 ppm/°C is achieved. This is 6 to 10 times less than in state-ofthe-art implementations. Furthermore, the calibration technique also compensates drifts due to mechanical stresses and ageing.
EPFL2005
