A multi-sensor microsystem for autonomous data gathering.

This dissertation describes the development of an integrated microsystem, 0.5cc in volume, which can autonomously acquire and store data from multiple sensors on and off the system platform. This work is motivated by the desire to gather environmental and biological data using systems that offer substantial reconfigurability and performance and yet are extremely compact. To overcome obstacles to microsystem miniaturization, advances have been made in microsystem packaging and interconnect, sensor-readout circuitry, and microsystem architecture. These advances are demonstrated in a microsystem that incorporates a commercial microcontroller (2 MIPS with 16-bit on-chip ADC), a 16-megabit flash memory, and a custom sensor-interface chip. The interface chip includes a 0.4--3.2mV/fF capacitive-readout circuit, a 1.5mV/ohm resistive-readout circuit, and four low-noise biopotential amplifiers with 100x gain. Serial communication from the microcontroller configures this chip and instructs circuit blocks to be powered only during sampling. Although designed for a flexible sensor population, the microsystem has been configured with on-board capacitive pressure/temperature/humidity sensors as well as off-board neural/EMG electrodes. The microsystem components are integrated into 9.5mm x 7.6mm x 2.0mm (0.16cc) (0.5cc including the battery) using a silicon platform fabricated with microelectromechanical systems (MEMS) technology. The double-sided platform implements air-isolated through-wafer interconnect structures with 25ohms series resistance and 850fF parasitic capacitance, deep-etched cavities for mounting and wire-bonding chips within the platform profile, and a reconfigurable microconnection approach that uses automatic proximity detection to initiate localized solder bonding between off-board sensor leads and thermally-isolated substrate pads with 50--400 micron pitch. Microsystem testing has demonstrated acquisition and storage of 14-bit capacitive sensor data with 0.1--1000s measurement intervals and 8-bit neural-recording data up to 11kHz. Operating under stored-program control, the microsystem performs automatic capacitive sensor calibration using a binary search algorithm and the programmable 40pF 10-bit reference capacitor. It also uses threshold-based spike detection software to accomplish neural data compression, gaining an eightfold increase in recording time and a 67% power savings. With system-level power management and flexible sleep modes, the system can operate from two silver-oxide coin cells for periods of 30 minutes to over one year, consuming an average of 3--20mW for 1--10kHz neural recording and 0.05mW for environmental measurements below 10Hz.Ph.D.Applied SciencesBiomedical engineeringElectrical engineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/125139/2/3186684.pd