A Common Architecture for Processing Data from Thin Film Sensor Arrays Tailored to 3-D Applications

Abstract

Thin film sensor arrays on ICs are studied as an example of 3-D integration. Various thin film sensor arrays are studied, with emphasis on systems integratable on silicon. 3-D integration offers the chance to have all the data from the sensors in the sensor arrays available to the processing layer at once. The data from the sensors is read vertically directly into the processing layer. With the motivation to identify the needs of various thin film sensors and to exploit the available parallelism in data, thin film sensors are categorized into two - those with similar sensors in the arrays or those with different sensors in the sensor array. A common architecture to address both the categories is proposed, designed and implemented. Such a generic processing layer design also helps independant development of sensor arrays (of either categories) and finally deposited on such a generic processing layer. Also any commercial off the shelf generic microprocessor can not be used for such a project. The reasons being that the generic microprocessors were not designed keeping in mind the vertical integration and hence parallelism in data. Nor are they designed to handle the specific needs of sensor arrays. The modSIMD processor (proposed and implemented in this thesis) exploits the 3-D integration aspects and the needs of sensor arrays. modSIMD stands for a modified SIMD (Single Instruction Multiple Data) architecture. In the common addressing mode, it behaves like a SIMD processor and works on various data elements parallely. In the specific addressing mode, it behaves like a SISD (Single Instruction Single Data) processor. First the advantages of 3-D integration are studied. The 3-D system architecture is compared to a 2-D system architecture with similar processing architecture. The 3-D architecture is seen to dissipate more power but is more efficient in area and speed. Then the processing architecture is put to test with applications from both the categories. It outperforms an ARM microprocessor in SIMD applications, thus exploiting the parallelism in data. modSIMD also works on applications from both the categories, thus developing such a common architecture is made possible.