Low-power digital processor for wireless sensor networks

Abstract

In order to make sensor networks cost-effective and practical, the electronic components of a wireless sensor node need to run for months to years on the same battery. This thesis explores the design of a low-power digital processor for these sensor nodes, employing techniques such as hardwired algorithms, lowered supply voltages, clock gating and subsystem shutdown. Prototypes were built on both a FPGA and ASIC platform, in order to verify functionality and characterize power consumption. The resulting 0.18[micro]m silicon fabricated in National Semiconductor Corporation's process was operational for supply voltages ranging from 0.5V to 1.8V. At the lowest operating voltage of 0.5V and a frequency of 100KHz, the chip performs 8 full-accuracy FFT computations per second and draws 1.2nJ of total energy per cycle. Although this energy/cycle metric does not surpass existing low-energy processors demonstrated in literature or commercial products, several low-power techniques are suggested that could drastically improve the energy metrics of a future implementation.