Shahin Lotfabadi
Ryerson University
February, 2013
Abstract: Power is a significant design constraint for implementing efficient
portable biomedical applications. Operating transistors in the
subthreshold region can significantly reduce power consumption; it,
however, also reduces performance. While this performance reduction
can be significant in many applications, the low frequency nature of
biosignals makes subthreshold region a good candidate for implementing
biomedical applications. In this work, we investigate the feasibility
of designing a specialized FPGA for implementing portable biomedical
applications. In particular, we perform a case study on the
performance of the Burg algorithm, a widely used biomedical signal
processing algorithm, to deter- mine the minimum operating frequency
required for the processing of biosignals in real time. Based on the
requirement, the trade-off between power consumption and performance
is measured for FPGA routing resources operating in the subthreshold
region. It is found that operating FPGA routing resources in the
subthreshold region can significantly reduce power consumption while
allowing the Burg algorithm to operate in real time. For the 32 nm
Predictive Technology Model studied in this work, we observed a power
reduction of 197.7 times (which corresponds to a power-delay-product
reduction of 10.78 times) for operating FPGA routing tracks in the
subthreshold region under a supply voltage of 0.4 V. Under this
voltage, the FPGA can operate at 2.0 MHz while allowing signals to
propagate unregistered through 45 routing tracks. Furthermore, the 2.0
MHz operating frequency meets the realtime requirement of the Burg
algorithm for processing 20,000 samples per second.