Welcome to Biomedbench
BiomedBench is a benchmark suite targeting low-power wearables.
It proposes eight representative state-of-the-art patient-monitoring applications with varying requirements regarding computation, control flow, memory, signal acquisition, and active-to-idle ratio.
The applications are end-to-end, featuring the complete application cycle from signal acquisition to the final inference, highlighting the importance of the idle and acquisition phases in the total energy footprint.
Why do we need a benchmark suite for wearables?
While other suites touch on low-power embedded or biomedical domains, none nail the specifics like BiomedBench for low-power wearable applications. BiomedBench benchmarks cover all phases of wearable applications—idle, acquisition, and processing—incorporating diverse workloads, memory needs, and real-time demands. In addition to the benchmarking, BiomedBench, coupled with an application characterization by metrics, gives the complete picture of the challenges in the biomedical wearable domain and paves the way toward application-driven hardware design in the domain.
Wearable application profile
Our application characterization
Eight end-to-end wearable applications
How state-of-the-art platforms perform
ANNOUNCEMENTS
GitHub
Take a look at our GitHub repository!
We have open-sourced the C/C++ code for all applications!
We include different versions that we ported to multiple low-power platforms.
Clone and run the wearable applications on your device!