HEAL-WEAR: An Ultra-Low Power Heterogeneous System for Bio-Signal Analysis
Abstract:
Personalized healthcare devices enable low-cost, unobtrusive and long-term acquisition of clinically relevant biosignals. These appliances, termed wireless body sensor nodes (WBSNs), are fostering a revolution in health monitoring for patients affected by chronic ailments. Nowadays, WBSNs often embed complex digital processing routines, which must be performed within an extremely tight energy budget. Addressing this challenge, in this paper, we introduce a novel computing architecture devoted to the ultra-low power analysis of biosignals. Its heterogeneous structure comprises multiple processors interfaced with a shared acceleration resource, implemented as a coarse-grained reconfigurable array (CGRA). The CGRA mesh effectively supports the execution of the intensive loops that characterize biosignal analysis applications, while requiring a low reconfiguration overhead. Moreover, both the processors and the reconfigurable fabric feature single-instruction/multiple data (SIMD) execution modes to increase efficiency when multiple data streams are concurrently processed. The run-time behavior on the system is orchestrated by a lightweight hardware mechanism, which concurrently synchronizes processors for SIMD execution and regulates access to the reconfigurable accelerator. By jointly leveraging run-time reconfiguration and SIMD execution, the illustrated heterogeneous system achieves, when executing complex biosignal analysis applications, speedups of up to 11.3× on the considered kernels and up to 37.2% overall energy savings, with respect to an ultra-low power multi-core platform, which does not feature CGRA acceleration.
Existing System:
A key requirement for WBSN is that they must operate for extended periods of time while relying on small batteries, thus requiring a high energy efficiency. In this context, the energy bottleneck of most WBSNs, which only perform acquisition and transmission, usually resides in the wireless communication stage, because the slow dynamics of bio-signals allow their acquisition while employing little energy [6]. A striking alternative is embodied by “smart” WBSNs, which perform advanced on-board Digital Signal Processing (DSP) to extract high-level relevant features from acquisitions [7]. In this approach, only features (as opposed to samples) are sent through the energy-hungry wireless link, potentially resulting in large energy gains [8], [9]. Nonetheless, these benefits can only be leveraged by performing the DSP stage itself within a small energy envelope. In fact, thanks to progresses in the design of domain-specific analog-to-digital converters [10], [11] and wireless protocols [12], DSP tends to dominate the energy budget of smart WBSNs [13], so that any increase in its efficiency has a tangible impact at the system level.
Proposed System:
An orthogonal strategy to maximize energy efficiency is the use of dedicated hardware blocks (custom instructions [34] or dedicated accelerators [35]) to efficiently support computationally-intensive segments of applications.While this strategy can result in orders-of-magnitude power reductions, it is also inflexible, as each block can perform a single function. These characteristics are even more pre-eminent when accelerators are shared by multiple cores [36], [37], because requests for accelerated functions must be arbitrated. In this context, reconfigurable solutions are good candidates to couple the efficiency typical of dedicated hardware with the degree of flexibility required by a wide variety of computational kernels. However, bit-level reconfigurable arrays (such as FPGAs) present huge overheads in terms of area and power consumption, with respect to fixed-function ASICs [24]. CGRAs dramatically reduce these overheads by being programmable only at the operation level, which allows efficient mapping of kernels [22], [38]. Indeed, in [23] the use of CGRAs is advocated based on energy efficiency considerations, while in [39] a coarse-grained array was proposed for the efficient analysis of EEG acquisitions. Scheduling the kernels on CGRA meshes is not straightforward, as operations must be assigned to a spatially distributed RC element, as well as to a temporally defined execution cycle.
Conclusion:
Energy efficiency is a major concern in the design of digital systems across the computing landscape. It is especially important in the context of Internet of Things (IoT) appliances, where the power budget is drastically limited. To achieve the required ultra-low-power operating levels, a careful optimization of the architectural components is mandatory. For such optimization to be effective, it must be domain-specific, i.e.: it has to take into account and exploit the characteristics of target workloads. Herein, we addressed this complex endeavour by proposing the HEAL-WEAR platform, a heterogeneous architecture devoted to bio-signal processing applications. In this domain, the run-time execution profile of applications is often divided between control-dominated phases and computationally-intensive ones, presenting with compact loops (kernels). The illustrated HEAL-WEAR platform can efficiently support both: the former on multiple ultra-low power processing cores, the latter by employing a coarsegrained reconfigurable array interfaced to the cores as a shared acceleration resource. Moreover, bio-signal analysis applications often process multiple input sources in parallel. We leverage this characteristic as an energy-saving feature by supporting SIMD modes in the processor as well as in CGRA. The resource management of HEAL-WEAR is performed with a low-overhead strategy, based on a dedicated Instruction Set Extension, which is interpreted by a lightweight hardware synchronizer to orchestrate the run-time execution of biosignal processing applications.
References:
[1] D. Evans. (Apr. 2011). The Internet of Things: How the Next Evolution of the Internet Is Changing Everything. [Online]. Available: http://www.cisco.com/c/dam/en_us/about/ac79/docs/innov/ IoT_IBSG_0411FINAL.pdf
[2] MEP Heart Group. (Aug. 2016). Cardiovascular Diseases Facts and Figures. [Online]. Available: http://www.mepheartgroup.eu/index.php/factsa- figures
[3] World Health Organization. (Aug. 2016). Cardiovascular Diseases. [Online]. Available: http://www.who.int/topics/cardiovascular _diseases/en
[4] P. Bonato, “Wearable sensors and systems,” IEEE Eng. Med. Biol. Mag., vol. 29, no. 3, pp. 25–36, May 2010.
[5] R. Vecht, M. A. Gatzoulis, and N. Peters, ECG Diagnosis in Clinical Practice. London, U.K.: Springer-Verlag, 2009.
[6] F. Zhang, J. Holleman, and B. P. Otis, “Design of ultra-low power biopotential amplifiers for biosignal acquisition applications,” IEEE Trans. Biomed. Circuits Syst., vol. 6, no. 4, pp. 344–355, Aug. 2012.
[7] R. Braojos et al., “Ultra-low power design of wearable cardiac monitoring systems,” in Proc. 51st ACM/EDAC/IEEE Design Autom. Conf. (DAC), Jun. 2014, pp. 1–6.
[8] F. Chen, F. Lim, O. Abari, A. Chandrakasan, and V. Stojanovic, “Energyaware design of compressed sensing systems for wireless sensors under performance and reliability constraints,” IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 60, no. 3, pp. 650–661, Mar. 2013.
[9] H. Mamaghanian, N. Khaled, D. Atienza, and P. Vandergheynst, “Compressed sensing for real-time energy-efficient ECG compression on wireless body sensor nodes,” IEEE Trans. Biomed. Eng., vol. 58, no. 9, pp. 2456–2466, Sep. 2011.
[10] D. C. Daly and A. P. Chandrakasan, “A 6-bit, 0.2 V to 0.9 V highly digital flash ADC with comparator redundancy,” IEEE J. Solid-State Circuits, vol. 44, no. 11, pp. 3030–3038, Nov. 2009.