Heterogeneous and Inexact: Maximizing Power Efficiency of Edge Computing Sensors for Health Monitoring Applications

In the Internet-of-Things (IoT) era, there is an increasing trend to enable intelligent behavior in edge computing sensors. Thus, a new generation of smart wearable devices for health monitoring is being developed, able to perform complex Digital Signal Processing (DSP) routines that extract features of clinical relevance from the acquired data. These new edge computing sensors for personalized healthcare must operate within a tight energy envelope; addressing the ensuing challenge, we herein introduce an inexact and heterogeneous edge computing architecture, specifically tailored to the bio-DSP domain. We observe that bio-signal analysis applications present task-level parallelism, intensive computational hotspots and a high degree of resilience towards errors. These characteristics drive our new bio-DSP edge node architecture design composed of multiple processing cores, a Coarse-Grained Reconfigurable Array (CGRA) accelerator, and hardware-software co-design support to become resilient to a non-zero probability of bit-flips at runtime. All these characteristics enable our new bio-DSP architecture to operate with an ultra-low voltage operating point. Indeed our results indicate that the energy benefits attained from the inclusion of all these characteristics in bio-DSP architectures are more than additive: task parallelism is harnessed both at the processor and the accelerator level, and the high tolerance of the CGRA towards voltage down-scaling is exploited to further decrease the IoT edge bio-DSP system energy envelope.

FPGAs for the Masses: Affordable Hardware Synthesis from Domain-Specific Languages

Nithin George

Field Programmable Gate Arrays (FPGAs) have the unique ability to be configured into application-specific architectures that are well suited to specific computing problems. This enables them to achieve performances and energy efficiencies that outclass other processor-based architectures, such as Chip Multiprocessors (CMPs), Graphic Processing Units (GPUs) and Digital Signal Processors (DSPs). Despite this, FPGAs are yet to gain widespread adoption, especially among application and software developers, because of their laborious application development process that requires hardware design expertise. In some application areas, domain-specific hardware synthesis tools alleviate this problem by using a Domain-Specific Language (DSL) to hide the low-level hardware details and also improve productivity of the developer. Additionally, these tools leverage domain knowledge to perform optimizations and produce high-quality hardware designs. While this approach holds great promise, the significant effort and cost of developing such domain-specific tools make it unaffordable in many application areas. In this thesis, we develop techniques to reduce the effort and cost of developing domain-specific hardware synthesis tools. To demonstrate our approach, we develop a toolchain to generate complete hardware systems from high-level functional specifications written in a DSL. Firstly, our approach uses language embedding and type-directed staging to develop a DSL and compiler in a cost-effective manner. To further reduce effort, we develop this compiler by composing reusable optimization modules, and integrate it with existing hardware synthesis tools. However, most synthesis tools require users to have hardware design knowledge to produce high-quality results. Therefore, secondly, to facilitate people without hardware design skills to develop domain-specific tools, we develop a methodology to generate high-quality hardware designs from well known computational patterns, such as map, zipWith, reduce and foreach; computational patterns are algorithmic methods that capture the nature of computation and communication and can be easily understood and used without expert knowledge. In our approach, we decompose the DSL specifications into constituent computational patterns and exploit the properties of these patterns, such as degree of parallelism, interdependence between operations and data-access characteristics, to generate high-quality hardware modules to implement them, and compose them into a complete system design. Lastly, we extended our methodology to automatically parallelize computations across multiple hardware modules to benefit from the spatial parallelism of the FPGA as well as overcome performance problems caused by non-sequential data access patterns and long access latency to external memory. To achieve this, we utilize the data-access properties of the computational patterns to automatically identify synchronization requirements and generate such multi-module designs from the same high-level functional specifications. Driven by power and performance constraints, today the world is turning to reconfigurable technology (i.e., FPGAs) to meet the computational needs of tomorrow. In this light, this work addresses the cardinal problem of making tomorrow's computing infrastructure programmable to application developers.

EPFL2016

Architecture of a Real-Time Platform Independant GPS L1 Software Receiver

Personal digital assistants or mobile phones applications are not anymore restricted to multimedia or wireless communications, but have been extended to handle Global Positioning System (GPS) functionalities. Consequently, the growing market of GPS capable mobile devices is driving the interest of software receiver solutions as they provide several advantages with respect to traditional hardware implementations. First, they share the same system resources such as the processor, embedded memory and power with other system units, reducing both the size and the costs of their integration. Second, they can be easily reprogrammed – via a firmware update – for incorporating the latest developments, such as the exploitation of the future satellites signals or some improved multipath mitigation techniques. Finally, they offer a more flexible solution for rapid research and development as compared to conventional hardware receivers where the chip design is fixed and obtained after a long integration process. With the increasing performance of modern processors, it becomes now feasible to implement in software a multi-channel GPS receiver operating in real-time. However, a major problem with the software architecture is the large computing resources required for the digital signal processing. Former studies have demonstrated that a straightforward transposition of traditional hardware based architectures into software would lead to an amount of integer operations which is not suitable for today's fastest computers. From this observation, several strategies have been proposed in the literature in order to reduce the complexity of the receiver operations. The first one relies on the utilization of advanced microprocessor instructions set which provides the capability of processing vectors of data by operating on multiple integer values at the same time. This results in significant gains in execution speed, but also severely limits the portability of the code, since the operations are tied to specific processors architectures. Another alternative consists in exploiting the native bitwise representation of the signal. The data bits are stored in separate vectors on which logical parallel operations can be performed. The objective is to take advantage of the universality, high parallelism, and speed of the bitwise operations for which a single integer operation translates into a few simple parallel logical relations. However, the inherent drawback of the bitwise processing is the lack of flexibility as the complexity becomes bit-depth dependent. This thesis has been carried out in the framework of a two-year industrial project (2007-2009) in collaboration with U-blox AG in Thalwil. It aimed to the realization of a multi-channel, platform-independent real-time GPS L1 software receiver. The main challenge of this project consisted in providing real-time performances while keeping the portability of the code to make the receiver suitable for any type of software implementation. In that sense, new techniques and algorithms have been developed for optimizing the processing chain in order to lower the processor load. The main idea consists in regrouping data which share the same characteristics, and process them in batches instead of sequentially. This way, it becomes possible to progressively reduce the data throughput and consequently the amount of operations to perform. A completely new receiver architecture has been proposed and validated through the realization of a functional prototype, thus demonstrating the feasibility of the concept.

EPFL2010

Hardware / Software Architectural and Technological Exploration for Energy-Efficient and Reliable Biomedical Devices

Loris Gérard Duch

Nowadays, the ubiquity of smart appliances in our everyday lives is increasingly strengthening the links between humans and machines. Beyond making our lives easier and more convenient, smart devices are now playing an important role in personalized healthcare delivery. This technological breakthrough is particularly relevant in a world where population aging and unhealthy habits have made non-communicable diseases the first leading cause of death worldwide according to international public health organizations. In this context, smart health monitoring systems termed Wireless Body Sensor Nodes (WBSNs), represent a paradigm shift in the healthcare landscape by greatly lowering the cost of long-term monitoring of chronic diseases, as well as improving patients' lifestyles. WBSNs are able to autonomously acquire biological signals and embed on-node Digital Signal Processing (DSP) capabilities to deliver clinically-accurate health diagnoses in real-time, even outside of a hospital environment. Energy efficiency and reliability are fundamental requirements for WBSNs, since they must operate for extended periods of time, while relying on compact batteries. These constraints, in turn, impose carefully designed hardware and software architectures for hosting the execution of complex biomedical applications. In this thesis, I develop and explore novel solutions at the architectural and technological level of the integrated circuit design domain, to enhance the energy efficiency and reliability of current WBSNs. Firstly, following a top-down approach driven by the characteristics of biomedical algorithms, I perform an architectural exploration of a heterogeneous and reconfigurable computing platform devoted to bio-signal analysis. By interfacing a shared Coarse-Grained Reconfigurable Array (CGRA) accelerator, this domain-specific platform can achieve higher performance and energy savings, beyond the capabilities offered by a baseline multi-processor system. More precisely, I propose three CGRA architectures, each contributing differently to the maximization of the application parallelization. The proposed Single, Multi and Interleaved-Datapath CGRA designs allow the developed platform to achieve substantial energy savings of up to 37%, when executing complex biomedical applications, with respect to a multi-core-only platform. Secondly, I investigate how the modeling of technology reliability issues in logic and memory components can be exploited to adequately adjust the frequency and supply voltage of a circuit, with the aim of optimizing its computing performance and energy efficiency. To this end, I propose a novel framework for workload-dependent Bias Temperature Instability (BTI) impact analysis on biomedical application results quality. Remarkably, the framework is able to determine the range of safe circuit operating frequencies without introducing worst-case guard bands. Experiments highlight the possibility to safely raise the frequency up to 101% above the maximum obtained with the classical static timing analysis. Finally, through the study of several well-known biomedical algorithms, I propose an approach allowing energy savings by dynamically and unequally protecting an under-powered data memory in a new way compared to regular error protection schemes. This solution relies on the Dynamic eRror compEnsation And Masking (DREAM) technique that reduces by approximately 21% the energy consumed by traditional error correction codes.

EPFL2018