Search

Current edge computing systems are deployed in highly complex application scenarios with dynamically changing requirements. In order to provide the expected performance and energy efficiency values in these situations, the use of heterogeneous hardware/software platforms at the edge has become widespread. However, these computing platforms still suffer from the lack of unified software-driven programming models to efficiently deploy multi-purpose hardware-accelerated solutions. In parallel, edge computing systems also face another huge challenge: operating under multiple conditions that were not taken into account during any of the design stages. Moreover, these conditions may change over time, forcing self-adaptation mechanisms to become a must. This paper presents an integrated architecture to exploit hardware-accelerated data-parallel models and transparent hardware/software multithreading. In particular, the proposed architecture leverages the ARTICo3 framework and ReconOS to allow developers to select the most suitable programming model to deploy their edge computing applications onto run-time reconfigurable hardware devices. An evolvable hardware system is used as an additional architectural component during validation, providing support for continuous lifelong learning in smart edge computing scenarios. In particular, the proposed setup exhibits online learning capabilities that include learning by imitation from software-based reference algorithms. Experimental results show the benefits of the proposed approach, exposing different run-time tradeoffs (e.g., computing performance versus functional correctness of the evolved solutions), and highlighting the benefits of using scalable data-parallel models to perform circuit evolution under dynamically changing application scenarios. [ABSTRACT FROM AUTHOR]

Deep neural networks (DNNs) are penetrating into a broad spectrum of applications and replacing manual algorithmic implementations, including the radio frequency communications domain with classical signal processing algorithms. However, the high throughput (gigasamples per second) and low latency requirements of this application domain pose a significant hurdle for adopting computationally demanding DNNs. In this article, we explore highly specialized DNN inference accelerator approaches on field-programmable gate arrays (FPGAs) for RadioML modulation classification. Using an automated end-to-end flow for the generation of the FPGA solution, we can easily explore a spectrum of solutions that optimize for different design targets, including accuracy, power efficiency, resources, throughput, and latency. By leveraging reduced precision arithmetic and customized streaming dataflow, we demonstrate a solution that meets the application requirements and outperforms alternative FPGA efforts by 3.5× in terms of throughput. Against modern embedded graphics processing units (GPUs), we measure $>\mathsf{10}\times$>10× higher throughput and $>\mathsf{100}\times$>100× lower latency under comparable accuracy and power envelopes. [ABSTRACT FROM AUTHOR]

Approximate circuits (AxCs) tradeoff computational accuracy against improvements in hardware area, delay, or energy consumption. IP core vendors who wish to create such circuits need to convince consumers of the resulting approximation quality. As a solution, we propose proof-carrying AxCs. The vendor creates an approximate IP core together with a certificate that proves the approximation quality. The proof certificate is bundled with the approximate IP core and sent off to the consumer. The consumer can formally verify the approximation quality of the IP core at a fraction of the typical computational cost for formal verification. In this brief, we first make the case for proof-carrying AxCs and then demonstrate the feasibility of the approach by a set of synthesis experiments using an exemplary approximation framework. [ABSTRACT FROM AUTHOR]

Reconfigurable caches offer an intriguing opportunity to tailor cache behavior to applications for better run-times and energy consumptions. While one may adapt structural cache parameters such as cache and block sizes, we adapt the memory-address-to-cache-index mapping function to the needs of an application. Using a LEON3 embedded multi-core processor with reconfigurable cache mappings, a metaheuristic search procedure, and MiBench applications, we show in this work how to accurately compare non-deterministic performances of applications and how to use this information to implement an optimization procedure that evolves application-specific cache mappings for the LEON3 multi-core processor. [ABSTRACT FROM AUTHOR]

Radiation tolerance in FPGAs is an important field of research particularly for reliable computation in electronics used in aerospace and satellite missions. The motivation behind this research is the degradation of reliability in FPGA hardware due to single-event effects caused by radiation particles. Redundancy is a commonly used technique to enhance the fault-tolerance capability of radiation-sensitive applications. However, redundancy comes with an overhead in terms of excessive area consumption, latency, and power dissipation. Moreover, the redundant circuit implementations vary in structure and resource usage with the redundancy insertion algorithms as well as number of used redundant stages. The radiation environment varies during the operation time span of the mission depending on the orbit and space weather conditions. Therefore, the overheads due to redundancy should also be optimized at run-time with respect to the current radiation level. In this paper, we propose a technique called Dynamic Reliability Management (DRM) that utilizes the radiation data, interprets it, selects a suitable redundancy level, and performs the run-time reconfiguration, thus varying the reliability levels of the target computation modules. DRM is composed of two parts. The design-time tool flow of DRM generates a library of various redundant implementations of the circuit with different magnitudes of performance factors. The run-time tool flow, while utilizing the radiation/error-rate data, selects a required redundancy level and reconfigures the computation module with the corresponding redundant implementation. Both parts of DRM have been verified by experimentation on various benchmarks. The most significant finding we have from this experimentation is that the performance can be scaled multiple times by using partial reconfiguration feature of DRM, e.g., 7.7 and 3.7 times better performance results obtained for our data sorter and matrix multiplier case studies compared with static reliability management techniques. Therefore, DRM allows for maintaining a suitable trade-off between computation reliability and performance overhead during run-time of an application. [ABSTRACT FROM AUTHOR]

Profiling applications on a heterogeneous compute node is challenging since the way to retrieve data from the resources and interpret them varies between resource types and manufacturers. This holds especially true for measuring the energy consumption. In this paper we present Ampehre, a novel open source measurement framework that allows developers to gather comparable measurements from heterogeneous compute nodes, e.g., nodes comprising CPU, GPU, and FPGA. We explain the architecture of Ampehre and detail the measurement process on the example of energy measurements on CPU and GPU. To characterize the probing effect, we quantitatively analyze the trade-off between the accuracy of measurements and the CPU load imposed by Ampehre. Based on this analysis, we are able to specify reasonable combinations of sampling periods for the different resource types of a compute node. [ABSTRACT FROM AUTHOR]

Monte Carlo Tree Search evaluates positions with the help of a playout policy. If the playout policy evaluates a position wrong then there are cases where the tree search has difficulties to find the correct move due to the large search space. This paper explores adaptive playout policies which improve the playout policy during a tree search. With the help of policy gradient reinforcement learning techniques we optimize the playout policy to give better evaluations. We tested the algorithm in Computer Go and measured an increase in playing strength of more than 100 ELO. The resulting program was able to deal with difficult test cases which are known to pose a problem for Monte Carlo Tree Search. [ABSTRACT FROM AUTHOR]

The reconfigurable mesh is a parallel model of computation, which exploits a massive amount of rather simple processing elements connected through a reconfigurable interconnection network. During the last decades, the model received strong interest and many researchers have devised algorithms for it. However, most of this work focuses on theoretical aspects. Due to some idealistic modeling assumptions only a few attempts have been made to implement the model and to study the practical use of reconfigurable meshes. In this paper, we leverage the reconfigurable mesh model to study potential architectures and programming models for future many-cores. We design a reconfigurable mesh in form of a scalable soft core array with a reconfigurable interconnect and implement it on FPGA technology in order to create a prototype platform. We present an overall hardware/software tool flow for generating and programming reconfigurable mesh prototypes. The new language ARMLang and a corresponding compiler facilitate the programming of the massively parallel processor arrays. To our knowledge, this work is the first practical study of word-level reconfigurable meshes. To analyze the performance of our implementation we study four algorithmic kernels from the application domains arithmetic, sorting, graph algorithms and imaging. For each kernel, we devise a reconfigurable mesh program in ARMLang, compile it to our soft core array and measure its runtime depending on the mesh size. Then, we compare the runtimes to two sequential implementations of the algorithms, which are executed on two single core systems. The results show that many-cores leveraging the reconfigurable mesh model can efficiently use a vast number of processing elements and that, for the chosen algorithms, they come close to optimally parallelized programs. [ABSTRACT FROM AUTHOR]