FPGA clusters that consist of multiple FPGA boards have been gaining interest in recent times. Massively parallel processing with a stand-alone heterogeneous FPGA cluster with SoC- style FPGAs and mid-scale FPGAs is promising with cost-performance benefit. Here, we propose such a heterogeneous FPGA cluster with FiC and M-KUBOS cluster. FiC consists of multiple boards, mounting middle scale Xilinx's FPGAs and DRAMs, which are tightly coupled with high-speed serial links. In addition, M-KUBOS boards are connected to FiC for ensuring high IO data transfer bandwidth. As an example of massively parallel processing, here we implement genomic pattern search. Next-generation sequencing (NGS) technology has revolutionized biological system related research by its high-speed, scalable and massive throughput. To analyze the genomic data, short read mapping technique is used where short Deoxyribonucleic acid (DNA) sequences are mapped relative to a known reference sequence. Although several pattern matching techniques are available, FM-index based pattern search is perfectly suitable for this task due to the fastest mapping from known indices. Since matching can be done in parallel for different data, the massively parallel computing which distributes data, executes in parallel and gathers the results can be applied. We also implement a data compression method where about 10 times reduction in data size is achieved. We found that a M-KUBOS board matches four FiC boards, and a system with six M-KUBOS boards and 24 FiC boards achieved 30 times faster than the software based implementation.

We developed a PYNQ cluster that consists of economical Zynq boards, called M-KUBOS, that are interconnected through low-cost high-performance GTH serial links. For the software environment, we employed the PYNQ open-source software platform. The PYNQ cluster is anticipated to be a multi-access edge computing (MEC) server for 5G mobile networks. We implemented the ResNet-50 inference accelerator on the PYNQ cluster for image recognition of MEC applications. By estimating the execution time of each ResNet-50 layer, layers of ResNet-50 were divided into multiple boards so that the execution time of each board would be as equal as possible for efficient pipeline processing. Owing to the PYNQ cluster in which FPGAs were directly connected by high-speed serial links, stream processing without network bottlenecks and pipeline processing between boards were readily realized. The implementation on 4 boards achieved 292 GOPS performance, 75.1 FPS throughput, and 7.81 GOPS/W power efficiency. It achieved 17 times faster speed and 130 times more power efficiency compared to the implementation on the CPU, and 5.8 times more power efficiency compared to the implementation on the GPU.

The multi-FPGA system known as, the Flow-in-Cloud (FiC) system, is composed of mid-range FPGAs that are directly interconnected by high-speed serial links. FiC is currently being developed as a server for multi-access edge computing (MEC), which is one of the core technologies of 5G. Because the applications of MEC are sometimes timing-critical, a static time division multiplexing (STDM) network has been used on FiC. However, the STDM network exhibits the disadvantage of decreasing link utilization, especially under light traffic. To solve this problem, we propose a hybrid router that combines packet switching for low-priority communication and STDM for high-priority communication. In our hybrid network, the packet switching uses slots that are unused by the STDM; therefore, best-effort communication by packet switching and QoS guarantee communication by the STDM can be used simultaneously. Furthermore, to improve each link utilization under a low network traffic load, we propose a dynamic communication switching algorithm. In our algorithm, each router monitors the network load metrics, and according to the metrics, timing-critical tasks select the STDM according to the metrics only when congestion occurs. This can achieve both QoS guarantee and efficient utilization of each link with a small resource overhead. In our evaluation, the dynamic algorithm was up to 24.6% faster on the execution time with a high network load compared to the packet switching on a real multi-FPGA system with 24 boards.

Multi-FPGA systems have gained attention because of their high performance and power efficiency. A multi-FPGA system called Flow-in-Cloud (FiC) is currently being developed as an accelerator of multi-access edge computing (MEC). FiC consists of multiple mid-range FPGAs tightly connected by high-speed serial links. Since time-critical jobs are assumed in MEC, a circuit-switched network with static time-division multiplexing (STDM) switches has been implemented on FiC. This paper investigates techniques of enhancing the interconnection performance of FiC. Unlike switching fabrics for Network on Chips or parallel machines, economical multi-FPGA systems, such as FiC, use Xilinx Aurora IP and FireFly cables with multiple lanes. We adopted the link aggregation and the slot distribution for using multiple lanes. To mitigate the bottleneck between an STDM switch and user logic, we also propose a multi-ejection STDM switch. We evaluated various combinations of our techniques by using three practical applications on an FiC prototype with 24 boards. When the number of slots is large and transferred data size is small, the slot distribution was sometimes more effective, while the link aggregation was superior for other most cases. Our multi-ejection STDM switch mitigated the bottleneck in ejection ports and successfully reduced the number of time slots. As a result, by combining the link aggregation and multi-ejection STDM switch, communication performance improved up to 7.50 times with few additional resources. Although the performance of the fast Fourier transform with the highest communication ratio could not be enhanced by using multiple boards when a lane was used, 1.99 times performance improvement was achieved by using 8 boards with four lanes and our multi-ejection switch compared with a board.

FPGA (Field Programmable Gate Array) based accelerators are attracting significant interest in cloud computing systems. Combining multi-FPGA systems with cloud computing brings a new perspective to the reconfigurable computing research. However, the multi-tenancy of a multi-FPGA system has not been fully discussed in the previous researches. In this paper, we propose a multi-tenant resource management system, named FiC-RM, for a multi-FPGA cloud system. FiC-RM provides users with a set of FPGA resources according to their requirements and allows them to exclusively access FPGA boards and the interconnection network. To achieve this, we propose a placement algorithm which is a key to efficiently share the limited resources. We demonstrate FiC-RM controls a practical scale multi-FPGA system. Moreover, Our simulation study shows that our placement algorithm achieved 3 to 4% improvement in the average resource usage and a 20-second reduction in the response time, compared to other existing naive algorithms.

Recurrent neural networks (RNNs) have been proven effective for sequence-based tasks thanks to their capability to process temporal information. In real-world systems, deep RNNs are more widely used to solve complicated tasks such as large-scale speech recognition and machine translation. However, the implementation of deep RNNs on traditional hardware platforms is inefficient due to long-range temporal dependence and irregular computation patterns within RNNs. This inefficiency manifests itself in the proportional increase in the latency of RNN inference with respect to the number of layers of deep RNNs on CPUs and GPUs. Previous work has focused mostly on optimizing and accelerating individual RNN cells. To make deep RNN inference fast and efficient, we propose an accelerator based on a multi-FPGA platform called Flow-in-Cloud (FiC). In this work, we show that the parallelism provided by the multi-FPGA system can be taken advantage of to scale up the inference of deep RNNs, by partitioning a large model onto several FPGAs, so that the latency stays close to constant with respect to increasing number of RNN layers. For single-layer and four-layer RNNs, our implementation achieves 31x and 61x speedup compared with an Intel CPU.

Multi-FPGA systems, which consist of multiple FPGAs and a printed circuit board connecting them, are useful and important tools for prototyping large scale circuits, including SoCs. In this paper, we propose a method for optimizing inter-FPGA signal transmission to accelerate the system frequency of multi-FPGA prototyping systems and shorten prototyping time. Compared with the number of I/O pins of an FPGA, the number of I/O signals between FPGAs usually becomes very large. Thus, time-multiplexed I/Os are used to resolve the problem. On the other hand, they introduce large delays to inter-FPGA I/O signals, and much lower the system frequency. To reduce the degradation of the system frequency, we have proposed a method for optimally selecting signals to be time-multiplexed and signals not to be time-multiplexed. However, this method assumes that there exist physical connections (i.e., wires on the printed circuit board) between every pair of FPGAs, and cannot handle I/O signals between a pair of FPGAs that have no physical connections between them. Thus, in this paper, we propose a method for obtaining indirect inter-FPGA routes for such I/O signals, and then combine the indirect routing method and the time-multiplexed signal selection method to realize effective time-multiplexing of inter-FPGA I/O signals on systems with various topologies.

In this paper, we extend the circuit partitioning algorithm which we have proposed for multi-FPGA systems and present a new algorithm in which the delay of each critical signal path is within a specified upper bound imposed on it. The core of the presented algorithm is recursive bipartitioning of a circuit. The bipartitioning procedure consists of three stages: 0) detection of critical paths; 1) bipartitioning of a set of primary inputs and outputs; and 2) bipartitioning of a set of logic-blocks. In 0), the algorithm computes the lower bounds of delays for paths with path delay constraints and detects the critical paths based on the difference between the lower and upper bound dynamically in every bipartitioning procedure. The delays of the critical paths are reduced with higher priority. In 1), the algorithm attempts to assign the primary inputs and outputs on each critical path to one chip so that the critical path does not cross between chips. Finally in 2), the algorithm not only decreases the number of crossings between chips but also assigns the logic-blocks on each critical path to one chip by exploiting a network flow technique. The algorithm has been implemented and applied to MCNC PARTITIONING 93 benchmark circuits. The experimental results demonstrate that it resolves almost all path delay constraints with maintaining the maximum number of required I/O blocks per chip small compared with conventional alogorithms.