GPU-based platforms provide high computation throughput for large mini-batch deep neural network computations. However, a large batch size may not be ideal for some situations, such as aiming at low latency, training on edge/mobile devices, partial retraining for personalization, and having irregular input sequence lengths. GPU performance suffers from low utilization especially for small-batch recurrent neural network (RNN) applications where sequential computations are required. In this article, we propose a hybrid architecture, called FARNN, which combines a GPU and an FPGA to accelerate RNN computation for small batch sizes. After separating RNN computations into GPU-efficient and GPU-inefficient tasks, we design special FPGA computation units that accelerate the GPU-inefficient RNN tasks. FARNN off-loads the GPU-inefficient tasks to the FPGA. We evaluate FARNN with synthetic RNN layers of various configurations on the Xilinx UltraScale+ FPGA and the NVIDIA P100 GPU in addition to evaluating it with real RNN applications. The evaluation result indicates that FARNN outperforms the P100 GPU platform for RNN training by up to 4.2$\times {}$with small batch sizes, long input sequences, and many RNN cells per layer.

Recently, OpenCL has been emerging as a programming model for energy-efficient FPGA accelerators. However, the state-of-the-art OpenCL frameworks for FPGAs suffer from poor performance and usability. This paper proposes a high-level synthesis framework of OpenCL for FPGAs, called SOFF. It automatically synthesizes a datapath to execute many OpenCL kernel threads in a pipelined manner. It also synthesizes an efficient memory subsystem for the datapath based on the characteristics of OpenCL kernels. Unlike previous high-level synthesis techniques, we propose a formal way to handle variable latency instructions, complex control flows, OpenCL barriers, and atomic operations that appear in real-world OpenCL kernels. SOFF is the first OpenCL framework that correctly compiles and executes all applications in the SPEC ACCEL benchmark suite except three applications that require more FPGA resources than are available. In addition, SOFF achieves the speedup of 1.33 over Intel FPGA SDK for OpenCL without any explicit user annotation or source code modification.

Artificial linking of two isolated culture dishes is a fascinating means of investigating interactions among multiple groups of microbes or fungi. We examined artificial interaction between two isolated dishes containing Euglena cells, which are photophobic to strong blue light. The spatial distribution of swimming Euglena cells in two micro-aquariums in the dishes was evaluated as a set of new measures: the trace momentums (TMs). The blue light patterns next irradiated onto each dish were deduced from the set of TMs using digital or analogue feedback algorithms. In the digital feedback experiment, one of two different pattern-formation rules was imposed on each feedback system. The resultant cell distribution patterns satisfied the two rules with an and operation, showing that cooperative interaction was realized in the interlink feedback. In the analogue experiment, two dishes A and B were interlinked by a feedback algorithm that illuminated dish A (B) with blue light of intensity proportional to the cell distribution in dish B (A). In this case, a distribution pattern and its reverse were autonomously formed in the two dishes. The autonomous formation of a pair of reversal patterns reflects a type of habitat separation realized by competitive interaction through the interlink feedback. According to this study, interlink feedback between two or more separate culture dishes enables artificial interactions between isolated microbial groups, and autonomous cellular distribution patterns will be achieved by correlating various microbial species, despite environmental and spatial scale incompatibilities. The optical interlink feedback is also useful for enhancing the performance of Euglena-based soft biocomputing.