It is well known that many types of artificial neural networks, including recurrent networks, can achieve a high classification accuracy even with low-precision weights and activations. The reduction in precision generally yields much more efficient hardware implementations in regards to hardware cost, memory requirements, energy, and achievable throughput. In this paper, we present the first systematic exploration of this design space as a function of precision for Bidirectional Long Short-Term Memory (BiLSTM) neural network. Specifically, we include an in-depth investigation of precision vs. accuracy using a fully hardware-aware training flow, where during training quantization of all aspects of the network including weights, input, output and in-memory cell activations are taken into consideration. In addition, hardware resource cost, power consumption and throughput scalability are explored as a function of precision for FPGA-based implementations of BiLSTM, and multiple approaches of parallelizing the hardware. We provide the first open source HLS library extension of FINN [1] for parameterizable hardware architectures of LSTM layers on FPGAs which offers full precision flexibility and allows for parameterizable performance scaling offering different levels of parallelism within the architecture. Based on this library, we present an FPGA-based accelerator for BiLSTM neural network designed for optical character recognition, along with numerous other experimental proof points for a Zynq UltraScale+ XCZU7EV MPSoC within the given design space.

No abstract

No abstract

Recurrent Neural Networks, in particular One-dimensional and Multidimensional Long Short-Term Memory (1D-LSTM and MD-LSTM) have achieved state-of-the-art classification accuracy in many applications such as machine translation, image caption generation, handwritten text recognition, medical imaging and many more. However, high classification accuracy comes at high compute, storage, and memory bandwidth requirements, which make their deployment challenging, especially for energy-constrained platforms such as portable devices. In comparison to CNNs, not so many investigations exist on efficient hardware implementations for 1D-LSTM especially under energy constraints, and there is no research publication on hardware architecture for MD-LSTM. In this article, we present two novel architectures for LSTM inference: a hardware architecture for MD-LSTM, and a DRAM-based Processing-in-Memory (DRAM-PIM) hardware architecture for 1D-LSTM. We present for the first time a hardware architecture for MD-LSTM, and show a trade-off analysis for accuracy and hardware cost for various precisions. We implement the new architecture as an FPGA-based accelerator that outperforms NVIDIA K80 GPU implementation in terms of runtime by up to 84× and energy efficiency by up to 1238× for a challenging dataset for historical document image binarization from DIBCO 2017 contest, and a well known MNIST dataset for handwritten digits recognition. Our accelerator demonstrates highest accuracy and comparable throughput in comparison to state-of-the-art FPGA-based implementations of multilayer perceptron for MNIST dataset. Furthermore, we present a new DRAM-PIM architecture for 1D-LSTM targeting energy efficient compute platforms such as portable devices. The DRAM-PIM architecture integrates the computation units in a close proximity to the DRAM cells in order to maximize the data parallelism and energy efficiency. The proposed DRAM-PIM design is 16.19 × more energy efficient as compared to FPGA implementation. The total chip area overhead of this design is 18 % compared to a commodity 8 Gb DRAM chip. Our experiments show that the DRAM-PIM implementation delivers a throughput of 1309.16 GOp/s for an optical character recognition application.

In recent years, $$\hbox {optical character recognition (OCR)}$$ optical character recognition (OCR) systems have been used to digitally preserve historical archives. To transcribe historical archives into a machine-readable form, first, the documents are scanned, then an $$\hbox {OCR}$$ OCR is applied. In order to digitize documents without the need to remove them from where they are archived, it is valuable to have a portable device that combines scanning and $$\hbox {OCR}$$ OCR capabilities. Nowadays, there exist many commercial and open-source document digitization techniques, which are optimized for contemporary documents. However, they fail to give sufficient text recognition accuracy for transcribing historical documents due to the severe quality degradation of such documents. On the contrary, the anyOCR system, which is designed to mainly digitize historical documents, provides high accuracy. However, this comes at a cost of high computational complexity resulting in long runtime and high power consumption. To tackle these challenges, we propose a low power energy-efficient accelerator with real-time capabilities called iDocChip, which is a configurable hybrid hardware-software programmable $$\hbox {System-on-Chip (SoC)}$$ System-on-Chip (SoC) based on anyOCR for digitizing historical documents. In this paper, we focus on one of the most crucial processing steps in the anyOCR system: Text and Image Segmentation, which makes use of a multi-resolution morphology-based algorithm. Moreover, an optimized $$\hbox {FPGA}$$ FPGA -based hybrid architecture of this anyOCR step along with its optimized software implementations are presented. We demonstrate our results on multiple embedded and general-purpose platforms with respect to runtime and power consumption. The resulting hardware accelerator outperforms the existing anyOCR by 6.2$$\times$$ × , while achieving 207$$\times$$ × higher energy-efficiency and maintaining its high accuracy.