Even with generational improvements in DRAM technology, memory access latency still remains the major bottleneck for application accelerators, primarily due to limitations in memory interface IPs which cannot fully account for variations in target applications, the algorithms used, and accelerator architectures. Since developing memory controllers for different applications is time-consuming, this paper introduces a modular and programmable memory controller that can be configured for different target applications on available hardware resources. The proposed memory controller efficiently supports cache-line accesses along with bulk memory transfers. The user can configure the controller depending on the available logic resources on the FPGA, memory access pattern, and external memory specifications. The modular design supports various memory access optimization techniques including, request scheduling, internal caching, and direct memory access. These techniques contribute to reducing the overall latency while maintaining high sustained bandwidth. We implement the system on a state-of-the-art FPGA and evaluate its performance using two widely studied domains: graph analytics and deep learning workloads. We show improved overall memory access time up to 58% on CNN and GCN workloads compared with commercial memory controller IPs.
With the expansion of the semiconductor supply chain, the use of recycled field-programmable gate arrays (FPGAs) has become a serious concern. Several methods for detecting recycled FPGAs by analyzing the ring oscillator (RO) frequencies have been proposed; however, most assume the known fresh FPGAs (KFFs) as the training data in machine-learning-based classification. In this study, we propose a novel recycled FPGA detection method based on an unsupervised anomaly detection scheme when there are few or no KFFs available. As the RO frequencies in the neighboring logic blocks on an FPGA are simi…
University of Southern California（南加州大学）；US Army Research Lab（美国陆军研究实验室）
Tensor decomposition has become an essential tool in many data science applications. Sparse Matricized Tensor Times Khatri-Rao Product (MTTKRP) is the pivotal kernel in tensor decomposition algorithms that decompose higher-order real-world large tensors into multiple matrices. Accelerating MTTKRP can speed up the tensor decomposition process immensely. Sparse MTTKRP is a challenging kernel to accelerate due to its irregular memory access characteristics. Implementing accelerators on Field Programmable Gate Array (FPGA) for kernels such as MTTKRP is attractive due to the energy efficiency and t…
The traditional USB 3.0 communication based on FPGA uses an external chip as a USB PHY or a USB controller including a USB PHY. This paper realizes a USB 3.0 controller by using FPGA resources, in which FPGA logic realizes a serial interface engine, and an FPGA internal transceiver is as a USB PHY. Used slices percent after implementation is 4.59% in Kintex-7 325t. The test result shows that the speed of USB 3.0 is more than 320 MB/s bulk-in and bulk-out transfers.
Graphics Processing Units (GPUs) are currently the dominating programmable architecture for Deep Learning (DL) accelerators. The adoption of Field Programmable Gate Arrays (FPGAs) in DL accelerators is however getting momentum. In this paper, we demonstrate that Direct Hardware Mapping (DHM) of a Convolutional Neural Network (CNN) on an embedded FPGA substantially outperforms a GPU implementation in terms of energy efficiency and execution time. However, DHM is highly resource intensive and cannot fully substitute the GPU when implementing a state-of-the-art CNN. We thus propose a hybrid FPGA-…
This paper presents BP-BedRock, the open-source cache coherence protocol and system implemented within the BlackParrot 64-bit RISC-V multicore processor. BP-BedRock implements the BedRock directory-based MOESIF cache coherence protocol and includes two different open-source coherence protocol engines, one FSM-based and the other microcode programmable. Both coherence engines support coherent uncacheable access to cacheable memory and L1-based atomic read-modify-write operations. Fitted within the BlackParrot multicore, BP-BedRock has been silicon validated in a GlobalFoundries 12nm FinFET proc…
Demands on Field-Programmable Gate Array (FPGA) data transport have been increasing over the years as frame sizes and refresh rates increase. As the bandwidths requirements increase the ability to implement data transport protocol layers using "soft" programmable logic becomes harder and start to require harden IP blocks implementation. To reduce the number of physical links and interconnects, it is common for data acquisition systems to require interleaving of streams on the same link (e.g. streaming data and streaming register access). This paper presents a way to leverage existing FPGA hard…
