Search for: All records

With reduced data reuse and parallelism, recent convolutional neural networks (CNNs) create new challenges for FPGA acceleration. Systolic arrays (SAs) are efficient, scalable architectures for convolutional layers, but without proper optimizations, their efficiency drops dramatically for reasons: 1) the different dimensions within same-type layers, 2) the different convolution layers especially transposed and dilated convolutions, and 3) CNN’s complex dataflow graph. Furthermore, significant overheads arise when integrating FPGAs into machine learning frameworks. Therefore, we present a flexible, composable architecture called FlexCNN, which delivers high computation efficiency by employing dynamic tiling, layer fusion, and data layout optimizations. Additionally, we implement a novel versatile SA to process normal, transposed, and dilated convolutions efficiently. FlexCNN also uses a fully-pipelined software-hardware integration that alleviates the software overheads. Moreover, with an automated compilation flow, FlexCNN takes a CNN in the ONNX representation, performs a design space exploration, and generates an FPGA accelerator. The framework is tested using three complex CNNs: OpenPose, U-Net, and E-Net. The architecture optimizations achieve 2.3 × performance improvement. Compared to a standard SA, the versatile SA achieves close-to-ideal speedups, with up to 15.98 × and 13.42 × for transposed and dilated convolutions, with a 6% average area overhead. The pipelined integration leadsmore »to a 5 × speedup for OpenPose.« less

Dense matrix multiply (MM) serves as one of the most heavily used kernels in deep learning applications. To cope with the high computation demands of these applications, heterogeneous architectures featuring both FPGA and dedicated ASIC accelerators have emerged as promising platforms. For example, the AMD/Xilinx Versal ACAP architecture combines general-purpose CPU cores and programmable logic (PL) with AI Engine processors (AIE) optimized for AI/ML. An array of 400 AI Engine processors executing at 1 GHz can theoretically provide up to 6.4 TFLOPs performance for 32-bit floating-point (fp32) data. However, machine learning models often contain both large and small MM operations. While large MM operations can be parallelized efficiently across many cores, small MM operations typically cannot. In our investigation, we observe that executing some small MM layers from the BERT natural language processing model on a large, monolithic MM accelerator in Versal ACAP achieved less than 5% of the theoretical peak performance. Therefore, one key question arises: How can we design accelerators to fully use the abundant computation resources under limited communication bandwidth for end-to-end applications with multiple MM layers of diverse sizes? We identify the biggest system throughput bottleneck resulting from the mismatch of massive computation resources of onemore »monolithic accelerator and the various MM layers of small sizes in the application. To resolve this problem, we propose the CHARM framework to compose multiple diverse MM accelerator architectures working concurrently towards different layers within one application. CHARM includes analytical models which guide design space exploration to determine accelerator partitions and layer scheduling. To facilitate the system designs, CHARM automatically generates code, enabling thorough onboard design verification. We deploy the CHARM framework for four different deep learning applications, including BERT, ViT, NCF, MLP, on the AMD/Xilinx Versal ACAP VCK190 evaluation board. Our experiments show that we achieve 1.46 TFLOPs, 1.61 TFLOPs, 1.74 TFLOPs, and 2.94 TFLOPs inference throughput for BERT, ViT, NCF, MLP, respectively, which obtain 5.40x, 32.51x, 1.00x and 1.00x throughput gains compared to one monolithic accelerator.« less

Ever-growing edge applications often require short processing latency and high energy efficiency to meet strict timing and power budget. In this work, we propose that the compact long short-term memory (LSTM) model can approximate conventional acausal algorithms with reduced latency and improved efficiency for real-time causal prediction, especially for the neural signal processing in closed-loop feedback applications. We design an LSTM inference accelerator by taking advantage of the fine-grained parallelism and pipelined feedforward and recurrent updates. We also propose a bit-sparse quantization method that can reduce the circuit area and power consumption by replacing the multipliers with the bit-shift operators. We explore different combinations of pruning and quantization methods for energy-efficient LSTM inference on datasets collected from the electroencephalogram (EEG) and calcium image processing applications. Evaluation results show that our proposed LSTM inference accelerator can achieve 1.19 GOPS/mW energy efficiency. The LSTM accelerator with 2-sbit/16-bit sparse quantization and 60% sparsity can reduce the circuit area and power consumption by 54.1% and 56.3%, respectively, compared with a 16-bit baseline implementation.

Adopting FPGA as an accelerator in datacenters is becoming mainstream for customized computing, but the fact that FPGAs are hard to program creates a steep learning curve for software programmers. Even with the help of high-level synthesis (HLS) , accelerator designers still have to manually perform code reconstruction and cumbersome parameter tuning to achieve optimal performance. While many learning models have been leveraged by existing work to automate the design of efficient accelerators, the unpredictability of modern HLS tools becomes a major obstacle for them to maintain high accuracy. To address this problem, we propose an automated DSE framework— AutoDSE —that leverages a bottleneck-guided coordinate optimizer to systematically find a better design point. AutoDSE detects the bottleneck of the design in each step and focuses on high-impact parameters to overcome it. The experimental results show that AutoDSE is able to identify the design point that achieves, on the geometric mean, 19.9× speedup over one CPU core for MachSuite and Rodinia benchmarks. Compared to the manually optimized HLS vision kernels in Xilinx Vitis libraries, AutoDSE can reduce their optimization pragmas by 26.38× while achieving similar performance. With less than one optimization pragma per design on average, we are making progress towardsmore »democratizing customizable computing by enabling software programmers to design efficient FPGA accelerators.« less