1 March 2024 in IEEE Transactions on Circuits and Systems I: Regular Papers
This paper presents the design and development of Signed Carry Disregard Multiplier (SCDM8), a family of signed approximate multipliers tailored for integration into Convolutional Neural Networks (CNNs). Extensive experiments were conducted on popular pre-trained CNN models, including VGG16, VGG19, ResNet101, ResNet152, MobileNetV2, InceptionV3, and ConvNeXt-T to evaluate the trade-off between accuracy and approximation. The results demonstrate that ACE-CNN outperforms other configurations, offering a favorable balance between accuracy and computational efficiency. In our experiments, when applied to VGG16, SCDM8 achieves an average reduction in power consumption of 35% with a marginal decrease in accuracy of only 1.5%. Similarly, when incorporated into ResNet152, SCDM8 yields an energy saving of 42% while sacrificing only 1.8% in accuracy. ACE-CNN provides the first approximate version of ConvNeXt which yields up to 72% energy improvement at the price of less than only 1.3% Top-1 accuracy. These results highlight the suitability of SCDM8 as an approximation method across various CNN models. Our analysis shows that the ACE-CNN outperforms state-of-the-art approaches in accuracy, energy efficiency, and computation precision for image classification tasks in CNNs. Our study investigated the resiliency of CNN models to approximate multipliers, revealing that ResNet101 demonstrated the highest resiliency with an average difference in the accuracy of 0.97%, whereas LeNet5 Inspired-CNN exhibited the lowest resiliency with an average difference of 2.92%. These findings aid in selecting energy-efficient approximate multipliers for CNN-based systems, and contribute to the development of energy-efficient deep learning systems by offering an effective approximation technique for multipliers in CNNs. The proposed SCDM8 family of approximate multipliers opens new avenues for efficient deep learning applications, enabling significant energy savings with virtually no loss in accuracy.
30 September 2024 in ACM Transactions on Reconfigurable Technology and Systems
In recent years, the adoption of FPGAs in datacenters has increased, with a growing number of users choosing High-Level Synthesis (HLS) as their preferred programming method. While HLS simplifies FPGA programming, one notable challenge arises when scaling up designs for modern datacenter FPGAs that comprise multiple dies. The extra delays introduced due to die crossings and routing congestion can significantly degrade the frequency of large designs on these FPGA boards. Due to the gap between HLS design and physical design, it is challenging for HLS programmers to analyze and identify the root causes, and fix their HLS design to achieve better timing closure. Recent efforts have aimed to address these issues by employing coarse-grained floorplanning and pipelining strategies on task-parallel HLS designs where multiple tasks run concurrently and communicate through FIFO stream channels. However, many applications are not streaming friendly and many existing accelerator designs heavily rely on buffer channel based communication between tasks.In this work, we take a step further to support a task-parallel programming model where tasks can communicate via both FIFO stream channels and buffer channels. To achieve this goal, we design and implement the PASTA framework, which takes a large task-parallel HLS design as input and automatically generates a high-frequency FPGA accelerator via HLS and physical design co-optimization. Our framework introduces a latency-insensitive buffer channel design, which supports memory partitioning and ping-pong buffering while remaining compatible with vendor HLS tools. On the frontend, we provide an easy-to-use programming model for utilizing the proposed buffer channel; while on the backend, we implement efficient placement and pipelining strategies for the proposed buffer channel. To validate the effectiveness of our framework, we test it on four widely used Rodinia HLS benchmarks and two real-world accelerator designs and show an average frequency improvement of 25%, with peak improvements of up to 89% on AMD/Xilinx Alveo U280 boards compared to Vitis HLS baselines.
6 March 2026 in ACM Transactions on Reconfigurable Technology and Systems
Sparse Matrix-Dense Matrix Multiplication (SpMM) is a critical operation in scientific computing, machine learning, and graph analytics. However, accelerating SpMM on FPGAs presents major challenges due to irregular memory access patterns and imbalanced workload distribution. In this work, we address a fundamental bottleneck in SpMM acceleration on High Bandwidth Memory (HBM)-equipped FPGAs: workload imbalance among processing elements (PEs). Additionally, we mitigate a scalability barrier present in state-of-the-art designs—namely, the tight coupling between PEs and HBM channels for dense matrix access. Furthermore, we provide an automated design space exploration framework. We propose HiSpMM, a high-performance SpMM accelerator architecture that introduces Dense Row Sharing to mitigate PE under-utilization by distributing heavy-row computations, a decoupled HBM access mechanism to allow independent scaling of PEs and memory bandwidth, and an automation tool that optimizes design parameters according to matrix structure-specific properties and user-defined hardware constraints. Our design achieves a geomean of \(5.81\times\) speedup and \(5.75\times\) energy efficiency improvement for imbalanced matrices compared to state-of-the-art designs, while also maintaining competitive performance for balanced matrices on the AMD/Xilinx U280 HBM FPGA board. Our HiSpMM project will be open sourced in the near future at https://github.com/SFU-HiAccel/HiSpMM .