Computational memory (CM) is a promising approach for accelerating inference on neural networks (NN) by using enhanced memories that, in addition to storing data, allow computations on them. One of the main challenges of this approach is defining a hardware/software interface that allows a compiler to map NN models for efficient execution on the underlying CM accelerator. This is a non-trivial task because efficiency dictates that the CM accelerator is explicitly programmed as a dataflow engine where the execution of the different NN layers form a pipeline. In this paper, we present our work towards a software stack for executing ML models on such a multi-core CM accelerator. We describe an architecture for the hardware and software, and focus on the problem of implementing the appropriate control logic so that data dependencies are respected. We propose a solution to the latter that is based on polyhedral compilation.
Convolutional neural networks (CNNs) are the core of most state-of-the-art deep learning algorithms specialized for object detection and classification. CNNs are both computationally complex and embarrassingly parallel. Two properties that leave room for potential software and hardware optimizations for embedded systems. Given a programmable hardware accelerator with a CNN oriented custom instructions set, the compiler's task is to exploit the hardware's full potential, while abiding with the hardware constraints and maintaining generality to run different CNN models with varying workload prop…
There is increasing demand to bring machine learning capabilities to low power devices. By integrating the computational power of machine learning with the deployment capabilities of low power devices, a number of new applications become possible. In some applications, such devices will not even have a battery, and must rely solely on energy harvesting techniques. This puts extreme constraints on the hardware, which must be energy efficient and capable of tolerating interruptions due to power outages. Here, as a representative example, we propose an in-memory support vector machine learning ac…
The excellent performance of modern deep neural networks (DNNs) comes at an often prohibitive training cost, limiting the rapid development of DNN innovations and raising various environmental concerns. To reduce the dominant data movement cost of training, process in-memory (PIM) has emerged as a promising solution as it alleviates the need to access DNN weights. However, state-of-the-art PIM DNN training accelerators employ either analog/mixed signal computing which has limited precision or digital computing based on a memory technology that supports limited logic functions and thus requires…
Building efficient embedded deep learning systems requires a tight co-design between DNN algorithms, memory hierarchy, and dataflow. However, owing to the large degrees of freedom in the design space, finding an optimal solution through the implementation of individual design points becomes infeasible. Recently, several estimation frameworks for fast design space exploration (DSE) have emerged, yet they either suffer from long runtimes or a limited exploration space. This work introduces ZigZag, a memory-centric rapid DNN accelerator DSE framework which extends the DSE with uneven mapping oppo…
Hardware accelerators are key to the efficiency and performance of system-on-chip (SoC) architectures. With high-level synthesis (HLS), designers can easily obtain several performance-cost trade-off implementations for each component of a complex hardware accelerator. However, navigating this design space in search of the Pareto-optimal implementations at the system level is a hard optimization task. We present COSMOS, an automatic methodology for the design-space exploration (DSE) of complex accelerators, that coordinates both HLS and memory optimization tools in a compositional way. First, t…
Many applications are increasingly requiring numerical simulations for solving complex problems. Most of these numerical algorithms are massively parallel and often implemented on parallel high-performance computers. However, classic CPU-based platforms suffers due to the demand for higher resolutions and the exponential growth of data. FPGAs offer a powerful and flexible alternative that can host accelerators to complement such platforms. Developing such application-specific accelerators is still challenging because it is hard to provide efficient code for hardware synthesis. In this paper, w…
Convolutional Neural Networks (CNNs) are rapidly gaining popularity in varied fields. Due to their increasingly deep and computationally heavy structures, it is difficult to deploy them on energy constrained mobile applications. Hardware accelerators such as FPGAs have come up as an attractive alternative. However, with the limited on-chip memory and computation resources of FPGA, meeting the high memory throughput requirement and exploiting the parallelism of CNNs is a major challenge. We propose a high-performance FPGA based architecture - Depth Concatenation and Inter-Layer Fusion based Con…
Sparse matrix-vector multiplication (SpMV) multiplies a sparse matrix with a dense vector. SpMV plays a crucial role in many applications, from graph analytics to deep learning. The random memory accesses of the sparse matrix make accelerator design challenging. However, high bandwidth memory (HBM) based FPGAs are a good fit for designing accelerators for SpMV. In this paper, we present Serpens, an HBM based accelerator for general-purpose SpMV.Serpens features (1) a general-purpose design, (2) memory-centric processing engines, and (3) index coalescing to support the efficient processing of a…
We propose AIDA, an inference engine for accelerating fully-connected (FC) layers of Deep Neural Network (DNN). AIDA is an associative in-memory processor, where the bulk of data never leaves the confines of the memory arrays, and processing is performed in-situ. AIDA area and energy efficiency strongly benefit from sparsity and lower arithmetic precision. We show that AIDA outperforms the state of art inference accelerator, EIE, by 14.5x (peak performance) and 2.5x (throughput).
Residual block is a very common component in recent state-of-the art CNNs such as EfficientNet or EfficientDet. Shortcut data accounts for nearly 40% of feature-maps access in ResNet152 [8]. Most of the previous DNN compilers, accelerators ignore the shortcut data optimization. This paper presents ShortcutFusion, an optimization tool for FPGA-based accelerator with a reuse-aware static memory allocation for shortcut data, to maximize on-chip data reuse given resource constraints. From TensorFlow DNN models, the proposed design generates instruction sets for a group of nodes which uses an optim…
We propose a generic algorithmic building block to accelerate training of machine learning models on heterogeneous compute systems. Our scheme allows to efficiently employ compute accelerators such as GPUs and FPGAs for the training of large-scale machine learning models, when the training data exceeds their memory capacity. Also, it provides adaptivity to any system's memory hierarchy in terms of size and processing speed. Our technique is built upon novel theoretical insights regarding primal-dual coordinate methods, and uses duality gap information to dynamically decide which part of the da…
In current convolutional neural network (CNN) accelerators, communication (i.e., memory access) dominates the energy consumption. This work provides comprehensive analysis and methodologies to minimize the communication for CNN accelerators. For the off-chip communication, we derive the theoretical lower bound for any convolutional layer and propose a dataflow to reach the lower bound. This fundamental problem has never been solved by prior studies. The on-chip communication is minimized based on an elaborate workload and storage mapping scheme. We in addition design a communication-optimal CN…
Deformable convolutional networks have demonstrated outstanding performance in object recognition tasks with an effective feature extraction. Unlike standard convolution, the deformable convolution decides the receptive field size using dynamically generated offsets, which leads to an irregular memory access. Especially, the memory access pattern varies both spatially and temporally, making static optimization ineffective. Thus, a naive implementation would lead to an excessive memory footprint. In this paper, we present a novel approach to accelerate deformable convolution on FPGA. First, we …
We present a vision for the Erudite architecture that redefines the compute and memory abstractions such that memory bandwidth and capacity become first-class citizens along with compute throughput. In this architecture, we envision coupling a high-density, massively parallel memory technology like Flash with programmable near-data accelerators, like the streaming multiprocessors in modern GPUs. Each accelerator has a local pool of storage-class memory that it can access at high throughput by initiating very large numbers of overlapping requests that help to tolerate long access latency. The a…
Many state-of-the-art ML results have been obtained by scaling up the number of parameters in existing models. However, parameters and activations for such large models often do not fit in the memory of a single accelerator device; this means that it is necessary to distribute training of large models over multiple accelerators. In this work, we propose PipeDream-2BW, a system that supports memory-efficient pipeline parallelism. PipeDream-2BW uses a novel pipelining and weight gradient coalescing strategy, combined with the double buffering of weights, to ensure high throughput, low memory foo…
Technische Universität Wien（维也纳工业大学）；Institute of Computer Engineering（计算机工程研究所）
Enabling high energy efficiency is crucial for embedded implementations of deep learning. Several studies have shown that the DRAM-based off-chip memory accesses are one of the most energy-consuming operations in deep neural network (DNN) accelerators, and thereby limit the designs from achieving efficiency gains at the full potential. DRAM access energy varies depending upon the number of accesses required as well as the energy consumed per-access. Therefore, searching for a solution towards the minimum DRAM access energy is an important optimization problem. Towards this, we propose the ROMA…
Flexibility at hardware level is the main driving force behind adaptive systems whose aim is to realise microarhitecture deconfiguration 'online'. This feature allows the software/hardware stack to tolerate drastic changes of the workload in data centres. With emerge of FPGA reconfigurablity this technology is becoming a mainstream computing paradigm. Adaptivity is usually accompanied by the high-level tools to facilitate multi-dimensional space exploration. An essential aspect in this space is memory orchestration where on-chip and off-chip memory distribution significantly influences the arc…
