Abstract-High-performance computing systems are moving towards 2.5D and 3D memory hierarchies, based on High Bandwidth Memory (HBM) and Hybrid Memory Cube (HMC) to mitigate the main memory bottlenecks. This trend is also creating new opportunities to revisit near-memory computation. In this paper, we propose a flexible processor-in-memory (PIM) solution for scalable and energy-efficient execution of deep convolutional networks (ConvNets), one of the fastest-growing workloads for servers and high-end embedded systems. Our codesign approach consists of a network of Smart Memory Cubes (modular extensions to the standard HMC) each augmented with a many-core PIM platform called NeuroCluster. NeuroClusters have a modular design based on NeuroStream coprocessors (for Convolution-intensive computations) and general-purpose RISC-V cores. In addition, a DRAM-friendly tiling mechanism and a scalable computation paradigm are presented to efficiently harness this computational capability with a very low programming effort. NeuroCluster occupies only 8% of the total logic-base (LoB) die area in a standard HMC and achieves an average performance of 240 GFLOPS for complete execution of full-featured state-of-the-art (SoA) ConvNets within a power budget of 2.5 W. Overall 11 W is consumed in a single SMC device, with 22.5 GFLOPS/W energy-efficiency which is 3.5X better than the best GPU implementations in similar technologies. The minor increase in system-level power and the negligible area increase make our PIM system a cost-effective and energy efficient solution, easily scalable to 955 GFLOPS with a small network of just four SMCs.

3D integration of solid-state memories and logic, as demonstrated by the Hybrid Memory Cube (HMC), offers major opportunities for revisiting near-memory computation and gives new hope to mitigate the power and performance losses caused by the "memory wall". Several publications in the past few years demonstrate this renewed interest. In this paper we present the first exploration steps towards design of the Smart Memory Cube (SMC), a new Processor-in-Memory (PIM) architecture that enhances the capabilities of the logic-base (LoB) die in HMC. An accurate simulation environment called SMCSim has been developed, along with a full featured software stack. The key contribution of this work is full system analysis of near memory computation including high-level software to low-level firmware and hardware layers, considering offloading and dynamic overheads caused by the operating system (OS), cache coherence, and memory management. A zero-copy pointer passing mechanism has been devised to allow low overhead data sharing between the host and the PIM. Benchmarking results demonstrate up to 2X performance improvement in comparison with the host Systemon-Chip (SoC), and around 1.5X against a similar host-side accelerator. Moreover, by scaling down the voltage and frequency of PIM's processor it is possible to reduce energy by around 70 % and 55 % in comparison with the host and the accelerator, respectively.

Hybrid memory cube (HMC) has promised to improve bandwidth, power consumption, and density for the next-generation main memory systems. In addition, 3-D integration gives a second shot for revisiting near memory computation to fill the gap between processors and memories. In this paper, we study the required infrastructure inside the HMC to support near memory computation in a modular and flexible fashion. We propose a fully backward compatible extension to the standard HMC called the smart memory cube, and design a high bandwidth, low latency, and Advanced eXtensible Interface-4.0 compatible logic base (LoB) interconnect to serve the huge bandwidth demand by the HMCs serial links, and to provide extra bandwidth to a generic processor-in-memory (PIM) device embedded in the LoB. This interconnect features a novel address scrambling mechanism for the reduction in the vault/bank conflicts and robust operation even in the presence of pathological traffic patterns. Our cycle accurate simulation results demonstrate that this interconnect can easily meet the demands of the latest HMC specifications (up to 205 GB/s read bandwidth with 4 serial links and 32 memory vaults for injected random traffic). It further shown that the default addressing scheme of the HMC (low interleaving) is not reliable enough and operates poorly in the presence of specific traffic patterns from real applications. This is while the proposed scrambling mechanism operates robustly even in those cases. The interference between the PIM traffic and the main links is shown to be negligible when the number of PIM ports is limited to 2, requesting up to 64 GB/s without pushing the system into saturation. Finally, logic synthesis with Synopsys Design Compiler confirms that our interconnect is implementable and effective in terms of power, area, and timing (power consumption less than 5 mW up to 1 GHz and area less than 0.4 mm 2 ).

No abstract

Shared tightly coupled data memories are key architectural elements for building multi-core clusters in programmable accelerators and embedded systems, as they provide a convenient shared memory abstraction while avoiding cache coherence overheads. The performance of these memories largely depends on the architecture of the interconnect used between processing elements (PEs) and memory banks. The advent of three-dimensional (3D) technology has provided new opportunities to increase design modularity and reduce latency and manufacturing cost. In this study, the authors propose two 3D network architectures: C-logarithmic interconnect (LIN) and Distributed logarithmic interconnect (D-LIN) (designed in synthesisable RTL), which allow modular stacking of multiple L1 memory dies over a multi-core cluster with a limited number of PEs. The authors have used two through-silicon-via technologies: the state-of-the-art micro-bumps and the promising and dense Cu-Cu direct bonding. The overhead of electrostatic discharge protection circuits has been considered, as well. Architectural simulation results demonstrate that, in processor-to-L1-memory context, C-LIN and D-LIN perform significantly better than traditional network-on-chips and simple time-division multiplexing buses. Furthermore, post-layout results show that the proposed 3D architectures achieve comparable speed against their 2D counterparts, whereas enabling modularity: from 256 kB to 2 MB L1 memory configurations with a single mask set.