Search

To efficiently deploy machine learning applications to the edge, compute-in-memory (CIM) based hardware accelerator is a promising solution with improved throughput and energy efficiency. Instant-on inference is further enabled by emerging non-volatile memory technologies such as resistive random access memory (RRAM). This paper reviews the recent progresses of the RRAM based CIM accelerator design. First, the multilevel states RRAM characteristics are measured from a test vehicle to examine the key device properties for inference. Second, a benchmark is performed to study the scalability of the RRAM CIM inference engine and the feasibility towards monolithic 3D integration that stacks RRAM arrays on top of advanced logic process node. Third, grand challenges associated with in-situ training are presented. To support accurate and fast in-situ training and enable subsequent inference in an integrated platform, a hybrid precision synapse that combines RRAM with volatile memory (e.g. capacitor) is designed and evaluated at system-level. Prospects and future research needs are discussed.

Compute-in-memory (CIM) is an attractive solution to address the “memory wall” challenges for the extensive computation in deep learning hardware accelerators. For custom ASIC design, a specific chip instance is restricted to a specific network during runtime. However, the development cycle of the hardware is normally far behind the emergence of new algorithms. Although some of the reported CIM-based architectures can adapt to different deep neural network (DNN) models, few details about the dataflow or control were disclosed to enable such an assumption. Instruction set architecture (ISA) could support high flexibility, but its complexity would be an obstacle to efficiency. In this article, a runtime reconfigurable design methodology of CIM-based accelerators is proposed to support a class of convolutional neural networks running on one prefabricated chip instance with ASIC-like efficiency. First, several design aspects are investigated: (1) the reconfigurable weight mapping method; (2) the input side of data transmission, mainly about the weight reloading; and (3) the output side of data processing, mainly about the reconfigurable accumulation. Then, a system-level performance benchmark is performed for the inference of different DNN models, such as VGG-8 on a CIFAR-10 dataset and AlexNet GoogLeNet, ResNet-18, and DenseNet-121 on an ImageNet dataset to measure the trade-offs between runtime reconfigurability, chip area, memory utilization, throughput, and energy efficiency.

Physarum polycephalum, a unicellular and multiheaded slime mould, can form highly efficient networks connecting separated food sources during the process of foraging. These adaptive networks exhibit a unique characteristic in that they are optimized without the control of a central consciousness. Inspired by this phenomenon, we present an efficient exploration and navigation strategy for a swarm of robots, which exploits cooperation and self-organisation to overcome the limited abilities of the individual robots. The task faced by the robots consists in the exploration of an unknown environment in order to find a path between two distant target areas. For the proposed algorithm (EAIPP), we experimentally present robustness tests and obstacle tests conducted to analyse the performance of our algorithm and compare the proposed algorithm with other swarm robot foraging algorithms that also focus on the path formation task. This work has certain significance for the research of swarm robots and Physarum polycephalum. For the research of swarm robotics, our algorithm not only can lead multirobot as a whole to overcome the limitations of very simple individual agents but also can offer better performance in terms of search efficiency and success rate. For the research of Physarum polycephalum, this work is the first one combining swarm robots and Physarum polycephalum. It also reveals the potential of the Physarum polycephalum foraging principle in multirobot systems.

Compute-in-memory (CIM) is a promising computing paradigm to accelerate the inference of deep neural network (DNN) algorithms due to its high processing parallelism and energy efficiency. Prior CIM-based DNN accelerators mostly consider full custom design, which assumes that all the weights are stored on-chip. For lightweight smart edge devices, this assumption may not hold. In this article, CIM-based DNN accelerators are designed and benchmarked under different chip area constraints. First, a scheduling strategy and dataflow for DNN inference is investigated when only part of the weights can be stored on-chip. Two weight reload schemes are evaluated: 1) reload partial weights and reuse input/output feature maps and 2) load a batch of input and reuse the partial weights on-chip across the batch. Then, system-level performance benchmark is performed for the inference of ResNet-18 on ImageNet data set. The design tradeoffs with different area constraints, dataflow, and device technologies [static random access memory (SRAM) versus ferroelectric field-effect transistor (FeFET)] are discussed.

Compute-in-memory (CIM) with emerging non-volatile memories (eNVMs) is time and energy efficient for deep neural network (DNN) inference. However, challenges still remain for DNN in-situ training with eNVMs due to the asymmetric weight update behavior, high programming latency and energy consumption. To overcome these challenges, a hybrid precision synapse combining eNVMs with capacitor has been proposed. It leverages the symmetric and fast weight update in the volatile capacitor, as well as the non-volatility and large dynamic range of the eNVMs. In this article, DNN in-situ training architecture with hybrid precision synapses is proposed and system level benchmarked is conducted. First, the circuit modules required for in-situ training with hybrid precision synapses are designed and the system architecture is proposed. Then, the impact of different weight precision configurations, weight transfer interval and limited capacitor retention time on training accuracy is investigated by incorporating hardware properties into Tensorflow simulation. Finally, the system-level benchmark is conducted at 32nm technology node in the modified NeuroSim simulator for hybrid precision synapse, in comparison with the baseline designs that are solely based on eNVMs or SRAM technology. The benchmark results show that CIM accelerator based on hybrid precision synapse achieves at least 3.07x and 2.89x better energy efficiency for training compared with its eNVM counterparts and SRAM technology at 32nm node, respectively. 227x and 33.8x better energy efficiency are obtained when compared to GPU and TPU. The scaling trend of hybrid precision synapse is projected towards 7nm node and comparison with state-of-the-art 7nm SRAM technology is made.

Resistive RAM (RRAM) has been presented as a promising memory technology toward deep neural network (DNN) hardware design, with nonvolatility, high density, high ON/OFF ratio, and compatibility with logic process. However, prior RRAM works for DNNs have shown limitations on parallelism for in-memory computing, array efficiency with large peripheral circuits, multilevel analog operation, and demonstration of monolithic integration. In this article, we propose circuit-/device-level optimizations to improve the energy and density of RRAM-based in-memory computing architectures. We report experimental results based on prototype chip design of 128 × 64 RRAM arrays and CMOS peripheral circuits, where RRAM devices are monolithically integrated in a commercial 90-nm CMOS technology. We demonstrate the CMOS peripheral circuit optimization using input-splitting scheme and investigate the implication of higher low resistance state on energy efficiency and robustness. Employing the proposed techniques, we demonstrate RRAM-based in-memory computing with up to 116.0 TOPS/W energy efficiency and 84.2% CIFAR-10 accuracy. Furthermore, we investigate four-level programming with single RRAM device, and report the system-level performance and DNN accuracy results using circuit-level benchmark simulator NeuroSim.

Fault diagnosis is of great significance to improve the production efficiency and accuracy of industrial robots. Compared with the traditional gradient descent algorithm, the extreme learning machine (ELM) has the advantage of fast computing speed, but the input weights and the hidden node biases that are obtained at random affects the accuracy and generalization performance of ELM. However, the level-based learning swarm optimizer algorithm (LLSO) can quickly and effectively find the global optimal solution of large-scale problems, and can be used to solve the optimal combination of large-scale input weights and hidden biases in ELM. This paper proposes an extreme learning machine with a level-based learning swarm optimizer (LLSO-ELM) for fault diagnosis of industrial robot RV reducer. The model is tested by combining the attitude data of reducer gear under different fault modes. Compared with ELM, the experimental results show that this method has good stability and generalization performance.

In-memory computing with analog nonvolatile memories can accelerate the in situ training of deep neural networks. Recently, we proposed a synaptic cell of a ferroelectric transistor (FeFET) with two CMOS transistors (2T1F) that exploit the hybrid precision for training and inference, which overcomes the challenges of nonlinear and asymmetric weight update and achieves nearly software comparable training accuracy at the algorithm level. In this paper, we further present the circuit-level benchmark results of this hybrid precision synapse in terms of area, latency, and energy. The corresponding array architecture is presented and the array-level operations are illustrated. The benchmark is conducted by multilayer-perceptron (MLP) + NeuroSim framework with comparison to other capacitor-assisted (e.g., 3T1C + 2PCM) hybrid precision cell. The design tradeoffs and scalability are discussed between different implementations.

Compute-in-memory (CIM) with emerging non-volatile memories (eNVMs) is time and energy efficient for deep neural network (DNN) inference. However, challenges still remain for in-situ DNN training with eNVMs due to the asymmetric weight update behavior, high programming latency and energy consumption. To overcome these challenges, a hybrid precision synapse combining eNVMs with capacitor has been proposed. It leverages the symmetric and fast weight update in the volatile capacitor, as well as the non-volatility and large dynamic range of the eNVMs. In this paper, in-situ DNN training architecture with hybrid precision synapses is proposed and benchmarked with the modified NeuroSim simulator. First, all the circuit modules required for in-situ training with hybrid precision synapses are designed. Then, the impact of weight transfer interval and limited capacitor retention time on training accuracy is investigated by incorporating hardware properties into Tensorflow simulation. Finally, a system-level benchmark is conducted for hybrid precision synapse compared with baseline design that is solely based on eNVMs.

On-chip learning with compute-in-memory (CIM) paradigm has become popular in machine learning hardware design in the recent years. However, it is hard to achieve high on-chip learning accuracy due to the high nonlinearity in the weight update curve of emerging nonvolatile memory (eNVM) based analog synapse devices. Although digital synapse devices offer good learning accuracy, the row-by-row partial sum accumulation leads to high latency. In this paper, the methods to solve the aforementioned issues are presented with a device-to-algorithm level optimization. For analog synapses, novel hybrid precision synapses with good linearity and more advanced training algorithms are introduced to increase the on-chip learning accuracy. The latency issue for digital synapses can be solved by using parallel partial sum read-out scheme. All these features are included into the recently released MLP + NeuroSimV3.0, which is an in-house developed device-to-system evaluation framework for neuro-inspired accelerators based on CIM paradigm.

Swarm robotics refers to artificial swarm systems composed of a large number of autonomous mobile robots with relatively simple structures and functions. One of the basic problems of swarm robotics involves the target search process, which entails a cooperative search using limited perception and local interaction of robots under a self-organizing mechanism. When communication is limited, the connectivity of swarm robotics may decrease, leading to the failure of the target search task. This article describes a new target search method based on the robot chain model and the elimination mechanism. The proposed method allows the target search task to be completed efficiently and reliably while maintaining the connectivity of the robots. Experimental results show that the proposed algorithm offers better performance than conventional techniques in terms of search speed and success rate, and provides an effective method for solving the target search problem using swarm robotics in limited-communication environments.

Low throughput has been a critical issue in extreme ultraviolet (EUV) patterning due to the difficulty in increasing light source power. This limitation has driven the need for photoresists with better throughput which unfortunately come with higher line edge roughness (LER). In this work, the possibility of relaxing LER requirements for metal layer patterned by EUV lithography (EUVL) is studied. Single patterning and litho-etch litho-etch (LELE) patterning with EUVL are considered. To assess the impact of LER on design yield, analytical and simulation based modeling approaches are developed, which consider the LER induced metal wire shorts/opens and the enhanced time dependent dielectrics breakdown (TDDB) for metal wires with different geometries. The impact of LER on wire delay is studied by Elmore’s delay model.