Digital neuromorphic technology: current and future prospects.

Digital approaches to brain-inspired computing have advanced apace over recent years - where is the state-of-the-art and what does the future hold?

FPGA-based fast bin-ratio spiking ensemble network for radioisotope identification.

In this work, we demonstrate the training, conversion, and implementation flow of an FPGA-based bin-ratio ensemble spiking neural network applied for radioisotope identification. The combination of techniques including learned step quantisation (LSQ) and pruning facilitated the implementation by compressing the network's parameters down to 30% yet retaining the accuracy of 97.04% with an accuracy loss of less than 1%. Meanwhile, the proposed ensemble network of 20 3-layer spiking neural networks (SNNs), which incorporates 1160 spiking neurons, only needs 334 µs for a single inference with the given clock frequency of 100 MHz. Under such optimisation, this FPGA implementation in an Artix-7 board consumes 157 µJ per inference by estimation.

BitBrain and Sparse Binary Coincidence (SBC) memories: Fast, robust learning and inference for neuromorphic architectures.

We present an innovative working mechanism (the SBC memory) and surrounding infrastructure (BitBrain) based upon a novel synthesis of ideas from sparse coding, computational neuroscience and information theory that enables fast and adaptive learning and accurate, robust inference. The mechanism is designed to be implemented efficiently on current and future neuromorphic devices as well as on more conventional CPU and memory architectures. An example implementation on the SpiNNaker neuromorphic platform has been developed and initial results are presented. The SBC memory stores coincidences between features detected in class examples in a training set, and infers the class of a previously unseen test example by identifying the class with which it shares the highest number of feature coincidences. A number of SBC memories may be combined in a BitBrain to increase the diversity of the contributing feature coincidences. The resulting inference mechanism is shown to have excellent classification performance on benchmarks such as MNIST and EMNIST, achieving classification accuracy with single-pass learning approaching that of state-of-the-art deep networks with much larger tuneable parameter spaces and much higher training costs. It can also be made very robust to noise. BitBrain is designed to be very efficient in training and inference on both conventional and neuromorphic architectures. It provides a unique combination of single-pass, single-shot and continuous supervised learning; following a very simple unsupervised phase. Accurate classification inference that is very robust against imperfect inputs has been demonstrated. These contributions make it uniquely well-suited for edge and IoT applications.

Error driven synapse augmented neurogenesis.

Capturing the learning capabilities of the brain has the potential to revolutionize artificial intelligence. Humans display an impressive ability to acquire knowledge on the fly and immediately store it in a usable format. Parametric models of learning, such as gradient descent, focus on capturing the statistical properties of a data set. Information is precipitated into a network through repeated updates of connection weights in the direction gradients dictate will lead to less error. This work presents the EDN (Error Driven Neurogenesis) algorithm which explores how neurogenesis coupled with non-linear synaptic activations enables a biologically plausible mechanism to immediately store data in a one-shot, online fashion and readily apply it to a task without the need for parameter updates. Regression (auto-mpg) test error was reduced more than 135 times faster and converged to an error around three times smaller compared to gradient descent using ADAM optimization. EDN also reached the same level of performance in wine cultivar classification 25 times faster than gradient descent and twice as fast when applied to MNIST and the inverted pendulum (reinforcement learning).

Quantization Framework for Fast Spiking Neural Networks.

Compared with artificial neural networks (ANNs), spiking neural networks (SNNs) offer additional temporal dynamics with the compromise of lower information transmission rates through the use of spikes. When using an ANN-to-SNN conversion technique there is a direct link between the activation bit precision of the artificial neurons and the time required by the spiking neurons to represent the same bit precision. This implicit link suggests that techniques used to reduce the activation bit precision of ANNs, such as quantization, can help shorten the inference latency of SNNs. However, carrying ANN quantization knowledge over to SNNs is not straightforward, as there are many fundamental differences between them. Here we propose a quantization framework for fast SNNs (QFFS) to overcome these difficulties, providing a method to build SNNs with enhanced latency and reduced loss of accuracy relative to the baseline ANN model. In this framework, we promote the compatibility of ANN information quantization techniques with SNNs, and suppress "occasional noise" to minimize accuracy loss. The resulting SNNs overcome the accuracy degeneration observed previously in SNNs with a limited number of time steps and achieve an accuracy of 70.18% on ImageNet within 8 time steps. This is the first demonstration that SNNs built by ANN-to-SNN conversion can achieve a similar latency to SNNs built by direct training.

Linking Brain Structure, Activity, and Cognitive Function through Computation.

Understanding the human brain is a "Grand Challenge" for 21st century research. Computational approaches enable large and complex datasets to be addressed efficiently, supported by artificial neural networks, modeling and simulation. Dynamic generative multiscale models, which enable the investigation of causation across scales and are guided by principles and theories of brain function, are instrumental for linking brain structure and function. An example of a resource enabling such an integrated approach to neuroscientific discovery is the BigBrain, which spatially anchors tissue models and data across different scales and ensures that multiscale models are supported by the data, making the bridge to both basic neuroscience and medicine. Research at the intersection of neuroscience, computing and robotics has the potential to advance neuro-inspired technologies by taking advantage of a growing body of insights into perception, plasticity and learning. To render data, tools and methods, theories, basic principles and concepts interoperable, the Human Brain Project (HBP) has launched EBRAINS, a digital neuroscience research infrastructure, which brings together a transdisciplinary community of researchers united by the quest to understand the brain, with fascinating insights and perspectives for societal benefits.

Comparison of Artificial and Spiking Neural Networks on Digital Hardware.

Despite the success of Deep Neural Networks-a type of Artificial Neural Network (ANN)-in problem domains such as image recognition and speech processing, the energy and processing demands during both training and deployment are growing at an unsustainable rate in the push for greater accuracy. There is a temptation to look for radical new approaches to these applications, and one such approach is the notion that replacing the abstract neuron used in most deep networks with a more biologically-plausible spiking neuron might lead to savings in both energy and resource cost. The most common spiking networks use rate-coded neurons for which a simple translation from a pre-trained ANN to an equivalent spike-based network (SNN) is readily achievable. But does the spike-based network offer an improvement of energy efficiency over the original deep network? In this work, we consider the digital implementations of the core steps in an ANN and the equivalent steps in a rate-coded spiking neural network. We establish a simple method of assessing the relative advantages of rate-based spike encoding over a conventional ANN model. Assuming identical underlying silicon technology we show that most rate-coded spiking network implementations will not be more energy or resource efficient than the original ANN, concluding that more imaginative uses of spikes are required to displace conventional ANNs as the dominant computing framework for neural computation.

Stochastic rounding and reduced-precision fixed-point arithmetic for solving neural ordinary differential equations.

Although double-precision floating-point arithmetic currently dominates high-performance computing, there is increasing interest in smaller and simpler arithmetic types. The main reasons are potential improvements in energy efficiency and memory footprint and bandwidth. However, simply switching to lower-precision types typically results in increased numerical errors. We investigate approaches to improving the accuracy of reduced-precision fixed-point arithmetic types, using examples in an important domain for numerical computation in neuroscience: the solution of ordinary differential equations (ODEs). The Izhikevich neuron model is used to demonstrate that rounding has an important role in producing accurate spike timings from explicit ODE solution algorithms. In particular, fixed-point arithmetic with stochastic rounding consistently results in smaller errors compared to single-precision floating-point and fixed-point arithmetic with round-to-nearest across a range of neuron behaviours and ODE solvers. A computationally much cheaper alternative is also investigated, inspired by the concept of dither that is a widely understood mechanism for providing resolution below the least significant bit in digital signal processing. These results will have implications for the solution of ODEs in other subject areas, and should also be directly relevant to the huge range of practical problems that are represented by partial differential equations. This article is part of a discussion meeting issue 'Numerical algorithms for high-performance computational science'.

Real-time cortical simulation on neuromorphic hardware.

Real-time simulation of a large-scale biologically representative spiking neural network is presented, through the use of a heterogeneous parallelization scheme and SpiNNaker neuromorphic hardware. A published cortical microcircuit model is used as a benchmark test case, representing ≈1 mm2 of early sensory cortex, containing 77 k neurons and 0.3 billion synapses. This is the first hard real-time simulation of this model, with 10 s of biological simulation time executed in 10 s wall-clock time. This surpasses best-published efforts on HPC neural simulators (3 × slowdown) and GPUs running optimized spiking neural network (SNN) libraries (2 × slowdown). Furthermore, the presented approach indicates that real-time processing can be maintained with increasing SNN size, breaking the communication barrier incurred by traditional computing machinery. Model results are compared to an established HPC simulator baseline to verify simulation correctness, comparing well across a range of statistical measures. Energy to solution and energy per synaptic event are also reported, demonstrating that the relatively low-tech SpiNNaker processors achieve a 10 × reduction in energy relative to modern HPC systems, and comparable energy consumption to modern GPUs. Finally, system robustness is demonstrated through multiple 12 h simulations of the cortical microcircuit, each simulating 12 h of biological time, and demonstrating the potential of neuromorphic hardware as a neuroscience research tool for studying complex spiking neural networks over extended time periods. This article is part of the theme issue 'Harmonizing energy-autonomous computing and intelligence'.

SpiNNTools: The Execution Engine for the SpiNNaker Platform.

SpiNNaker is a massively parallel distributed architecture primarily focused on real time simulation of spiking neural networks. The largest realization of the architecture consists of one million general purpose processors, making it the largest neuromorphic computing platform in the world at the present time. Utilizing these processors efficiently requires expert knowledge of the architecture to generate executable code and to harness the potential of the unique inter-processor communications infra-structure that lies at the heart of the SpiNNaker architecture. This work introduces a software suite called SpiNNTools that can map a computational problem described as a graph into the required set of executables, application data and routing information necessary for simulation on this novel machine. The SpiNNaker architecture is highly scalable, giving rise to unique challenges in mapping the problem to the machines resources, loading the generated files to the machine and subsequently retrieving the results of simulation. In this paper we describe these challenges in detail and the solutions implemented.

Efficient Reward-Based Structural Plasticity on a SpiNNaker 2 Prototype.

Advances in neuroscience uncover the mechanisms employed by the brain to efficiently solve complex learning tasks with very limited resources. However, the efficiency is often lost when one tries to port these findings to a silicon substrate, since brain-inspired algorithms often make extensive use of complex functions, such as random number generators, that are expensive to compute on standard general purpose hardware. The prototype chip of the second generation SpiNNaker system is designed to overcome this problem. Low-power advanced RISC machine (ARM) processors equipped with a random number generator and an exponential function accelerator enable the efficient execution of brain-inspired algorithms. We implement the recently introduced reward-based synaptic sampling model that employs structural plasticity to learn a function or task. The numerical simulation of the model requires to update the synapse variables in each time step including an explorative random term. To the best of our knowledge, this is the most complex synapse model implemented so far on the SpiNNaker system. By making efficient use of the hardware accelerators and numerical optimizations, the computation time of one plasticity update is reduced by a factor of 2. This, combined with fitting the model into to the local static random access memory (SRAM), leads to 62% energy reduction compared to the case without accelerators and the use of external dynamic random access memory (DRAM). The model implementation is integrated into the SpiNNaker software framework allowing for scalability onto larger systems. The hardware-software system presented in this paper paves the way for power-efficient mobile and biomedical applications with biologically plausible brain-inspired algorithms.

Classification and regression of spatio-temporal signals using NeuCube and its realization on SpiNNaker neuromorphic hardware.

OBJECTIVE: The objective of this work is to use the capability of spiking neural networks to capture the spatio-temporal information encoded in time-series signals and decode them without the use of hand-crafted features and vector-based learning and the realization of the spiking model on low-power neuromorphic hardware. APPROACH: The NeuCube spiking model was used to classify different grasp movements directly from raw surface electromyography signals (sEMG), the estimations of the applied finger forces as well as the classification of two motor imagery movements from raw electroencephalography (EEG). In a parallel investigation, the designed spiking decoder was implemented on SpiNNaker neuromorphic hardware, which allows low-energy real-time processing. MAIN RESULTS: Experimental results reveal a better classification accuracy using the NeuCube model compared to traditional machine learning methods. For sEMG classification, we reached a training accuracy of 85% and a test accuracy of 84.8%, as well as less than 19% of relative root mean square error (rRMSE) when estimating finger forces from six subjects. For the EEG classification, a mean accuracy of 75% was obtained when tested on raw EEG data from nine subjects from the existing 2b dataset from 'BCI competition IV'. SIGNIFICANCE: This work provides a proof of concept for a successful implementation of the NeuCube spiking model on the SpiNNaker neuromorphic platform for raw sEMG and EEG decoding, which could chart a route ahead for a new generation of portable closed-loop and low-power neuroprostheses.

Memory-Efficient Deep Learning on a SpiNNaker 2 Prototype.

The memory requirement of deep learning algorithms is considered incompatible with the memory restriction of energy-efficient hardware. A low memory footprint can be achieved by pruning obsolete connections or reducing the precision of connection strengths after the network has been trained. Yet, these techniques are not applicable to the case when neural networks have to be trained directly on hardware due to the hard memory constraints. Deep Rewiring (DEEP R) is a training algorithm which continuously rewires the network while preserving very sparse connectivity all along the training procedure. We apply DEEP R to a deep neural network implementation on a prototype chip of the 2nd generation SpiNNaker system. The local memory of a single core on this chip is limited to 64 KB and a deep network architecture is trained entirely within this constraint without the use of external memory. Throughout training, the proportion of active connections is limited to 1.3%. On the handwritten digits dataset MNIST, this extremely sparse network achieves 96.6% classification accuracy at convergence. Utilizing the multi-processor feature of the SpiNNaker system, we found very good scaling in terms of computation time, per-core memory consumption, and energy constraints. When compared to a X86 CPU implementation, neural network training on the SpiNNaker 2 prototype improves power and energy consumption by two orders of magnitude.

sPyNNaker: A Software Package for Running PyNN Simulations on SpiNNaker.

This work presents sPyNNaker 4.0.0, the latest version of the software package for simulating PyNN-defined spiking neural networks (SNNs) on the SpiNNaker neuromorphic platform. Operations underpinning realtime SNN execution are presented, including an event-based operating system facilitating efficient time-driven neuron state updates and pipelined event-driven spike processing. Preprocessing, realtime execution, and neuron/synapse model implementations are discussed, all in the context of a simple example SNN. Simulation results are demonstrated, together with performance profiling providing insights into how software interacts with the underlying hardware to achieve realtime execution. System performance is shown to be within a factor of 2 of the original design target of 10,000 synaptic events per millisecond, however SNN topology is shown to influence performance considerably. A cost model is therefore developed characterizing the effect of network connectivity and SNN partitioning. This model enables users to estimate SNN simulation performance, allows the SpiNNaker team to make predictions on the impact of performance improvements, and helps demonstrate the continued potential of the SpiNNaker neuromorphic hardware.

Parallel Distribution of an Inner Hair Cell and Auditory Nerve Model for Real-Time Application.

This paper summarizes recent efforts in implementing a model of the ear's inner hair cell and auditory nerve on a neuromorphic hardware platform, the SpiNNaker machine. This exploits the massive parallelism of the target architecture to obtain real-time modeling of a biologically realistic number of human auditory nerve fibres. We show how this model can be integrated with additional modules that simulate previous stages of the early auditory pathway running on the same hardware architecture, thus producing a full-scale spiking auditory nerve output from a single sound stimulus. The results of the SpiNNaker implementation are shown to be comparable with a MATLAB version of the same model algorithms, while removing the inherent performance limitations associated with an increase in auditory model scale that are seen in the conventional computer simulations. Finally, we outline the potential for using this system as part of a full-scale, real-time digital model of the complete human auditory pathway on the SpiNNaker platform.

Structural Plasticity on the SpiNNaker Many-Core Neuromorphic System.

The structural organization of cortical areas is not random, with topographic maps commonplace in sensory processing centers. This topographical organization allows optimal wiring between neurons, multimodal sensory integration, and performs input dimensionality reduction. In this work, a model of topographic map formation is implemented on the SpiNNaker neuromorphic platform, running in realtime using point neurons, and making use of both synaptic rewiring and spike-timing dependent plasticity (STDP). In agreement with Bamford et al. (2010), we demonstrate that synaptic rewiring refines an initially rough topographic map over and beyond the ability of STDP, and that input selectivity learnt through STDP is embedded into the network connectivity through rewiring. Moreover, we show the presented model can be used to generate topographic maps between layers of neurons with minimal initial connectivity, and stabilize mappings which would otherwise be unstable through the inclusion of lateral inhibition.

Behavioral Learning in a Cognitive Neuromorphic Robot: An Integrative Approach.

We present here a learning system using the iCub humanoid robot and the SpiNNaker neuromorphic chip to solve the real-world task of object-specific attention. Integrating spiking neural networks with robots introduces considerable complexity for questionable benefit if the objective is simply task performance. But, we suggest, in a cognitive robotics context, where the goal is understanding how to compute, such an approach may yield useful insights to neural architecture as well as learned behavior, especially if dedicated neural hardware is available. Recent advances in cognitive robotics and neuromorphic processing now make such systems possible. Using a scalable, structured, modular approach, we build a spiking neural network where the effects and impact of learning can be predicted and tested, and the network can be scaled or extended to new tasks automatically. We introduce several enhancements to a basic network and show how they can be used to direct performance toward behaviorally relevant goals. Results show that using a simple classical spike-timing-dependent plasticity (STDP) rule on selected connections, we can get the robot (and network) to progress from poor task-specific performance to good performance. Behaviorally relevant STDP appears to contribute strongly to positive learning: "do this" but less to negative learning: "don't do that." In addition, we observe that the effect of structural enhancements tends to be cumulative. The overall system suggests that it is by being able to exploit combinations of effects, rather than any one effect or property in isolation, that spiking networks can achieve compelling, task-relevant behavior.

Spiking neural networks for computer vision.

State-of-the-art computer vision systems use frame-based cameras that sample the visual scene as a series of high-resolution images. These are then processed using convolutional neural networks using neurons with continuous outputs. Biological vision systems use a quite different approach, where the eyes (cameras) sample the visual scene continuously, often with a non-uniform resolution, and generate neural spike events in response to changes in the scene. The resulting spatio-temporal patterns of events are then processed through networks of spiking neurons. Such event-based processing offers advantages in terms of focusing constrained resources on the most salient features of the perceived scene, and those advantages should also accrue to engineered vision systems based upon similar principles. Event-based vision sensors, and event-based processing exemplified by the SpiNNaker (Spiking Neural Network Architecture) machine, can be used to model the biological vision pathway at various levels of detail. Here we use this approach to explore structural synaptic plasticity as a possible mechanism whereby biological vision systems may learn the statistics of their inputs without supervision, pointing the way to engineered vision systems with similar online learning capabilities.

Performance Comparison of the Digital Neuromorphic Hardware SpiNNaker and the Neural Network Simulation Software NEST for a Full-Scale Cortical Microcircuit Model.

The digital neuromorphic hardware SpiNNaker has been developed with the aim of enabling large-scale neural network simulations in real time and with low power consumption. Real-time performance is achieved with 1 ms integration time steps, and thus applies to neural networks for which faster time scales of the dynamics can be neglected. By slowing down the simulation, shorter integration time steps and hence faster time scales, which are often biologically relevant, can be incorporated. We here describe the first full-scale simulations of a cortical microcircuit with biological time scales on SpiNNaker. Since about half the synapses onto the neurons arise within the microcircuit, larger cortical circuits have only moderately more synapses per neuron. Therefore, the full-scale microcircuit paves the way for simulating cortical circuits of arbitrary size. With approximately 80, 000 neurons and 0.3 billion synapses, this model is the largest simulated on SpiNNaker to date. The scale-up is enabled by recent developments in the SpiNNaker software stack that allow simulations to be spread across multiple boards. Comparison with simulations using the NEST software on a high-performance cluster shows that both simulators can reach a similar accuracy, despite the fixed-point arithmetic of SpiNNaker, demonstrating the usability of SpiNNaker for computational neuroscience applications with biological time scales and large network size. The runtime and power consumption are also assessed for both simulators on the example of the cortical microcircuit model. To obtain an accuracy similar to that of NEST with 0.1 ms time steps, SpiNNaker requires a slowdown factor of around 20 compared to real time. The runtime for NEST saturates around 3 times real time using hybrid parallelization with MPI and multi-threading. However, achieving this runtime comes at the cost of increased power and energy consumption. The lowest total energy consumption for NEST is reached at around 144 parallel threads and 4.6 times slowdown. At this setting, NEST and SpiNNaker have a comparable energy consumption per synaptic event. Our results widen the application domain of SpiNNaker and help guide its development, showing that further optimizations such as synapse-centric network representation are necessary to enable real-time simulation of large biological neural networks.

Neuromodulated Synaptic Plasticity on the SpiNNaker Neuromorphic System.

SpiNNaker is a digital neuromorphic architecture, designed specifically for the low power simulation of large-scale spiking neural networks at speeds close to biological real-time. Unlike other neuromorphic systems, SpiNNaker allows users to develop their own neuron and synapse models as well as specify arbitrary connectivity. As a result SpiNNaker has proved to be a powerful tool for studying different neuron models as well as synaptic plasticity-believed to be one of the main mechanisms behind learning and memory in the brain. A number of Spike-Timing-Dependent-Plasticity(STDP) rules have already been implemented on SpiNNaker and have been shown to be capable of solving various learning tasks in real-time. However, while STDP is an important biological theory of learning, it is a form of Hebbian or unsupervised learning and therefore does not explain behaviors that depend on feedback from the environment. Instead, learning rules based on neuromodulated STDP (three-factor learning rules) have been shown to be capable of solving reinforcement learning tasks in a biologically plausible manner. In this paper we demonstrate for the first time how a model of three-factor STDP, with the third-factor representing spikes from dopaminergic neurons, can be implemented on the SpiNNaker neuromorphic system. Using this learning rule we first show how reward and punishment signals can be delivered to a single synapse before going on to demonstrate it in a larger network which solves the credit assignment problem in a Pavlovian conditioning experiment. Because of its extra complexity, we find that our three-factor learning rule requires approximately 2× as much processing time as the existing SpiNNaker STDP learning rules. However, we show that it is still possible to run our Pavlovian conditioning model with up to 1 × 104 neurons in real-time, opening up new research opportunities for modeling behavioral learning on SpiNNaker.