John V. Arthur

Michael V. DeBole, Rathinakumar Appuswamy, Neil McGlohon et al. · ibm-research

A vertically integrated, end-to-end, research prototype system combines 288 NorthPole neural inference accelerator cards, offline training algorithms, a high-performance runtime stack, and a containerized inference pipeline to deliver a scalable and efficient cloud inference service. The system delivers 115 peta-ops at 4-bit integer precision and 3.7 PB/s of memory bandwidth across 18 2U servers, while consuming only 30 kW of power and weighing 730 kg in a 0.67 m^2 42U rack footprint. The system can run 3 simultaneous instances of the 8-billion-parameter open-source IBM Granite-3.3-8b-instruct model at 2,048 context length with 28 simultaneous users and a per-user inter-token latency of 2.8 ms. The system is scalable, modular, and reconfigurable, supporting various model sizes and context lengths, and is ideal for deploying agentic workflows for enterprise AI applications in existing data center (cloud, on-prem) environments. For example, the system can support 18 instances of a 3-billion-parameter model or a single instance of a 70-billion-parameter model.

Andrew S. Cassidy, Guillaume Garreau, Jay Sivagnaname et al.

We show for invertible problems that transform data from a source domain (for example, Logic Condition Tables (LCTs)) to a destination domain (for example, Hardware Description Language (HDL) code), an approach of using Large Language Models (LLMs) as a lossless encoder from source to destination followed by as a lossless decoder back to the source, comparable to lossless compression in information theory, can mitigate most of the LLM drawbacks of hallucinations and omissions. Specifically, using LCTs as inputs, we generate the full HDL for a two-dimensional network-on-chip router (13 units, 1500-2000 lines of code) using seven different LLMs, reconstruct the LCTs from the auto-generated HDL, and compare the original and reconstructed LCTs. This approach yields significant productivity improvements, not only confirming correctly generated LLM logic and detecting incorrectly generated LLM logic but also assisting developers in finding design specification errors.

Jeffrey L Mckinstry, Davis R. Barch, Deepika Bablani et al.

Low precision networks in the reinforcement learning (RL) setting are relatively unexplored because of the limitations of binary activations for function approximation. Here, in the discrete action ATARI domain, we demonstrate, for the first time, that low precision policy distillation from a high precision network provides a principled, practical way to train an RL agent. As an application, on 10 different ATARI games, we demonstrate real-time end-to-end game playing on low-power neuromorphic hardware by converting a sequence of game frames into discrete actions.

Jeffrey L. McKinstry, Steven K. Esser, Rathinakumar Appuswamy et al.

To realize the promise of ubiquitous embedded deep network inference, it is essential to seek limits of energy and area efficiency. To this end, low-precision networks offer tremendous promise because both energy and area scale down quadratically with the reduction in precision. Here we demonstrate ResNet-18, -34, -50, -152, Inception-v3, Densenet-161, and VGG-16bn networks on the ImageNet classification benchmark that, at 8-bit precision exceed the accuracy of the full-precision baseline networks after one epoch of finetuning, thereby leveraging the availability of pretrained models. We also demonstrate ResNet-18, -34, -50, -152, Densenet-161, and VGG-16bn 4-bit models that match the accuracy of the full-precision baseline networks -- the highest scores to date. Surprisingly, the weights of the low-precision networks are very close (in cosine similarity) to the weights of the corresponding baseline networks, making training from scratch unnecessary. We find that gradient noise due to quantization during training increases with reduced precision, and seek ways to overcome this noise. The number of iterations required by SGD to achieve a given training error is related to the square of (a) the distance of the initial solution from the final plus (b) the maximum variance of the gradient estimates. Therefore, we (a) reduce solution distance by starting with pretrained fp32 precision baseline networks and fine-tuning, and (b) combat gradient noise introduced by quantization by training longer and reducing learning rates. Sensitivity analysis indicates that these simple techniques, coupled with proper activation function range calibration to take full advantage of the limited precision, are sufficient to discover low-precision networks, if they exist, close to fp32 precision baseline networks. The results herein provide evidence that 4-bits suffice for classification.

Rathinakumar Appuswamy, Tapan Nayak, John Arthur et al.

We derive a relationship between network representation in energy-efficient neuromorphic architectures and block Toplitz convolutional matrices. Inspired by this connection, we develop deep convolutional networks using a family of structured convolutional matrices and achieve state-of-the-art trade-off between energy efficiency and classification accuracy for well-known image recognition tasks. We also put forward a novel method to train binary convolutional networks by utilising an existing connection between noisy-rectified linear units and binary activations.

Paul Merolla, Rathinakumar Appuswamy, John Arthur et al.

Recent results show that deep neural networks achieve excellent performance even when, during training, weights are quantized and projected to a binary representation. Here, we show that this is just the tip of the iceberg: these same networks, during testing, also exhibit a remarkable robustness to distortions beyond quantization, including additive and multiplicative noise, and a class of non-linear projections where binarization is just a special case. To quantify this robustness, we show that one such network achieves 11% test error on CIFAR-10 even with 0.68 effective bits per weight. Furthermore, we find that a common training heuristic--namely, projecting quantized weights during backpropagation--can be altered (or even removed) and networks still achieve a base level of robustness during testing. Specifically, training with weight projections other than quantization also works, as does simply clipping the weights, both of which have never been reported before. We confirm our results for CIFAR-10 and ImageNet datasets. Finally, drawing from these ideas, we propose a stochastic projection rule that leads to a new state of the art network with 7.64% test error on CIFAR-10 using no data augmentation.

Steven K. Esser, Paul A. Merolla, John V. Arthur et al.

Deep networks are now able to achieve human-level performance on a broad spectrum of recognition tasks. Independently, neuromorphic computing has now demonstrated unprecedented energy-efficiency through a new chip architecture based on spiking neurons, low precision synapses, and a scalable communication network. Here, we demonstrate that neuromorphic computing, despite its novel architectural primitives, can implement deep convolution networks that i) approach state-of-the-art classification accuracy across 8 standard datasets, encompassing vision and speech, ii) perform inference while preserving the hardware's underlying energy-efficiency and high throughput, running on the aforementioned datasets at between 1200 and 2600 frames per second and using between 25 and 275 mW (effectively > 6000 frames / sec / W) and iii) can be specified and trained using backpropagation with the same ease-of-use as contemporary deep learning. For the first time, the algorithmic power of deep learning can be merged with the efficiency of neuromorphic processors, bringing the promise of embedded, intelligent, brain-inspired computing one step closer.

Srinjoy Das, Bruno Umbria Pedroni, Paul Merolla et al.

Restricted Boltzmann Machines and Deep Belief Networks have been successfully used in a wide variety of applications including image classification and speech recognition. Inference and learning in these algorithms uses a Markov Chain Monte Carlo procedure called Gibbs sampling. A sigmoidal function forms the kernel of this sampler which can be realized from the firing statistics of noisy integrate-and-fire neurons on a neuromorphic VLSI substrate. This paper demonstrates such an implementation on an array of digital spiking neurons with stochastic leak and threshold properties for inference tasks and presents some key performance metrics for such a hardware-based sampler in both the generative and discriminative contexts.