Efficient Deployment of Transformer Models in Analog In-Memory Computing Hardware

Analog in-memory computing (AIMC) has emerged as a promising solution to overcome the von Neumann bottleneck, accelerating neural network computations and improving computational efficiency. While AIMC has demonstrated success with architectures such as CNNs, MLPs, and RNNs, deploying transformer-based models using AIMC presents unique challenges. Transformers are expected to handle diverse downstream tasks and adapt to new user data or instructions after deployment, which requires more flexible approaches to suit AIMC constraints. In this paper, we propose a novel method for deploying pre-trained transformer models onto AIMC hardware. Unlike traditional approaches requiring hardware-aware training, our technique allows direct deployment without the need for retraining the original model. Instead, we utilize lightweight, low-rank adapters -- compact modules stored in digital cores -- to adapt the model to hardware constraints. We validate our approach on MobileBERT, demonstrating accuracy on par with, or even exceeding, a traditional hardware-aware training approach. Our method is particularly appealing in multi-task scenarios, as it enables a single analog model to be reused across multiple tasks. Moreover, it supports on-chip adaptation to new hardware constraints and tasks without updating analog weights, providing a flexible and versatile solution for real-world AI applications. Code is available.

Neuromorphic Wireless Split Computing with Multi-Level Spikes

Inspired by biological processes, neuromorphic computing utilizes spiking neural networks (SNNs) to perform inference tasks, offering significant efficiency gains for workloads involving sequential data. Recent advances in hardware and software have demonstrated that embedding a few bits of payload in each spike exchanged between the spiking neurons can further enhance inference accuracy. In a split computing architecture, where the SNN is divided across two separate devices, the device storing the first layers must share information about the spikes generated by the local output neurons with the other device. Consequently, the advantages of multi-level spikes must be balanced against the challenges of transmitting additional bits between the two devices. This paper addresses these challenges by investigating a wireless neuromorphic split computing architecture employing multi-level SNNs. For this system, we present the design of digital and analog modulation schemes optimized for an orthogonal frequency division multiplexing (OFDM) radio interface. Simulation and experimental results using software-defined radios provide insights into the performance gains of multi-level SNN models and the optimal payload size as a function of the quality of the connection between a transmitter and receiver.

In-Context Learned Equalization in Cell-Free Massive MIMO via State-Space Models

Sequence models have demonstrated the ability to perform tasks like channel equalization and symbol detection by automatically adapting to current channel conditions. This is done without requiring any explicit optimization and by leveraging not only short pilot sequences but also contextual information such as long-term channel statistics. The operating principle underlying automatic adaptation is in-context learning (ICL), an emerging property of sequence models. Prior art adopted transformer-based sequence models, which, however, have a computational complexity scaling quadratically with the context length due to batch processing. Recently, state-space models (SSMs) have emerged as a more efficient alternative, affording a linear inference complexity in the context size. This work explores the potential of SSMs for ICL-based equalization in cell-free massive MIMO systems. Results show that selective SSMs achieve comparable performance to transformer-based models while requiring approximately eight times fewer parameters and five times fewer floating-point operations.

Baseline Drift Tolerant Signal Encoding for ECG Classification with Deep Learning

Common artefacts such as baseline drift, rescaling, and noise critically limit the performance of machine learningbased automated ECG analysis and interpretation. This study proposes Derived Peak (DP) encoding, a non-parametric method that generates signed spikes corresponding to zero crossings of the signals first and second-order time derivatives. Notably, DP encoding is invariant to shift and scaling artefacts, and its implementation is further simplified by the absence of userdefined parameters. DP encoding was used to encode the 12-lead ECG data from the PTB-XL dataset (n=18,869 participants) and was fed to 1D-ResNet-18 models trained to identify myocardial infarction, conductive deficits and ST-segment abnormalities. Robustness to artefacts was assessed by corrupting ECG data with sinusoidal baseline drift, shift, rescaling and noise, before encoding. The addition of these artefacts resulted in a significant drop in accuracy for seven other methods from prior art, while DP encoding maintained a baseline AUC of 0.88 under drift, shift and rescaling. DP achieved superior performance to unencoded inputs in the presence of shift (AUC under 1mV shift: 0.91 vs 0.62), and rescaling artefacts (AUC 0.91 vs 0.79). Thus, DP encoding is a simple method by which robustness to common ECG artefacts may be improved for automated ECG analysis and interpretation.

Neuromorphic In-Context Learning for Energy-Efficient MIMO Symbol Detection

In-context learning (ICL), a property demonstrated by transformer-based sequence models, refers to the automatic inference of an input-output mapping based on examples of the mapping provided as context. ICL requires no explicit learning, i.e., no explicit updates of model weights, directly mapping context and new input to the new output. Prior work has proved the usefulness of ICL for detection in MIMO channels. In this setting, the context is given by pilot symbols, and ICL automatically adapts a detector, or equalizer, to apply to newly received signals. However, the implementation tested in prior art was based on conventional artificial neural networks (ANNs), which may prove too energy-demanding to be run on mobile devices. This paper evaluates a neuromorphic implementation of the transformer for ICL-based MIMO detection. This approach replaces ANNs with spiking neural networks (SNNs), and implements the attention mechanism via stochastic computing, requiring no multiplications, but only logical AND operations and counting. When using conventional digital CMOS hardware, the proposed implementation is shown to preserve accuracy, with a reduction in power consumption ranging from $5.4\times$ to $26.8\times$, depending on the model sizes, as compared to ANN-based implementations.

Stochastic Spiking Attention: Accelerating Attention with Stochastic Computing in Spiking Networks

Spiking Neural Networks (SNNs) have been recently integrated into Transformer architectures due to their potential to reduce computational demands and to improve power efficiency. Yet, the implementation of the attention mechanism using spiking signals on general-purpose computing platforms remains inefficient. In this paper, we propose a novel framework leveraging stochastic computing (SC) to effectively execute the dot-product attention for SNN-based Transformers. We demonstrate that our approach can achieve high classification accuracy ($83.53\%$) on CIFAR-10 within 10 time steps, which is comparable to the performance of a baseline artificial neural network implementation ($83.66\%$). We estimate that the proposed SC approach can lead to over $6.3\times$ reduction in computing energy and $1.7\times$ reduction in memory access costs for a digital CMOS-based ASIC design. We experimentally validate our stochastic attention block design through an FPGA implementation, which is shown to achieve $48\times$ lower latency as compared to a GPU implementation, while consuming $15\times$ less power.

Bayesian Inference Accelerator for Spiking Neural Networks

Bayesian neural networks offer better estimates of model uncertainty compared to frequentist networks. However, inference involving Bayesian models requires multiple instantiations or sampling of the network parameters, requiring significant computational resources. Compared to traditional deep learning networks, spiking neural networks (SNNs) have the potential to reduce computational area and power, thanks to their event-driven and spike-based computational framework. Most works in literature either address frequentist SNN models or non-spiking Bayesian neural networks. In this work, we demonstrate an optimization framework for developing and implementing efficient Bayesian SNNs in hardware by additionally restricting network weights to be binary-valued to further decrease power and area consumption. We demonstrate accuracies comparable to Bayesian binary networks with full-precision Bernoulli parameters, while requiring up to $25\times$ less spikes than equivalent binary SNN implementations. We show the feasibility of the design by mapping it onto Zynq-7000, a lightweight SoC, and achieve a $6.5 \times$ improvement in GOPS/DSP while utilizing up to 30 times less power compared to the state-of-the-art.

Performance Evaluation of Neuromorphic Hardware for Onboard Satellite Communication Applications

Spiking neural networks (SNNs) implemented on neuromorphic processors (NPs) can enhance the energy efficiency of deployments of artificial intelligence (AI) for specific workloads. As such, NP represents an interesting opportunity for implementing AI tasks on board power-limited satellite communication spacecraft. In this article, we disseminate the findings of a recently completed study which targeted the comparison in terms of performance and power-consumption of different satellite communication use cases implemented on standard AI accelerators and on NPs. In particular, the article describes three prominent use cases, namely payload resource optimization, onboard interference detection and classification, and dynamic receive beamforming; and compare the performance of conventional convolutional neural networks (CNNs) implemented on Xilinx's VCK5000 Versal development card and SNNs on Intel's neuromorphic chip Loihi 2.

Noise Adaptor in Spiking Neural Networks

Recent strides in low-latency spiking neural network (SNN) algorithms have drawn significant interest, particularly due to their event-driven computing nature and fast inference capability. One of the most efficient ways to construct a low-latency SNN is by converting a pre-trained, low-bit artificial neural network (ANN) into an SNN. However, this conversion process faces two main challenges: First, converting SNNs from low-bit ANNs can lead to ``occasional noise" -- the phenomenon where occasional spikes are generated in spiking neurons where they should not be -- during inference, which significantly lowers SNN accuracy. Second, although low-latency SNNs initially show fast improvements in accuracy with time steps, these accuracy growths soon plateau, resulting in their peak accuracy lagging behind both full-precision ANNs and traditional ``long-latency SNNs'' that prioritize precision over speed. In response to these two challenges, this paper introduces a novel technique named ``noise adaptor.'' Noise adaptor can model occasional noise during training and implicitly optimize SNN accuracy, particularly at high simulation times $T$. Our research utilizes the ResNet model for a comprehensive analysis of the impact of the noise adaptor on low-latency SNNs. The results demonstrate that our method outperforms the previously reported quant-ANN-to-SNN conversion technique. We achieved an accuracy of 95.95\% within 4 time steps on CIFAR-10 using ResNet-18, and an accuracy of 74.37\% within 64 time steps on ImageNet using ResNet-50. Remarkably, these results were obtained without resorting to any noise correction methods during SNN inference, such as negative spikes or two-stage SNN simulations. Our approach significantly boosts the peak accuracy of low-latency SNNs, bringing them on par with the accuracy of full-precision ANNs. Code will be open source.

Towards Efficient and Trustworthy AI Through Hardware-Algorithm-Communication Co-Design

Artificial intelligence (AI) algorithms based on neural networks have been designed for decades with the goal of maximising some measure of accuracy. This has led to two undesired effects. First, model complexity has risen exponentially when measured in terms of computation and memory requirements. Second, state-of-the-art AI models are largely incapable of providing trustworthy measures of their uncertainty, possibly `hallucinating' their answers and discouraging their adoption for decision-making in sensitive applications. With the goal of realising efficient and trustworthy AI, in this paper we highlight research directions at the intersection of hardware and software design that integrate physical insights into computational substrates, neuroscientific principles concerning efficient information processing, information-theoretic results on optimal uncertainty quantification, and communication-theoretic guidelines for distributed processing. Overall, the paper advocates for novel design methodologies that target not only accuracy but also uncertainty quantification, while leveraging emerging computing hardware architectures that move beyond the traditional von Neumann digital computing paradigm to embrace in-memory, neuromorphic, and quantum computing technologies. An important overarching principle of the proposed approach is to view the stochasticity inherent in the computational substrate and in the communication channels between processors as a resource to be leveraged for the purpose of representing and processing classical and quantum uncertainty.