Alex Yakovlev

Bimal Bhattarai, Ole-Christoffer Granmo, Lei Jiao et al.

In this paper, we introduce a sparse Tsetlin Machine (TM) with absorbing Tsetlin Automata (TA) states. In brief, the TA of each clause literal has both an absorbing Exclude- and an absorbing Include state, making the learning scheme absorbing instead of ergodic. When a TA reaches an absorbing state, it will never leave that state again. If the absorbing state is an Exclude state, both the automaton and the literal can be removed from further consideration. The literal will as a result never participates in that clause. If the absorbing state is an Include state, on the other hand, the literal is stored as a permanent part of the clause while the TA is discarded. A novel sparse data structure supports these updates by means of three action lists: Absorbed Include, Include, and Exclude. By updating these lists, the TM gets smaller and smaller as the literals and their TA withdraw. In this manner, the computation accelerates during learning, leading to faster learning and less energy consumption.

Samuel Prescott, Adrian Wheeldon, Rishad Shafik et al.

There is a need for machine learning models to evolve in unsupervised circumstances. New classifications may be introduced, unexpected faults may occur, or the initial dataset may be small compared to the data-points presented to the system during normal operation. Implementing such a system using neural networks involves significant mathematical complexity, which is a major issue in power-critical edge applications. This paper proposes a novel field-programmable gate-array infrastructure for online learning, implementing a low-complexity machine learning algorithm called the Tsetlin Machine. This infrastructure features a custom-designed architecture for run-time learning management, providing on-chip offline and online learning. Using this architecture, training can be carried out on-demand on the \ac{FPGA} with pre-classified data before inference takes place. Additionally, our architecture provisions online learning, where training can be interleaved with inference during operation. Tsetlin Machine (TM) training naturally descends to an optimum, with training also linked to a threshold hyper-parameter which is used to reduce the probability of issuing feedback as the TM becomes trained further. The proposed architecture is modular, allowing the data input source to be easily changed, whilst inbuilt cross-validation infrastructure allows for reliable and representative results during system testing. We present use cases for online learning using the proposed infrastructure and demonstrate the energy/performance/accuracy trade-offs.

Omar Ghazal, Tian Lan, Shalman Ojukwu et al.

The modern implementation of machine learning architectures faces significant challenges due to frequent data transfer between memory and processing units. In-memory computing, primarily through memristor-based analog computing, offers a promising solution to overcome this von Neumann bottleneck. In this technology, data processing and storage are located inside the memory. Here, we introduce a novel approach that utilizes floating-gate Y-Flash memristive devices manufactured with a standard 180 nm CMOS process. These devices offer attractive features, including analog tunability and moderate device-to-device variation; such characteristics are essential for reliable decision-making in ML applications. This paper uses a new machine learning algorithm, the Tsetlin Machine (TM), for in-memory processing architecture. The TM's learning element, Automaton, is mapped into a single Y-Flash cell, where the Automaton's range is transferred into the Y-Flash's conductance scope. Through comprehensive simulations, the proposed hardware implementation of the learning automata, particularly for Tsetlin machines, has demonstrated enhanced scalability and on-edge learning capabilities.

Tian Lan, Rishad Shafik, Alex Yakovlev

Machine learning fits model parameters to approximate input-output mappings, predicting unknown samples. However, these models often require extensive arithmetic computations during inference, increasing latency and power consumption. This paper proposes a digital-time-domain computing approach for Tsetlin machine (TM) inference process to address these challenges. This approach leverages a delay accumulation mechanism to mitigate the costly arithmetic sums of classes and employs a Winner-Takes-All scheme to replace conventional magnitude comparators. Specifically, a Hamming distance-driven time-domain scheme is implemented for multi-class TMs. Furthermore, differential delay paths, combined with a leading-ones-detector logarithmic delay compression digital-time-domain scheme, are utilised for the coalesced TMs, accommodating both binary-signed and exponential-scale delay accumulation issues. Compared to the functionally equivalent, post-implementation digital TM architecture baseline, the proposed architecture demonstrates orders-of-magnitude improvements in energy efficiency and throughput.

Shengyu Duan, Marcos L. L. Sartori, Rishad Shafik et al.

Increasing demands for adaptability, privacy, and security at the edge have persistently pushed the frontiers for a new generation of machine learning (ML) algorithms with training and inference capabilities on-chip. Weightless Neural Network (WNN) is such an algorithm that is principled on lookup table based simple neuron structures. As a result, it offers architectural benefits, such as low-latency, low-complexity inference, compared to deep neural networks that depend heavily on multiply-accumulate operations. However, traditional WNNs rely on memorization-based one-shot training, which either leads to overfitting and reduced accuracy or requires tedious post-training adjustments, limiting their effectiveness for efficient on chip training. In this work, we propose TsetlinWiSARD, a training approach for WNNs that leverages Tsetlin Automata (TAs) to enable probabilistic, feedback-driven learning. It overcomes the overfitting of WiSARD's one-shot training with iterative optimization, while maintaining simple, continuous binary feedback for efficient on-chip training. Central to our approach is a field programmable gate array (FPGA)-based training architecture that delivers state-of-the-art accuracy while significantly improving hardware efficiency. Our approach provides over 1000x faster training when compared with the traditional WiSARD implementation of WNNs. Further, we demonstrate 22% reduced resource usage, 93.3% lower latency, and 64.2% lower power consumption compared to FPGA-based training accelerators implementing other ML algorithms.

Mohamed Tarraf, Alex Chan, Alex Yakovlev et al.

Binary Neural Networks (BNNs) offer a low-complexity and energy-efficient alternative to traditional full-precision neural networks by constraining their weights and activations to binary values. However, their discrete, highly non-linear behavior makes them difficult to explain, validate and formally verify. As a result, BNNs remain largely opaque, limiting their suitability in safety-critical domains, where causal transparency and behavioral guarantees are essential. In this work, we introduce a Petri net (PN)-based framework that captures the BNN's internal operations as event-driven processes. By "eventizing" their operations, we expose their causal relationships and dependencies for a fine-grained analysis of concurrency, ordering, and state evolution. Here, we construct modular PN blueprints for core BNN components including activation, gradient computation and weight updates, and compose them into a complete system-level model. We then validate the composed PN against a reference software-based BNN, verify it against reachability and structural checks to establish 1-safeness, deadlock-freeness, mutual exclusion and correct-by-construction causal sequencing, before we assess its scalability and complexity at segment, component, and system levels using the automated measurement tools in Workcraft. Overall, this framework enables causal introspection of transparent and event-driven BNNs that are amenable to formal reasoning and verification.

Tousif Rahman, Gang Mao, Sidharth Maheshwari et al.

System-on-Chip Field-Programmable Gate Arrays (SoC-FPGAs) offer significant throughput gains for machine learning (ML) edge inference applications via the design of co-processor accelerator systems. However, the design effort for training and translating ML models into SoC-FPGA solutions can be substantial and requires specialist knowledge aware trade-offs between model performance, power consumption, latency and resource utilization. Contrary to other ML algorithms, Tsetlin Machine (TM) performs classification by forming logic proposition between boolean actions from the Tsetlin Automata (the learning elements) and boolean input features. A trained TM model, usually, exhibits high sparsity and considerable overlapping of these logic propositions both within and among the classes. The model, thus, can be translated to RTL-level design using a miniscule number of AND and NOT gates. This paper presents MATADOR, an automated boolean-to-silicon tool with GUI interface capable of implementing optimized accelerator design of the TM model onto SoC-FPGA for inference at the edge. It offers automation of the full development pipeline: model training, system level design generation, design verification and deployment. It makes use of the logic sharing that ensues from propositional overlap and creates a compact design by effectively utilizing the TM model's sparsity. MATADOR accelerator designs are shown to be up to 13.4x faster, up to 7x more resource frugal and up to 2x more power efficient when compared to the state-of-the-art Quantized and Binary Deep Neural Network implementations.

Gang Mao, Tousif Rahman, Sidharth Maheshwari et al.

The increased demand for data privacy and security in machine learning (ML) applications has put impetus on effective edge training on Internet-of-Things (IoT) nodes. Edge training aims to leverage speed, energy efficiency and adaptability within the resource constraints of the nodes. Deploying and training Deep Neural Networks (DNNs)-based models at the edge, although accurate, posit significant challenges from the back-propagation algorithm's complexity, bit precision trade-offs, and heterogeneity of DNN layers. This paper presents a Dynamic Tsetlin Machine (DTM) training accelerator as an alternative to DNN implementations. DTM utilizes logic-based on-chip inference with finite-state automata-driven learning within the same Field Programmable Gate Array (FPGA) package. Underpinned on the Vanilla and Coalesced Tsetlin Machine algorithms, the dynamic aspect of the accelerator design allows for a run-time reconfiguration targeting different datasets, model architectures, and model sizes without resynthesis. This makes the DTM suitable for targeting multivariate sensor-based edge tasks. Compared to DNNs, DTM trains with fewer multiply-accumulates, devoid of derivative computation. It is a data-centric ML algorithm that learns by aligning Tsetlin automata with input data to form logical propositions enabling efficient Look-up-Table (LUT) mapping and frugal Block RAM usage in FPGA training implementations. The proposed accelerator offers 2.54x more Giga operations per second per Watt (GOP/s per W) and uses 6x less power than the next-best comparable design.

Runar Helin, Ole-Christoffer Granmo, Mayur Kishor Shende et al.

Data modeling using Tsetlin machines (TMs) is all about building logical rules from the data features. The decisions of the model are based on a combination of these logical rules. Hence, the model is fully transparent and it is possible to get explanations of its predictions. In this paper, we present a probability score for TM predictions and develop new techniques for uncertainty quantification to increase the explainability further. The probability score is an inherent property of any TM variant and is derived through an analysis of the TM learning dynamics. Simulated data is used to show a clear connection between the learned TM probability scores and the underlying probabilities of the data. A visualization of the probability scores also reveals that the TM is less confident in its predictions outside the training data domain, which contrasts the typical extrapolation phenomenon found in Artificial Neural Networks. The paper concludes with an application of the uncertainty quantification techniques on an image classification task using the CIFAR-10 dataset, where they provide new insights and suggest possible improvements to current TM image classification models.

Tousif Rahman, Gang Mao, Bob Pattison et al.

Embedded Field-Programmable Gate Arrays (eFPGAs) allow for the design of hardware accelerators of edge Machine Learning (ML) applications at a lower power budget compared with traditional FPGA platforms. However, the limited eFPGA logic and memory significantly constrain compute capabilities and model size. As such, ML application deployment on eFPGAs is in direct contrast with the most recent FPGA approaches developing architecture-specific implementations and maximizing throughput over resource frugality. This paper focuses on the opposite side of this trade-off: the proposed eFPGA accelerator focuses on minimizing resource usage and allowing flexibility for on-field recalibration over throughput. This allows for runtime changes in model size, architecture, and input data dimensionality without offline resynthesis. This is made possible through the use of a bitwise compressed inference architecture of the Tsetlin Machine (TM) algorithm. TM compute does not require any multiplication operations, being limited to only bitwise AND, OR, NOT, summations and additions. Additionally, TM model compression allows the entire model to fit within the on-chip block RAM of the eFPGA. The paper uses this accelerator to propose a strategy for runtime model tuning in the field. The proposed approach uses 2.5x fewer Look-up-Tables (LUTs) and 3.38x fewer registers than the current most resource-fugal design and achieves up to 129x energy reduction compared with low-power microcontrollers running the same ML application.

Shengyu Duan, Rishad Shafik, Alex Yakovlev

The Tsetlin Machine (TM) is a novel alternative to deep neural networks (DNNs). Unlike DNNs, which rely on multi-path arithmetic operations, a TM learns propositional logic patterns from data literals using Tsetlin automata. This fundamental shift from arithmetic to logic underpinning makes TM suitable for empowering new applications with low-cost implementations. In TM, literals are often included by both positive and negative clauses within the same class, canceling out their impact on individual class definitions. This property can be exploited to develop compressed TM models, enabling energy-efficient and high-throughput inferences for machine learning (ML) applications. We introduce a training approach that incorporates excluded automata states to sparsify TM logic patterns in both positive and negative clauses. This exclusion is iterative, ensuring that highly class-correlated (and therefore significant) literals are retained in the compressed inference model, ETHEREAL, to maintain strong classification accuracy. Compared to standard TMs, ETHEREAL TM models can reduce model size by up to 87.54%, with only a minor accuracy compromise. We validate the impact of this compression on eight real-world Tiny machine learning (TinyML) datasets against standard TM, equivalent Random Forest (RF) and Binarized Neural Network (BNN) on the STM32F746G-DISCO platform. Our results show that ETHEREAL TM models achieve over an order of magnitude reduction in inference time (resulting in higher throughput) and energy consumption compared to BNNs, while maintaining a significantly smaller memory footprint compared to RFs.

Svein Anders Tunheim, Yujin Zheng, Lei Jiao et al.

We present an all-digital programmable machine learning accelerator chip for image classification, underpinning on the Tsetlin machine (TM) principles. The TM is an emerging machine learning algorithm founded on propositional logic, utilizing sub-pattern recognition expressions called clauses. The accelerator implements the coalesced TM version with convolution, and classifies booleanized images of 28$\times$28 pixels with 10 categories. A configuration with 128 clauses is used in a highly parallel architecture. Fast clause evaluation is achieved by keeping all clause weights and Tsetlin automata (TA) action signals in registers. The chip is implemented in a 65 nm low-leakage CMOS technology, and occupies an active area of 2.7 mm$^2$. At a clock frequency of 27.8 MHz, the accelerator achieves 60.3k classifications per second, and consumes 8.6 nJ per classification. This demonstrates the energy-efficiency of the TM, which was the main motivation for developing this chip. The latency for classifying a single image is 25.4 $μ$s which includes system timing overhead. The accelerator achieves 97.42%, 84.54% and 82.55% test accuracies for the datasets MNIST, Fashion-MNIST and Kuzushiji-MNIST, respectively, matching the TM software models.

Yefan Zeng, Shengyu Duan, Rishad Shafik et al.

The Tsetlin Machine (TM) offers high-speed inference on resource-constrained devices such as CPUs. Its logic-driven operations naturally lend themselves to parallel execution on modern CPU architectures. Motivated by this, we propose an efficient software implementation of the TM by leveraging instruction-level bitwise operations for compact model representation and accelerated processing. To further improve inference speed, we introduce an early exit mechanism, which exploits the TM's AND-based clause evaluation to avoid unnecessary computations. Building upon this, we propose a literal Reorder strategy designed to maximize the likelihood of early exits. This strategy is applied during a post-training, pre-inference stage through statistical analysis of all literals and the corresponding actions of their associated Tsetlin Automata (TA), introducing negligible runtime overhead. Experimental results using the gem5 simulator with an ARM processor show that our optimized implementation reduces inference time by up to 96.71% compared to the conventional integer-based TM implementations while maintaining comparable code density.

Shengyu Duan, Marcos L. L. Sartori, Rishad Shafik et al.

Population count (popcount) is a crucial operation for many low-complexity machine learning (ML) algorithms, including Tsetlin Machine (TM)-a promising new ML method, particularly well-suited for solving classification tasks. The inference mechanism in TM consists of propositional logic-based structures within each class, followed by a majority voting scheme, which makes the classification decision. In TM, the voters are the outputs of Boolean clauses. The voting mechanism comprises two operations: popcount for each class and determining the class with the maximum vote by means of an argmax operation. While TMs offer a lightweight ML alternative, their performance is often limited by the high computational cost of popcount and comparison required to produce the argmax result. In this paper, we propose an innovative approach to accelerate and optimize these operations by performing them in the time domain. Our time-domain implementation uses programmable delay lines (PDLs) and arbiters to efficiently manage these tasks through delay-based mechanisms. We also present an FPGA design flow for practical implementation of the time-domain popcount, addressing delay skew and ensuring that the behavior matches that of the model's intended functionality. By leveraging the natural compatibility of the proposed popcount with asynchronous architectures, we demonstrate significant improvements in an asynchronous TM, including up to 38% reduction in latency, 43.1% reduction in dynamic power, and 15% savings in resource utilization, compared to synchronous TMs using adder-based popcount.

Omar Ghazal, Wei Wang, Shahar Kvatinsky et al.

The increasing demand for processing large volumes of data for machine learning models has pushed data bandwidth requirements beyond the capability of traditional von Neumann architecture. In-memory computing (IMC) has recently emerged as a promising solution to address this gap by enabling distributed data storage and processing at the micro-architectural level, significantly reducing both latency and energy. In this paper, we present the IMPACT: InMemory ComPuting Architecture Based on Y-FlAsh Technology for Coalesced Tsetlin Machine Inference, underpinned on a cutting-edge memory device, Y-Flash, fabricated on a 180 nm CMOS process. Y-Flash devices have recently been demonstrated for digital and analog memory applications, offering high yield, non-volatility, and low power consumption. The IMPACT leverages the Y-Flash array to implement the inference of a novel machine learning algorithm: coalesced Tsetlin machine (CoTM) based on propositional logic. CoTM utilizes Tsetlin automata (TA) to create Boolean feature selections stochastically across parallel clauses. The IMPACT is organized into two computational crossbars for storing the TA and weights. Through validation on the MNIST dataset, IMPACT achieved 96.3% accuracy. The IMPACT demonstrated improvements in energy efficiency, e.g., 2.23X over CNN-based ReRAM, 2.46X over Neuromorphic using NOR-Flash, and 2.06X over DNN-based PCM, suited for modern ML inference applications.

Omar Ghazal, Simranjeet Singh, Tousif Rahman et al.

In-memory computing for Machine Learning (ML) applications remedies the von Neumann bottlenecks by organizing computation to exploit parallelism and locality. Non-volatile memory devices such as Resistive RAM (ReRAM) offer integrated switching and storage capabilities showing promising performance for ML applications. However, ReRAM devices have design challenges, such as non-linear digital-analog conversion and circuit overheads. This paper proposes an In-Memory Boolean-to-Current Inference Architecture (IMBUE) that uses ReRAM-transistor cells to eliminate the need for such conversions. IMBUE processes Boolean feature inputs expressed as digital voltages and generates parallel current paths based on resistive memory states. The proportional column current is then translated back to the Boolean domain for further digital processing. The IMBUE architecture is inspired by the Tsetlin Machine (TM), an emerging ML algorithm based on intrinsically Boolean logic. The IMBUE architecture demonstrates significant performance improvements over binarized convolutional neural networks and digital TM in-memory implementations, achieving up to a 12.99x and 5.28x increase, respectively.

Rishad Shafik, Tousif Rahman, Adrian Wheeldon et al.

Energy efficiency is a crucial requirement for enabling powerful artificial intelligence applications at the microedge. Hardware acceleration with frugal architectural allocation is an effective method for reducing energy. Many emerging applications also require the systems design to incorporate interpretable decision models to establish responsibility and transparency. The design needs to provision for additional resources to provide reachable states in real-world data scenarios, defining conflicting design tradeoffs between energy efficiency. is challenging. Recently a new machine learning algorithm, called the Tsetlin machine, has been proposed. The algorithm is fundamentally based on the principles of finite-state automata and benefits from natural logic underpinning rather than arithmetic. In this paper, we investigate methods of energy-frugal artificial intelligence hardware design by suitably tuning the hyperparameters, while maintaining high learning efficacy. To demonstrate interpretability, we use reachability and game-theoretic analysis in two simulation environments: a SystemC model to study the bounded state transitions in the presence of hardware faults and Nash equilibrium between states to analyze the learning convergence. Our analyses provides the first insights into conflicting design tradeoffs involved in energy-efficient and interpretable decision models for this new artificial intelligence hardware architecture. We show that frugal resource allocation coupled with systematic prodigality between randomized reinforcements can provide decisive energy reduction while also achieving robust and interpretable learning.

Adrian Wheeldon, Alex Yakovlev, Rishad Shafik

We present a hardware design for the learning datapath of the Tsetlin machine algorithm, along with a latency analysis of the inference datapath. In order to generate a low energy hardware which is suitable for pervasive artificial intelligence applications, we use a mixture of asynchronous design techniques - including Petri nets, signal transition graphs, dual-rail and bundled-data. The work builds on previous design of the inference hardware, and includes an in-depth breakdown of the automaton feedback, probability generation and Tsetlin automata. Results illustrate the advantages of asynchronous design in applications such as personalized healthcare and battery-powered internet of things devices, where energy is limited and latency is an important figure of merit. Challenges of static timing analysis in asynchronous circuits are also addressed.

Dainius Jenkus, Fei Xia, Rishad Shafik et al.

Web servers scaled across distributed systems necessitate complex runtime controls for providing quality of service (QoS) guarantees as well as minimizing the energy costs under dynamic workloads. This paper presents a QoS-aware runtime controller using horizontal scaling (node allocation) and vertical scaling (resource allocation within nodes) methods synergistically to provide adaptation to workloads while minimizing the power consumption under QoS constraint (i.e., response time). A horizontal scaling determines the number of active nodes based on workload demands and the required QoS according to a set of rules. Then, it is coupled with vertical scaling using transfer Q-learning, which further tunes power/performance based on workload profile using dynamic voltage/frequency scaling (DVFS). It transfers Q-values within minimally explored states reducing exploration requirements. In addition, the approach exploits a scalable architecture of the many-core server allowing to reuse available knowledge from fully or partially explored nodes. When combined, these methods allow to reduce the exploration time and QoS violations when compared to model-free Q-learning. The technique balances design-time and runtime costs to maximize the portability and operational optimality demonstrated through persistent power reductions with minimal QoS violations under different workload scenarios on heterogeneous multi-processing nodes of a server cluster.

Jie Lei, Tousif Rahman, Rishad Shafik et al.

The emergence of Artificial Intelligence (AI) driven Keyword Spotting (KWS) technologies has revolutionized human to machine interaction. Yet, the challenge of end-to-end energy efficiency, memory footprint and system complexity of current Neural Network (NN) powered AI-KWS pipelines has remained ever present. This paper evaluates KWS utilizing a learning automata powered machine learning algorithm called the Tsetlin Machine (TM). Through significant reduction in parameter requirements and choosing logic over arithmetic based processing, the TM offers new opportunities for low-power KWS while maintaining high learning efficacy. In this paper we explore a TM based keyword spotting (KWS) pipeline to demonstrate low complexity with faster rate of convergence compared to NNs. Further, we investigate the scalability with increasing keywords and explore the potential for enabling low-power on-chip KWS.

Adrian Wheeldon, Alex Yakovlev, Rishad Shafik et al.

Modern internet of things (IoT) devices leverage machine learning inference using sensed data on-device rather than offloading them to the cloud. Commonly known as inference at-the-edge, this gives many benefits to the users, including personalization and security. However, such applications demand high energy efficiency and robustness. In this paper we propose a method for reduced area and power overhead of self-timed early-propagative asynchronous inference circuits, designed using the principles of learning automata. Due to natural resilience to timing as well as logic underpinning, the circuits are tolerant to variations in environment and supply voltage whilst enabling the lowest possible latency. Our method is exemplified through an inference datapath for a low power machine learning application. The circuit builds on the Tsetlin machine algorithm further enhancing its energy efficiency. Average latency of the proposed circuit is reduced by 10x compared with the synchronous implementation whilst maintaining similar area. Robustness of the proposed circuit is proven through post-synthesis simulation with 0.25 V to 1.2 V supply. Functional correctness is maintained and latency scales with gate delay as voltage is decreased.

K. Darshana Abeyrathna, Ole-Christoffer Granmo, Rishad Shafik et al.

Due to the high energy consumption and scalability challenges of deep learning, there is a critical need to shift research focus towards dealing with energy consumption constraints. Tsetlin Machines (TMs) are a recent approach to machine learning that has demonstrated significantly reduced energy usage compared to neural networks alike, while performing competitively accuracy-wise on several benchmarks. However, TMs rely heavily on energy-costly random number generation to stochastically guide a team of Tsetlin Automata to a Nash Equilibrium of the TM game. In this paper, we propose a novel finite-state learning automaton that can replace the Tsetlin Automata in TM learning, for increased determinism. The new automaton uses multi-step deterministic state jumps to reinforce sub-patterns. Simultaneously, flipping a coin to skip every $d$'th state update ensures diversification by randomization. The $d$-parameter thus allows the degree of randomization to be finely controlled. E.g., $d=1$ makes every update random and $d=\infty$ makes the automaton completely deterministic. Our empirical results show that, overall, only substantial degrees of determinism reduces accuracy. Energy-wise, random number generation constitutes switching energy consumption of the TM, saving up to 11 mW power for larger datasets with high $d$ values. We can thus use the new $d$-parameter to trade off accuracy against energy consumption, to facilitate low-energy machine learning.

Sergey Mileiko, Thanasin Bunnam, Fei Xia et al.

Neural Networks (NNs) are steering a new generation of artificial intelligence (AI) applications at the micro-edge. Examples include wireless sensors, wearables and cybernetic systems that collect data and process them to support real-world decisions and controls. For energy autonomy, these applications are typically powered by energy harvesters. As harvesters and other power sources which provide energy autonomy inevitably have power variations, the circuits need to robustly operate over a dynamic power envelope. In other words, the NN hardware needs to be able to function correctly under unpredictable and variable supply voltages. In this paper, we propose a novel NN design approach using the principle of pulse width modulation (PWM). PWM signals represent information with their duty cycle values which may be made independent of the voltages and frequencies of the carrier signals. We design a PWM-based perceptron which can serve as the fundamental building block for NNs, by using an entirely new method of realising arithmetic in the PWM domain. We analyse the proposed approach building from a 3x3 perceptron circuit to a complex multi-layer NN. Using handwritten character recognition as an exemplar of AI applications, we demonstrate the power elasticity, resilience and efficiency of the proposed NN design in the presence of functional and parametric variations including large voltage variations in the power supply.