We propose a novel approach to build photonic physical unclonable functions (PPUFs) by leveraging Mach-Zehnder interferometer (MZI)-based computing mesh architectures. Statistical properties and robustness to machine-learning attacks show promising results for strong PPUFs.

Photonic Neural Networks (PNNs) still depend on electronic components for tasks such as memory storage and control, as a result, they are susceptible to faults that can compromise model robustness. In this study, we present the first robustness analysis of PNNs with respect to memory faults in their stored parameters. To mitigate these issues, we propose a hardening solution with zero memory overhead based on Triple Modular Redundancy (TMR) that leads to increased accuracy under faulty conditions.

Physically unclonable functions (PUFs) have emerged as a promising hardware security primitive, and recent years have seen growing interest in realizing PUFs on photonic integrated circuits (PICs). We classify existing PIC-based PUFs into electrically configured and optically encoded designs. Notably, security properties remain insufficiently analyzed, with mathematical unclonability often assumed without supporting argument. To illustrate this issue, we introduce a modeling attack against the subclass of optically encoded PUFs that operate in a linear optical regime. We show that, for a representative simulated PUF construction, observing as few as 200 challenge-response pairs allows learning the challenge-response mapping. Under canonical modeling assumptions, linear optically encoded PUFs are therefore mathematically clonable, calling for a careful assessment of their suitability for authentication protocols that rely on mathematical unclonability.

Reservoir computing offers a versatile approach to temporal information processing by exploiting the dynamics of nonlinear systems to enable learning with a linear readout. Photonic hardware is particularly attractive for this paradigm, thanks to ultrafast dynamics and massive parallelism across multiple accessible degrees of freedom of light. In this thematic review, we chart the development of photonic reservoir computing across hardware platforms, from delay-based architectures and integrated photonic circuit implementations to free-space systems. We also discuss how the input and output layers influence the performance of photonic reservoirs. We present the explored learning algorithms that go beyond linear regression, and list some popular benchmark tasks and shed light on some key applications that have been investigated in the literature. Finally, we discuss emerging unconventional implementations. The overall goal of this review is to provide a panoramic overview of photonic reservoir computing, useful for both practitioners and newcomers to the field.

Multilevel phase shifters are key components in photonic integrated circuits. A major requirement in many applications is achieving non-volatile operation and low insertion loss simultaneously. Electrically, multiple phase levels can be encoded by controlling the heater power and employing different microheater architectures to induce varying degrees of phase-change material (PCM) amorphization, thereby modulating the device's optical properties and the amplitude and phase of the propagating field. Here, we explore a platform based on the integration of a low-loss PCM, namely GeSe, sandwiched between microheaters and silicon waveguides. To achieve a large number of levels, we modify standard straight microheaters and propose a segmented heater design that breaks the heater's symmetry along the propagation direction. We numerically demonstrate under pulse-width/pulse-amplitude modulation (PWM/PAM) that multilevel phase shifts can be achieved due to non-uniform heating in the GeSe PCM layer. However, the resulting phase levels for the basic configuration are highly abrupt because the constant power dissipation along the light propagation direction, associated with a uniform cross-section, does not permit a wide range of amorphization patterns. The proposed segmented heater, whose width gradually increases in steps along the light propagation direction, allows overcoming this limitation. This configuration enables the encoding of 164 well-spaced phase levels between 0 and π (>7-bit resolution), facilitated by smoother amorphization arising from the combined effects of non-uniform heating across segments and within each segment, while maintaining an insertion loss of only 0.6 dB in the fully crystalline state (worst case).

Matrix-vector multiplications (MVMs) are essential for a wide range of applications, particularly in modern machine learning and quantum computing. In photonics, there is growing interest in developing architectures capable of performing linear operations with high speed, low latency, and minimal loss. Traditional interferometric photonic architectures, such as the Clements design, have been extensively used for MVM operations. However, as these architectures scale, improving stability and robustness becomes critical. In this paper, we introduce a novel photonic braid interferometer architecture that outperforms both the Clements and Fldzhyan designs in these aspects. Using numerical simulations, we evaluate the performance of these architectures under ideal conditions and systematically introduce non-idealities such as insertion losses, beam splitter imbalances, and crosstalk. The results demonstrate that the braid architecture offers superior robustness due to its symmetrical design and reduced layer count. Further analysis shows that the braid architecture is particularly advantageous in large-scale implementations, delivering better performance as the size of the interferometer increases. We also assess the footprint and total insertion losses of each architecture. Although waveguide crossings in the braid architecture slightly increase the footprint and insertion loss, recent advances in crossing technology significantly minimize these effects. Our study suggests that the braid architecture is a robust solution for photonic neuromorphic computing, maintaining high fidelity in realistic conditions where imperfections are inevitable.

We propose a novel hybrid mode interferometer (HMI) leveraging the interference of hybridized TE–TM modes in a silicon-on-insulator (SOI) waveguide integrated with a GeSe phase change material (PCM) layer. The SOI waveguide’s dimensions are optimized to support the hybridization of the fundamental transverse magnetic (TM0) and the first higher transverse electric (TE1) mode. This design allows for efficient and nearly equal power coupling between these two modes, resulting in high-contrast interference when starting from the amorphous PCM state. The PCM’s phase transition induces a differential change in the modal effective index, enabling high-contrast transmittance modulation. Our numerical simulations demonstrate a multilevel transmission with a high contrast of nearly 14 dB when the amorphous region’s length is varied incrementally, enabling multi-bit storage. The transmittance is maximized in the fully crystalline state with an insertion loss below 0.1 dB. The HMI can also operate as a quasi-pure phase shifter when partially amorphized, making it suitable for Mach–Zehnder interferometers. These characteristics make the proposed device a promising candidate for applications in photonic memories and neuromorphic computing.

Phase-change material (PCM)-based non-volatile multilevel phase shifters are key components in photonic integrated circuits. Electrically, multiple phase levels can be encoded by controlling the heater power and employing different microheater architectures to induce varying degrees of PCM amorphization. However, encoding a large number of levels is not straightforward. In this work, we first investigate a phase shifter structure based on a GeSe PCM integrated on top of a silicon-on-insulator waveguide, employing a simple rectangular-shaped heater under pulse-width modulation (PWM). We numerically demonstrate that multilevel phase shifts can be achieved because of non-uniform heating in the GeSe PCM layer. However, the resulting phase levels for this basic configuration are highly non-linear because of the uniform power dissipation along the light propagation direction characterized by the same cross-section. To overcome this limitation, we designed a novel PCM-based phase shifter with a segmented heater whose width gradually increases along the light propagation direction. This configuration enables the encoding of hundreds of well-spaced phase levels between 0 and π, facilitated by smoother amorphization arising from the combined effects of non-uniform heating across segments and within each segment, while achieving an insertion loss of only 0.6 dB in the worst case. Furthermore, when evaluating both heater architectures under pulse amplitude modulation (PAM) at a fixed pulse duration, we observe behavior consistent with the trends observed for PWM, confirming the superior performance of the segmented heater design.

In this paper, we show that complex-valued Photonic Neural Networks (PNNs) may yield unfair decisions based on encoding strategy and feature pairing, even when these networks are trained to similar accuracy levels. © 2024 The Author(s)

This roadmap consolidates recent advances while exploring emerging applications, reflecting the remarkable diversity of hardware platforms, neuromorphic concepts, and implementation philosophies reported in the field. It emphasizes the critical role of cross-disciplinary collaboration in this rapidly evolving field.

Fabio Pavanello and co-authors discuss the importance of security layers in computer systems, particularly in the context of the Horizon Europe NEUROPULS project, which focuses on innovative security solutions based on novel neuromorphic architectures and PUF-based security layers. Security layers are critical to protecting computer systems from threats. They often rely on cryptographic protocols that use secret keys, which are typically stored in memory. However, storing such sensitive data in non-volatile digital memory can pose risks, especially if exploited through hardware vulnerabilities. To address this, the Horizon Europe NEUROPULS project is exploring novel solutions based on integrated photonics

Pierre Noé, Benoît Cluzel, Stéphane Malhouitre, and Benoît Charbonnier, discuss phase change materials for reconfigurable photonic integrated circuits. Phase-change materials (PCMs) have gained increasing interest over the past decade for their potential in photonic applications. This article reviews their properties, key advantages over competing reconfigurable photonic technologies, and the challenges limiting their widespread adoption.

In this work, we discuss our vision for neuromorphic accelerators based on integrated photonics within the framework of the Horizon Europe NEUROPULS project. Augmented integrated photonic architectures that leverage phase-change and III-V materials for optical computing will be presented. A CMOS-compatible platform will be discussed that integrates these materials to fabricate photonic neuromorphic architectures, along with a gem5-based simulation platform to model accelerator operation once it is interfaced with a RISC-V processor. This simulation platform enables accurate system-level accelerator modeling and benchmarking in terms of key metrics such as speed, energy consumption, and footprint.

This study presents a comparative examination of state-of-the-art resiliency approaches of Convolutional, Spiking, and Photonic neural networks (CNNs, SNNs, PNNs), their fault and error models, and the main fault tolerance techniques.

Photonic time-delay reservoir computing schemes usually employ an input mask as a means of performance enhancement. However, input masking usually necessitates a domain conversion, requiring a signal to be treated before sending it to the reservoir. More recent implementations have explored further ways of performance enhancement, whether through operating in the asynchronous regime, or by using post-filtering approaches. In this numerical study, we analyze the task-independent performance of a passive integrated photonic reservoir, and show that it can achieve good results on some benchmark tasks in the absence of an input mask. We also consider the effects of post-filtering and operating in the asynchronous regime through a parameter space exploration. The proposed scheme enables ultra-fast processing speeds while simultaneously reducing the associated power and complexity costs of the associated electronics. We compare the obtained results with the case of using a mask, and also with other schemes from the literature, showing comparable performance on the investigated tasks.

As AI continues to grow in importance, in order to reduce its carbon footprint and utilization of computer resources, numerous alternatives are under investigation to improve its hardware building blocks. In particular, in convolutional neural networks (CNNs), the convolution function represents the most important operation and one of the best targets for optimization. A new approach to convolution had recently emerged using optics, phase-change materials (PCMs) and stochastic computing, but is thus far limited to unsigned operands. In this paper, we propose an extension in which the convolutional kernels are signed, using mixed-polarity bitstreams. We present a proof of validity for our method, while also showing that, in simulation and under similar operating conditions, our approach is less affected by noise than the common approach in the literature.

Phase-change materials (PCMs) have been growing in interest over the last decade for photonic applications. In this article, we will firstly review their properties and their key benefits with respect to concurring technologies for reconfigurable photonic devices and systems. Then, we will highlight some key open challenges PCMs are currently facing for their ubiquitous adoption. Finally, we will provide some potential routes for addressing these challenges with a focus on current activities in the Grenoble (France) region.

Photonic neural networks (PNNs) are gaining significant interest in the research community due to their potential for high parallelization, low latency, and energy efficiency. PNNs compute using light, which leads to several differences in implementation when compared to electronics, such as the need to represent input features in the photonic domain before feeding them into the network. In this encoding process, it is common to combine multiple features into a single input to reduce the number of inputs and associated devices, leading to smaller and more energy-efficient PNNs. Although this alters the network’s handling of input data, its impact on PNNs remains understudied. This paper addresses this open question, investigating the effect of commonly used encoding strategies that combine features on the performance and learning capabilities of PNNs. Here, using the concept of feature importance, we develop a mathematical methodology for analyzing feature combination. Through this methodology, we demonstrate that encoding multiple features together in a single input determines their relative importance, thus limiting the network’s ability to learn from the data. However, given some prior knowledge of the data, this can also be leveraged for higher accuracy. By selecting an optimal encoding method, we achieve up to a 12.3% improvement in the accuracy of PNNs trained on the Iris dataset compared to other encoding techniques, surpassing the performance of networks where features are not combined. These findings highlight the importance of carefully choosing the encoding to the accuracy and decision-making strategies of PNNs, particularly in size or power constrained applications.

The study of regularity in signals can be of great importance, typically in medicine to analyse electrocardiogram (ECG) or electromyography (EMG) signals, but also in climate studies, finance or security. In this work we focus on security primitives such as Physical Unclonable Functions (PUFs) or Pseudo-Random Number Generators (PRNGs). Such primitives must have a high level of complexity or entropy in their responses to guarantee enough security for their applications. There are several ways of assessing the complexity of their responses, especially in the binary domain. With the development of analog PUFs such as optical (photonic) PUFs, it would be useful to be able to assess their complexity in the analog domain when designing them, for example, before converting analog signals into binary. In this numerical study, we decided to explore the potential of the disentropy of autocorrelation as a measure of complexity for security primitives as PUFs, TRNGs or PRNGs with analog output or responses. We compare this metric to others used to assess regularities in analog signals such as Approximate Entropy (ApEn) and Fuzzy Entropy (FuzEn). We show that the disentropy of autocorrelation is able to differentiate between well-known PRNGs and non-optimised or bad PRNGs in the analog and binary domain with a better contrast than ApEn and FuzEn. Next, we show that the disentropy of autocorrelation is able to detect small patterns injected in PUFs responses and then we applied it to photonic PUFs simulations.

In the contemporary security landscape, the incorporation of photonics has emerged as a transformative force, unlocking a spectrum of possibilities to enhance the resilience and effectiveness of security primitives. This integration represents more than a mere technological augmentation; it signifies a paradigm shift towards innovative approaches capable of delivering security primitives with key properties for low-power systems. This not only augments the robustness of security frameworks, but also paves the way for novel strategies that adapt to the evolving challenges of the digital age. This paper discusses the security layers and related services that will be developed, modeled, and evaluated within the Horizon Europe NEUROPULS project. These layers will exploit novel implementations for security primitives based on physical unclonable functions (PUFs) using integrated photonics technology. Their objective is to provide a series of services to support the secure operation of a neuromorphic photonic accelerator for edge computing applications.