A novel class of programmable integrated photonic circuits has emerged over the past years, strongly driven by approaches to tackle unsolved computing problems in the optical domain. Photonic neuromorphic and quantum computing are examples of optical systems implemented in complex photonic circuits, which are reconfigured before and during operation. However, a key building block to enable efficient reconfigurable optical network architectures is still missing: a non-volatile optical phase shifter. Here we demonstrate such an element—compatible with silicon photonics—based on the monolithic integration of BaTiO 3 thin films with silicon waveguides. By manipulating ferroelectric domains in BaTiO 3 with electrical control signals, we achieve analogue and non-volatile optical phase tuning with no absorption changes. We demonstrate an eight-level long-term-stable photonic device with non-destructive optical readout and switching energy as low as 4.6 pJ. With our results, an analogue non-volatile photonic element is added to the integrated photonics toolbox, enabling a new generation of power-efficient programmable photonic circuits.

Energy Efficiency of Artificial Intelligence (AI) workloads increasingly becomes a challenge as i) they are being adopted by a growing community of industries and ii) the AI models used are growing tremendously in complexity. We can identify certain common operations which contribute to the bulk of the computations of these workloads and thus also to the overall energy footprint. Besides data-transport and generic multiply-accumulate operations, convolutions with relatively small kernels constitute a substantial part of today’s AI workloads. In this paper we will investigate potential and limitations of optical convolutional processors for AI workloads to improve their energy efficiency. We underline our findings with a thorough system analysis and with simulation and measurement results of a sequential lattice filter type optical convolutional processor on silicon.

We demonstrate a non-volatile optical memory element integrated in silicon photonics for low-power reconfigurable photonic circuits and neural networks. Stable transmission states are set by manipulating ferroelectric domains in BaTiO3 films embedded in photonic waveguides.

Dedicated technology platforms gain interest for enhancing the performance and efficiency of neuromorphic computing. We demonstrate integrated optic devices for convolutional signal processing and neural network training.

Neuromorphic systems are designed with careful consideration of the physical properties of the computational substrate they use. Neuromorphic engineers often exploit physical phenomena to directly implement a desired functionality, enabled by "the isomorphism between physical processes in different media" (Douglas et al., 1995). This bottom-up design methodology could be described as matching computational primitives to physical phenomena. In this paper, we propose a top-down counterpart to the bottom-up approach to neuromorphic design. Our top-down approach, termed "bias matching," is to match the inductive biases required in a learning system to the hardware constraints of its implementation; a well-known example is enforcing translation equivariance in a neural network by tying weights (replacing vector-matrix multiplications with convolutions), which reduces memory requirements. We give numerous examples from the literature and explain how they can be understood from this perspective. Furthermore, we propose novel network designs based on this approach in the context of collaborative filtering. Our simulation results underline our central conclusions: additional hardware constraints can improve the predictions of a Machine Learning system, and understanding the inductive biases that underlie these performance gains can be useful in finding applications for a given constraint.

We demonstrate for the first time the heterogeneous co-integration of Si photonics, BTO/Si for high-speed modulation and III-V materials for photodetection and emission. We show light coupling with losses <; 0.5 dB between the different functional layers.

Photonics offers exciting opportunities for neuromorphic computing. This paper specifically reviews the prospects of integrated optical solutions for accelerating inference and training of artificial neural networks. Calculating the synaptic function, thereof, is computationally very expensive and does not scale well on state-of-the-art computing platforms. Analog signal processing, using linear and nonlinear properties of integrated optical devices, offers a path toward substantially improving performance and power efficiency of these artificial intelligence workloads. The ability of integrated photonics to operate at very high speeds opens opportunities for time-critical real-time applications, while chip-level integration paves the way to cost-effective manufacturing and assembly.

We demonstrate PIC-based non-volatile optical synaptic elements, an essential building block in large non-von Neumann circuits realized in integrated photonics. The impact of non-idealities on the performance of a photonic recurrent neural networks is evaluated.

Analog signal processing is one of the promising paths to enhance performance and power efficiency of neural network inference and training. Recently, analog optical neuromorphic computing concepts also gained increasing interest. We discuss properties of photonic systems in view of neuromorphic computing, applications thereof and present corresponding devices and subsystems.

Photonic integrated circuits (PICs) operating at cryogenic temperatures are fundamental building blocks required to achieve scalable quantum computing and cryogenic computing technologies1,2. Silicon PICs have matured for room-temperature applications, but their cryogenic performance is limited by the absence of efficient low-temperature electro-optic modulation. Here we demonstrate electro-optic switching and modulation from room temperature down to 4 K by using the Pockels effect in integrated barium titanate (BaTiO3) devices3. We investigate the temperature dependence of the nonlinear optical properties of BaTiO3, showing an effective Pockels coefficient of 200 pm V−1 at 4 K. The fabricated devices show an electro-optic bandwidth of 30 GHz, ultralow-power tuning that is 109 times more efficient than thermal tuning, and high-speed data modulation at 20 Gbps. Our results demonstrate a missing component for cryogenic PICs, removing major roadblocks for the realization of cryogenic-compatible systems in the field of quantum computing, supercomputing and sensing, and for interfacing those systems with instrumentation at room temperature. The integration of barium titanate thin films with silicon-based waveguides enables the operation of efficient electro-optic switches and modulators at temperatures as low as 4 K, with potential applications in quantum computing and cryogenic computing technologies.

The complex linear network analyzer (COLNA) python package analytically computes the propagation of complex valued signals in networks with linear nodes and directed, complex valued and delayed edges (Fig. 1). COLNA offers an easy and well-documented interface, which allows users to quickly build network models and understand their behaviour. Its main purpose is to compute coherent wave propagation through linear photonic circuits, but COLNA might be useful in any research area where signal propagation through linear complex networks is of practical relevance.

Integrated electrical and photonic circuits (PIC) operating at cryogenic temperatures are fundamental building blocks required to achieve scalable quantum computing, and cryogenic computing technologies. Optical interconnects offer better performance and thermal insulation than electrical wires and are imperative for true quantum communication. Silicon PICs have matured for room temperature applications but their cryogenic performance is limited by the absence of efficient low temperature electro-optic (EO) modulation. While detectors and lasers perform better at low temperature, cryogenic optical switching remains an unsolved challenge. Here we demonstrate EO switching and modulation from room temperature down to 4 K by using the Pockels effect in integrated barium titanate (BaTiO3)-based devices. We report the nonlinear optical (NLO) properties of BaTiO3 in a temperature range which has previously not been explored, showing an effective Pockels coefficient of 200 pm/V at 4 K. We demonstrate the largest EO bandwidth (30 GHz) of any cryogenic switch to date, ultra-low-power tuning which is 10^9 times more efficient than thermal tuning, and high-speed data modulation at 20 Gbps. Our results demonstrate a missing component for cryogenic PICs. It removes major roadblocks for the realisation of novel cryogenic-compatible systems in the field of quantum computing and supercomputing, and for interfacing those systems with the real world at room-temperature.

Integrated photonics technology offers great potential for applications in neuromorphic systems. We discuss two examples; integrated photonic non-volatile optical weights and a photonic non-volatile memory based analog accelerator for the inference and training of deep neural networks.

We demonstrate the monolithic integration of CMOS-compatible ultralow capacitance hybrid III-V/Si photodetectors and test these devices up to 32Gbps NRZ. The lateral photodiodes are suitable for ultrafast optical communication without using a transimpedance amplifier.

Reservoir computing (RC) is a promising implementation of a non-von Neumann computing architecture. It is well-suited to solve time-dependent and dynamic problems like speech recognition or bitwise operations. In principal integrated photonic RC offers energy-efficient, high-speed signal processing capabilities. However, current photonic RC systems perform the linear combination of the complex reservoir states in software. Software weights are ideal (high precision, no drift), but limit the processing speed and energy efficiency of the RC system. Here, we discuss a hardware implementation of photonic weights compatible with silicon photonic RC architectures. First, we demonstrate non-volatile synaptic weights based on ferroelectric barium titanate (BTO) thin films. Second, we explore how imperfections in these hardware weights impact the reservoir performance.

We demonstrate the first electro-optic switch operating at cryogenic temperatures of 4 K with a high electro-optic bandwidth of >18 GHz. Our novel technology exploits the Pockels effect in barium titanate thin films co-integrated with silicon photonics and offers low losses, pure phase modulation, and sub-pW electro-optic tuning.