The integration of artificial intelligence systems into daily applications like speech recognition and autonomous driving rapidly increases the amount of data generated and processed. However, satisfying the hardware requirements with the conventional von Neumann architecture remains challenging due to the von Neumann bottleneck. Therefore, new architectures inspired by the working principles of the human brain are developed, and they are called neuromorphic computing. The key principles of neuromorphic computing are in-memory computing to reduce data shuffling and parallelization to decrease computation time. One promising framework for neuromorphic computing is phase-change photonics. By switching to the optical domain, parallelization is inherently possible by wavelength division multiplexing, and high modulation speeds can be deployed. Non-volatile phase-change materials are used to perform multiplications and non-linear operations in an energetically efficient manner. Here, we present two prototypes of neuromorphic photonic computation units based on chalcogenide phase-change materials. First is a neuromorphic hardware accelerator designed to carry out matrix vector multiplication in convolutional neural networks. Due to the neuromorphic architecture, this prototype can already operate at tera-multiply-accumulate per second speeds. Second is an all-optical spiking neuron, which can serve as a building block for large-scale artificial neural networks. Here, the whole computation is carried out in the optical domain, and the device only needs an electrical interface for data input and readout.

The integration of artificial intelligence (AI) systems in the daily life greatly increases the amount of data generated and processed. In addition to the large computational power required, the hardware needs to be compact and energy efficient. One promising approach to fulfill those requirements is phase-change material based photonic neuromorphic computing that enables in-memory computation and a high degree of parallelization. In the following, we present an optimized layout of a photonic tensor core (PTC) which is designed to perform real valued matrix vector multiplications and operates at telecommunication wavelengths. We deploy the well-studied phase-change material Ge2Sb2Te5 (GST) as an optical attenuator to perform single positive valued multiplications. In order to generalize the multiplication to arbitrary real factors, we develop a novel symmetric multiplication unit which directly includes a reference-computation branch. The variable GST attenuator enables a modulation depth of 5 dB over a wavelength range of 100 nm with a wavelength dependency below 0.8 dB. The passive photonic circuit itself ensures equal coupling to the main-computation and reference-computation branch over the complete wavelength range. For the first time, we integrate wavelength multiplexers (MUX) together with a photonic crossbar array on-chip, paving the way towards fully integrated systems. The MUX are crucial for the PTC since they enable multiple computational channels in a single photonic crossbar array. We minimize the crosstalk between the channels by designing Bragg scattering based MUX. By cascading, we achieve an extinction ratio larger than 61 dB while the insertion loss is below 1 dB.

Nonvolatile photonic memory cells are basic building blocks for neuromorphic hardware enabling the realization of all‐optical synapses and artificial neurons. These devices commonly exploit chalcogenide phase‐change materials, which are evanescently coupled to photonic waveguides, and provide fast write/erase speeds and large storage capacity. Here, we report for the first time the programming of a nonvolatile photonic memory cell based on Ag3In4Sb76Te17 (AIST) which is capable of mimicking biphasic synapses. We evaluate the underlying mechanism of biphasic behavior of AIST cells based on numerical simulations and correlate to experimental findings. Switching dynamics demonstrate enhanced performance with a post‐excitation dead time as short as 12.8 ns. Based on AIST double cells, we demonstrate reversible multilevel switching between 45 unique synaptic weights for long‐term depression (LTD) and long‐term potentiation (LTP). The observed biphasic programming and excellent switching performance render AIST‐based photonic memory cells promising for artificial neural networks and neuromorphic photonic computing hardware.

With more and more aspects of modern life and scientific tools becoming digitized, the amount of data being generated is growing exponentially. Fast and efficient statistical processing, such as identifying correlations in big datasets, is therefore becoming increasingly important, and this, on account of the various compute bottlenecks in modern digital machines, has necessitated new computational paradigms. Here, we demonstrate one such novel paradigm, via the development of an integrated phase-change photonics engine. The computational memory engine exploits the accumulative property of Ge 2 Sb 2 Te 5 phase-change cells and wavelength division multiplexing property of optics in delivering fully parallelized and colocated temporal correlation detection computations. We investigate this property and present an experimental demonstration of identifying real-time correlations in data streams on the social media platform Twitter and high-traffic computing nodes in data centers. Our results demonstrate the use case of high-speed integrated photonics in accelerating statistical analysis methods.

Integrated neuromorphic photonic circuits aim to power complex artificial neural networks (ANNs) in an energy and time efficient way by exploiting the large bandwidth and the low loss of photonic structures. However, scaling photonic circuits to match the requirements of modern ANNs still remains challenging. In this perspective, we give an overview over the usual sizes of matrices processed in ANNs and compare them with the capability of existing photonic matrix processors. To address shortcomings of existing architectures, we propose a time multiplexed matrix processing scheme which virtually increases the size of a physical photonic crossbar array without requiring any additional electrical post-processing. We investigate the underlying process of time multiplexed incoherent optical accumulation and achieve accumulation accuracy of 98.9% with 1 ns pulses. Assuming state of the art active components and a reasonable crossbar array size, this processor architecture would enable matrix vector multiplications with 16,000 × 64 matrices all optically on an estimated area of 51.2 mm2, while performing more than 110 trillion multiply and accumulate operations per second.