Harish Bhaskaran, FREng

Neuromorphic photonic processors are redefining the boundaries of classical computing by enabling high-speed multidimensional information processing within the memory. Memristors, the backbone of neuromorphic processors, retain their state after programming without static power consumption. Among them, electro-optic memristors are of great interest, as they enable dual electrical–optical functionality that bridges the efficiency of electronics and the bandwidth of photonics. However, efficient, scalable, and CMOS-compatible implementations of electro-optic memristors are still lacking. Here, we devise electro-optic memristors by structuring the phase-change material as a nanoscale constriction, geometrically confining the electrically generated heat profile to overlap with the optical field, thus achieving programmability and readability in both the electrical and optical domains. We demonstrate sub-10 pJ electrical switching energy and a high electro-optical modulation efficiency of 0.15 nJ/dB. Our work opens up opportunities for high-performance and energy-efficient integrated electro-optic neuromorphic computing.

Plastic self-adaptation, nonlinear recurrent dynamics and multi-scale memory are desired features in hardware implementations of neural networks, because they enable them to learn, adapt, and process information similarly to the way biological brains do. In this work, these properties occurring in arrays of photonic neurons are experimentally demonstrated. Importantly, this is realized autonomously in an emergent fashion, without the need for an external controller setting weights and without explicit feedback of a global reward signal. Using a hierarchy of such arrays coupled to a backpropagation-free training algorithm based on simple logistic regression, a performance of 98.2% is achieved on the MNIST task, a popular benchmark task looking at classification of written digits. The plastic nodes consist of silicon photonics microring resonators covered by a patch of phase-change material that implements nonvolatile memory. The system is compact, robust, and straightforward to scale up through the use of multiple wavelengths. Moreover, it constitutes a unique platform to test and efficiently implement biologically plausible learning schemes at a high processing speed.

An on-chip asymmetric directional coupler (DC) can convert fundamental modes to higher-order modes and is one of the core components of mode-division multiplexing (MDM) technology. In this study, we propose that waveguides of the asymmetric DC can be trimmed by silicon ion implantation to tune the effective refractive index and facilitate mode conversion into higher-order modes. Through this method of tuning, transmission changes of up to 18 dB have been realized with one ion implantation step. In addition, adjusting the position of the ion implantation on the waveguide can provide a further degree of control over the transmission into the resulting mode. The results of this work present a promising new route for the development of high-efficiency, low-loss mode converters for integrated photonic platforms, and aim to facilitate the application of MDM technology in emerging photonic neuromorphic computing.

Advancements in optical coherence control1–5 have unlocked many cutting-edge applications, including long-haul communication, light detection and ranging (LiDAR) and optical coherence tomography6–8. Prevailing wisdom suggests that using more coherent light sources leads to enhanced system performance and device functionalities9–11. Our study introduces a photonic convolutional processing system that takes advantage of partially coherent light to boost computing parallelism without substantially sacrificing accuracy, potentially enabling larger-size photonic tensor cores. The reduction of the degree of coherence optimizes bandwidth use in the photonic convolutional processing system. This breakthrough challenges the traditional belief that coherence is essential or even advantageous in integrated photonic accelerators, thereby enabling the use of light sources with less rigorous feedback control and thermal-management requirements for high-throughput photonic computing. Here we demonstrate such a system in two photonic platforms for computing applications: a photonic tensor core using phase-change-material photonic memories that delivers parallel convolution operations to classify the gaits of ten patients with Parkinson’s disease with 92.2% accuracy (92.7% theoretically) and a silicon photonic tensor core with embedded electro-absorption modulators (EAMs) to facilitate 0.108 tera operations per second (TOPS) convolutional processing for classifying the Modified National Institute of Standards and Technology (MNIST) handwritten digits dataset with 92.4% accuracy (95.0% theoretically).