Diffractive optical neural networks (DONNs) are emerging as high‐throughput and energy‐efficient hardware platforms to perform all‐optical machine learning (ML) in machine vision systems. However, the current demonstrated applications of DONNs are largely image classification tasks, which undermine the prospect of developing and utilizing such hardware for other ML applications. Herein, the deployment of an all‐optical reconfigurable DONNs system for scientific computing is demonstrated numerically and experimentally, including guiding two‐dimensional quantum material synthesis, predicting the properties of two‐dimensional quantum materials and small molecular cancer drugs, predicting the device response of nanopatterned integrated photonic power splitters, and the dynamic stabilization of an inverted pendulum with reinforcement learning. Despite a large variety of input data structures, a universal feature engineering approach is developed to convert categorical input features to images that can be processed in the DONNs system. The results open up new opportunities for employing DONNs systems for a broad range of ML applications.
Multilayer diffractive optical neural networks (DONNs) can perform machine learning (ML) tasks at the speed of light with low energy consumption. Decreasing the number of diffractive layers can reduce inevitable material and diffraction losses to improve system performance, and incorporating compact devices can reduce the system footprint. However, current analytical DONN models cannot accurately describe such physical systems. Here we show the ever-ignored effects of interlayer reflection and interpixel interaction on the deployment performance of DONNs through full-wave electromagnetic simulations and terahertz (THz) experiments. We demonstrate that the drop of handwritten digit classification accuracy due to reflection is negligible with conventional low-index THz polymer materials, while it can be substantial with high-index materials. We further show that one- and few-layer DONN systems can achieve high classification accuracy, but there is a trade-off between accuracy and model-system matching rate because of the fast-varying spatial distribution of optical responses in diffractive masks. Deep DONNs can break down such a trade-off because of reduced mask spatial complexity. Our results suggest that new accurate and trainable DONN models are needed to advance the development and deployment of compact DONN systems for sophisticated ML tasks.
more » « less- Award ID(s):
- 1936729
- NSF-PAR ID:
- 10389196
- Publisher / Repository:
- Optical Society of America
- Date Published:
- Journal Name:
- Optics Letters
- Volume:
- 48
- Issue:
- 2
- ISSN:
- 0146-9592; OPLEDP
- Page Range / eLocation ID:
- Article No. 219
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
more » « less
-
This paper presents a design approach for the modeling and simulation of ultra-low power (ULP) analog computing machine learning (ML) circuits for seizure detection using EEG signals in wearable health monitoring applications. In this paper, we describe a new analog system modeling and simulation technique to associate power consumption, noise, linearity, and other critical performance parameters of analog circuits with the classification accuracy of a given ML network, which allows to realize a power and performance optimized analog ML hardware implementation based on diverse application-specific needs. We carried out circuit simulations to obtain non-idealities, which are then mathematically modeled for an accurate mapping. We have modeled noise, non-linearity, resolution, and process variations such that the model can accurately obtain the classification accuracy of the analog computing based seizure detection system. Noise has been modeled as an input-referred white noise that can be directly added at the input. Device process and temperature variations were modeled as random fluctuations in circuit parameters such as gain and cut-off frequency. Nonlinearity was mathematically modeled as a power series. The combined system level model was then simulated for classification accuracy assessments. The design approach helps to optimize power and area during the development of tailored analog circuits for ML networks with the ability to potentially trade power and performance goals while still ensuring the required classification accuracy. The simulation technique also enables to determine target specifications for each circuit block in the analog computing hardware. This is achieved by developing the ML hardware model, and investigating the effect of circuit nonidealities on classification accuracy. Simulation of an analog computing EEG seizure detection block shows a classification accuracy of 91%. The proposed modeling approach will significantly reduce design time and complexity of large analog computing systems. Two feature extraction approaches are also compared for an analog computing architecture.more » « less
-
more » « less
Abstract The incorporation of high‐performance optoelectronic devices into photonic neuromorphic processors can substantially accelerate computationally intensive matrix multiplication operations in machine learning (ML) algorithms. However, the conventional designs of individual devices and system are largely disconnected, and the system optimization is limited to the manual exploration of a small design space. Here, a device‐system end‐to‐end design methodology is reported to optimize a free‐space optical general matrix multiplication (GEMM) hardware accelerator by engineering a spatially reconfigurable array made from chalcogenide phase change materials. With a highly parallelized integrated hardware emulator with experimental information, the design of unit device to directly optimize GEMM calculation accuracy is achieved by exploring a large parameter space through reinforcement learning algorithms, including deep Q‐learning neural network, Bayesian optimization, and their cascaded approach. The algorithm‐generated physical quantities show a clear correlation between system performance metrics and device specifications. Furthermore, physics‐aware training approaches are employed to deploy optimized hardware to the tasks of image classification, materials discovery, and a closed‐loop design of optical ML accelerators. The demonstrated framework offers insights into the end‐to‐end and co‐design of optoelectronic devices and systems with reduced human supervision and domain knowledge barriers.
-
more » « less
Abstract Diffractive optical neural networks have shown promising advantages over electronic circuits for accelerating modern machine learning (ML) algorithms. However, it is challenging to achieve fully programmable all‐optical implementation and rapid hardware deployment. Here, a large‐scale, cost‐effective, complex‐valued, and reconfigurable diffractive all‐optical neural networks system in the visible range is demonstrated based on cascaded transmissive twisted nematic liquid crystal spatial light modulators. The employment of categorical reparameterization technique creates a physics‐aware training framework for the fast and accurate deployment of computer‐trained models onto optical hardware. Such a full stack of hardware and software enables not only the experimental demonstration of classifying handwritten digits in standard datasets, but also theoretical analysis and experimental verification of physics‐aware adversarial attacks onto the system, which are generated from a complex‐valued gradient‐based algorithm. The detailed adversarial robustness comparison with conventional multiple layer perceptrons and convolutional neural networks features a distinct statistical adversarial property in diffractive optical neural networks. The developed full stack of software and hardware provides new opportunities of employing diffractive optics in a variety of ML tasks and in the research on optical adversarial ML.
-
more » « less
Abstract Multilevel diffractive lenses (MDLs) have emerged as an alternative to both conventional diffractive optical elements (DOEs) and metalenses for applications ranging from imaging to holographic and immersive displays. Recent work has shown that by harnessing structural parametric optimization of DOEs, one can design MDLs to enable multiple functionalities like achromaticity, depth of focus, wide-angle imaging, etc. with great ease in fabrication. Therefore, it becomes critical to understand how fabrication errors still do affect the performance of MDLs and numerically evaluate the trade-off between efficiency and initial parameter selection, right at the onset of designing an MDL, i.e., even before putting it into fabrication. Here, we perform a statistical simulation-based study on MDLs (primarily operating in the THz regime) to analyse the impact of various fabrication imperfections (single and multiple) on the final structure as a function of the number of ring height levels. Furthermore, we also evaluate the performance of these same MDLs with the change in the refractive index of the constitutive material. We use focusing efficiency as the evaluation criterion in our numerical analysis; since it is the most fundamental property that can be used to compare and assess the performance of lenses (and MDLs) in general designed for any application with any specific functionality.
