An official website of the United States government
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
Digitized Phase‐Change Material Heterostack for Transmissive Diffractive Optical Neural Network
All‐optical and fully reconfigurable transmissive diffractive optical neural network (DONN) architectures emerge as high‐throughput and energy‐efficient machine learning (ML) hardware accelerators in broad applications. However, current device and system implementations have limited performance. In this work, a novel transmissive diffractive device architecture, a digitized phase‐change material (PCM) heterostack, which consists of multiple nonvolatile PCM layers with different thicknesses, is demonstrated. Through this architecture, the advantages of PCM electrical and optical properties can be leveraged and challenges associated with multilevel operations in a single PCM layer can be mitigated. Through proof‐of‐concept experiments, the electrical tuning of one PCM layer is demonstrated in a transmissive spatial light modulation device, and thermal analysis guides the design of multilayer devices and DONN systems to avoid thermal cross talk if individual heterostacks are assembled into an array. Further, a heterostack containing three PCM layers is designed based on experimental results to produce a large‐phase modulation range and uniform coverage, and the ML performance of DONN systems with the designed heterostack is evaluated. The developed device architecture is practically feasible and scalable for future energy‐efficient, fast‐reconfigured, and compact transmissive DONN systems.
more »
« less
- PAR ID:
- 10600198
- Publisher / Repository:
- Wiley Blackwell (John Wiley & Sons)
- Date Published:
- Journal Name:
- Advanced Photonics Research
- Volume:
- 6
- Issue:
- 6
- ISSN:
- 2699-9293
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Phase Change Memory (PCM) is an attractive candidate for main memory, as it offers non-volatility and zero leakage power while providing higher cell densities, longer data retention time, and higher capacity scaling compared to DRAM. In PCM, data is stored in the crystalline or amorphous state of the phase change material. The typical electrically controlled PCM (EPCM), however, suffers from longer write latency and higher write energy compared to DRAM and limited multi-level cell (MLC) capacities. These challenges limit the performance of data-intensive applications running on computing systems with EPCMs. Recently, researchers demonstrated optically controlled PCM (OPCM) cells with support for 5bits/cellin contrast to 2bits/cellin EPCM. These OPCM cells can be accessed directly with optical signals that are multiplexed in high-bandwidth-density silicon-photonic links. The higher MLC capacity in OPCM and the direct cell access using optical signals enable an increased read/write throughput and lower energy per access than EPCM. However, due to the direct cell access using optical signals, OPCM systems cannot be designed using conventional memory architecture. We need a complete redesign of the memory architecture that is tailored to the properties of OPCM technology. This article presents the design of a unified network and main memory system called COSMOS that combines OPCM and silicon-photonic links to achieve high memory throughput. COSMOS is composed of a hierarchical multi-banked OPCM array with novel read and write access protocols. COSMOS uses an Electrical-Optical-Electrical (E-O-E) control unit to map standard DRAM read/write commands (sent in electrical domain) from the memory controller on to optical signals that access the OPCM cells. Our evaluation of a 2.5D-integrated system containing a processor and COSMOS demonstrates2.14 ×average speedup across graph and HPC workloads compared to an EPCM system. COSMOS consumes3.8×lower read energy-per-bit and5.97×lower write energy-per-bit compared to EPCM. COSMOS is the first non-volatile memory that provides comparable performance and energy consumption as DDR5 in addition to increased bit density, higher area efficiency, and improved scalability.more » « less
-
Abstract Reconfigurability of photonic integrated circuits (PICs) has become increasingly important due to the growing demands for electronic–photonic systems on a chip driven by emerging applications, including neuromorphic computing, quantum information, and microwave photonics. Success in these fields usually requires highly scalable photonic switching units as essential building blocks. Current photonic switches, however, mainly rely on materials with weak, volatile thermo‐optic or electro‐optic modulation effects, resulting in large footprints and high energy consumption. As a promising alternative, chalcogenide phase‐change materials (PCMs) exhibit strong optical modulation in a static, self‐holding fashion, but the scalability of present PCM‐integrated photonic applications is still limited by the poor optical or electrical actuation approaches. Here, with phase transitions actuated by in situ silicon PIN diode heaters, scalable nonvolatile electrically reconfigurable photonic switches using PCM‐clad silicon waveguides and microring resonators are demonstrated. As a result, intrinsically compact and energy‐efficient switching units operated with low driving voltages, near‐zero additional loss, and reversible switching with high endurance are obtained in a complementary metal‐oxide‐semiconductor (CMOS)‐compatible process. This work can potentially enable very large‐scale CMOS‐integrated programmable electronic–photonic systems such as optical neural networks and general‐purpose integrated photonic processors.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 Advancements in nanofabrication processes have propelled nonvolatile phase change materials (PCMs) beyond storage‐class applications. They are now making headway in fields such as photonic integrated circuits (PIC), free‐space optics, and plasmonics. This shift is owed to their distinct electrical, optical, and thermal properties between their different atomic structures, which can be reversibly switched through thermal stimuli. However, the reliability of PCM‐based optical components is not yet on par with that of storage‐class devices. This is in part due to the challenges in maintaining a uniform temperature distribution across the PCM volume during phase transformation, which is essential to mitigate stress and element segregation as the device size exceeds a few micrometers. Understanding thermal transport in PCM‐based devices is thus crucial as it dictates not only the durability but also the performance and power consumption of these devices. This article reviews recent advances in the development of PCM‐based photonic devices from a thermal transport perspective and explores potential avenues to enhance device reliability. The aim is to provide insights into how PCM‐based technologies can evolve beyond storage‐class applications, maintain their functionality, and achieve longer lifetimes.more » « less
