A blueprint for precise and fault-tolerant analog neural networks.

Analog computing has reemerged as a promising avenue for accelerating deep neural networks (DNNs) to overcome the scalability challenges posed by traditional digital architectures. However, achieving high precision using analog technologies is challenging, as high-precision data converters are costly and impractical. In this work, we address this challenge by using the residue number system (RNS) and composing high-precision operations from multiple low-precision operations, thereby eliminating the need for high-precision data converters and information loss. Our study demonstrates that the RNS-based approach can achieve ≥99% FP32 accuracy with 6-bit integer arithmetic for DNN inference and 7-bit for DNN training. The reduced precision requirements imply that using RNS can achieve several orders of magnitude higher energy efficiency while maintaining the same throughput compared to conventional analog hardware with the same precision. We also present a fault-tolerant dataflow using redundant RNS to protect the computation against noise and errors inherent within analog hardware.

Delocalized photonic deep learning on the internet's edge.

Advanced machine learning models are currently impossible to run on edge devices such as smart sensors and unmanned aerial vehicles owing to constraints on power, processing, and memory. We introduce an approach to machine learning inference based on delocalized analog processing across networks. In this approach, named Netcast, cloud-based "smart transceivers" stream weight data to edge devices, enabling ultraefficient photonic inference. We demonstrate image recognition at ultralow optical energy of 40 attojoules per multiply (<1 photon per multiply) at 98.8% (93%) classification accuracy. We reproduce this performance in a Boston-area field trial over 86 kilometers of deployed optical fiber, wavelength multiplexed over 3 terahertz of optical bandwidth. Netcast allows milliwatt-class edge devices with minimal memory and processing to compute at teraFLOPS rates reserved for high-power (>100 watts) cloud computers.

Dual slot-mode NOEM phase shifter.

Photonic system component counts are increasing rapidly, particularly in CMOS-compatible silicon photonics processes. Large numbers of cascaded active photonic devices are difficult to implement when accounting for constraints on area, power dissipation, and response time. Plasma dispersion and the thermo-optic effect, both available in CMOS-compatible silicon processes, address a subset of these criteria. With the addition of a few back-end-of-line etch processing steps, silicon photonics platforms can support nano-opto-electro-mechanical (NOEM) phase shifters. Realizing NOEM phase shifters that operate at CMOS-compatible voltages (≤ 1.2 V) and with low insertion loss remains a challenge. Here, we introduce a novel NOEM phase shifter fabricated alongside 90 nanometer transistors that imparts 5.63 radians phase shift at 1.08 volts bias over an actuation length of 25µm with an insertion loss of less than 0.04 dB and 3 dB bandwidth of 0.26 MHz.

Large-alphabet encoding for higher-rate quantum key distribution.

The manipulation of high-dimensional degrees of freedom provides new opportunities for more efficient quantum information processing. It has recently been shown that high-dimensional encoded states can provide significant advantages over binary quantum states in applications of quantum computation and quantum communication. In particular, high-dimensional quantum key distribution enables higher secret-key generation rates under practical limitations of detectors or light sources, as well as greater error tolerance. Here, we demonstrate high-dimensional quantum key distribution capabilities both in the laboratory and over a deployed fiber, using photons encoded in a high-dimensional alphabet to increase the secure information yield per detected photon. By adjusting the alphabet size, it is possible to mitigate the effects of receiver bottlenecks and optimize the secret-key rates for different channel losses. This work presents a strategy for achieving higher secret-key rates in receiver-limited scenarios and marks an important step toward high-dimensional quantum communication in deployed fiber networks.

A MoTe2-based light-emitting diode and photodetector for silicon photonic integrated circuits.

One of the current challenges in photonics is developing high-speed, power-efficient, chip-integrated optical communications devices to address the interconnects bottleneck in high-speed computing systems. Silicon photonics has emerged as a leading architecture, in part because of the promise that many components, such as waveguides, couplers, interferometers and modulators, could be directly integrated on silicon-based processors. However, light sources and photodetectors present ongoing challenges. Common approaches for light sources include one or few off-chip or wafer-bonded lasers based on III-V materials, but recent system architecture studies show advantages for the use of many directly modulated light sources positioned at the transmitter location. The most advanced photodetectors in the silicon photonic process are based on germanium, but this requires additional germanium growth, which increases the system cost. The emerging two-dimensional transition-metal dichalcogenides (TMDs) offer a path for optical interconnect components that can be integrated with silicon photonics and complementary metal-oxide-semiconductors (CMOS) processing by back-end-of-the-line steps. Here, we demonstrate a silicon waveguide-integrated light source and photodetector based on a p-n junction of bilayer MoTe2, a TMD semiconductor with an infrared bandgap. This state-of-the-art fabrication technology provides new opportunities for integrated optoelectronic systems.

Programmable dispersion on a photonic integrated circuit for classical and quantum applications.

We demonstrate a large-scale tunable-coupling ring resonator array, suitable for high-dimensional classical and quantum transforms, in a CMOS-compatible silicon photonics platform. The device consists of a waveguide coupled to 15 ring-based dispersive elements with programmable linewidths and resonance frequencies. The ability to control both quality factor and frequency of each ring provides an unprecedented 30 degrees of freedom in dispersion control on a single spatial channel. This programmable dispersion control system has a range of applications, including mode-locked lasers, quantum key distribution, and photon-pair generation. We also propose a novel application enabled by this circuit - high-speed quantum communications using temporal-mode-based quantum data locking - and discuss the utility of the system for performing the high-dimensional unitary optical transformations necessary for a quantum data locking demonstration.