Massachusetts Institute of Technology

Columbia University

Harvard College

Ntt Research, Inc.

Raytheon Bbn Technologies Corp.

National Tech. & Eng. Solutions of Sandia, LLC

The USA as Represented by the Sec. of the Army

The USA as Represented by the Secy. of the Army

Dirk Robert Englund

Nicholas Christopher Harris

Jacob C Mower

Marin Soljacic

Scott A Skirlo

Mihika Prabhu

Hyeongrak Choi

Jacques Johannes Carolan

Christopher Louis Panuski

Ryan Hamerly

Matthew Edwin Trusheim

Yichen Shen

Donggyu Kim

Greg Steinbrecher

Mikkel Heuck

Simon Verghese

Mikhail D Lukin

Jeffrey Scott Herd

Paul William Juodawlkis

Hannah Anne Clevenson

Mihir Pant

Yi Yang

Cheryl Marie Sorace-Agaskar

Ian Robert Christen

Gregory R Steinbrecher

Saumil Bandyopadhyay

Cheng Peng

Michael P Walsh

Edward H Chen

Hannes Bernien

Harry Jay Levine

Ahmed Omran

Alexander Keesling Contreras

Jordan Goldstein

Kurt Jacobs

Jeehwan Kim

Darius Bunandar

Michael W Geis

Chee Wei Wong

Mark Alexander Hollis

Karl Kimon Berggren

Matt Eichenfield

Kin Chung Fong

Ruonan Han

Jonathan Hodges

Seth Lloyd

Norman Y Yao

Richard J Molnar

Franco N C Wong

Prineha Narang

Faraz Najafi

Xiaolong Hu

Tomas Neuman

Mohamed I Ibrahim

Donald Fahey

Christopher Foy

Stefan Ivanov Krastanov

Carlos Errando Herranz

Eric Alexander Bersin

Adrian Johannes Menssen

Artur Hermans

Jonathan Hoffman

Hamza Raniwala

Jelena Notaros

Zaijun Chen

Cosmo Lupo

Travis Wade

Chaitanya Rastogi

Fan Meng

Tim Schroder

Noel Heng Loon Wang

Ren-Jye Shiue

Ophir Gaathon

Xuetao Gan

Uttara Chakraborty

Igal Bayn

Luozhou Li

Pierre Desjardins

Noel Wan

Tsung-Ju Lu

Liane Sarah Beland Bernstein

Alexander Sludds

Laura Kim

The exponential growth in deep learning models is challenging existing computing hardware. Optical neural networks (ONNs) accelerate machine learning tasks with potentially ultrahigh bandwidth and nearly no loss in data movement. Scaling up ONNs involves improving scalability, energy efficiency, compute density, and inline nonlinearity. However, realizing all these criteria remains an unsolved challenge. Here, we demonstrate a three-dimensional spatial time-multiplexed ONN architecture based on dense arrays of microscale vertical cavity surface emitting lasers (VCSELs). The VCSELs, coherently injection-locked to a leader laser, operate at gigahertz data rates with a 7T-phase-shift voltage on the 10-millivolt level. Optical nonlinearity is incorporated into the ONN with no added energy cost using coherent detection of optical interference between VCSELs.

VCSEL-based coherent scalable deep learning

An atom control architecture based on VIS-IR photonic integrated circuit (PIC) technology is characterized by (1) visible (VIS) and near-infrared (IR) wavelength operation, (2) channel counts extensible beyond 1000s of individually addressable atoms, (3) high intensity modulation extinction and (4) repeatability compatible with low gate errors, and (5) fast switching times. A 16-channel SiN-based APIC with (5.8±0.4) ns response times and <−30 dB extinction ratio at a wavelength of 780 nm. Based on a complementary metal-oxide-semiconductor (CMOS) fabrication process, this atom-control PIC (APIC) technology can be used for atomic, molecular, and optical physics and emerging applications, from quantum computers with cold atoms or ions to quantum networks with solid-state color centers. This APIC technology is especially suitable for scalable quantum information processing based on optically programmable atomic systems.

High-speed spatial light modulation and quantum control

A system for optically modulating a plurality of optical channels includes a power delivery module adapted to convert a coherent light beam into a plurality of optical channels, at least one optical modulator, optically coupled to the power delivery module, the at least one optical modulator adapted to optically modulate each of the plurality of the optical channels, and a vacuum chamber having a trapping plane therein, the vacuum chamber adapted to generate an addressable array of trapped particles at the trapping plane, wherein each of the plurality of optical channels is optically coupled to at least one of the trapped particles of the addressable array.

System and method for multiplexed optical addressing of atomic memories

Programmable photonic circuits of reconfigurable interferometers can be used to implement arbitrary operations on optical modes, providing a flexible platform for accelerating tasks in quantum simulation, signal processing, and artificial intelligence. A major obstacle to scaling up these systems is static fabrication error, where small component errors within each device accrue to produce significant errors within the circuit computation. Mitigating errors usually involves numerical optimization dependent on real-time feedback from the circuit, which can greatly limit the scalability of the hardware. Here, we present a resource-efficient, deterministic approach to correcting circuit errors by locally correcting hardware errors within individual optical gates. We apply our approach to simulations of large-scale optical neural networks and infinite impulse response filters implemented in programmable photonics, finding that they remain resilient to component error well beyond modern day process tolerances. Our error correction process can be used to scale up programmable photonics within current fabrication processes.

Error correction for programmable photonics

The typical approach to transfer quantum information between two superconducting quantum computers is to transduce the quantum information into the optical regime at the first superconducting quantum computer, transmit the quantum information in the optical regime to the second superconducting quantum computer, and then transduce the quantum information back into the microwave regime at the second superconducting quantum computer. However, direct microwave-to-optical and optical-to-microwave transduction have low fidelity due to the low microwave-optical coupling rates and added noise. These problems compound in consecutive microwave-to-optical and optical-to-microwave transduction steps. We break this rate-fidelity trade-off by heralding end-to-end entanglement with one detected photon and teleportation. In contrast to cascaded direct transduction, our technology absorbs the low optical-microwave coupling efficiency into the entanglement heralding step. Our approach unifies and simplifies entanglement generation between superconducting devices and other physical modalities in quantum networks.

Optically heralded entanglement of superconducting systems in quantum networks

Component errors prevent linear photonic circuits from being scaled to large sizes. These errors can be compensated by programming the components in an order corresponding to nulling operations on a target matrix X through Givens rotations X→TX, X→XT. Nulling is implemented on hardware through measurements with feedback, in a way that builds up the target matrix even in the presence of hardware errors. This programming works with unknown errors and without internal sources or detectors in the circuit. Modifying the photonic circuit architecture can reduce the effect of errors still further, in some cases even rendering the hardware asymptotically perfect in the large-size limit. These modifications include adding a third directional coupler or crossing after each Mach-Zehnder interferometer in the circuit and a photonic implementation of the generalized FFT fractal. The configured photonic circuit can be used for machine learning, quantum photonics, prototyping, optical switching/multicast networks, microwave photonics, or signal processing.

Self-configuration and error correction in linear photonic circuits

Quantum information processing involves entangling large numbers of qubits, which can be realized as defect centers in a solid-state host. The qubits can be implemented as individual unit cells, each with its own control electronics, that are arrayed in a cryostat. Free-space control and pump beams address the qubit unit cells through a cryostat window. The qubit unit cells emit light in response to these control and pump beams and microwave pulses applied by the control electronics. The emitted light propagates through free space to a mode mixer, which interferes the optical modes from adjacent qubit unit cells for heralded Bell measurements. The qubit unit cells are small (e.g., 10 μm square), so they can be tiled in arrays of up to millions, addressed by free-space optics with micron-scale spot sizes. The processing overhead for this architecture remains relatively constant, even with large numbers of qubits, enabling scalable large-scale quantum information processing.

Freely scalable quantum computing using a 2D atomic emitter array with massively parallel optical interconnects

The next-generation of optoelectronic systems will require efficient optical signal transfer between many discrete photonic components integrated onto a single substrate. While modern assembly processes can easily integrate thousands of electrical components onto a single board, photonic assembly is far more challenging due to the wavelength-scale alignment tolerances required. Here we address this problem by introducing a self-aligning photonic coupler insensitive to x, y, z displacement and angular misalignment. The self-aligning coupler provides a translationally invariant evanescent interaction between waveguides by intersecting them at an angle, which enables a lateral and angular alignment tolerance fundamentally larger than non-evanescent approaches such as edge coupling. This technology can function as a universal photonic connector interfacing photonic integrated circuits and microchiplets across different platforms. For example, it can be used in a self-aligning photonic circuit board that can be assembled more easily, with larger misalignment tolerances, than other complex optoelectronic systems.

Self-aligning photonic interconnections for photonic integrated circuits

Chemical-shift nuclear magnetic resonance (NMR) spectroscopy involves measuring the effects of chemical bonds in a sample on the resonance frequencies of nuclear spins in the sample. Applying a magnetic field to the sample causes the sample nuclei to emit alternating current magnetic fields that can be detected with color centers, which can act as very sensitive magnetometers. Cryogenically cooling the sample increases the sample's polarization, which in turn enhances the NMR signal strength, making it possible to detect net nuclear spins for very small samples. Flash-heating the sample or subjecting it to a magic-angle-spinning magnetic field (instead of a static magnetic field) eliminates built-in magnetic field inhomogeneities, improving measurement sensitivity without degrading the sample polarization. Tens to hundreds of small, cryogenically cooled sample chambers can be integrated in a semiconductor substrate interlaced with waveguides that contain color centers for optically detected magnetic resonance measurements of the samples' chemical-shift NMR frequencies.

Cryogenic integrated circuits architecture for multiplexed chemical-shift NMR

A method of generating uniform large-scale optical focus arrays (LOT As) with a phase spatial light modulator (SLM) includes identifying and removing undesired phase rotation in the iterative Fourier transform algorithm (IFTA), thereby producing computer-generated holograms of highly uniform LOT As using a reduced number of iterations as compared to a weighted Gerchberg-Saxton algorithm. The method also enables a faster compensation of optical system-induced LOT A intensity inhomogeneity than the conventional IFTA.

Large-scale uniform optical focus array generation with a phase spatial light modulator

A spectrally multiplexed quantum repeater (SMuQR) based on spatially arrayed nodes of frequency-multiplexed multi-qubit registers uses the natural inhomogeneous distribution of optical transition frequencies in solid state defect centers. This distribution enables spectrally selective, individual addressing of large numbers of defect centers within an optical diffraction limited spot along a long cavity or waveguide. The spectral selection relies on frequency shifting an incident optical field at a rate as fast as once per defect center lifetime. The defect centers are resonant at visible frequencies and emit visible single photons which are down-converted to a wavelength compatible with long-distance transmission via conventional optical fiber. The down-converted photons are all at the same telecommunications wavelength, with the different spectral bins mapped to different temporal bins to preserve the multiplexing in the time domain, for distribution to other nodes in the quantum network.

Spectrally multiplexed solid state quantum emitters and memories for quantum repeaters

An optical neural network is constructed based on photonic integrated circuits to perform neuromorphic computing. In the optical neural network, matrix multiplication is implemented using one or more optical interference units, which can apply an arbitrary weighting matrix multiplication to an array of input optical signals. Nonlinear activation is realized by an optical nonlinearity unit, which can be based on nonlinear optical effects, such as saturable absorption. These calculations are implemented optically, thereby resulting in high calculation speeds and low power consumption in the optical neural network.

Apparatus and methods for optical neural network

A spatial light modulator (SLM) comprised of a 2D array of optically-controlled semiconductor nanocavities can have a fast modulation rate, small pixel pitch, low pixel tuning energy, and millions of pixels. Incoherent pump light from a control projector tunes each PhC cavity via the free-carrier dispersion effect, thereby modulating the coherent probe field emitted from the cavity array. The use of high-Q/V semiconductor cavities enables energy-efficient all-optical control and eliminates the need for individual tuning elements, which degrade the performance and limit the size of the optical surface. Using this technique, an SLM with 10pixels, micron-order pixel pitch, and GHz-order refresh rates could be realized with less than 1 W of pump power.

All-optical spatial light modulator

Many of the features of neural networks for machine learning can naturally be mapped into the quantum optical domain by introducing the quantum optical neural network (QONN). A QONN can be performed to perform a range of quantum information processing tasks, including newly developed protocols for quantum optical state compression, reinforcement learning, black-box quantum simulation and one way quantum repeaters. A QONN can generalize from only a small set of training data onto previously unseen inputs. Simulations indicate that QONNs are a powerful design tool for quantum optical systems and, leveraging advances in integrated quantum photonics, a promising architecture for next generation quantum processors.

Quantum optical neural networks

Growing community of inventors

Brookline, MA, United States of America

Dirk Robert Englund