Nonvolatile Electrically Reconfigurable Integrated Photonic Switch Enabled by a Silicon PIN Diode Heater

Advanced Materials, Journal Year: 2020, Volume and Issue: 32(31)

Published: June 26, 2020

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 switching units as essential building blocks. Current switches, however, mainly rely materials with weak, volatile thermo-optic or electro-optic modulation effects, resulting large footprints high energy consumption. As promising alternative, chalcogenide phase-change (PCMs) exhibit strong optical static, self-holding fashion, but scalability present PCM-integrated applications is still limited poor electrical actuation approaches. Here, phase transitions actuated situ silicon PIN diode heaters, nonvolatile electrically reconfigurable switches using PCM-clad waveguides microring resonators are demonstrated. result, intrinsically compact energy-efficient operated low driving voltages, near-zero additional loss, reversible endurance obtained complementary metal-oxide-semiconductor (CMOS)-compatible process. This work can potentially enable very large-scale CMOS-integrated programmable such neural networks general-purpose processors.

Language: Английский

---

Quantum optical neural networks

npj Quantum Information, Journal Year: 2019, Volume and Issue: 5(1)

Published: July 17, 2019

Abstract Physically motivated quantum algorithms for specific near-term hardware will likely be the next frontier in information science. Here, we show how many of features neural networks machine learning can naturally mapped into optical domain by introducing network (QONN). Through numerical simulation and analysis train QONN to perform a range processing tasks, including newly developed protocols state compression, reinforcement learning, black-box simulation, one-way repeaters. We consistently demonstrate that our system generalize from only small set training data onto inputs which it has not been trained. Our results indicate QONNs are powerful design tool systems and, leveraging advances integrated photonics, promising architecture next-generation processors.

Language: Английский

---

Wafer-scale silicon photonic switches beyond die size limit

Optica, Journal Year: 2019, Volume and Issue: 6(4), P. 490 - 490

Published: April 11, 2019

Fast optical switches have been proposed as a promising alternative to enable continual scaling of data centers with increasing size and rates. Silicon photonics is compelling platform for large-scale integrated photonic switches, leveraging advanced manufacturing foundries electronic circuits. In the past decade, port counts silicon increased steadily 128×128. Further switch constrained by maximum reticle (2–3 cm) lithography tools. Here, we propose use wafer-scale integration overcome die limit. As proof concept demonstration, fabricated 240×240 lithographically stitching 3×3 array identical 80×80 blocks across boundaries. Stitching loss substantially reduced (0.004 dB) tapering waveguide width 10 μm. The on 4 cm×4 cm chip exhibits on-chip 9.8 dB, an ON/OFF ratio 70 dB, switching times less than 400 ns. To our knowledge, this largest ever reported. loss-to-port count (0.04 dB/port) also lowest.

Language: Английский

---

High-speed programmable photonic circuits in a cryogenically compatible, visible–near-infrared 200 mm CMOS architecture

Nature Photonics, Journal Year: 2021, Volume and Issue: 16(1), P. 59 - 65

Published: Dec. 13, 2021

Recent advances in photonic integrated circuits (PICs) have enabled a new generation of "programmable many-mode interferometers" (PMMIs) realized by cascaded Mach Zehnder Interferometers (MZIs) capable universal linear-optical transformations on N input-output optical modes. PMMIs serve critical functions quantum information processing, quantum-enhanced sensor networks, machine learning and other applications. However, PMMI implementations reported to date rely thermo-optic phase shifters, which limit applications due slow response times high power consumption. Here, we introduce large-scale platform, based 200 mm CMOS process, that uses aluminum nitride (AlN) piezo-optomechanical actuators coupled silicon (SiN) waveguides, enabling low-loss propagation with modulation at greater than 100 MHz the visible near-infrared wavelengths. Moreover, vanishingly low holding-power consumption piezo-actuators enables these PICs operate cryogenic temperatures, paving way for fully device architecture range

Language: Английский

---

Experimentally realized in situ backpropagation for deep learning in photonic neural networks

Science, Journal Year: 2023, Volume and Issue: 380(6643), P. 398 - 404

Published: April 27, 2023

Neural networks are widely deployed models across many scientific disciplines and commercial endeavors ranging from edge computing sensing to large-scale signal processing in data centers. The most efficient well-entrenched method train such is backpropagation, or reverse-mode automatic differentiation. To counter an exponentially increasing energy budget the artificial intelligence sector, there has been recent interest analog implementations of neural networks, specifically nanophotonic for which no backpropagation demonstration exists. We design mass-manufacturable silicon photonic that alternately cascade our custom designed "photonic mesh" accelerator with digitally implemented nonlinearities. These reconfigurable meshes program computationally intensive arbitrary matrix multiplication by setting physical voltages tune interference optically encoded input propagating through integrated Mach-Zehnder interferometer networks. Here, using packaged chip, we demonstrate situ first time solve classification tasks evaluate a new protocol keep entire gradient measurement update device domain, improving on past theoretical proposals. Our made possible introducing three changes typical meshes: (1) measurements at optical "grating tap" monitors, (2) bidirectional propagation automated fiber switch, (3) universal generation readout amplitude phase. After training, achieves accuracies similar digital equivalents even presence systematic error. findings suggest training paradigm photonics-accelerated based entirely popular technique.

Language: Английский

---

Scalable reservoir computing on coherent linear photonic processor

Communications Physics, Journal Year: 2021, Volume and Issue: 4(1)

Published: Feb. 10, 2021

Abstract Photonic neuromorphic computing is of particular interest due to its significant potential for ultrahigh speed and energy efficiency. The advantage photonic hardware lies in ultrawide bandwidth parallel processing utilizing inherent parallelism. Here, we demonstrate a scalable on-chip implementation simplified recurrent neural network, called reservoir computer, using an integrated coherent linear processor. In contrast previous approaches, both the input weights are encoded spatiotemporal domain by processing, which enables ultrafast beyond electrical bandwidth. As device can process multiple wavelength inputs over telecom C-band simultaneously, use optical (~5 terahertz) as computational resource. Experiments standard benchmarks showed good performance chaotic time-series forecasting image classification. considered be able perform 21.12 tera multiplication–accumulation operations per second (MAC ∙ s −1 ) each reach petascale computation on single chip division multiplexing. Our results challenging conventional Turing–von Neumann machines, they confirm great towards peta-scale super-computing chip.

Language: Английский

---

Hardware error correction for programmable photonics

Optica, Journal Year: 2021, Volume and Issue: 8(10), P. 1247 - 1247

Published: Aug. 18, 2021

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

Language: Английский

---

Aluminium nitride integrated photonics: a review

Nanophotonics, Journal Year: 2021, Volume and Issue: 10(9), P. 2347 - 2387

Published: June 18, 2021

Abstract Integrated photonics based on silicon has drawn a lot of interests, since it is able to provide compact solution for functional devices, and its fabrication process compatible with the mature complementary metal-oxide-semiconductor (CMOS) technology. In meanwhile, material itself few limitations, including an indirect bandgap 1.1 eV, transparency wavelength >1.1 μm, insignificant second-order nonlinear optical property. Aluminum nitride (AlN), as CMOS-compatible material, can overcome these limitations. It wide 6.2 broad window covering from ultraviolet mid-infrared, significant effect. Furthermore, also exhibits piezoelectric pyroelectric effects, which enable be utilized optomechanical devices photodetectors, respectively. this review, recent research works integrated AlN in past decade have been summarized. The related properties covered. After that, demonstrated linear emitters, metasurfaces, are presented. Last but not least, summary future outlook AlN-based provided.

Language: Английский

---

Optical Computing: Status and Perspectives

Nanomaterials, Journal Year: 2022, Volume and Issue: 12(13), P. 2171 - 2171

Published: June 24, 2022

For many years, optics has been employed in computing, although the major focus and remains to be on connecting parts of computers, for communications, or more fundamentally systems that have some optical function element (optical pattern recognition, etc.). Optical digital computers are still evolving; however, a variety components can eventually lead true such as logic gates, switches, neural networks, spatial light modulators previously developed discussed this paper. High-performance off-the-shelf accurately simulate construct complicated photonic devices systems. These advancements under unusual circumstances: photonics is an emerging tool next generation computing hardware, while recent advances empowered design, modeling, creation new class with unparalleled challenges. Thus, review status perspectives shows technology offers incredible developments computational efficiency; only separately implemented operations known so far, launch world's first commercial processing system was recently announced. Most likely, computer not put into mass production because there no good solutions transistors, memory, much acceptance break huge inertia proven technologies electronics.

Language: Английский

---

Reconfigurable Neuromorphic Computing: Materials, Devices, and Integration

Advanced Materials, Journal Year: 2023, Volume and Issue: 35(51)

Published: June 7, 2023

Abstract Neuromorphic computing has been attracting ever‐increasing attention due to superior energy efficiency, with great promise promote the next wave of artificial general intelligence in post‐Moore era. Current approaches are, however, broadly designed for stationary and unitary assignments, thus encountering reluctant interconnections, power consumption, data‐intensive that domain. Reconfigurable neuromorphic computing, an on‐demand paradigm inspired by inherent programmability brain, can maximally reallocate finite resources perform proliferation reproducibly brain‐inspired functions, highlighting a disruptive framework bridging gap between different primitives. Although relevant research flourished diverse materials devices novel mechanisms architectures, precise overview remains blank urgently desirable. Herein, recent strides along this pursuit are systematically reviewed from material, device, integration perspectives. At material device level, one comprehensively conclude dominant reconfigurability, categorized into ion migration, carrier phase transition, spintronics, photonics. Integration‐level developments reconfigurable also exhibited. Finally, perspective on future challenges is discussed, definitely expanding its horizon scientific communities.

Language: Английский

---