An official website of the United States government
Implementing scalable and effective synaptic networks will enable neuromorphic computing to deliver on its promise of revolutionizing computing. RRAM represents the most promising technology for realizing the fully connected synapse network: By using programmable resistive elements as weights, RRAM can modulate the strength of synapses in a neural network architecture. Oscillatory Neural Networks (ONNs) that are based on phase-locked loop (PLL) neurons are compatible with the resistive synapses but otherwise rather impractical. In this paper, A PLL-free ONN is implemented in 28 nm CMOS and compared to its PLL-based counterpart. Our silicon results show that the PLL-free architecture is compatible with resistive synapses, addresses practical implementation issues for improved robustness, and demonstrates favorable energy consumption compared to state-of-the-art NNs.
more »
« less
Parity–time symmetric optical neural networks
Optical neural networks (ONNs), implemented on an array of cascaded Mach–Zehnder interferometers (MZIs), have recently been proposed as a possible replacement for conventional deep learning hardware. They potentially offer higher energy efficiency and computational speed when compared to their electronic counterparts. By utilizing tunable phase shifters, one can adjust the output of each of MZI to enable emulation of arbitrary matrix–vector multiplication. These phase shifters are central to the programmability of ONNs, but they require a large footprint and are relatively slow. Here we propose an ONN architecture that utilizes parity–time (PT) symmetric couplers as its building blocks. Instead of modulating phase, gain–loss contrasts across the array are adjusted as a means to train the network. We demonstrate that PT symmetric ONNs (PT-ONNs) are adequately expressive by performing the digit-recognition task on the Modified National Institute of Standards and Technology dataset. Compared to conventional ONNs, the PT-ONN achieves a comparable accuracy (67% versus 71%) while circumventing the problems associated with changing phase. Our approach may lead to new and alternative avenues for fast training in chip-scale ONNs.
more »
« less
- PAR ID:
- 10304997
- Publisher / Repository:
- Optical Society of America
- Date Published:
- Journal Name:
- Optica
- Volume:
- 8
- Issue:
- 10
- ISSN:
- 2334-2536
- Format(s):
- Medium: X Size: Article No. 1328
- Size(s):
- Article No. 1328
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Millimeter wave (mmW) communications is viewed as the key enabler of 5G cellular networks due to vast spectrum availability that could boost peak rate and capacity. Due to increased propagation loss in mmW band, transceivers with massive antenna array are required to meet a link budget, but their power consumption and cost become limiting factors for commercial systems. Radio designs based on hybrid digital and analog array architectures and the usage of radio frequency (RF) signal processing via phase shifters have emerged as potential solutions to improve radio energy efficiency and deliver performances close to the conventional digital antenna arrays. In this paper, we provide an overview of the state-of-the-art mmW massive antenna array designs and comparison among three array architectures, namely digital array, partially-connected hybrid array (sub-array), and fully-connected hybrid array. The comparison of performance, power, and area for these three architectures is performed for three representative 5G downlink use cases, which cover a range of pre-beamforming signal-to-noise-ratios (SNR) and multiplexing regimes. This is the first study to comprehensively model and quantitatively analyze all design aspects and criteria including: 1) optimal linear precoder, 2) impact of quantization error in digital-to-analog converter (DAC) and phase shifters, 3) RF signal distribution network, 4) power and area estimation based on state-of-the-art mmW circuits including baseband digital precoding, digital signal distribution network, high-speed DACs, oscillators, mixers, phase shifters, RF signal distribution network, and power amplifiers. Our simulation results show that the fully-digital array architecture is the most power and area efficient compared against optimized designs for sub-array and hybrid array architectures. Our analysis shows that digital array architecture benefits greatly from multi-user multiplexing. The analysis also reveals that sub-array architecture performance is limited by reduced beamforming gain due to array partitioning, while the system bottleneck of the fully-connected hybrid architecture is the excessively complicated and power hungry RF signal distribution network.more » « less
-
Abstract Phased arrays have been a cornerstone of non-destructive evaluation, sonar communications, and medical imaging for years. Conventional arrays work by imparting a static phase gradient across a set of transducers to steer a self-created wavefront in a desired direction. Most recently, space-time-periodic (STP) phased arrays have been explored in the context of multi-harmonic wave beaming. Owing to the STP phase profile, multiple scattered harmonics of a single-frequency input are generated which propagate simultaneously in different directional lanes. Each of these lanes is characterized by a principal angle and a distinct frequency signature that can be computationally predicted. However, owing to the Hermitian (real) nature of the spatiotemporal phase gradient, waves emergent from the array are still bound to propagate simultaneously along up- and down-converted directions with a perfectly symmetric energy distribution. Seeking to push this boundary, this paper presents a class of non-Hermitian STP phased arrays which exercise a degree of unprecedented control over the transmitted waves through an interplay between gain, loss, and coupling between its individual components. A complex phase profile under two special symmetries, parity-time (PT) and anti-PT, is introduced that enables the modulation of the amplitude of various harmonics and decouples up- and down-converted harmonics of the same order. We show that these arrays provide on-demand suppression of either up- or down-converted harmonics at an exceptional point—a degeneracy in the parameter space where the system’s eigenvalues and eigenvectors coalesce. An experimental prototype of the non-Hermitian array is constructed to illustrate the selective directional suppression via time-transient measurements of the out-of-plane displacements of an elastic substrate via laser vibrometry. The theory of non-Hermitian phased arrays and their experimental realization unlock rich opportunities in precise elastoacoustic wave manipulation that can be tailored for a diverse range of engineering applications.more » « less
-
When using ultra-wideband signaling on massive multiple-input multiple-output (mMIMO) systems, the electromagnetic wave incurs an extra delay (across the array elements) comparable to or larger than the symbol duration, which translates into a shift in beam direction known as the beam squint effect. The beam squinting problem degrades the array gain and reduces the system capacity. This paper proposes a novel transceiver design based on lens antenna subarray and analog subband filters to compensate for the beam squinting effect. Specifically, the proposed design chunks the wideband signal from the phase shifters into groups of narrowband signals and controls their squints through an exhaustive search-based switching/precoding mechanism under the lenses. Furthermore, a simplified, thresholded search-based precoding algorithm is proposed, which demonstrates good performance while significantly minimizing complexity. The proposed system is analyzed in terms of beam gain, complexity, power consumption, and capacity. The numerical results demonstrate significant performance enhancement for the proposed system design as compared to the conventional mMIMO system with an uncompensated beam squinting problem.more » « less
-
Parity-time (PT) symmetry was first studied in quantum mechanical systems with a non-Hermitian Hamiltonian whose observables are real-valued. Most existing designs of PT symmetric systems in electronics, optics, and acoustics rely on an exact balance of loss and gain in the media to achieve PT symmetry. However, the dispersive behavior of most loss and gain materials restricts the frequency range where the system is PT symmetric. This makes it challenging to access the exceptional points of the system to observe the PT symmetric transition dynamics. Here, we propose a new path to realize PT symmetric systems based on gyroscopic effects instead of using loss and gain units. We demonstrate that PT symmetry and the occurrence of exceptional points are preserved for inversive, counter-rotating gyroscopic systems even with dispersive sub-units. In a gyroscopic system with two circular rings rotating in opposite directions at the same speed, the spontaneous symmetry breaking across the exceptional points results in a phase transition from a moving maximum deformation location to a motionless maximum point. The motionless maximum point occurs despite the externally imposed rotation of the two rings. The results set the foundation to study nonlinear dispersive physics in PT symmetric systems, including solitary waves and inelastic wave scattering.more » « less
